机器学习学术速递[6.9]

AI 总结：将Transformer数据传播建模为概率测度空间上的非线性控制系统，证明高斯分布在自注意力与仿射前馈层下保持高斯性，从而降维为双线性控制系统，并揭示与Riccati方程的联系。

链接：https://arxiv.org/abs/2606.07600

机构：Friedrich–Alexander University Erlangen–Nürnberg（弗里德里希-亚历山大大学埃尔朗根-纽伦堡）； Research Council of Norway（挪威研究理事会）

作者：Albert Alcalde, Zhengping Ji, Enrique Zuazua

英文摘要：We formulate data propagation through the Transformer, the machine learning architecture powering large language models, as a nonlinear control system on the space of probability measures. For the mean-field Transformer model with self-attention and affine feed-forward layers, we prove that Gaussian distributions remain exactly Gaussian along the induced flow. This invariance reduces the infinite-dimensional measure dynamics to a finite-dimensional bilinear control system governing the evolution of the mean and covariance, reformulates the expressive capacity of Transformers as a reachability problem for prescribed Gaussian moments, and reveals a novel connection with Riccati-type equations from classical filtering and control. For time-varying controls, we prove exact finite-time reachability of any target Gaussian distribution whose covariance matrix has the same rank as the initial one, this rank constraint being an intrinsic invariant of the dynamics. For time-invariant parameters, we derive explicit spectral conditions leading either to asymptotic stability toward positive-definite equilibria or to finite-time blow-up of the covariance. Numerical experiments complement the theory by showing that practical Transformers with Gaussian inputs remain close to moment-matched Gaussian distributions through early and intermediate layers, while Transformers with prescribed attention matrices reproduce the predicted covariance regimes: bounded evolution in stabilizing configurations and blow-up in destabilizing ones.

AI 总结：提出拉普拉斯-傅里叶神经算子（LFNO），通过双分支架构显式分解系统动力学为瞬态和稳态分量，在九个基准上超越现有算子，提升稳定性和可解释性。

链接：https://arxiv.org/abs/2606.07601

作者：Jeongun Ha, Sanga Yoon, Donghun Lee

英文摘要：We introduce the Laplace-Fourier Neural Operator (LFNO), a unified framework for modeling dynamical systems across transient and steady-state regimes by integrating the spectral advantages of Laplace and Fourier Neural Operators. LFNO employs a dual-branch architecture that explicitly decomposes system dynamics into transient and steady-state components. We evaluate LFNO on nine benchmarks, including three ODE systems (Duffing, Lorenz, and Pendulum) and six PDE systems (Euler-Bernoulli beam, Heat, Reaction-diffusion, Brusselator, Burgers, and Navier-Stokes). LFNO significantly outperforms existing operators on ODE systems, where transient dynamics dominate, and consistently surpasses LNO while achieving performance competitive with FNO on PDE benchmarks. Furthermore, LFNO offers improved stability and physical interpretability through its component-wise decomposition. These results demonstrate that LFNO provides a robust and unified approach for learning complex dynamical systems across multiple temporal scales.

AI 总结：提出基于投影的贡献权重度量，结合注意力权重、值向量大小和方向对齐，更准确识别关键令牌，并揭示注意力汇的主动抑制功能。

链接：https://arxiv.org/abs/2606.07604

作者：Harry Jake Cunningham, Nicola Muca Cirone

英文摘要： Analyzing attention weights has become a standard approach for interpreting the information flow of Large Language Models (LLMs). However, this approach has significant limitations as it neglects the geometric properties of the value vectors being aggregated. To address this gap, we introduce \emph{Contribution Weights}, a projection-based metric that quantifies a token's influence by accounting for it's attention weight, value magnitude, and directional alignment with the layer output. We demonstrate that contribution weights provide a more faithful measure of token importance, consistently outperforming attention-based metrics in identifying semantically critical tokens across different decoder-only models, tasks, and datasets. Further, our metric enables novel mechanistic analysis of \emph{attention sinks}. While previous work characterized sinks as passive repositories for excess attention, we reveal they serve an active functional role, suppressing information through a convex relationship between sink rate and output norm, stabilizing representations by opposing the semantic drift of low-confidence tokens.

AI 总结：提出DOG-DPO框架，将偏好对表示为模型表示空间中的方向，通过几何分解和多样性覆盖选择子集，仅用11%数据即可恢复大部分安全增益。

链接：https://arxiv.org/abs/2606.07678

机构：University of Southern California（南加州大学）； Iowa State University（爱荷华州立大学）； University of Wisconsin–Madison（威斯康星大学麦迪逊分校）； UT Austin（德克萨斯大学奥斯汀分校）； Independent Researcher（独立研究员）； University of Notre Dame（圣母大学）

作者：Yi Nian, Tiankai Yang, Yudi Zhang, Qi Pan, Zelong Xu, Shenzhe Zhu, Qingqing Luan, Yue Huang, Xiangliang Zhang, Yue Zhao

英文摘要：Safety alignment for large language models relies on preference data, but current pipelines often train on large, redundant datasets. Existing data selection methods typically score each preference pair independently, collapsing directional preference information into scalar quality or diversity scores. This sample-centric view is especially limiting in multi-dataset settings, where shared safety directions coexist with dataset-specific residual risks. We propose DOG-DPO, a training-free data selection framework that treats preference pairs as structured geometric signals. DOG-DPO first represents each preference pair as a direction in model representation space. It then decomposes multi-dataset preference geometry into a global anchor subspace and dataset-specific residual subspaces. Finally, it selects subsets by maximizing diversity-based coverage, encouraging broad, non-redundant coverage of alignment directions before DPO training. Across six safety benchmarks and two model backbones, DOG-DPO achieves a strong utility-robustness trade-off using only 11% of the preference pairs. It recovers most of the safety gains of full-data training while remaining entirely teacher-free, training-free, and substantially faster than representative selection baselines.

AI 总结：提出双域谱滤波网络DSFNet，通过特征域和空间域谱算子分解空间-模态交互，显式建模跨变量耦合与异质空间依赖，结合外部门控机制自适应调节时间动态，在五个真实交通数据集上MAE降低3.21%-10.16%。

链接：https://arxiv.org/abs/2606.07695

机构：Southeast University（东南大学）； Jiangsu Province Collaborative Innovation Center of Modern Urban Traffic Technologies（江苏省现代城市交通技术协同创新中心）； City University of Hong Kong（香港城市大学）； School of Mathematics, Southeast University（东南大学数学学院）； Systems Research Institute of the Polish Academy of Sciences（波兰科学院系统研究所）； Luoyang Normal University（洛阳师范学院）； Purple Mountain Laboratories（紫金山实验室）； AGH University of Krakow（AGH科技大学）

作者：Yongchao Li, Yang Li, Zhuoxuan Li, Jun Chen, Chu Zhang, Jinde Cao, Leszek Rutkowski

英文摘要：Multi-Modality Spatio-Temporal Forecasting (MoSTF) extends traditional spatio-temporal forecasting by incorporating diverse traffic modalities. Despite significant recent strides in spatio-temporal modeling, existing approaches often fail to explicitly model the coupling relationships between different modality variables. Accurate MoSTF is challenging, as it requires modeling (1) temporal dynamic heterogeneity under exogenous influences and (2) heterogeneous spatial dependencies alongside complex cross-variable couplings. To address these challenges, we propose the Dual-Domain Spectral Filtering Network (DSFNet). Our framework employs dual-domain spectral filtering to capture heterogeneous spatial patterns and explicitly model the relationships between variables. Unlike graph-based message passing or dense attention over node-modality pairs, DSFNet factorizes space-modality interactions into feature-domain and spatial-domain spectral operators, enabling scalable modeling of nonlocal dependencies and cross-modality couplings. Furthermore, we introduce an external gating mechanism to adaptively regulate temporal dynamics under external influences. We validate our method through extensive experiments on five representative real-world traffic datasets. Compared with the second-best baselines, DSFNet reduces MAE by 3.21%-10.16% across these datasets. The results demonstrate that DSFNet significantly outperforms existing state-of-the-art baselines in accuracy while exhibiting efficiency and robustness.

AI 总结：提出WhiFlash，首个统一自回归与扩散并行草稿的跨范式推测解码方法，通过细粒度路由和缓存优化实现高达69.6%的吞吐量提升。

链接：https://arxiv.org/abs/2606.07710

机构：Samsung AI Center, Cambridge, UK（三星AI中心，剑桥，英国）

作者：Young D. Kwon, Miles Williams, Rui Li, Alexandros Kouris, Stylianos I. Venieris

英文摘要： The autoregressive nature of large language models (LLMs) remains a significant bottleneck for inference, particularly in complex agentic workloads. While speculative decoding (SD) accelerates inference, current approaches rely on static drafting paradigms, utilising either autoregressive drafting models for reasoning or diffusion-based parallel drafting models for structured outputs. We empirically find that drafting accuracy fluctuates dramatically within a single sequence, leaving significant performance unrealised by static paradigms and coarse-grained routing. To address this volatility, we introduce WhiFlash, the first cross-paradigm SD method that unifies autoregressive and diffusion-based parallel drafting under a single token-level controller. WhiFlash adopts a fine-grained routing mechanism that employs either a lightweight entropy-based or a learned neural policy, both parametrised to provide a tunable balance between expected token gain and latency. To make high-frequency switching computationally viable, we introduce novel cache-management optimisations, Lazy Catch-up and KV-only Prefill, reducing switching overhead to below 7% of per-round latency. By capitalising on the complementary strengths of fundamentally distinct drafting architectures, WhiFlash achieves significantly higher acceptance lengths, yielding category-specific throughput gains of up to 69.6% over the state-of-the-art autoregressive EAGLE-3 and 37.3% over the diffusion-based DFlash.

AI 总结：在自由验证器域上，使用无教师自训练（STaR）和批评者指导的选择，发现自训练放大模型能力但不复合，通过 Pass@$K$ 交叉诊断证实。

链接：https://arxiv.org/abs/2606.07856

机构：Federal University of Rio de Janeiro（里约热内卢联邦大学）

作者：Igor Lima Strozzi

英文摘要：When a language model trains on its own verified outputs, does it acquire capability beyond its base, or merely get better at expressing capability the base already had? We make the question decidable with a teacher-free "constellation" -- a generator, a learned critic, and a free exact verifier -- on a FlashFill-style "trapdoor" DSL, where verified (problem, solution) pairs are cheap to synthesize, hard to invert, and free to check exactly. Everything runs on one 4-bit Qwen3-4B on a single 24 GB GPU, with no model in the loop larger than the base. We report three findings. (i) Critic-guided selection beats verifier-filtered best-of-$k$ by $+9.1$ pp ($6/6$ seeds), with the entire gain localized to tasks where candidates disagree on held-out inputs. (ii) Per-round STaR self-training raises the ceiling but never accelerates -- the gain tracks remaining headroom and decelerates across $K=4$ independent training trajectories. (iii) The domain has no clean zero-capability frontier, so the usual "$0\% \to$ climb $=$ emergence" test is invalid here. A measured pass@$K$ crossover settles the diagnosis: the trained model wins at the operating budget (pass@$8$) but the base overtakes it at a large budget (pass@$64$) on every trajectory, so self-training concentrates probability mass rather than expanding reach. This is amplification, not compounding. ($K=4$ is indicative, not yet a robust across-trajectory CI.)

AI 总结：提出PACI方法，通过局部梯度累积控制版本漂移，实现无气泡异步流水线并行，在GPT风格语言模型预训练中匹配同步1F1B-flush的稳定性和困惑度，吞吐量完全利用，训练时间至准确率提升达1.69倍。

链接：https://arxiv.org/abs/2606.07881

机构：Technion - Israel Institute of Technology（以色列理工学院）； Ben-Gurion University of the Negev（本·古里安大学）

作者：Itay Elam, Eliron Rahimi, Avi Mendelson, Chaim Baskin

英文摘要：Pipeline parallelism is essential for training large neural networks, but existing schedules trade off throughput, memory, and optimization consistency. Synchronous pipelines preserve forward/backward weight consistency but suffer from bubbles; asynchronous pipelines remove bubbles but introduce weight-version mismatch, typically requiring weight stashing, prediction, or correction mechanisms. We introduce PACI (Pipeline Asynchronous training with Controlled Inconsistency), a bubble-free asynchronous pipeline method that bounds forward/backward version drift without weight stashing, prediction, additional parameter copies, or global synchronization. The key idea is to use local gradient accumulation as a version-control mechanism: by slowing parameter-version evolution relative to pipeline delay, PACI limits the number of optimizer updates crossed by any micro-batch while preserving steady-state utilization. In GPT-style language-model pretraining, PACI matches the stability and final perplexity of synchronous 1F1B-flush, retains the same peak memory footprint, achieves fully utilized pipeline throughput, and improves training time-to-accuracy by up to $1.69\times$ over the fastest flush baseline. These results show that forward/backward inconsistency need not be eliminated: when explicitly bounded, it can be safely traded for substantial efficiency gains.

AI 总结：基于ChainzRule的导数控制网络通过逐层雅可比惩罚，在表格和NLP任务中实现低数据高性能，梯度尾比作为泛化诊断指标。

链接：https://arxiv.org/abs/2606.07908

作者：Rowan Martnishn

英文摘要：Derivative-controlled networks based on ChainzRule (CR) combine cubic polynomial layers with a lightweight forward-mode per-layer Jacobian penalty (DREG). In this second paper of a multi-part series, we evaluate the generalization properties of CR across data regimes. We ablate the shape of the DREG coefficient schedule, demonstrating that the optimal annealing range depends on representation noise. On the Pima Diabetes dataset, CR achieves strong low-data performance and maintains a consistent accuracy advantage over baselines from 5\% to 100\% training data, supported by exceptionally stable gradient tail ratios ($\sim$1.01--1.02 vs. 1.07--1.09 for ReLU networks). Extensions to SST-5 show competitive or superior results in both frozen-embedding and BERT fine-tuned regimes, including outperforming prior BERT baselines despite substantially less training data. These results are statistically significant: CR achieves superior accuracy over the strongest published baselines we could identify on both datasets ($p < 0.05$). These results establish that layer-wise derivative control induces a structural inductive bias toward low-frequency, stable representations that generalizes robustly across tabular and NLP domains, data volumes, and representation qualities. The gradient tail ratio serves as a reliable, label-free diagnostic of generalization capability.

AI 总结： 针对大型生成式AI模型可解释性差、计算成本高的问题，提出局部化机器学习架构，通过降低带宽、提高节点表达能力来提升可解释性和效率，并评估了多种硬件实现方案的适用性。

链接：https://arxiv.org/abs/2606.07998

作者：Ian Seet, Jonas Bozenhard, Simon Osterman

英文摘要：Recent advances in generative AI, especially powerful Large Language Models (LLMs) and Large Reasoning Models (LRMs), raise concerns over the interpretability, safety and sustainability of these large and opaque AI models. The power of such architectures is derived not only from the scalability of deep neural networks, but also massively parallel hardware such as GPU clusters. The diffuse nature of deep neural networks gives them great function-approximation capability when provided with sufficient training data but imposes a cost in interpretability and computational efficiency. Observing that localised machine learning (ML) models tend to be more interpretable and computationally efficient than deep neural networks on small datasets, we reason by analogy that similar advantages may apply to specific localised hardware ML architectures. We argue that localised architectures with lower bandwidth but higher expressivity per node have the potential to be fundamentally more interpretable than deep neural networks running on GPU clusters while remaining competitive for smaller datasets. We then evaluate the suitability of various hardware ML paradigms for implementing such localised architectures and evaluate their per-node expressivity, energy efficiency and practical maturity of the technology required.

AI 总结：本文揭示注意力汇聚（attention sink）可对应两种不同机制：自适应空操作（adaptive nop）和广播（broadcast），并据此提出诊断方法，证明门控（gating）和寄存器（register）等干预分别针对不同机制，组合使用效果更佳。

链接：https://arxiv.org/abs/2606.08105

机构：Kempner Institute（肯普纳研究所）； Harvard University（哈佛大学）

作者：Lukas Fesser, Mozes Jacobs, Thomas Fel, Andy Keller, Sham Kakade

英文摘要：When attention concentrates on a single token, a sink, what is the model actually computing? Attention sinks are ubiquitous in softmax transformers, yet this shared visual signature can hide fundamentally different algorithms. We show that visually similar sink patterns can reflect two distinct mechanisms: {i} adaptive nop, where a head suppresses its update by routing to a null token, and {ii} broadcast, where a sink aggregates and redistributes global information. In that case, sinks serve an analogous role: a safe destination when there is nothing useful to compute. Proposed interventions like gating or registers work because they implicitly target one or the other, revealing a duality between method and assumed mechanism: gating implicitly assumes nop; registers implicitly assume broadcast. Each mechanism leaves distinct traces (nop sinks exhibit negligible value norms; broadcast sinks induce low-rank outputs) which we formalize on synthetic tasks and use to derive practical diagnostics. Applied to pretrained vision transformers, these diagnostics reveal that both mechanisms exist at scale: sinks transition from CLS in early layers to patches in deeper layers, and concentrate in specialized heads. Strikingly, register tokens, designed for broadcast, are repurposed to also serve nop, confirming that neither intervention alone suffices. Combining gating with registers yields complementary gains in stability and performance. Overall, we find that the same attention pattern can reflect two very different computations and effective intervention requires first asking what the model is actually computing.

AI 总结：提出FLaG模块，通过实FFT变换、可学习潜在查询的频谱分量汇总、通道门控和时域重建，实现跨域令牌聚合，在AMP预测、图像分类和文本分类任务上取得提升。

链接：https://arxiv.org/abs/2606.08191

作者：Kewei Li, Rongying Zhang, Xueli Wang, Xiwen Gong, Zhongjian Wang, Lan Huang, Ruochi Zhang, Fengfeng Zhou

英文摘要：Token aggregation is a common bottleneck in models that map token representations to sample-level predictions, yet most pooling methods operate only in the original token domain. We propose FLaG, a plug-in aggregation module that transforms token representations with the real FFT, summarizes spectral components with learnable latent queries, applies a channel-wise gate, and reconstructs enhanced time-domain tokens for final pooling. We evaluate FLaG on antimicrobial peptide (AMP) activity prediction with ESM2, image classification with ResNet18 on CIFAR-10 and CIFAR-100, and text classification with RoBERTa on IMDB and GLUE. FLaG achieves its clearest gains on the ESM2-8M antimicrobial peptide tasks and on CIFAR-100, while remaining competitive with strong text baselines on IMDB and GLUE. Then we probe its behavior on the AMP setting with band knockouts, gate summaries, residue perturbations, latent-query readouts, and structure-proxy stratification. We find that low-frequency bands contribute the most overall, and the remaining higher-band pattern is more sample-specific. The gate acts as a broadly shared spectral reweighting stage and the cross-attention patterns are sample-specific with mild query-wise differentiation, and higher-helix peptides exhibit stronger average spectral sensitivity in both bacteria. The supplementary materials, source code and data are released at this https URL and this https URL.

AI 总结：提出CVAformer框架，通过因果干预解耦变量中的动态和不变成分，消除对齐中的混杂偏差，在多种预测场景下达到或超越最先进性能。

链接：https://arxiv.org/abs/2606.08262

作者：Kexuan Zhang, Xiaobei Zou, Cesare Alippi, Gary G. Yen, Yang Tang

英文摘要： Recent advances in Large Language Models (LLMs) have opened new possibilities for time series forecasting by enabling alignment between temporal patterns and pretrained word embeddings. However, most LLM-based methods overlook the heterogeneous nature of time series, where dynamic fluctuations and invariant semantics are entangled. This entanglement introduces spurious correlations during the alignment, as dynamic components act as confounders by simultaneously influencing invariant components and the resulting aligned embeddings. To address this issue, a variable-level alignment framework CVAformer is proposed. CVAformer explicitly disentangles each variable into invariant and dynamic components just before alignment, and applies causal intervention to mitigate the confounding effect of the dynamics. To better support variable-level alignment, CVAformer replaces the standard causal attention in LLMs with a non-causal attention mechanism that captures interactions among variables at each time step. Extensive experiments across long-term, short-term, few-shot, and zero-shot forecasting settings indicate that CVAformer matches or exceeds state-of-the-art performance on most datasets, and in some cases achieves notably better accuracy. Experimental results validate the effectiveness of variable-level alignment and dynamic disentanglement in CVAformer, offering a new perspective for LLM-based time series tasks.

AI 总结：提出GENERIC-FNO，首个在函数空间直接嵌入非平衡热力学完整GENERIC结构的神经算子，通过秩一投影精确满足退化条件，实现能量守恒与熵产生，在超分辨率下保持结构保证。

链接：https://arxiv.org/abs/2606.08343

机构：University of Illinois at Chicago（伊利诺伊大学芝加哥分校）； Georgia Tech Research Institute（佐治亚理工学院研究所）

作者：Jason Sulskis, Sathya Ravi

英文摘要：We introduce GENERIC-FNO, the first neural operator to embed the full GENERIC (metriplectic) structure of nonequilibrium thermodynamics -- reversible, energy-conserving dynamics and irreversible, entropy-producing dynamics coupled through the degeneracy conditions -- directly in function space. Existing structure-preserving neural operators enforce at most a single conservation law or reversible (Hamiltonian) structure, while thermodynamically consistent learning has been confined to finite-dimensional, graph, or particle systems. GENERIC-FNO closes this gap: it learns the energy and entropy functionals as neural operators and parameterizes the Poisson and friction operators as diagonal Fourier multipliers sandwiched between rank-one projections that enforce the degeneracy conditions exactly, by construction, with no penalty term, update projection, or residual. The degeneracy identities hold to machine precision (residuals ~10^-13) for any initialization, dimension, or resolution, so the continuous-time dynamics conserve the learned energy and produce entropy exactly; the explicit time stepping adds only a small O(dt^2) drift (per-step residual ~10^-6). We further note that the (E,S,L,M) decomposition of a given flow is not unique, and introduce a gauge-invariant dissipation diagnostic separating reversible from dissipative dynamics independently of the learned functionals. Across three operator backbones (1D/2D FNOs and DeepONet) and four PDEs spanning reversible, dissipative, and mixed regimes, GENERIC-FNO preserves its exact structural guarantees zero-shot across a 4x super-resolution range (64 to 256), recovers the ground-truth ordering of physical dissipation, and is competitive with strong unconstrained and energy-penalized baselines, outperforming them on several dissipative and mixed problems at comparable or fewer parameters.

AI 总结：提出INNSteer框架，通过可逆神经网络将LLM激活映射到潜在空间进行线性控制，再逆变换回原空间，实现非线性、输入依赖的激活引导，在多个模型和基准上优于现有方法。

链接：https://arxiv.org/abs/2606.08454

机构：Indiana University Bloomington（印第安纳大学伯明顿分校）

作者：Tuc Nguyen, Thai Le

英文摘要：Activation steering provides a lightweight inference-time mechanism for controlling large language models (LLMs) by modifying their internal activation vectors toward desired behaviors. Most existing methods compute a fixed steering direction in the original activation space, typically from pairs of contrastive examples using mean differences, linear probes, or arbitrary separability criteria. While effective to a certain extent, these methods treat behavioral control as a global, linear, additive offset: the same direction is applied across inputs, and behaviors are linearly separable. This can be restrictive when behavioral features vary nonlinearly across the activation space or lie on curved and anisotropic manifolds, where the optimal intervention may be input-dependent. To address this limitation, we propose INNSteer, a nonlinear activation steering framework based on invertible latent transformations. Rather than searching for a better steering vector in the original representation space, INNSteer learns a lightweight invertible neural network $\phi$ that maps an LLM's activations into a latent space where behavioral classes are more amenable to linear control. At inference time, activations are mapped through $\phi$, steered in the latent space, and mapped back through the exact inverse transformation $\phi^{-1}$. This makes a simple latent-space translation become a nonlinear, input-dependent intervention in the original activation space. Across experiment settings on multiple LLM families, scales, behavioral traits, and safety benchmarks, INNSteer consistently improves model control over linear, transport-based, and nonlinear steering baselines while largely preserving generation fluency.

AI 总结：针对预训练大型时间序列模型微调时因非凸损失景观导致过拟合的问题，提出平滑全微调（SFF）方法，通过随机初始化辅助模型插值平滑损失景观，提升可训练性，在八个代表性模型上取得一致改进。

链接：https://arxiv.org/abs/2606.08578

机构：Shanghai Key Laboratory of Data Science（上海市数据科学重点实验室）； College of Computer Science and Artificial Intelligence（计算机科学与人工智能学院）； Fudan University（复旦大学）

作者：Xu Zhang, Peang Wang, Wei Wang

英文摘要： Recently, large time series models (LTSMs) have gained increasing attention due to their similarities to large language models, including flexible context length, scalability, and task generality, outperforming advanced task-specific models. However, prior studies indicate that pre-trained LTSMs may exhibit a poorly conditioned non-convex loss landscape, leading to limited trainability. As a result, direct fine-tuning tends to cause overfitting and suboptimal performance, sometimes even worse than training from scratch, substantially diminishing the benefits of pre-training. To overcome this limitation, we propose Smoothed Full Fine-tuning (SFF), a novel fine-tuning technology. Specifically, we construct an auxiliary LTSM via random initialization to obtain a smoother loss landscape, and then linearly interpolate its weights with those of the pre-trained model to smooth the original landscape. This process improves trainability while preserving pre-trained knowledge, thereby enabling more effective downstream fine-tuning. From an optimization perspective, SFF perturbs sharp minima without significantly harming flat regions, facilitating escape from poor local basins toward smoother and more generalizable solutions. Extensive experiments on benchmark datasets demonstrate consistent improvements across eight representative LTSMs, including Timer, TimesFM, MOMENT, UniTS, MOIRAI, Chronos, TTMs, and Sundial, on diverse downstream tasks. The code is available at the link: this https URL.

AI 总结：针对强化学习中的训练-推理差异问题，提出差异约束马尔可夫决策过程（DCMDP），通过拉格朗日松弛自适应平衡性能提升与差异控制，实现稳定高效训练。

链接：https://arxiv.org/abs/2606.08779

机构：Hong Kong University of Science and Technology（香港科技大学）； Zhejiang University（浙江大学）； Tianjin University（天津大学）

作者：Jiashun Liu, Runze Liu, Xu Wan, Jing Liang, Hongyao Tang, Ling Pan

英文摘要：Reinforcement Learning (RL) has emerged as a pivotal post-training paradigm, yet it frequently suffers from unpredictable sub-optimum performance or even training collapses. Recent findings attribute these failures to a hidden train-inference discrepancy (or mismatch), stemming from the disparate underlying engines and architecture. We find that the training policy can actively self-correct such a discrepancy when provided with an appropriate learning signal. Then, we further empirically identify a discrepancy tolerance region: within this region, aggressively narrowing the discrepancy can suppress policy exploration and reduce learning efficiency, whereas outside this region, reducing excessive discrepancy improves optimization consistency and raises the achievable local performance ceiling. According to such findings, we formulate this problem as a Discrepancy-Constrained Markov Decision Process (DCMDP), where reward maximization is coupled with a constraint that aligns training-Inference behavior, achieving stable dual-objective optimization. To adaptively balance performance improvement and discrepancy control, we introduce a Lagrangian relaxation mechanism that dynamically adjusts the relative weight of the two objectives according to the current degree of discrepancy violation. This enables stable dual-objective optimization: the policy is allowed to explore freely within the tolerance region, while being guided back when the discrepancy exceeds the safe boundary. Empirically, DCMDP significantly improves the performance of 8B dense model (Qwen-3-8b) and 30B Mixture-of-Expert model (Qwen-3-30bA3b), and enables a heterogeneous training paradigm, where LLMs can be optimized in high-fidelity training setup while being explicitly aligned for low-cost, resource-constrained inference deployment.

AI 总结：提出基于并行样本集内在统计量（长度调整尾熵）的推理时扩展方法，通过后验候选排序和步骤级重采样，无需外部验证即可提升开放领域任务性能。

链接：https://arxiv.org/abs/2606.08850

机构：MIT（麻省理工学院）； Red Hat（红帽公司）； IBM（IBM公司）

作者：Giorgio Giannone, Mustafa Eyceoz, Shabana Baig, Shivchander Sudalairaj, Anna C. Doris, Faez Ahmed, Akash Srivastava, Kai Xu

英文摘要：Inference-Time Scaling (ITS) has largely succeeded in verifiable domains like math and coding, where cheap verification enables scalable output selection. However, extending ITS to tasks prone to systematic failure - driven by faulty initial assumptions or unmet multidimensional constraints - typically relies on costly external solvers or brittle, model-based verifiers. Our key insight is that the intrinsic statistics of parallel sample sets, specifically length-adjusted tail entropy, provide a robust discriminative signal for solution quality without access to ground truth. Crucially, these statistics serve as a difficulty gate for adaptive compute allocation, dynamically routing problems across scaling regimes. First, Intrinsic Selection (iS) ranks candidates post-hoc, matching consensus-based algorithms across three domains and improving engineering design selection by 20% over pass@1 baselines. Second, Intrinsic Particle Filtering (iPF) generalizes this to step-level resampling, guiding generation toward high-confidence reasoning trajectories to improve pass@1 by 6.1 points on average on hard math problems. Finally, Particle Distillation (dPF) injects privileged guidance via early logit blending and KL-guided resampling, steering generation past systematic reasoning errors to satisfy expert rubrics, yielding up to 26.5% gains on complex clinical responses. Our pipeline applies seamlessly across broad-purpose, domain-specialized, and multimodal architectures, successfully extending ITS to open-ended domains without requiring trained reward models or exact ground-truth verification.

AI 总结：提出sGPO方法，通过少量推理计算预估查询难度，自适应分配训练预算，将训练计算量降低三倍，同时保持或提升性能。

链接：https://arxiv.org/abs/2606.08854

机构：Red Hat（红帽）； IBM

作者：Shivchander Sudalairaj, Kai Xu, Akash Srivastava, Giorgio Giannone

英文摘要： Standard Reinforcement Learning with Verifiable Rewards (RLVR) training allocates a fixed rollout budget to every query, without regard for what each query's difficulty means for the current policy. This leads to two symmetric failure modes: easy queries produce near-zero advantage because the policy already solves them, while unsolvable queries produce no signal because the policy never solves them. Both regimes waste training FLOPs without contributing to a learning gradient. We introduce sorted Group Policy Optimization (sGPO), a compute-efficient strategy that trades a small budget of inference FLOPs for a large reduction in wasted training FLOPs. The key insight is that cheap inference compute can serve as a single offline proxy for query difficulty. By generating a small batch of parallel samples per query under the initial policy, we obtain a model-aware empirical success rate. This motivates setting the training rollout group size to the inverse of this success rate, a practical rule that maximizes sample efficiency by extracting the most advantage per generated rollout. This single profiling pass simultaneously drives data filtering (removing trivial queries and sub-sampling unsolvable ones), adaptive group size allocation, and curriculum construction (scheduling queries from easy to hard). sGPO matches or exceeds baseline performance while reducing total training compute by a factor of three, with the upfront inference profiling cost included.

AI 总结：提出反向相干性概念，通过拟逆鞅理论证明隐藏状态序列几乎必然收敛，并设计正则化方法，在多个任务中实现更早稳定和更低误差。

链接：https://arxiv.org/abs/2606.08934

机构：Institute of Statistical Science, Academia Sinica（中央研究院统计科学研究所）

作者：Yuan-chin Ivan Chang

英文摘要：Recurrent neural networks maintain a hidden state $h_t$, but its probabilistic meaning is often unclear. We study hidden-state stability through \emph{backward coherence}: the extent to which $h_t$ can be reconstructed from $h_{t+1}$ by a learned backward projector $g_\phi$. Under contraction and summable backward drift, the hidden-state sequence forms a quasi-reverse-martingale. This yields almost-sure convergence, rates under mixing, an interpretable limiting representation, finite pathwise stopping times, and a theoretical framework for time-uniform confidence sequences. Simulations support the theory. Backward-coherence regularisation reduces the empirical quasi-martingale total $\hat Q$ by $43$--$58%$, reaches stability $28$--$44%$ earlier than an unregularised RNN, and gives tracking-error recovery consistent with geometric bounds. Additional tests confirm echo-state forgetting rates bounded by $\rho$ and verify the increment-sum tube $R_t$ with $100%$ simultaneous coverage, although $R_t$ is conservative; in practice, the defect-tail proxy $\hat Q_t$ is the more useful monitor. The backward-coherence loss is also equivalent to minimising a Kullback--Leibler divergence in a Gaussian backward model, linking the method to variational inference. Extensions cover $\phi$-mixing inputs, change-point tracking, and finite-sample concentration. Three real-data studies further validate the approach. On PhysioNet 2012 ICU data, the Reverse Martingale RNN (RMRNN) matches RNN mortality-prediction AUC while reaching stable representations 13 hours earlier. On FRED-MD, it reduces one-month-ahead forecast error by about fourfold under concept drift. On UCI Human Activity Recognition, it maintains lower post-transition tracking error with geometric decay. The guarantees apply under the stated assumptions; universality is not claimed.

AI 总结：本文发现模加法任务中网络表示呈现二维循环几何而非神经坍缩的单纯形等角紧框架，通过层间非均匀训练、子空间锁定后的相位对齐动力学和复杂度优势分析解释了这一现象。

链接：https://arxiv.org/abs/2606.08985

机构：Academy of Mathematics and Systems Science, Chinese Academy of Sciences（中国科学院数学与系统科学研究院）； School of Mathematical Sciences, University of Chinese Academy of Sciences（中国科学院大学数学科学学院）； Shanghai Institute for Mathematics and Interdisciplinary Sciences (SIMIS)（上海数学与交叉学科研究院）； Key Laboratory of Systems Health Science of Zhejiang Province, School of Life Science, Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences（浙江省系统健康科学重点实验室，中国科学院大学杭州高等研究院生命科学学院）

作者：Hu Tan, Kuo Gai, Shihua Zhang

英文摘要：While neural collapse (NC) predicts that a $K$-class-balanced classifier should organize terminal representations as a $(K-1)$-dimensional simplex equiangular tight frame (ETF), modular addition consistently enters a different regime: networks compress to a two-dimensional cyclic geometry in which both classifier weights and token embeddings lie on circles. We refine the explanation of this phenomenon in three directions. First, we formalize a layerwise non-uniform training mechanism: downstream classifier weights are driven by dense cross-entropy gradients into a rank-2 equiangular configuration before upstream embeddings fully reorganize, and once this classifier plane forms, backpropagated feature gradients constrain embedding motion to the same plane while weight decay suppresses orthogonal components. Second, after this subspace locking, the induced in-plane dynamics admit an entropy-regularized transport interpretation on $S^1$; combined with modular-addition labels, this reduces embedding formation to phase alignment, whose minimizers are single-frequency characters of $\mathbb{Z}/P\mathbb{Z}$ and hence equal-angle points on a circle. Third, we quantify why this solution prevails over NC: a simplex ETF gains only an $O(1)$ advantage in cross-entropy, whereas the cyclic rank-2 solution enjoys a $\Theta(K)$ advantage under Schatten or weight-decay surrogates, yielding a critical threshold $\lambda_{\mathrm{crit}} = \Theta(1/K)$. Our results explain both why classifier weights move first and why embeddings subsequently align with them, showing that grokking on modular arithmetic is governed not by maximal separation alone but by a task-structured trade-off between separation, symmetry, and complexity.

AI 总结：本文通过小数据实验研究两阶段后训练中Stage-1（SFT或OPD）的作用，发现其主要影响策略熵状态，但对最终性能影响有限。

链接：https://arxiv.org/abs/2606.09059

作者：Jianxiong Shen

英文摘要： Two-stage post-training -- a Stage-1 warm-start (supervised fine-tuning, SFT, or on-policy distillation, OPD) followed by Stage-2 reinforcement learning (RL) -- is increasingly used for vision-language models (VLMs). We ask what Stage-1 actually controls in a small-data study using Qwen2.5-VL-7B with a same-modality 72B VLM teacher for OPD. First, the three warm-starts reach a narrow $53$--$54\%$ band on Geometry3K internal validation, consistent with the narrow range reported by recent specialized methods; this setup provides little evidence that Stage-1 changes the in-domain endpoint. Second, a matched-recipe, early-stopped SFT improves out-of-domain MathVista by $+2.1$ points, reversing the $-9.5$-point drop of an over-trained variant. The clearest difference is the \emph{entropy regime}: OPD enters RL with substantially higher policy entropy than either SFT initialization, and the separation remains visible through the available trajectories. At the in-domain initialization, OPD also has higher answer diversity and pass@16 ($+2.0$ to $+5.2$ points over SFT), although problem-level bootstrap intervals show that the smaller contrast is uncertain. The advantage is absent after RL (endpoint pass@16 values within $1.1$ points) and on MathVista (six models within $1.2$ points). Our contribution is therefore a bounded empirical characterization: Stage-1 is strongly associated with the entropy regime in this setup, but the downstream payoff is small, localized, and not evidence that OPD is a better RL warm-start.

AI 总结：提出基于 Hessian 预处理的神经 Legendre-Fenchel 变换方法，通过仿射变形改善病态函数的共轭计算，提高收敛速度和数值精度。

链接：https://arxiv.org/abs/2606.09077

机构：École Normale Supérieure Paris-Saclay（巴黎-萨克雷高等师范学校）； Sony Computer Science Laboratories Inc.（索尼计算机科学实验室公司）

作者：Basile Plus-Gourdon, Frank Nielsen

英文摘要：The Legendre-Fenchel (LF) transform is a fundamental tool in convex analysis and machine learning that maps lower semi-continuous functions to their convex conjugates. In practice, when closed-form formula are not available for expressing convex conjugates of given functions, one must approximate them using various techniques. One recent such versatile numerical method is the deep Legendre transform method which relies on neural networks although it remains challenging particularly for tackling ill-conditioned functions. This work builds on the reformulation of the LF transform as a projective polarity. A notable property of this framework is its affine invariance. We leverage this affine invariance to introduce a Hessian-based preconditioning strategy. Specifically, we apply an affine deformation around a minimizer so that the second-order Taylor approximation of the function coincides with the canonical paraboloid, whose conjugation map is the identity. A residual network initialized near the identity can then learn this simplified mapping, while the original conjugation map is recovered through the inverse deformation. The proposed preconditioning incurs only a modest computational overhead, consisting of a single eigendecomposition during initialization and two matrix-vector multiplications per query. Experiments on a diverse set of convex functions, including high-dimensional benchmarks, demonstrate improved convergence rates and enhanced numerical accuracy of the conjugation, with particularly significant gains for ill-conditioned problems. Finally, we discuss the scope of applicability of our proposed method and highlight several of its limitations.

AI 总结：针对过程奖励模型因训练数据不平衡导致的虚假高评分偏见，提出PRISM框架，通过对比步骤级比较和前瞻策略生成的难负样本，结合难度感知课程学习优化，显著降低假阳性率并提升推理准确性。

链接：https://arxiv.org/abs/2606.09078

机构：University of Maryland（马里兰大学）； Amazon（亚马逊）； University of Central Florida（中佛罗里达大学）

作者：Aakriti Agrawal, Souradip Chakraborty, Armin Saghafian, Nihal Sharma, Rizal Fathony, Nam H Nguyen, C. Bayan Bruss, Amrit Singh Bedi, Furong Huang

英文摘要：Process Reward Models (PRMs) improve credit assignment for reasoning by providing step-level feedback. However, we identify a hidden bias in PRMs caused by severe imbalance in step-level training data. Standard cross-entropy training amplifies this bias, causing PRMs to overcredit plausible but incorrect steps and produce high false-positive rates. We show that these false positives have an asymmetric downstream effect: false negatives mainly slow exploration, whereas false positives actively steer Best-of-N selection, guided decoding, and policy optimization toward flawed reasoning. This suggests that PRM training should shift from pointwise label fitting to reliable relative comparisons. To address this, we propose PRISM (Precision Ranking for Improved Step Modeling), a policy-aware PRM training framework that learns from contrastive step-level comparisons and hard negatives generated by a temporal lookahead strategy, requiring no new human labels. We further use a difficulty-aware curriculum to optimize the contrastive step margin. Across PRMBench and ProcessBench, PRISM substantially reduces false positives (22% on PRMBench) and improves macro F1 over strong discriminative PRMs. When applied to policy optimization and search tasks, including guided decoding and Best-of-N selection, it consistently improves accuracy (up to 22% for guided decoding and 33% for Best-of-N) and robustness. More broadly, trustworthy process supervision is not just about assigning high rewards, but about rewarding the right reasoning for the right reasons.

AI 总结：针对策略蒸馏中异常状态导致梯度不稳定的问题，提出全局归一化蒸馏策略优化（GNDPO），通过将KL分数转化为批次级相对优势来稳定优化，提升多模态推理任务的训练鲁棒性和性能。

链接：https://arxiv.org/abs/2606.09091

机构：OPPO AI Center（OPPO AI中心）

作者：Dongze Hao, Zhiwei Jin, Chen Chen, Haonan Lu

英文摘要：On-policy distillation (OPD) has recently emerged as an important post-training paradigm. By using a stronger teacher model to provide dense, fine-grained supervision for sampled trajectories, OPD offers a clear advantage over reinforcement learning with verifiable rewards (RLVR), which typically depends on sparse binary or outcome-based environmental feedback. However, naive token-level distillation can suffer from gradient instability, due to magnitude misalignment in outlier states. To address this issue, we propose Globally Normalized Distillation Policy Optimization (GNDPO), a practical method that stabilizes optimization by transforming raw KL scores into batch-level relative advantages. This normalization effectively mitigates gradient explosions while retaining the benefits of token-level guidance. Experimental results show that GNDPO substantially improves training robustness and downstream performance across multimodal reasoning tasks. The code is released at this https URL.

AI 总结：提出一种受伊辛动力学启发的平衡传播框架，通过扩展相空间动力学替代耗散Hopfield松弛，加速收敛、提高噪声鲁棒性，并在MNIST等数据集上实现与反向传播相当的性能。

链接：https://arxiv.org/abs/2606.09112

机构：School of Artificial Intelligence, Beijing Normal University（北京师范大学人工智能学院）； Laboratory for Advanced Computing and Intelligence Engineering, Information Engineering University（信息工程大学先进计算与智能工程实验室）； School of Physical Science and Technology, Beijing University of Posts and Telecommunications（北京邮电大学物理科学与技术学院）

作者：Chen-Rui Fan, Bo Lu, Xing-Yu Wu, Tie-Jun Wang, Chuan Wang

英文摘要：The rapid evolution of artificial intelligence has led to substantial advances in deep neural networks. Nonetheless, conventional GPU-based training remains highly energy-demanding, motivating the exploration of physical dynamics and compatible energy-based learning schemes, such as equilibrium propagation (EP). EP-based training, however, frequently suffers from convergence to local minima due to phase-space contraction. Here we introduce an Ising-dynamics-inspired equilibrium-propagation framework in which dissipative Hopfield relaxation is replaced by an extended phase-space dynamics with conjugate variables. The resulting training paradigm keeps the local two-phase learning rule of EP while changing the physical route by which neural states reach equilibrium. We show that this dynamics lowers effective energy barriers, accelerates convergence, improves noise robustness, and trains deep convolutional Hopfield networks on MNIST, FashionMNIST, and CIFAR-10 with performance comparable to backpropagation.

AI 总结：本文利用相干伊辛机结合平衡传播训练基于能量的神经网络，并通过Adam优化器加速收敛，展示了在深层架构和卷积操作上的可扩展性，为下一代AI硬件提供了物理框架。

链接：https://arxiv.org/abs/2606.09117

机构：School of Artificial Intelligence, Beijing Normal University（北京师范大学人工智能学院）； Laboratory for Advanced Computing and Intelligence Engineering, Information Engineering University（信息工程大学先进计算与智能工程实验室）； China Mobile (Suzhou) Software Technology Company Limited（中移（苏州）软件技术有限公司）； School of Science, Beijing University of Posts and Telecommunications（北京邮电大学理学院）

作者：Chen-Rui Fan, Bo Lu, Zhi-Hong Zhang, Run-Qing Zhang, Jing-Wei Wen, Chuan Wang

英文摘要：While Ising machines serve as advanced physical solvers for the Ising model,enabling applications in combinatorial optimization and neural network training,their scalability for large-scale neural networks remains constrained by hardware connectivity limitations and suboptimal training methodologies. In this work,we leverage a Coherent Ising Machine (CIM) to train an energy-based neural network using Equilibrium Propagation, achieving performance comparable to existing software-based implementations. We further enhance the algorithm by integrating the Adam optimizer to solve for the ground state of a Hopfield energy network, significantly improving convergence speed and solution accuracy. Additionally, we demonstrate the scalability of our approach across deeper network architectures and convolutional operations. Our results highlight the potential of CIM dynamics as a scalable platform for training complex neural networks, offering a pathway toward energy-efficient implementations via analog circuits, optoelectronics, or integrated photonics. This work establishes a novel physical framework for next-generation AI hardware development.

AI 总结：针对大语言模型在精度关键领域（如技术图表和机械设计）中的幻觉问题，提出可编程几何DSL PyGeoX及分层基准PyGeoX-Bench，并设计饱和加性奖励（SAR）方法，将奖励分解为有界逐约束项，解决异常梯度掩盖问题，使8B模型在基准上达到与更大前沿系统竞争的水平。

链接：https://arxiv.org/abs/2606.09278

机构：Huawei Celia Team（华为Celia团队）

作者：Rafael Cabral, Pang Zixi, Ziyi Shou, Shen Xin

英文摘要：Large Language Models frequently hallucinate in precision-critical domains such as technical diagramming and mechanical design, where outputs must satisfy strict geometric constraints. We study open-ended geometric synthesis from natural language: translating free-form descriptions into precise constructions whose entities must simultaneously satisfy dozens of interacting constraints. To make this tractable, we release PyGeoX, a programmable geometric DSL that compiles declarative constraints into a differentiable loss, and PyGeoX-Bench, a stratified suite of 300 problems with per-constraint verifiable rewards. Using PyGeoX as a verifier, we identify a failure mode we call Outlier Gradient Masking: under global-norm rewards (any scheme that aggregates residuals through a single norm, for example, $\exp(-\mathrm{MSE})$), a single outlier constraint can nullify the learning signal across all others. To address this, we propose Saturating Additive Rewards (SAR), which decompose the reward into bounded per-constraint terms, preserving partial progress and ensuring consistent gradients even under severe violations. Against MSE-based rewards, the natural baseline for geometry solvers, SAR improves the hard-tier solving rate by $2.3\times$, and the resulting 8B model is competitive with much larger frontier systems on this benchmark. We release the engine, benchmark, and data at this https URL.

AI 总结：系统研究STGCN架构深度对交通预测性能与计算效率的影响，发现单块结构在多数数据集上达到最优或接近最优性能，且计算成本显著低于标准双块结构。

链接：https://arxiv.org/abs/2606.09539

机构：Technical University of Munich（慕尼黑工业大学）； The University of Queensland（昆士兰大学）

作者：Soban Nasir Lone, Mohamed Abouelela, Taeyoung Yu, Jiwon Kim, Constantinos Antoniou

英文摘要：Spatio-temporal graph neural networks (STGNNs) have become the dominant approach for traffic prediction, yet their computational requirements pose challenges for practical deployment in intelligent transportation systems (ITS). While recent work has proposed efficient alternatives to STGNNs, a fundamental question remains unexplored: are these architectures themselves over-parameterised? We examine this question using the Spatio-Temporal Graph Convolutional Network (STGCN), one of the most widely adopted models in this domain. Through systematic experiments across four diverse traffic datasets, we compare 1-block, 2-block (standard), and 3-block STGCN variants. Our findings reveal that the single-block architecture achieves optimal performance for short-term prediction (10 mins) on three of four datasets, while incurring only marginal degradation ($\leq$1.8% relative error) at longer horizons. Crucially, the 2-block variant incurs 61% higher CPU inference latency and 37% lower throughput relative to 1-block -- substantial overhead for resource-constrained ITS deployment. The 3-block architecture offers no favourable tradeoff, more than doubling computational cost for $

AI 总结：通过共激活聚类提出注意力头电路假设，并用因果消融验证闭包性，发现该方法在密集模型有效但在MoE模型失效，表明共激活仅是电路提议而非确认。

链接：https://arxiv.org/abs/2606.09607

作者：Yongzhong Xu

英文摘要：Interpretability increasingly treats groups of components, not individual units, as the basic object, and proposes to find them by clustering co-activation statistics. We ask whether such a cheap signal actually identifies an attention-head circuit. Adapting a sparse-autoencoder clustering recipe to attention heads -- but validating by causal ablation rather than reconstruction -- we cluster heads and then run a closure test: ablate the discovered community and compare per-example damage to matched-random controls. Across two dense 1B-scale models (Pythia 1B, OLMo 1B) and two input distributions, the communities pass closure. In a Mixture-of-Experts model (OLMoE-1B-7B), route-conditional clustering recovers a statistically real signal that nonetheless does not survive closure -- ablation improves loss, the wrong direction. Extending closure across training, attention-target selectivity and participation ratio decouple from function in both directions. We conclude that a cheap signal is a circuit proposal, not a confirmed circuit; closure is what separates them.

AI 总结：本文通过鲁棒性和可迁移性视角，证明 Muon 优化器相比 Adam 和 SGD 能学习到更鲁棒、更可迁移的特征，并通过理论分析支持了经验发现。

链接：https://arxiv.org/abs/2606.09658

机构：Yale University（耶鲁大学）； National University of Singapore（新加坡国立大学）； University of Chinese Academy of Sciences（中国科学院大学）； Academy of Mathematics and Systems Science, CAS（中国科学院数学与系统科学研究院）

作者：Tianyu Ruan, Fengzhuo Zhang, Shuche Wang, Shihua Zhang

英文摘要：Muon has recently emerged as a state-of-the-art optimizer for pretraining Large Language Models (LLMs) and vision classifiers. Despite its efficiency advantage over Adam and SGD, the feature-learning advantage of Muon remains unclear. This paper investigates Muon's feature-learning advantage through the lens of robustness and transferability. First, by evaluating pretrained models on corrupted images and texts, we show that features learned by Muon are consistently more robust than those learned by Adam and SGD across different architectures, including transformers and Convolutional Neural Networks (CNNs). Using trained layer-wise probes, we further show that this robustness advantage is reflected in larger logit margins across layers. Second, by training linear classifiers or fine-tuning full models from pretrained parameters on downstream tasks, we demonstrate that Muon-learned features transfer more effectively than those learned by Adam and SGD. This transferability advantage is further supported by the diversity of hidden states across layers, as measured by effective rank. Finally, in a representative classification problem with multi-component features, we prove that Muon attains larger margins and higher effective rank than Adam and SGD, providing theoretical support for our empirical findings.

AI 总结：提出扰动对比物理学习（PCPL）框架，通过对比物理系统在不同条件下的响应实现学习，无需外部处理器或反向传播，在弹簧网络和光子电路中验证了分类与模拟乘法任务。

链接：https://arxiv.org/abs/2606.09756

作者：Kyungeun Kim, Amanuel Anteneh, Israel Klich, Olivier Pfister, J. M. Schwarz

英文摘要： Responses to perturbations are key to understanding physical systems. The ability to contrast such responses by comparing how a system reacts under slightly different conditions provides a mechanism for learning. Here, we introduce Perturbative Contrastive Physical Learning (PCPL), a general framework in which learning emerges from measurable contrasts between physical states produced by controlled changes to inputs, boundary conditions, parameters, or interpreter functions. PCPL unifies and extends prior approaches: Equilibrium Propagation is rooted in contrasts between free and nudged equilibria in energy-based systems, while Frequency Propagation corresponds to contrasts extracted from sinusoidally driven, frequency-demodulated responses. We show that contrast-driven updates can reflect either local sensitivities or global inverse-problem structure, yet do not require centralized gradient computation. Instead, effective learning geometry emerges implicitly from the system's own physical response, allowing learning behavior to arise without an external processor or explicit backpropagation. We demonstrate PCPL in two platforms: (i) spring networks that update bond stiffness using measured displacements and forces, and (ii) continuous-variable photonic circuits trained via x quadrature measurements and finite-difference estimates of the Jacobian. Both platforms successfully learn classification tasks. We further show that a continuous-variable photonic circuit can be trained to implement analog multiplication, illustrating a step toward more autonomous physical learning systems.

AI 总结：提出拓扑神经算子(TNOs)，利用离散外微积分在细胞复形上实现跨维度耦合，并通过分层结构提升长程信息传播，在PDE基准上优于现有算子。

链接：https://arxiv.org/abs/2606.09806

机构：Imperial College London（伦敦帝国学院）； University of San Francisco（旧金山大学）

作者：Lennart Bastian, Samuel Leventhal, Mustafa Hajij, Tolga Birdal

英文摘要：We introduce Topological Neural Operators (TNOs), a principled framework for operator learning on cell complexes that lifts neural operators (NOs) from functions on points and/or edges to topological domains. TNOs represent data as features defined on cells of varying dimension and model their interactions through Discrete Exterior Calculus, enabling explicit cross-dimensional coupling via gradient-, curl-, and divergence-type operators. The key design principle is to decouple where information flows, as governed by fixed topological operators, from how it is transformed (which is learned), yielding models that respect the geometric support of physical quantities and expose conservation and compatibility structure. We further propose Hierarchical TNOs (HTNOs), which incorporate learned coarse complexes to propagate long-range and topology-dependent information. Our framework subsumes existing NOs as a special case, providing a unified perspective on operator learning across discretizations. Across a range of PDE benchmarks, including irregular-geometry flow problems, TNOs and HTNOs improve accuracy; controlled studies further isolate the benefits of native higher-rank and topological structure. Project page: this https URL

AI 总结：提出Query Lens方法，通过考虑编码器侧键特征和解码器侧值特征以及下游模块的间接效应，实现对稀疏自编码器特征更全面、忠实的解释。

链接：https://arxiv.org/abs/2606.07617

作者：Hwiyeong Lee, Ingyu Bang, Uiji Hwang, Hyelim Lim, Taeuk Kim

英文摘要：While sparse autoencoders provide features more interpretable than individual neurons, reliably characterizing them remains challenging. We propose Query Lens, which extends Logit Lens to enable more comprehensive and faithful interpretations of sparse features. By jointly considering encoder-side key features and decoder-side value features, we identify both the inputs that activate a feature and the outputs it promotes. We also account for indirect, module-mediated effects that arise when the feature is processed by downstream modules, going beyond the direct effect captured by Logit Lens. In experiments, we find that Query Lens yields coherent token signatures for features that remain uninterpretable under Logit Lens. Finally, we propose the Subspace Channel Hypothesis, suggesting that downstream modules read features through layer-specific subspaces.

AI 总结：针对自监督方法在潜在空间预测动力学时混淆慢噪声与信号的问题，本文分析其根源为跨轨迹采样负样本的对比目标，提出通过轨迹内采样负样本消除预测捷径，从而强制编码动力学相关变量。

链接：https://arxiv.org/abs/2606.07770

机构：Emory University（埃默里大学）

作者：Paarth Gulati, Ilya Nemenman

英文摘要：Self-supervised methods that learn representations and predict dynamics fully in the latent space, such as JEPA, have been shown to confuse slowly varying noise with the dynamical signals they aim to capture. Specifically, when noise features remain approximately constant within each trajectory, contrastive predictive objectives preferentially encode these features instead of the true latent variables governing the system. The learned representation then becomes dominated by trajectory-specific noise, so downstream performance degrades with noise strength and does not improve even as the number and duration of training trajectories increase. We argue that this failure is a property of the objective itself, shared by a long line of contrastive predictive objectives that sample negatives across trajectories. To illustrate this generality, we study the failure mode and its remedy in two settings: a standard SimCLR-style JEPA on a synthetic moving-dot dataset, and DySIB, a recently introduced method designed for extracting physically interpretable representations of dynamics, on movies of a rigid-body pendulum. When negatives are instead sampled within a single trajectory, the slow noise can no longer distinguish frames within that trajectory, removing the predictive shortcut. Training one encoder simultaneously on many such trajectories then forces it to encode the variables relevant for the dynamics, with longer trajectories yielding better representations even for strong slow noise. Our results point toward principles for designing contrastive predictive objectives in dynamical representation learning, especially for physical systems with noisy experimental observations.

AI 总结：提出LH-NeF框架，利用层次和局部性先验学习通用连续信号的分词表示，通过前馈编码替代元学习，内存减少42倍，批大小提升133倍，在图像、3D形状和气候场上匹配或超越多种基线。

链接：https://arxiv.org/abs/2606.08204

机构： Zuse Institute Berlin (ZIB)（柏林祖斯研究所）； Cartesia AI； Technische Universität Berlin（柏林工业大学）

作者：Alonso Urbano, David W. Romero, Max Zimmer, Sebastian Pokutta

英文摘要：Neural fields parameterize data as functions from coordinates to values, providing a unified framework for representation learning across modalities. Existing approaches are dominated by per-sample meta-learning, which scales poorly due to memory-intensive inner-loop optimization. The natural alternative -- feed-forward encoding -- typically introduces modality-specific assumptions, sacrificing the generality that makes learning with neural fields attractive. We argue that locality and hierarchy are useful priors for learning field representations that can be injected without compromising modality-agnosticism. We propose LH-NeF, a framework to learn general-purpose tokenized representations of continuous signals. A locality-preserving hierarchical encoder maps raw coordinate-value field observations to structured tokens, from which the field is reconstructed during training. By replacing meta-learning's inner loop with a single forward pass, LH-NeF uses 42$\times$ less memory and supports 133$\times$ larger batches than the strongest modality-agnostic baseline. Across images, 3D shapes, and climate fields, our learned representations match or exceed performance of modality-agnostic, modality-specific, and specialized generative neural field baselines on both reconstruction and downstream tasks.

AI 总结：提出统一框架分解概念对齐的两个轴（表征vs.概念、实例级vs.分布级），定义四种性质，并引入干预基准InterVenchA和耦合稀疏自编码器CoSAE，证明对齐是多目标问题。

链接：https://arxiv.org/abs/2606.09653

机构：Brown University（布朗大学）； ENS Paris Saclay（巴黎萨克雷高等师范学校）； CNRS（法国国家科学研究中心）； DEEL - IRT Saint Exupéry（DEEL - IRT 圣埃克苏佩里）； Goodfire

作者：Grégoire Dhimoïla, Victor Boutin, Agustin Martin Picard, Thomas Fel, Thomas Serre

英文摘要：Learned representations across models and modalities often exhibit striking structural similarities, suggesting shared underlying concept decompositions. However, concept alignment remains poorly defined: existing approaches optimize different objectives under the same terminology, obscuring what is actually aligned. We propose a unifying framework that decomposes alignment along two axes: what is aligned (representations vs. concepts) and at what level (instance-wise vs. distributional). This induces four corresponding properties -- instance-wise and distributional variants of translation and concept consistency -- and reveals precisely which of these guarantees existing methods provide. We further introduce \InterVenchA, an intervention-based benchmark that separately measures extraction quality, translation quality, and concept consistency. Through theory and experiments, we show that commonly assumed equivalences between alignment objectives fail in practice: optimizing one property does not reliably recover the others, and purely unsupervised objectives fail to recover meaningful instance-level alignment. We then propose the Coupled Sparse Autoencoder (CoSAE), which jointly enforces complementary alignment objectives. Strong alignment emerges only in this regime. Surprisingly, as little as 0.1\% paired data is sufficient to recover instance-level alignment when anchoring distributional objectives. Overall, our results show that concept alignment is fundamentally multi-objective: it must be defined, measured, and optimized as such.

AI 总结：受自监督学习启发，提出基于Fisher信息的度量ICR，分解特征为不变和残差成分，用于联合评估扩散模型的表示与生成能力，发现中间噪声水平下不变性最强且分类性能最佳，ICR可敏感检测训练中的记忆化。

链接：https://arxiv.org/abs/2606.09718

作者：Xiao Li, Yixuan Jia, Zekai Zhang, Xiang Li, Lianghe Shi, Jinxin Zhou, Zhihui Zhu, Liyue Shen, Qing Qu

英文摘要：Diffusion models have demonstrated remarkable generative capabilities and have also emerged as powerful self-supervised representation learners, yet the connection between these two abilities remains less explored. Drawing inspiration from self-supervised learning (SSL), we introduce a framework for jointly evaluating the representation and generation capabilities of diffusion models. Specifically, we decompose features into invariant and residual components and derive the Invariant Contamination Ratio (ICR), a Fisher-based metric that quantifies how residual variation contaminates invariant signal in feature space. We use this framework to analyze both discriminative and generative behavior of diffusion models. On the representation side, we find that invariance peaks at intermediate noise levels, which also yield the best downstream classification performance. On the generative side, we study how training transitions from genuine generalization to memorization in data-limited regimes, and show that ICR serves as a sensitive training-time indicator of early learning: increasing residual energy along Fisher directions marks the onset of memorization, detectable from training features alone without external evaluators or held-out test sets. Overall, our results show that diffusion models can be monitored from a self-supervised perspective through the geometry of their learned representations.

AI 总结：提出使用全息约简表示（HRR）的无监督解缠算法，利用HRR解绑操作提供归纳偏置，分离数据中的因子变化，并通过信息论分析证明其诱导近似独立的符号-值对。

链接：https://arxiv.org/abs/2606.09725

机构：Virginia Tech（弗吉尼亚理工大学）； Worcester Polytechnic Institute（伍斯特理工学院）

作者：Jhonny J. Velasquez Olivera, Christo K. Thomas, Walid Saad

英文摘要： Disentanglement, the separation of factors of variation in data using neural networks, remains a long-standing challenge in machine learning. Prior work has addressed this problem with variational autoencoders and generative adversarial networks that incorporate ideas from variational inference and information-theoretic constraints. In contrast to methods that rely on continuous representations, we propose a design that treats disentangled representations as symbolic structures, motivated by the compositional relationships among the concepts that make up samples from a distribution. However, learning discrete symbolic structures with neural networks while maintaining differentiability is difficult and often requires complex architectures. To address this, we introduce an unsupervised learning algorithm that uses holographic reduced representations (HRR) for neural disentanglement. We show that the HRR unbinding operation provides an inductive bias for separating factors and yields competitive results against baselines, as measured by latent traversals and disentanglement metrics. We complement these empirical findings with an information-theoretic analysis of the HRR unbinding channel. We prove that unbinding induces approximately independent symbol-value pairs and derive a per-slot capacity bound that quantifies how many distinct symbolic concepts can be reliably encoded, giving a quantitative account of the inductive bias toward disentanglement. The resulting representations differ from standard autoencoder-based models, in that their latent units are vectors that are summed together, rather than scalar dimensions of a low-dimensional latent vector. We show that this HRR representation is more robust to noise than other disentangled representations and maintains reconstruction quality across a range of SNRs.

AI 总结：提出SPIN框架，利用张量网络分解联合策略，将指数复杂度降为线性，并通过离线训练的神经符号管道实现边缘设备上的低延迟去中心化集群控制。

链接：https://arxiv.org/abs/2606.07557

作者：Zhaowen Fan

英文摘要：Decentralized multi-agent swarm coordination on resource-constrained edge platforms remains fundamentally bottlenecked by the exponential scaling of joint action spaces and high-latency communication overhead. This paper introduces the Swarm Policy Interference Network (SPIN) framework, an architectural paradigm that bypasses these limitations by modeling swarm topologies as a compressed tensor network. We factorize the joint policy tensors of local multi-agent cliques into Matrix Product State (MPS) chains, reducing the computational complexity of evaluation from an exponential $O(n^m)$ wall to a strictly linear $O(m \cdot n \cdot \chi^2)$ constraint. To bridge local continuous spatial geometry with this discrete algebraic backend without requiring power-intensive online training loops, we introduce a decoupled, hybrid neuro-symbolic control pipeline. Local multi-layered neural networks operate as structural coordination encoders, pre-trained offline to nonlinearly map hand-engineered geometric descriptors into abstract environmental target measures. At runtime, edge agents execute instantaneous behavioral adaptations by applying the Radon-Nikodým derivative directly as a zero-shot importance-reweighting filter. We validate the framework within a discrete-time multi-agent simulation sandbox spanning tracking, decentralized dispersion/area coverage, and multi-goal coordination regimes. Qualitative telemetry demonstrates that the integrated pipeline achieves stable target-directed motion, anti-collapse spatial spreading under decentralized constraints, and structured subgroup formation across multiple targets, providing a mathematically grounded route to tractable, low-power edge swarm intelligence.

AI 总结：提出频谱图强化学习框架，利用频谱图神经网络学习最优恢复策略，实现配电网故障实时近最优管理，在三个IEEE测试系统上验证了泛化能力。

链接：https://arxiv.org/abs/2606.07583

机构：Wayne State University（韦恩州立大学）； University of Texas at Dallas（德克萨斯大学达拉斯分校）

作者：Lihui Liu, Mucun Sun, Caisheng Wang

英文摘要：Self-healing smart grids can quickly adjust their network configuration during outages to minimize power disruptions. During an outage, several actions can be taken, such as network reconfiguration through switching operations and emergency load shedding. However, traditional machine learning methods for outage mitigation are not well suited for smart grids due to their slow response time and high computational cost. To address these challenges, recent studies have explored reinforcement learning to automatically perform network reconfiguration. In these approaches, the control policy is typically modeled using a graph neural network (GNN). However, conventional GNNs operate in the spatial domain and may fail to capture important relationships in the frequency domain. Frequency-domain information is particularly useful for modeling global structural patterns and system-wide interactions in power networks. In this paper, we propose a spectral graph reinforcement learning framework for outage management in distribution networks to enhance system resilience. Our model learns the optimal power restoration policy using a spectral graph neural network. We evaluate the proposed method on three modified IEEE test systems: the 13-bus, 34-bus, and 123-bus networks. Experimental results show that our approach achieves near-optimal performance in real time and generalizes well across a wide range of outage scenarios.

AI 总结：提出UNIQ方法，通过共形预测校准不确定性，实现状态自适应的保守性惩罚，在D4RL基准上以接近IQL的内存开销提升性能。

链接：https://arxiv.org/abs/2606.07592

机构：IIIT Delhi（印度德里国际信息技术学院）

作者：Aditya Upadhyay

英文摘要： Offline reinforcement learning requires careful conservatism to mitigate distribution shift, yet most existing methods apply a fixed penalty uniformly across all states regardless of local data coverage. We present UNIQ (Uncertainty-Informed Quantile), an offline RL method that introduces state-adaptive conservatism through conformally calibrated uncertainty estimation. Built on the Implicit Q-Learning (IQL) backbone, UNIQ trains a multi-expectile value ensemble, computes distribution-free uncertainty estimates using split conformal prediction, and maps the resulting signal to a state-dependent expectile that relaxes conservatism in well-covered regions while strengthening it in uncertain regions near the data frontier. On D4RL MuJoCo benchmarks, UNIQ consistently improves over IQL, with the largest gains observed on Walker2d and replay-heavy tasks. At the same time, UNIQ operates at near-IQL memory cost (approximately 250 MB peak VRAM), providing roughly a 10x reduction compared to EDAC. Rather than pursuing overall state-of-the-art performance, we position UNIQ as a practical mechanism contribution that improves the performance-efficiency trade-off in offline reinforcement learning.

AI 总结：针对LLM生成LEGO组装时出现的物理有效但几何语义错位问题，提出基于模型的数据选择方法和样本高效强化学习PVPO，结合体素空间几何奖励，提升结构、语义对齐和物理有效性。

链接：https://arxiv.org/abs/2606.07602

机构：HKUST(GZ)（香港科技大学（广州））； CUHK（香港中文大学）； ZODA

作者：Yuhuan Yuan, Zhouliang Yu, Minghao Liu, Weiyang Liu, Ge Lin Kan

英文摘要：LLM-based LEGO assembly generation requires both semantic grounding and physical feasibility. We identify a data-induced failure mode, PhysHack, in which the assemblies satisfy physical-validity constraints while producing structures that are geometrically misaligned, semantically inconsistent, or poorly calibrated. To address this challenge, we propose a model-based data selection approach that uses only a small fraction of the training data while improving physically grounded LEGO assembly generation. Building on the selected trajectories, we introduce PVPO, a sample-efficient reinforcement learning method that couples physical feasibility with voxel-space geometric rewards. Our results show that physical validity alone is an insufficient proxy for reliable physical reasoning: models can learn to generate valid structures without preserving semantic or geometric fidelity. Experiments across model backbones and test-time scaling settings demonstrate that PVPO improves structural and semantic alignment, physical validity, structural stability, and calibration, while reducing reliance on extensive post-hoc rejection sampling. In particular, results on calibration show that PVPO mitigates PhysHack by making test-time selection more predictive of semantic and structural quality.

AI 总结：针对语音感知大语言模型后训练中GRPO方法粗粒度信用分配问题，提出LEAF方法，通过回溯式树结构学习、高信息量边界选择和跨度级优势分配，在语音问答和翻译任务上超越GRPO。

链接：https://arxiv.org/abs/2606.07610

机构：University of Illinois, Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）

作者：Argyrios Gerogiannis, Yekaterina Yegorova, Mark Hasegawa-Johnson, Venugopal V. Veeravalli

英文摘要：State-of-the-art GRPO-style methods for speech-aware large language model post-training suffer from coarse credit assignment, broadcasting the same terminal-reward advantage to every token in a response. This ignores useful structure within rollout batches, where speech-conditioned completions often share prefixes before diverging at important decisions. We propose Low-rank Exploration with Adaptive Forking (LEAF), a retrospective tree-based RL method that recovers this structure without online branching or additional decoding. LEAF samples complete responses, selects high-surprisal boundaries, groups responses by shared prefixes, and assigns span-level advantages using descendant rewards. We theoretically justify LEAF's span-level credit assignment and boundary-selection design. Empirically, LEAF improves over GRPO across speech question answering and speech translation benchmarks under the same rollout and low-rank adaptation budget. Notably, smaller LEAF-trained models outperform current state-of-the-art, full-parameter baselines.

AI 总结：针对多目标强化学习中奖励学习异步性问题，提出轻量级动态加权机制SAW，利用变异系数实时调整各目标贡献，在GRPO和GDPO框架下提升训练效率和最终性能。

链接：https://arxiv.org/abs/2606.07705

机构：Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所）； University of Electronic Science and Technology of China（电子科技大学）

作者：Yuchen He, Baolong Bi, Shenghua Liu, Huaming Liao, Yuyao Ge, Bolin Wan, Siqian Tong, Juan Chen, Jiafeng Guo, Xueqi Cheng

英文摘要：Although multi-objective reinforcement learning (MORL) is central to aligning large language models with complex human preferences, the prevailing practice of static weighted summation overlooks a more fundamental phenomenon: reward learning is markedly asynchronous across objectives. Well-learned dimensions quickly produce homogeneous, low-variance signals whose residual noise contaminates the aggregated reward (in GRPO) or occupies a fixed share of the advantage budget (in GDPO), interfering with the scarce yet high-value signals carried by under-learned dimensions. To address this asynchrony, we propose Stage-Aware Dynamic Weighting (SAW), a lightweight, algorithm-agnostic dynamic weighting mechanism. SAW utilizes the coefficient of variation (CV) as a scale-invariant proxy for real-time informativeness, reweighting each dimension's reward or advantage contribution by its relative informativeness within the batch. Unlike gradient-based methods that require multiple forward and backward passes, SAW relies solely on batch-level statistics, introducing nearly negligible computational overhead. Experiments on tool-calling and text summarization tasks demonstrate that SAW consistently improves both training efficiency and final performance under both GRPO and GDPO frameworks, confirming it as a general-purpose plug-in for multi-reward LLM alignment. Our code is available at this https URL

AI 总结：提出CAAL框架，利用上下文信息和奖励预测动态选择主动学习策略，在公共数据集上优于现有基线方法。

链接：https://arxiv.org/abs/2606.07910

机构：University of Minnesota, Twin Cities（明尼苏达大学双城分校）； Amazon（亚马逊）

作者：Shao-An Yin, Jiacong Li, Tianpei Xie, Cecile Levasseur, Wojciech Kowalinski, Nicola Elia

英文摘要：The challenge with active learning algorithms is the uncertainty of the statistical distribution of unlabeled data, making it difficult to choose the best hand-crafted strategy. To address this, we introduced Contextual Adaptive Active Learning (CAAL). In CAAL, each "arm" represents a hand-crafted strategy. Unlike existing frameworks that select strategies based only on feedback from labeled data, we dynamically choose strategies for labeling batches of data using reward prediction with external context information. This general framework allows for customization with domain knowledge to design more effective rewards and context candidates. In addition, we experimentally show that CAAL outperforms the existing baseline adaptive strategy on public datasets using our reward and context design. Our results are consistent regardless of batch size in each iteration.

AI 总结：针对GRPO训练中均匀采样导致计算效率低的问题，提出CoDaPO方法，通过置信度和难度自适应重加权与重采样，在固定预算下提升可学习问题的发现，在12个基准上优于现有RL方法。

链接：https://arxiv.org/abs/2606.07950

作者：Zhanke Zhou, Xiangyu Lu, Chentao Cao, Brando Miranda, Tongliang Liu, Bo Han, Sanmi Koyejo

英文摘要：RL with verifiable rewards can substantially improve LLM reasoning, yet standard GRPO-style training often treats easy, hard, and learnable questions alike through uniform sampling and weighting, leading to inefficient compute allocation. We study GRPO by tracking token log-probabilities, group-normalized advantages, and the induced token-level update weights. This reveals three recurring dynamics as training proceeds: (1) confidence inflation, (2) advantage contraction, and (3) hierarchical convergence. These findings suggest that the utility of each update depends strongly on both question difficulty and the model's current competence. Motivated by this, we propose Confidence and Difficulty-adaptive Policy Optimization (CoDaPO), which assigns each question a bounded value from rollout confidence and empirical difficulty. CoDaPO then uses this value to reweight policy updates and resample high-value learnable questions within mini-batches, thereby increasing discovery within the learnable band under a fixed compute budget. Across twelve benchmarks, CoDaPO consistently improves accuracy over existing RL methods. Our code is publicly available at this https URL.

AI 总结：提出DICE框架，通过熵正则化均衡选择（HQRE）解决多智能体LLM协调中的不稳定性，实现线性收敛和有限贝叶斯遗憾，在11个基准上平均提升4.3-8.5个百分点。

链接：https://arxiv.org/abs/2606.08068

机构：University of Arizona（亚利桑那大学）； Hong Kong Baptist University（香港浸会大学）

作者：Yi Xie, Zhanke Zhou, Chentao Cao, Bo Liu, Bo Han

英文摘要：Multi-agent large language model (LLM) systems often fail to reliably outperform a single strong model equipped with best-of-N sampling. We argue that a core source of this instability is ill-posed equilibrium selection: current systems specify what information agents share, but not which coordination convention should be selected. We formalize a broad class of such systems as discounted incomplete-information Markov games and show that two common pathologies, oscillation between competing conventions and drift across them, can both induce unstable learning and linear Bayesian regret. To obtain a well-posed target, we introduce the Heterogeneous Quantal Response Equilibrium (HQRE), an entropy-regularized equilibrium concept with agent- and state-dependent temperatures. Under a monotonicity condition, HQRE is unique, admits linearly convergent mirror updates, and yields bounded Bayesian regret; the same condition yields rollout-measurable stability diagnostics. We instantiate this objective in two algorithms: DICE-PC, which coordinates frozen models through prompt-control actions, and DICE-FT, which performs parameter-efficient mirror fine-tuning. Across eleven benchmarks in four domains, DICE improves accuracy-cost trade-offs over strong within-class baselines; on reasoning and planning tasks, DICE-PC improves by 4.3 percentage points on average and DICE-FT by 8.5 points.

AI 总结：提出ConSteer-RL框架，将模型log概率的token级置信度信号融入GRPO，通过置信度感知奖励塑造机制惩罚过度自信错误并强化正确自信推理，在多个模型规模上平均提升2.3%-4.0%。

链接：https://arxiv.org/abs/2606.08088

机构：Xi'an Jiaotong University（西安交通大学）； University of Science and Technology of China（中国科学技术大学）

作者：Qing Miao, Yiming Zhao, Jing Yang, Chenxi Liu, Yuehai Chen, Yuewen Liu, Shaoyi Du, Badong Chen

英文摘要：Reinforcement Learning from Verifiable Rewards (RLVR) has recently become a key paradigm for improving the reasoning abilities of Large Language Models (LLMs), yet it remains limited by sparse binary rewards and its ignorance of model-internal uncertainty. In this paper, we propose ConSteer-RL, a simple yet effective framework that integrates token-level confidence signals derived from model log-probabilities into RLVR training. Specifically, building upon the Group Relative Policy Optimization (GRPO) framework, we construct a confidence-aware reward by aggregating per-token probabilities into a scalar confidence score and incorporating it into an awareness-based reward shaping mechanism that penalizes overconfident errors while reinforcing correct and confident reasoning. Experimental results demonstrate that ConSteer-RL consistently outperforms strong GRPO baselines, achieving average improvements of 2.3%-4.0% across different model scales.

AI 总结：针对同伴推荐招募中协变量依赖到达的现实问题，提出生成前沿规划（GFP），通过确定性备份和边际贪心分配实现高效规划，在模拟实验中优于基线方法。

链接：https://arxiv.org/abs/2606.08360

机构：Harvard University（哈佛大学）

作者：Lingkai Kong, Hezi Jiang, Andrew Ma, Keyu Wang, Akseli Kangaslahti, Milind Tambe

英文摘要：Peer-referral recruitment systems such as respondent-driven sampling are critical for studying and intervening on hidden populations affected by infectious diseases. To accelerate recruitment, public health agencies must adaptively allocate limited referral resources across multiple rounds, where current decisions shape both the number and the covariates of future recruits. Prior work makes this problem tractable by assuming that referrals are drawn i.i.d.\ from a homogeneous population, an assumption that ignores the homophily and shared context that drive real peer recruitment. We instead consider a more realistic model in which both referral capacity and the covariates of newly referred individuals are conditioned on the referrer, learned from data with a censored count model and a conditional generative model. The resulting planning problem is challenging because each candidate allocation induces a different distribution over future recruits. We propose \emph{Generative Frontier Planning} (GFP), a model-based planner that replaces per-step Monte-Carlo sampling with a deterministic backup over a latent covariate-coverage value surrogate. The surrogate is designed so that the expected value of the next frontier depends on the offspring generative model only through finite-dimensional summaries that are amortized offline, and so that the resulting per-round objective is monotone with diminishing returns. Together, these two properties make planning tractable: the deterministic backup eliminates Monte-Carlo sampling, and the diminishing-returns structure lets a marginal greedy allocation achieve a \((1-1/e)\)-approximation for the per-round problem. On a simulation environment calibrated to a real respondent-driven sampling dataset, GFP outperforms random, reinforcement-learning, and i.i.d.\ dynamic-programming baselines across four discount factors.

AI 总结：提出MO-PQUCB算法，通过主动查询获取用户偏好信号，结合Plackett-Luce模型和正则化UCB，解决多目标老虎机中偏好与奖励的耦合问题，实现更优的遗憾界。

链接：https://arxiv.org/abs/2606.08410

机构：The Ohio State University（俄亥俄州立大学）； University at Buffalo（布法罗大学）

作者：Linfeng Cao, Ming Shi, Ness B. Shroff

英文摘要：Personalized decision-making in multi-objective bandits requires learning user-specific trade-offs among competing objectives. Since arm utility depends on both unknown rewards and unknown preferences, existing methods infer preferences only from utility feedback, entangling preference learning with reward exploration. In practice, however, users often reveal their priorities through proactive conversational queries (e.g., "cheap and clean hotel"), yet this structured signal is not leveraged. We formalize a proactive query-based framework in which user queries provide structured preference signals. Modeling these signals via a Plackett-Luce subset choice model, we show that query-only learning is insufficient due to a fundamental shift-invariance barrier. To resolve this, we introduce MO-PQUCB, a hybrid algorithm that integrates query-based preference anchoring with bandit feedback through shift-invariant regularization and dual-exploration UCB. We prove that proactive queries accelerate preference estimation and yield improved regret scaling over prior preference-aware MO-MAB methods. Under corrupted queries, we further characterize statistical limits and design a robust estimator achieving near-optimal performance when the corruption is sparse. Experiments validate both theoretical and practical gains.

AI 总结：提出Aco2方法，通过上下文对比元强化学习，使四旋翼无人机在无需人工干预下自主完成不同载荷的抓取、运输和投递，并直接迁移到真实世界。

链接：https://arxiv.org/abs/2606.08533

机构：National Key Laboratory of Novel Software Technology, Nanjing University（南京大学计算机软件新技术国家重点实验室）； School of Artificial Intelligence, Nanjing University（南京大学人工智能学院）； Faculty of Robot Science and Engineering, Northeastern University（东北大学机器人科学与工程学院）； National Key Lab of Autonomous Intelligent Unmanned Systems, Beijing Institute of Technology（北京理工大学自主智能无人系统国家重点实验室）

作者：Lixuan Jin, Bingxuan Lan, Xinyi Bao, Xiangyuan Xie, Chunjie Zhang, Zheng Chen, Tianshuo Liu, Ruijie Tian, Jinyu Ru, Gang Wang, Lei Yuan, Yang Yu

英文摘要：Unmanned aerial vehicles (UAVs) are increasingly being deployed in logistics, service robotics, and other real-world applications, creating a growing demand for autonomous payload acquisition and delivery. Existing approaches typically assume pre-attached payloads or rely on specialized grippers, leaving versatile end-to-end aerial delivery largely unresolved, where different payloads induce highly variable flight dynamics, requiring a single policy to adapt online without manual calibration or explicit system identification. To this end, we study \textbf{A}utonomous \textbf{A}erial Manipulation via \textbf{Co}ntextual \textbf{Co}ntrastive Meta Reinforcement Learning (\textbf{\textit{Aco2}}), a fully autonomous aerial delivery setting in which a quadrotor equipped with a lightweight hook continuously picks up, transports, and delivers diverse handle-equipped objects between randomized locations, all without human intervention. First, we design a contextual observation encoder that infers a compact latent context from recent interaction history, enabling the policy to adapt online to payload-dependent dynamics. To further improve the quality of this context, we introduce a contrastive objective that structures the context embedding around task-relevant variations, improving generalization across diverse payloads without requiring explicit system identification. Trained entirely in simulation with extensive domain randomization, \textit{Aco2} can be directly deployed on a physical quadrotor without real-world fine-tuning.

AI 总结：提出在线强化学习算法RLDT，利用Stein变分梯度下降构建传输场，微调预训练流匹配策略，通过期望目标估计稳定训练，在连续控制任务中优于基线方法。

链接：https://arxiv.org/abs/2606.08602

机构：University of Pennsylvania（宾夕法尼亚大学）

作者：Boshu Lei, Kostas Daniilidis, Antonio Loquercio

英文摘要：We present an online reinforcement learning (RL) algorithm for fine-tuning flow-matching policies in continuous-control problems. Our key insight is to view RL-based policy improvement as a transport of action densities towards regions of high reward, which naturally aligns with the transport formulation of flow matching models. Prior methods either approximate the current or optimal policy distribution or resort to distillation, which introduces biased gradients or sacrifices multimodal modeling capacity. In contrast, our approach for RL with Density Transport, which we name \emph{RLDT}, constructs a transport field from a maximum-entropy RL objective using Stein Variational Gradient Descent (SVGD). Then, it finetunes a pretrained flow matching policy to align with this field. Training with this alignment objective is nontrivial because flow-matching policies generate actions via a multi-step process, making direct gradient-based optimization challenging. To overcome this challenge and stabilize training, we approximate policy actions from intermediate denoising steps via expected-target estimation. This allows the transport-field update to propagate into the network parameters without unstable backpropagation through time. Experimental results demonstrate that RLDT outperforms competitive baselines in reward quality and convergence speed. This performance holds across diverse continuous-control tasks, encompassing both dense and sparse rewards, as well as state- and vision-based long-horizon robot manipulation. The project webpage is \href{ this https URL }{ this https URL }.

AI 总结：提出SkillHone框架，通过持久决策历史记录诊断、修订和证据，实现智能体技能的持续演化，在开放网络深度研究基准上超越现有方法。

链接：https://arxiv.org/abs/2606.08671

机构：WeChat, Tencent Inc., China（腾讯微信，中国）

作者：Zhiwei Li, Yong Hu

英文摘要：Agent skills extend language-model agents with task-specific procedures, scripts, and references, but the tasks and environments they target continually change. Existing methods improve skills in bounded runs and retain only the final artifact, discarding the decision history that later agents need to interpret prior revisions, evaluations, and rejected alternatives. We introduce SkillHone, a harness for continual agent skill evolution grounded in persistent decision history. SkillHone pairs skill revisions with evaluation-side evidence that supplies practice feedback, recording structured histories of diagnoses, revisions, evidence, and outcomes. Role-separated subagents run candidate skills on practice probes with redacted reporting and propose revisions informed by prior decisions, enabling cross-session refinement without rediscovering past rationale. We evaluate SkillHone on deep-research benchmarks in a raw open-web setting, where agents are not given an integrated search stack and must organize retrieval through portable skills. We compare against a deep-research agent backed by commercial retrieval services. With Qwen3.6-35B-A3B as the evaluation-time backbone, the resulting skills outperform the deep-research agent by 15.8 points on GAIA and 3.2 points on WebWalkerQA-EN, while also exceeding prior skill-evolution methods.

AI 总结：针对在线学习中的近因效应，研究单遍滑动窗口流式多臂老虎机问题，提出纯探索和遗憾最小化算法，并给出记忆-遗憾权衡。

链接：https://arxiv.org/abs/2606.08977

机构：Johns Hopkins University（约翰霍普金斯大学）； Rensselaer Polytechnic Institute（伦斯勒理工学院）； Texas A&M University（德克萨斯农工大学）

作者：Vladimir Braverman, Chen Wang, Liudeng Wang, Samson Zhou

英文摘要：Motivated by the recency effect in online learning, we study algorithms for single-pass *sliding-window streaming multi-armed bandits (MABs)* in this paper. In this setting, we are given $n$ arms with unknown sub-Gaussian reward distributions and a parameter $W$. The arms arrive in a single-pass stream, and only the most recent $W$ arms are considered valid. The algorithm is required to perform pure exploration and regret minimization with limited memory, defined as the number of stored arms. The model is a natural extension of the streaming multi-armed bandits model (without the sliding window) that has been extensively studied in recent years. We provide a comprehensive analysis of both the pure exploration and regret minimization problems with the model. For pure exploration, we prove that finding the best arm is hard with sublinear memory while finding an approximate best arm admits an efficient algorithm. For regret minimization, we explore a new notion of regret and give sharp memory-regret trade-offs for any single-pass algorithm. We complement our theoretical results with experiments, demonstrating the trade-offs between sample, regret, and memory.

AI 总结：提出反事实传输流框架，通过检索高反馈轨迹构建局部偏好对，实现离线决策的保守轨迹细化，在D4RL基准上提升历史回报表现。

链接：https://arxiv.org/abs/2606.09115

作者：Lena Krieger, Xuan Zhao, Zhuo Cao, Qin Wang, Hanno Scharr, Ira Assent

英文摘要：Offline reinforcement learning (RL) offers a path to policy improvement from logged data alone, using historical returns or other measurable outcomes as world feedback. A key difficulty is improving observed behavior without extrapolating beyond what the offline data supports. We propose \emph{counterfactual transport flows}, a source-conditioned trajectory refinement framework for offline decision-making guided by world feedback. Given a low-feedback candidate trajectory, we construct local preference pairs from offline data by retrieving nearby trajectories in latent trajectory space with higher task-specific feedback, and use them as weak supervision for conservative refinement. The framework learns instance-specific refinement directions: at inference time, a refinement strength parameter controls how far the candidate trajectory is transported, enabling a trade-off between preserving the original behavior and applying stronger improvement. Experiments on D4RL benchmarks, including AntMaze and MuJoCo tasks, show that our method improves behavior from historical returns as world feedback, while providing interpretable trajectory-level refinement paths.

AI 总结：提出Claw-R1系统，通过网关服务器和数据池组件，将智能体交互步骤转化为结构化数据资产，支持实时检查、质量筛选和训练批次配置，解决智能体强化学习中数据生命周期管理问题。

链接：https://arxiv.org/abs/2606.09138

机构：State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China（中国科学技术大学认知智能国家重点实验室）

作者：Daoyu Wang, Mingyue Cheng, Qingchuan Li, Shuo Yu, Jie Ouyang, Qi Liu

英文摘要：Agentic reinforcement learning (RL) has become an important post-training paradigm for turning LLMs from static chatbots into interactive agents, giving rise to representative applications such as OpenClaw. Existing work mainly focuses on policy optimization algorithms and training frameworks, but pays less attention to the full data lifecycle of agent-environment interactions, from data production to training consumption. To bridge this gap, we present Claw-R1, an interactive step-level data middleware system for agentic RL. Claw-R1 connects heterogeneous agent runtimes with RL training backends through two core components: a Gateway Server and a Data Pool. The Gateway Server captures multi-turn interaction steps through a unified LLM API entry point, while the Data Pool organizes them into step-level records consisting of prompt IDs, response IDs, rewards and other metadata. In our demo, users can interactively inspect live trajectories, examine the state, action, and reward of each step, curate data by quality and readiness, and configure training-ready batches for different downstream RL algorithms. Overall, Claw-R1 treats agent interaction traces as managed data assets rather than temporary runtime logs. Through this demonstration, we hope to encourage the community to recognize the importance of data management in agentic RL. Our code is available at this https URL and the demonstration video can be found at link this https URL.

AI 总结：本文证明了一种无锚非参数汤普森采样算法在风险厌恶型多臂赌博机中达到实例依赖的渐近最优后悔界，适用于任意连续风险泛函，且仅需连续性条件，优于先前参数方法。

链接：https://arxiv.org/abs/2606.09191

作者：Joel Q. L. Chang

英文摘要：We prove that $\rho\text{-}\mathrm{NPTS}_{\mathrm{SG}}$, an anchor-free nonparametric Thompson Sampling algorithm for risk-averse bandits, achieves regret matching the instance-dependent lower bound to leading order in $\log n$, establishing it as asymptotically optimal for any continuous risk functional $\rho$ (CVaR, mean-variance, Sharpe ratio, distortion risk measures, and more) on the class of distributions with bounded density and sub-Gaussian tails, including Gaussian arms. Both this result and its bounded-support counterpart require only continuity of $\rho$: strictly weaker than the dominance condition of prior parametric Thompson Sampling results, and strictly weaker than the Lipschitz condition of UCB-type algorithms, yielding the first instance-optimal guarantees for non-Lipschitz functionals such as the Sharpe ratio without parametric reward assumptions. The bounded-support case is developed first as a stepping stone sharing the same proof structure. The key technical contributions are a discretisation lemma (bounded support) and a truncated discretisation lemma (sub-Gaussian tails), each projecting the growing-alphabet Dirichlet posterior onto a fixed grid via the Dirichlet aggregation property, holding all polynomial prefactors at fixed degree independent of sample size and breaking the super-exponential barrier that blocked prior proofs.

AI 总结：提出PBSD方法，通过贝叶斯校准的自蒸馏将稀疏最终奖励转化为细粒度步骤级信用信号，解决长程智能体任务中的信用分配问题，实验表明其提升领域内外性能并促进泛化。

链接：https://arxiv.org/abs/2606.09348

机构：School of AI, Shanghai Jiao Tong University（上海交通大学人工智能学院）； XYZ AI Lab（XYZ AI实验室）

作者：Yang Tian, Rui Wang, Xumeng Wen, Junjie Li, Shizhao Sun, Lei Song, Jiang Bian, Bo Zhao

英文摘要： Long-horizon agentic tasks pose a fundamental credit assignment challenge for outcome-base reinforcement learning: trajectory-level rewards verify final correctness but provide limited guidance on which intermediate reasoning steps or tool interactions contribute to the outcome. The difficulty is especially pronounced in multi-turn search agents, where successful trajectories may contain misleading actions and failed trajectories may contain valuable evidence-gathering steps. We propose PBSD (Privileged Bayesian Self-Distillation), a Bayes-calibrated self-distillation method for fine-grained credit assignment under sparse final rewards. PBSD measures trajectory quality through the posterior-to-prior probability ratio of the verified answer and applies Bayes' rule to convert this hard-to-estimate answer-side ratio into a tractable likelihood ratio between a standard student model and a privileged answer-conditioned teacher model. Autoregressive decomposition of this Bayesian evidence score yields turn-level signals that identify whether each intermediate turn supports or undermines the verified outcome. Consequently, PBSD provides a principled and elegant reweighting scheme that transforms sparse outcome supervision into Bayes-calibrated turn-level credit signals, while remaining fully compatible with standard policy optimization. Experiments demonstrate that PBSD consistently enhances performance across both in-domain and out-of-domain settings, and effectively transfers knowledge from short-context training to long-context inference, suggesting that its fine-grained credit assignment mechanism facilitates more effective policy learning and yields improved generalization.

AI 总结：提出推理竞技场框架，通过轨迹锦标赛将无梯度信号的非多样奖励组转化为相对奖励信号，结合Bradley-Terry模型高效整合强化学习，在数学和编码基准上平均提升7.6%，加速训练27%-41%。

链接：https://arxiv.org/abs/2606.09380

机构：University of Cambridge（剑桥大学）； Mistral AI

作者：Han Zhou, Adam X. Yang, Laurence Aitchison, Anna Korhonen, Albert Q. Jiang

英文摘要：Reinforcement learning with verifiable rewards (RLVR) has become a leading paradigm for improving the reasoning ability of large language models through outcome-based supervision. However, verifiable rewards frequently become uninformative at the group level: when all sampled traces of a given prompt receive identical rewards, group-relative advantage estimation provides no gradient signal, even though the traces may differ substantially in reasoning quality. We propose Reasoning Arena, an adaptive training framework that routes such non-diverse reward groups to a judge system instead of discarding them. Beyond examining the final answer, Reasoning Arena constructs trace tournaments, where reasoning traces are compared head-to-head to expose finer-grained preferences within the group, converting reasoning quality into rich relative reward signals. To make reward estimation efficient, rather than exhaustively comparing every pair, each new trace is evaluated against a small, dynamically updated pool of previously generated traces as anchors to efficiently establish a relative ranking. We then fit a Bradley-Terry model on the incomplete comparison graph, enabling scalable RL integration without quadratic pairwise comparisons. Empirical results demonstrate that Reasoning Arena consistently outperforms the RLVR baseline by 7.6% on average in competition mathematics and coding benchmarks. By converting otherwise wasted zero-advantage samples into useful gradient updates, our method accelerates training by 27% to 41%, saving nearly 50% of generation compute, and substantially improves overall reasoning performance.

AI 总结：针对上下文队列赌博机问题，提出三阶段算法CQB-η-2，通过仅在截止轮前进行随机探索，将队列长度遗憾从Õ(T^{-1/4})改进到Õ(T^{-1/2})，并证明该速率在最小最大意义下最优。

链接：https://arxiv.org/abs/2606.09668

机构：KAIST（韩国科学技术院）； Seoul National University（首尔大学）

作者：Seoungbin Bae, Dabeen Lee

英文摘要：Contextual queueing bandits provide a framework for learning to schedule heterogeneous jobs under unknown context-dependent service rates. Under stochastic contexts, existing algorithms achieve $\widetilde{\mathcal{O}}(T^{-1/4})$ queue length regret, defined as the expected difference between the learner's and oracle's queue lengths at horizon $T$. In this paper, we improve this rate to $\widetilde{\mathcal{O}}(T^{-1/2})$. The key observation is that random exploration is needed only up to a carefully chosen cutoff round, rather than throughout the entire horizon. We propose CQB-$\eta$-2, a three-phase algorithm: (i) pure random exploration to construct an initial estimator, (ii) $\eta$-random exploration combined with a UCB rule to continue learning while maintaining negative drift, and (iii) pure UCB after the exploration cutoff. Our proof decomposes the queue length regret at the cutoff round. Before the cutoff, negative drift suppresses queue length differences caused by suboptimal choices. After the cutoff, the first two phases provide sufficient random exploration samples, ensuring that UCB decisions incur small departure-rate gaps. Combining these two bounds yields queue length regret of order $\widetilde{\mathcal{O}}(T^{-1/2})$. We further prove a minimax lower bound of order $\Omega(T^{-1/2})$. The proof constructs two hard instances that are statistically indistinguishable up to the final service decision, and uses a queue-specific coupling argument to convert the resulting testing error into queue length regret. Together, our upper and lower bounds characterize the minimax dependence on the horizon $T$ up to logarithmic factors.

AI 总结：针对用户偏好和上下文分布随时间漂移的线性上下文随机多臂赌博机问题，提出Dri-MED算法，通过异方差回归处理非平稳噪声，实现实例相关的遗憾界和约束违规界。

链接：https://arxiv.org/abs/2606.09802

作者：Udvas Das, Waris Radji, Debabrota Basu, Odalric-Ambrym Maillard

英文摘要：We consider a variant of the linear contextual stochastic multi-armed bandits, where the learner must provide recommendations to a group of users, each having its personalized preference vector, and in the presence of context distributions that are drifting over time. Under practitioner-friendly assumptions, we reduce this setting to linear bandit with stationary mean but heteroskedastic and non-stationary noise. We further study the case when the learner must ensure the mean reward of each decision must exceed that of a baseline strategy $\boldsymbol{\pi}_0$ at each decision step. We introduce Dri-MED, an algorithm inspired from the linear version of the MED strategy, and carefully adapted to handle the non-stationary heteroskedastic noise. We show that the instance-dependent regret scales as $\tilde{\mathcal O}\left(\frac{\kappa}{\tilde{\Delta}}d^2(\log(T)\right)$, where $\tilde{\Delta}$ is the constraint-aware sub-optimality gap subject to policy $\pi_0$, with variance-aware multiplicative term $\kappa$ that we carefully handle using heteroskedastic regression. We further show Dri-MED enjoys $\tilde{\mathcal{O}}(d)$ expected constraint violations. Our numerical results suggest that Dri-MED significantly outperforms conservative baselines that ignores the drift and preference structure.

AI 总结：针对PPO等方法的硬裁剪或硬掩码在长尾词汇中分布偏移代理不佳的问题，提出DRPO，用平滑的优势加权二次正则化替代硬掩码，保持信任区域几何的同时提供连续梯度权重，提升训练稳定性和效率。

链接：https://arxiv.org/abs/2606.09821

机构：Tencent Hunyuan（腾讯混元）； UIUC（伊利诺伊大学厄巴纳-香槟分校）； NUS（新加坡国立大学）

作者：Jiarui Yao, Xiangxin Zhou, Penghui Qi, Wee Sun Lee, Liefeng Bo, Tianyu Pang

英文摘要：Reinforcement learning (RL) has become a key component of post-training large language models (LLMs). In practice, LLM RL is often off-policy because of training-inference mismatch and policy staleness, making trust-region control essential for stable optimization. Mainstream methods such as PPO and GRPO approximate this control with a ratio-clipping mechanism, but the importance ratio can be a poor proxy for distributional shift in long-tailed vocabularies. Recent work such as DPPO addresses this mismatch by replacing ratio-based clipping with a divergence-based mask, yielding a trust region defined by the sampled token's absolute probability shift. However, DPPO still relies on a hard mask: once a token crosses the trust-region boundary in a harmful direction, its gradient is discarded rather than corrected. To address this, we propose Divergence Regularized Policy Optimization (DRPO), which replaces the hard mask with a smooth advantage-weighted quadratic regularizer on policy shift. DRPO preserves the same trust-region geometry as DPPO while inducing bounded, continuous gradient weights that attenuate diverging updates and provide corrective signals beyond the boundary. Experiments across model scales, architectures, and precision settings show that DRPO improves the stability and efficiency of LLM RL training.

AI 总结：提出一种将次优基线策略嵌入强化学习训练的方法，通过逐步从基线策略向可学习策略转移代理权，提升训练效率并最终获得超越基线的独立策略。

链接：https://arxiv.org/abs/2606.09825

机构：Center for Engineering Systems and Sciences（工程系统与科学中心）； Central University（中央大学）； Sirius University of Science and Technology（天狼星科技大学）

作者：Anton Bolychev, Georgiy Malaniya, Sinan Ibrahim, Pavel Osinenko

英文摘要：Training reinforcement learning (RL) policies from scratch is costly: it requires careful reward and environment design, extensive tuning, and substantial computation. Yet many control problems already have a functional but suboptimal policy available as a baseline. This paper proposes a method for embedding such a baseline into the RL training process, simultaneously improving training efficiency relative to from-scratch methods and producing a learning policy that outperforms the baseline. At each step, the method arbitrates between the baseline policy and a trainable learning policy, initially relying strongly on the baseline policy and then progressively transferring agency to the learning policy. By the end of training, the learning policy is a standalone neural network that operates without baseline policy support. The paper formalizes what it means for the baseline policy to be functional: under this policy, the agent reaches a goal set and remains there with high probability. The proposed arbitration mechanism is designed to exploit this property during training, yielding high goal-reaching rates right from the beginning of training. A theoretical analysis provides a formal interpretation of this behavior under stated assumptions and extends it to the final baseline-free regime, where explicit lower bounds are derived for the goal-reaching probability of the standalone learning policy. Empirical results on continuous-control benchmarks show that the proposed method achieves returns that match or exceed those of competitive approaches, while maintaining the highest goal-reaching rates throughout training among the compared methods -- including in the final stage, where the learning policy operates without any baseline support.

AI 总结：针对城市级高频碳排放数据稀缺问题，提出TriHead-GAN，通过三头判别器联合监督分布真实性、跨变量依赖和步态平滑性，在多个数据集上优于主流基线并提升下游预测精度。

链接：https://arxiv.org/abs/2606.07569

作者：Zesen Wang, Lijuan Lan, Yonggang Li, Chunhua Yang

英文摘要：Accurate carbon emission monitoring is critical for climate policy and emerging regulatory mechanisms such as the EU Carbon Border Adjustment Mechanism, yet city-level high-frequency monitoring data remain extremely scarce, severely limiting data-hungry deep learning models. Time series generation is a natural remedy, but existing GAN and diffusion-based generators often provide limited explicit supervision for the domain structure of carbon emission data: they may match marginal distributional statistics while insufficiently preserving cross-variable correlations between CO$_2$ and co-emitted pollutants and meteorological factors, and tend to collapse the first-difference statistics of atmospheric measurements, producing sequences that are smooth on average but lack the realistic step-wise variability of the underlying signals. We propose TriHead-GAN, a Transformer-based adversarial framework whose triple-head discriminator jointly supervises three complementary aspects of the joint distribution: distributional authenticity via a Wasserstein critic, cross-variable dependency via leakage-free regression of the target variable, and step-wise temporal smoothness via adjacent-difference prediction. The generator combines global self-attention with local temporal convolution, per-step noise injection, and an anti-smoothing loss that matches first-difference statistics. Experiments on the self-collected Changsha Carbon dataset, two public carbon datasets (China, US), and the ETTh1 benchmark show that TriHead-GAN achieves favorable performance over mainstream baselines on the vast majority of settings, and that the resulting synthetic windows improve downstream forecasting accuracy in low-resource carbon monitoring scenarios.

AI 总结：提出DiffOR框架，将序数回归建模为连续生成任务，利用扩散模型通过迭代去噪恢复连续序数值，并设计双解耦策略（多尺度增量聚合与动态去噪感知）保留序数拓扑，在12个基准上超越现有方法。

链接：https://arxiv.org/abs/2606.07599

机构：Fudan University（复旦大学）； Kuaishou Technology（快手科技）； Shanghai University of Finance and Economics（上海财经大学）； Tongji University（同济大学）

作者：Hongxu Ma, Lin Wang, Chenghou Jin, Han Zhou, Jie Zhang, Xiaoyu Yang, Chunjie Chen, Jihong Guan, Shuigeng Zhou

英文摘要：Ordinal Regression (OR) aims to predict target values with inherent order, underpinning critical applications across diverse domains, from recommender systems to computer vision. Though having evolved from naive regression to discretization-based classification and generation, existing paradigms remain fundamentally constrained by quantization artifacts and the lack of global ordinal topological perception. These methods typically enforce rigid boundary delineations, failing to capture the non-stationary semantic transitions inherent to ordinal data. In this paper, we propose a novel paradigm where OR is formulated as a Continuous Generative Ordinal Regression task. Under the novel paradigm, we introduce DiffOR, a unified framework that leverages diffusion models to recover continuous ordinal values via iterative denoising, thereby enabling the dynamic learning of soft semantic transitions. To explicitly preserve ordinal topology, we devise a Dual-Decoupling Strategy: Spatially, Multi-scale Increment Aggregation decomposes targets into hierarchical continuous increments; Temporally, Dynamic Denoising Perception synchronizes denoising steps with feature frequencies, ensuring robust coarse-to-fine refinement. Theoretically, we show that the proposed method can significantly enhance both representation capability and mechanistic interpretability. Extensive experiments on 12 benchmarks across four domains validate DiffOR's consistent superiority over state-of-the-art methods, establishing a new standard that demonstrates strong potential as a general-purpose solution for universal ordinal regression.

AI 总结：提出SRT框架，通过解缠校正流将低分辨率时间序列重建为高分辨率，分解趋势与季节成分，利用隐式神经表示对齐分辨率，并引入跨分辨率注意力机制生成细节。

链接：https://arxiv.org/abs/2606.07605

机构：Bytedance（字节跳动）

作者：Jufang Duan, Shenglong Xiao, Yuren Zhang

英文摘要：Fine-grained time series data with high temporal resolution is critical for accurate analytics across a wide range of applications. However, the acquisition of such data is often limited by cost and feasibility. This problem can be tackled by reconstructing high-resolution signals from low-resolution inputs based on specific priors, known as super-resolution. While extensively studied in computer vision, directly transferring image super-resolution techniques to time series is not trivial. To address this challenge at a fundamental level, we propose Super-Resolution for Time series (SRT), a novel framework that reconstructs temporal patterns lost in low-resolution inputs via disentangled rectified flow. SRT decomposes the input into trend and seasonal components, aligns them to the target resolution using an implicit neural representation, and leverages a novel cross-resolution attention mechanism to guide the generation of high-resolution details. We further introduce SRT-large, a scaled-up version with extensive pre-training, which enables strong zero-shot super-resolution capability. Extensive experiments on nine public datasets demonstrate that SRT and SRT-large consistently outperform existing methods across multiple scale factors, showing both robust performance and the effectiveness of each component in our architecture.

AI 总结：提出scCBGM框架，通过概念瓶颈架构和解耦惩罚实现单细胞反事实编辑，在组合泛化和反事实预测上表现优异。

链接：https://arxiv.org/abs/2606.07760

作者：Alma Andersson, Aya Abdelsalam Ismail, Edward De Brouwer, Doron Haviv, Tommaso Biancalani, Kyunghyun Cho, Gabriele Scalia, Aïcha BenTaieb, Hector Corrada Bravo

英文摘要：Understanding cellular phenotypes and how they respond to perturbations is critical for disease biology and therapeutic design. Single-cell RNA sequencing enables characterization at cellular resolution, yet the combinatorial space of conditions makes exhaustive experimental mapping infeasible. We introduce single-cell Concept Bottleneck Generative Models (scCBGM), a framework for interpretable and precise counterfactual editing of individual cells. scCBGM adapts concept bottleneck architectures for single-cell data through decoder skip connections and a cross-covariance penalty that promotes disentanglement without dimensional constraints. We extend the framework to flow matching models, enabling concept-guided editing in both encoding-decoding and generation regimes. To enable rigorous evaluation, we develop a synthetic benchmark with ground-truth counterfactuals. Across multiple real datasets, scCBGM demonstrates superior performance in combinatorial generalization and counterfactual prediction, supported by cell-level validation on synthetic data and population-level benchmarks on real datasets.

AI 总结：针对扩散模型确定性概率流ODE大步长推理中的收缩陷阱问题，提出SteinDiff框架，通过Stein导出的几何感知残差校正机制正则化求解器更新，无需参考样本即可提升生成质量。

链接：https://arxiv.org/abs/2606.07835

作者：Shigui Li, Delu Zeng

英文摘要： A fundamental tension exists in the large-step inference of diffusion models via their deterministic probability flow ordinary differential equation (PF-ODE) trajectories, which we identify as the contractivity trap: efficient inference favors large step sizes, while aggressive steps and highly expressive denoisers can undermine contraction-based stability certificates for error suppression. To address this, we propose SteinDiff, a step-wise inference-time stabilization framework that employs Stein-derived corrections without requiring reference samples. Specifically, SteinDiff introduces a geometry-aware residual correction mechanism that regularizes large-step solver updates without retraining. To this end, we derive a closed-form Stein correction coefficient for step-wise solver adjustment, enabling reference-free adaptation to local data geometry. We further establish a score-controlled perturbation bound under distributional shifts and provide a complementary Stein perspective on EDM-style parameterizations. Extensive experiments demonstrate that SteinDiff mitigates severe artifacts and improves generative quality across large-step inference settings.

AI 总结：针对OLED分子光学性质可控生成中数据稀缺和条件控制可靠性有限的问题，提出基于GPT2的Token条件自回归语言模型，通过离散属性Token和多任务优化实现垂直吸收能和振子强度的定向生成，并在TDDFT级别评估分布保真度和可控性。

链接：https://arxiv.org/abs/2606.08221

机构：University of Toronto（多伦多大学）； Vector Institute for Artificial Intelligence（向量人工智能研究所）； Universidad Autónoma de Madrid（马德里自治大学）； Canadian Institute for Advanced Research (CIFAR)（加拿大高等研究院）； NVIDIA（英伟达）

作者：Haozhe Huang, Manuel Gonzalez Lastre, Hyun Suk Park, Jorge A. Campos-Gonzalez-Angulo, Xinjian Liu, Alán Aspuru-Guzik

英文摘要：Designing OLED molecules with targeted optical properties remains challenging due to the scarcity of high-quality data and the limited reliability of conditional control in generative models across chemical motifs. Here, we benchmark a token-conditioned autoregressive language model for OLED molecular generation in a realistic low-data regime. A GPT2 model is pretrained on large chemical corpora, augmented with discrete property tokens, and fine-tuned using multi-task optimisation. Conditioning targets vertical absorption energy and oscillator strength, with the HOMO-LUMO gap included as an auxiliary electronic descriptor. Generated molecules are evaluated at the TDDFT level to assess distributional fidelity and controllability. The generated library reproduces the dominant optical-property support of the training distribution while shifting towards lower molecular weight and fewer heavy atoms. Token-level control is consistently directional across conditioning bins, but is not fully orthogonal and exhibits local calibration irregularities. A chemotype-resolved analysis further shows that controllability depends strongly on local electronic environments: moderately conjugated aromatic-carbon motifs are associated with improved joint target satisfaction, whereas electron-withdrawing motifs, particularly aryl nitriles, show systematic red-shifting and reduced controllability. These results establish a quantitative benchmark for conditional OLED molecular generation and show that model reliability must be assessed in chemically meaningful subspaces rather than from aggregate property distributions alone.

AI 总结：提出基于二维正交小波基的分数函数参数化，通过数据分布矩分析揭示不同架构的归纳偏差，解释扩散模型中分数网络与数据分布的相互作用。

链接：https://arxiv.org/abs/2606.08309

机构：The Kempner Institute for the Study of Natural and Artificial Intelligence（肯普纳自然与人工智能研究所）； Harvard University（哈佛大学）

作者：Emma Finn, Binxu Wang, T. Anderson Keller, Demba E. Ba

英文摘要：Score-based generative models have had remarkable success over the last decade in generating a diverse set of visually plausible images. A variety of architectures including CNNs, U-Nets, and Transformers have been used as the score-approximation network in such diffusion modeling; however, to date, relatively little is known about how these architectural choices impact generative behavior. In this work, to provide insight into this area, we propose an analytically solvable parameterization of the score function using an expansion in a 2D orthogonal wavelet basis. In particular, we derive interpretable optimal score functions in terms of the moments of the data distribution. We use this parametrization to provide an architecture-agnostic, moment-based analysis that reveals which attributes of the data distribution tend to matter most for denoising. Our score machine is flexible enough to partially mimic the relevant inductive biases of multiple architectures, including U-Nets, and CNNs, taking a step towards understanding why different score architectures can exhibit distinct generative behavior. Since our score is solvable in terms of the moments of the data, we can begin to understand how the data distribution interacts with the score network to produce the behavior we observe in diffusion models.

AI 总结：提出DeCAF框架，将全原子共折叠扩散模型蒸馏为流图，仅需几步推理即可生成高质量样本，并通过奖励引导搜索提升采样质量。

链接：https://arxiv.org/abs/2606.08375

机构：Genesis Molecular AI； Massachusetts Institute of Technology（麻省理工学院）； Carnegie Mellon University（卡内基梅隆大学）； Imperial College London（伦敦帝国学院）； Mila

作者：Gianluca Scarpellini, Ron Shprints, Peter Holderrieth, Juno Nam, Pranav Murugan, Rafael Gómez-Bombarelli, Tommi Jaakola, Maruan Al-Shedivat, Nicholas Matthew Boffi, Avishek Joey Bose

英文摘要： All-atom generative modeling of 3D biomolecular complexes has emerged as the dominant paradigm for predicting the structure of proteins and protein-ligand systems. Generating structures at the atomic level of fidelity, however, typically requires expensive iterative diffusion rollouts, making both conventional deployment and inference-time search techniques computationally costly. In this paper, we introduce the Denoiser Cofolding All-Atom Flowmap (DeCAF) framework for distilling state-of-the-art all-atom cofolding models into all-atom flow maps that produce high-quality samples in only a few inference steps. We build DeCAF on a denoiser-based formulation of flow maps with endpoint losses that naturally support SE(3) rigid alignment, which we show is critical for training accurate models. We further derive a simple change of variables that lets DeCAF operate in the {\sigma}-space noise schedule of EDM-style architectures, enabling direct distillation from pretrained cofolding diffusion models. Equipped with DeCAF's flowmap lookahead, we introduce a purpose-built inference-time framework that improves sampling through reward-guided search. Empirically, DeCAF-Boltz statistically improves over Boltz-1x in both accuracy (RMSD) and physical validity scores of protein-ligand poses at strict NFE budgets on the challenging Runs N' Poses, while also showing a more optimal Pareto frontier across all inference compute budgets on PoseBusters. Distilling the state-of-the-art Pearl cofolding model, DeCAF-Pearl outperforms diffusion-based cofolding models and matches its teacher on success rate while using 5x fewer NFEs. We release our code at this https URL.

AI 总结：本文通过闭式解分析随机插值模型中的记忆化现象，揭示连续时间下确定性及随机生成过程均恢复训练样本，离散化与估计误差导致样本偏离，并给出过拟合与欠拟合的理论定义。

链接：https://arxiv.org/abs/2606.08554

作者：Yunchen Li, Shaohui Lin, Zhou Yu

英文摘要：This paper provides a theoretical account of memorization in stochastic interpolation models. By leveraging closed-form expressions for the optimal velocity field and the associated score function, we show that, in the continuous-time oracle setting, both deterministic and stochastic generation processes recover training samples. Under Euler discretization, generated samples remain centered around training samples, with deviations controlled by the step size. We further analyze generation in the presence of estimation errors and show that accumulated estimation errors control the endpoint deviation from the training set. These results imply that the generated sample admits a representation as a training sample perturbed by three controlled terms: a discretization-induced bound, an estimation-error-induced bound, and stochastic Gaussian noise. Based on this characterization, we provide theoretical definitions of overfitting and underfitting in generative models. Synthetic simulations support our theoretical findings.

AI 总结：提出Active Flow Expansion (ActFlow)方法，通过验证器反馈和主动探索扩展预训练流模型的生成集，覆盖更多有效设计空间，理论证明统计学习保证，在分子和蛋白质任务上优于现有方法。

链接：https://arxiv.org/abs/2606.08802

机构：ETH Zurich（苏黎世联邦理工学院）； ETH AI Center（ETH AI 中心）； University of Pennsylvania（宾夕法尼亚大学）； Caltech（加州理工学院）； FutureHouse

作者：Riccardo De Santi, Bruce Lee, Cristian Perez Jensen, Kimon Protopapas, Sophia Tang, Cheng-Hao Liu, Pranam Chatterjee, Yisong Yue, Andreas Krause

英文摘要：Standard flow and diffusion pre-training matches the distribution of available data (e.g., molecules), which often covers only a small fraction of the valid design space. In generative discovery, however, one aims to sample valid new-to-nature designs, assigned negligible probability under, and thus inaccessible to, standard models fitted to the observed data. To overcome this limitation, we depart from data distribution matching and view a generative model through its generable set: the region it covers with non-negligible probability. This allows to introduce a new learning principle for out-of-distribution flow modeling: enlarging a model's generable set to increase coverage of the valid design space. We propose Active Flow Expansion (ActFlow), a continued pre-training method that employs verifier feedback to expand a pre-trained model over new valid regions by iteratively adapting to synthetic data generated through active exploration in the learned flow representation. Theoretically, we establish to our knowledge first-of-their-kind statistical learning guarantees for out-of-distribution flow modeling, analyzing generable set expansion as a local-to-global reachability process over a learned representation. Empirically, we assess ActFlow with suitable out-of-distribution generative modeling metrics across small organic molecules, mid-sized drug-like molecules, therapeutic peptides, and protein sequence design tasks. Results show that ActFlow expands valid coverage far beyond the region modeled by the initial pre-trained model, significantly outperforming widely adopted synthetic flow pre-training methods.

AI 总结：提出自洽生成路径作为可容许局部变分输运校正的随机不动点，并引入随机不动点路径残差（R-FPR）来度量生成路径与校正之间的差距，为扩散、流、一步生成、VAE、GAN等模型提供残差控制原理。

链接：https://arxiv.org/abs/2606.08953

机构：PCA Lab, Key Lab of Intelligent Perception and Systems for High-Dimensional Information of Ministry of Education, School of Computer Science and Engineering, Nanjing University of Science and Technology（南京理工大学计算机科学与工程学院高维信息智能感知与系统教育部重点实验室PCA实验室）

作者：Lei Luo, Yingzhen Zhang, Jian Yang

英文摘要： Modern generative models often define an entire probability path from a simple prior to the data law, rather than only an endpoint map. Diffusion models follow stochastic denoising paths, flow matching learns transport fields, consistency and distillation methods compress paths into one or a few steps, adversarial models match terminal distributions, and VAEs generate through latent kernels. Existing unifying views mainly describe how such paths are constructed. We study a complementary question: when is a generated probability path self-consistent? We define a self-consistent generative path as a random fixed point of admissible local variational transport corrections. In this framework, a local correction is specified by a random variational transport operator combining a divergence or geometry term, an energy term, and a structural constraint. The framework contains random regularized optimal-transport proximal steps as a structured instance, while also allowing non-OT divergences, latent kernels, adversarial constraints, causal discrete kernels, and terminal one-step maps. The theory yields a random fixed-point path residual (R-FPR), which measures the gap between the actual generated path and an admissible local correction. We prove well-posedness, random fixed-point existence and attraction, non-contractive existence, residual-to-generation error bounds, empirical residual concentration, proxy perturbation bounds, continuous-time limits, and operator-level generalization with model-specific corollaries. The resulting theory turns endpoint matching into path self-consistency testing and provides a residual-control principle for diagnosing failures, regularizing training, and guiding adaptive sampling across diffusion, flow, one-step, VAE, GAN/WGAN, and autoregressive generators.

AI 总结：针对高维低样本量表格数据，提出BSTabDiff框架，通过将特征划分为潜在块并使用共享低维子单元变量生成每个块，结合扩散先验和copula依赖，实现稳定合成与可控基准生成。

链接：https://arxiv.org/abs/2606.09257

机构：West Virginia University（西弗吉尼亚大学）； The University of Utah（犹他大学）

作者：Al Zadid Sultan Bin Habib, Md Younus Ahamed, Prashnna Gyawali, Gianfranco Doretto, Donald A. Adjeroh

英文摘要：High-Dimensional Low-Sample Size (HDLSS) tabular domains (e.g., omics) are characterized by $n \ll m$, where $n$ = number of samples, and $m$ = number of features. Such domains often exhibit strong local correlation groups, sparse cross-group dependencies, heavy-tailed non-Gaussian marginals, heteroscedastic noise, and structured missingness, making direct density learning in $\mathbb{R}^m$ ill-conditioned since $n \ll m$. We propose BSTabDiff, a block-subunit generative framework that partitions the $m$ observed features into $M$ latent blocks ($M \ll m$) and generates each block via a shared low-dimensional subunit variable, concentrating global dependence learning in the compact block-latent space $\mathbb{R}^M$ while decoding to the full feature space with copula-driven dependence, flexible per-feature marginals, and explicit missingness mechanisms. BSTabDiff supports modern deep priors on block latents, including diffusion and normalizing flows, enabling stable synthesis and controllable benchmark generation in the HDLSS regime. Empirically, BSTabDiff produces more realistic and stable high-dimensional synthetic data when compared with unstructured tabular generators on HDLSS data.

AI 总结：针对潜空间贝叶斯优化中上下文学习模型与优化任务不匹配的问题，提出在分子VAE潜空间上定义合成优化任务进行持续预训练，并引入正则化器保持原始先验，显著提升分子优化性能。

链接：https://arxiv.org/abs/2606.09664

机构：Aalto University（阿尔托大学）

作者：Tuan A. Vu, Harri Lähdesmäki, Julien Martinelli

英文摘要：Bayesian optimization (BO) is a central tool for sample-efficient design, and latent-space Bayesian optimization (LSBO) extends it to structured objects such as molecules and proteins. In parallel, tabular foundation models such as TabPFN and TabICL now achieve state-of-the-art regression performance and are increasingly used as BO surrogates. Because their Bayesian behavior is induced by large synthetic pretraining collections, the composition of this pretraining distribution is crucial. LSBO creates a distinctive mismatch: the induced map from latent code to objective value differs markedly from the regression tasks used to train current in-context models. We address this mismatch by complementing the pretraining stage of tabular foundation model surrogates with synthetic optimization tasks defined on the latent space of a molecular VAE. The continued-pretraining objective features a regularizer that anchors the model to the original checkpoint, preserving its broad regression prior while avoiding overspecialization to the adaptation tasks. On held-out molecular optimization benchmarks, the resulting model achieves strong performance, supporting the relevance of LSBO-specific adaptation for in-context surrogates.

AI 总结：提出诊断理论，证明局部模型能否稳定外推取决于高斯平滑评分的准局部性，并引入有限深度局部流（FDLF）基准进行验证。

链接：https://arxiv.org/abs/2606.09705

机构：The University of Hong Kong（香港大学）； Department of Physics and HK Institute of Quantum Science & Technology（物理系与香港量子科学与技术研究所）

作者：Wenjie Xi

英文摘要：Scientific generative modeling often requires size transfer, where models trained on small systems are evaluated on larger ones. While translation-invariant architectures enable this evaluation, we show that architectural locality alone does not guarantee stable size extrapolation. Instead, stable extrapolation is governed by the quasi-locality of the Gaussian-smoothed score. Through Tweedie's formula, far-away perturbations can influence local score components via posterior covariance, meaning a local model succeeds only if its receptive field covers the smoothed score's response range. We formalize this mechanism, proving a size-uniform comparison theorem for local marginals under reverse diffusion. We also introduce Finite-Depth Local Flow (FDLF), a white-box diagnostic benchmark with exact scores, densities, and controllable response ranges. Empirically, we validate the interplay between spatial mixing, smoothed-score quasi-locality, and model receptive fields. Under spatial mixing, the smoothed score remains quasi-local relative to the receptive field, enabling stable extrapolation. Conversely, when spatial mixing weakens, the score's locality rapidly degrades, causing size transfer to fail.

AI 总结：本文揭示有界域上平稳核高斯过程边界方差膨胀的根本原因是核相关邻域截断，并证明该几何扭曲导致三类采集函数产生系统性偏差，提出无函数选择剖面诊断方法。

链接：https://arxiv.org/abs/2606.07561

机构：RISE Research Institutes of Sweden（瑞典RISE研究院）； Uppsala University（乌普萨拉大学）

作者：Maria Bånkestad, Sanna Jarl, Jens Sjölund

英文摘要：Gaussian processes with stationary kernels on bounded domains exhibit inflated posterior variance near the boundary. Despite being a long-recognized artifact in geostatistics and a source of over-exploration in Bayesian optimization, the causes and effects of boundary-induced acquisition bias are underexplored. We trace the root cause to a simple geometric mechanism: the truncation of the kernel correlation neighborhood at the domain boundary creates an observation-independent distortion that worsens with dimensionality. We show how this distortion manifests across three acquisition classes: variance maximization concentrates selections at the corners, whereas negative integrated posterior variance and expected predictive information gain move selections inward to axis-aligned interior shells. These patterns arise without reference to any objective function, meaning that acquisition behavior can be dominated by kernel geometry rather than the desired task-specific uncertainty. To quantify this, we introduce a function-free selection-profile diagnostic for arbitrary acquisitions, kernels, and bounded-domain geometries.

AI 总结：针对顺序过滤管道，在独立模型下证明按成本与拒绝概率递增比率排序可最小化期望总成本，并通过蒙特卡洛模拟验证其优于常见启发式方法。

链接：https://arxiv.org/abs/2606.07589

作者：Hrishikesh Paranjape, Abhishek Mandal, Xian Sun

英文摘要：Sequential filtering pipelines are a common design pattern in large-scale systems, where a large population of items is progressively reduced by a sequence of stages that each incur cost. Despite their prevalence in ranking systems, cascaded machine learning inference, and fraud detection, filter ordering is often determined by heuristics without formal guarantees. We formalize sequential filtering under an expected-cost objective and prove that, under an independence model, ordering filters by increasing ratio of cost to rejection probability minimizes expected total cost. Extensive Monte Carlo simulations show that the optimal ordering strictly dominates common heuristics across all runs, both in expectation and across the full distribution of outcomes.

AI 总结：提出用有限语义证书验证上下文条件语言模型行为，证明有限域线性任务族中确定性准则，并证明阈值涌现的反幻象定理，将阈值度量与语义置信度分离。

链接：https://arxiv.org/abs/2606.07623

机构：Bahcesehir University（巴切谢希尔大学）

作者：Faruk Alpay, Hamdi Alakkad

英文摘要：This paper develops a model-theoretic framework for verifying context-conditioned language-model behavior by replacing benchmark labels with finite semantic certificates. The first problem is finite determinacy: when do examples in a context force the answer to a query without changing model parameters? In finite-field linear task families, we prove an exact row-space criterion, compute the residual hypothesis count, derive full and query-local identification curves, and show that extracting a smallest forcing subcontext is NP-complete even for binary outputs. The second problem is threshold emergence: when does an apparent benchmark jump reflect a semantic transition rather than a discontinuity of the scoring map? We prove an anti-mirage theorem separating thresholded metrics from semantic confidence and give a rate-sensitive crossing bound for latent commitments becoming visible above threshold. The common semantic object is a confidence functional on definable events. We show that it is a Boolean probability measure, equivalently a Keisler measure on the relevant type space, whose measure-one formulas form a proper filter and whose Stone-space representation is invariant under definitional expansion. The resulting calculus provides finite context certificates, pair-separator hitting sets, query teaching dimension, prompt-preservation criteria, and scale-limit witnesses. Exact-arithmetic ancillary scripts reproduce the finite-field and threshold calculations and generate the data used by the figures.

AI 总结：本文研究全连接ReLU网络线性区域构成的复形，证明其连通图平均度上界为输入维度的两倍，且直径上界与输入维度无关。

链接：https://arxiv.org/abs/2606.07728

机构：University of Connecticut（康涅狄格大学）

作者：Blake B. Gaines, Jinbo Bi

英文摘要：It is well established that ReLU networks define continuous piecewise-linear functions, and that their linear regions are polyhedra in the input space. These regions form a complex that fully partitions the input space. The way these regions fit together is fundamental to the behavior of the network, as nonlinearities occur only at the boundaries where these regions connect. However, relatively little is known about the geometry of these complexes beyond bounds on the total number of regions, and calculating the complex exactly is intractable for most networks. In this work, we prove new theoretical results about these complexes that hold for all fully-connected ReLU networks, specifically about their connectivity graphs in which nodes correspond to regions and edges exist between each pair of regions connected by a face. We find that the average degree of this graph is upper bounded by twice the input dimension regardless of the width and depth of the network, and that the diameter of this graph has an upper bound that does not depend on input dimension, despite the number of regions increasing exponentially with input dimension. We corroborate our findings through experiments with networks trained on both synthetic and real-world data, which provide additional insight into the geometry of ReLU networks. Code to reproduce our results can be found at this https URL.

AI 总结：提出部分表现性预测框架，统一建模由模型部署引起的内生分布偏移和外部时间变化引起的外生偏移，并定义在线表现性稳定与最优性，分析重复训练等启发式方法的适应性条件。

链接：https://arxiv.org/abs/2606.07890

机构：Stanford University（斯坦福大学）

作者：Jaewook Lee, Tijana Zrnic

英文摘要：Performative prediction studies feedback loops that arise when predictive models are deployed in consequential domains. In these settings, deploying a model can change the population whose patterns the model aims to predict, inducing a distribution shift that is endogenous to the learning system. This perspective departs from classical treatments of distribution shift, where shifts are typically modeled as exogenous changes in the data-generating process. Yet, in practice, distribution shift is rarely one or the other. Predictive models may influence future data through the decisions they support, while the world itself continues to drift for reasons beyond the learner's control. We study partially performative prediction, a framework that captures both endogenous and exogenous sources of distribution shift. The framework generalizes performative prediction by allowing the data distribution to evolve both in response to the deployed model and according to an external, time-varying process. We extend the central notions of performative stability and performative optimality to this setting by defining their online analogues that track the evolving partially performative environment. We analyze practical learning heuristics, including repeated retraining, and characterize when they successfully adapt to partially performative environments.

AI 总结：针对强凸损失在线凸优化，提出噪声自适应高概率遗憾界，在完全信息下实现与噪声水平相关的乘性改进，并证明赌博反馈下遗憾与置信度的线性关系，同时为约束优化提供联合高概率保证。

链接：https://arxiv.org/abs/2606.08028

机构：Tsinghua Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生院）； College of Mathematics, Sichuan University（四川大学数学学院）

作者：Wentao Zhang, Yutong Zhang, Wentao Mo

英文摘要：We study high-probability regret bounds for online convex optimization (OCO) with strongly convex losses and establish three results that resolve open questions at the intersection of noise adaptivity, feedback structure, and constraint satisfaction. For the full-information setting with sub-Gaussian stochastic gradients, we prove a noise-adaptive high-probability regret bound in which the martingale deviation term scales with the noise level $\sigma$ rather than the gradient bound $G$, yielding a multiplicative improvement of $G/\sigma$ over the classical Azuma-Hoeffding baseline. Our analysis introduces an exponential supermartingale argument that bypasses the bounded-difference requirement of Freedman's inequality, enabling direct treatment of unbounded sub-Gaussian noise without truncation artifacts. For bandit feedback, we prove a minimax lower bound: the high-probability regret scales linearly in $\log(1/\delta)$, in contrast to the $\sqrt{\log(1/\delta)}$ confidence cost under full information. This constitutes a formal separation in the confidence cost of strongly convex OCO across feedback models. Regarding constrained OCO with stochastic constraints satisfying a Slater condition, we provide simultaneous high-probability guarantees for both cumulative regret and long-run constraint violation, achieving $\mathcal{O}(\sqrt{T\log(m/\delta)})$ regret and $\mathcal{O}(\sqrt{T}/(\zeta\delta) + m\sqrt{T\log(m/\delta)})$ violation. Synthetic experiments corroborate all theoretical predictions.

AI 总结：提出条件随机有序传输空间（CROTS），通过引入有序传输几何和条件风险泛函，解决分布学习中传输方向是否被允许的问题，并建立稳定性定理。

链接：https://arxiv.org/abs/2606.08113

机构：Nanjing University of Science and Technology（南京理工大学）； PCA Lab, Key Lab of Intelligent Perception and Systems for High-Dimensional Information of Ministry of Education（PCA实验室，教育部高维信息智能感知与系统重点实验室）； School of Computer Science and Engineering（计算机科学与工程学院）

作者：Lei Luo, Jian Yang

英文摘要：A small Wasserstein distance does not certify that a transformation is admissible. In evidence-constrained, semantic, causal, physical, monotone, or risk-sensitive learning, one must ask not only how far two probability laws are, but whether mass has moved in a direction allowed by available information. We introduce conditional random ordered transport spaces (CROTS), a class of \(L^0\)-valued spaces of random probability measures equipped with a Wasserstein ambient metric, a closed stochastic order, hard and soft ordered transport discrepancies, and a conditional risk functional for evaluating order violation under an evidence sigma-field. The central object is an order-admissible transport geometry for random measure-valued dynamics, distinct from cone-valued metrics, ordered Kantorovich constructions, random Wasserstein spaces alone, and model-specific residuals for generative paths. We develop the foundations of CROTS as a space theory for reliable distributional learning. The results include well-posedness and duality for hard and soft ordered transport, soft-to-hard variational convergence, measurability and completeness of the random lifted space, reductions to classical Wasserstein and ordered geometries, ordered geodesics, constrained barycenters and projections, conditional risk-transport duality, and separation of order-violating distributions. The main stability theorem shows that random learning dynamics may converge in the ambient Wasserstein metric while its local admissibility leakage follows a separate conditional order-risk recursion. The resulting asymptotic order-risk floor provides a mathematical language for evidence overreach, ordered distribution shift, robustness failure, and admissible distributional dynamics.

AI 总结：提出统一框架解释多领域数据混合中模型损失行为，基于能力竞争和噪声减少两个关键因素，在多个尺度上有效预测高性能混合。

链接：https://arxiv.org/abs/2606.08167

机构：Beijing Institute of Technology（北京理工大学）； IIIS, Tsinghua University（清华大学智能产业研究院）

作者：Rui Dai, Shuran Zheng

英文摘要：Recent research has established empirical scaling laws to predict model performance on multi-domain data mixtures. However, a theoretical understanding of these model loss behaviors remains absent. In this work, we propose a unified framework to explain the underlying mechanics of data mixing. Our approach extends theoretical perspectives originally developed for standard neural scaling laws (e.g., Kaplan and Chinchilla) to the multi-domain setting. Based on the distributional assumption that domains overlap on fundamental skills while diverging on specialized skills, we identify two key factors that govern the domain losses of models trained on different data mixtures: \textit{Capacity Competition}, where the allocation of finite model capacity couples domain losses globally, and \textit{Noise Reduction}, where optimal weights shift toward harder-to-learn domains to minimize overall noise. Empirical evaluations show that our framework outperforms existing baselines by fitting the loss landscape with a lower Mean Relative Error and identifying higher-performing training mixtures. Most importantly, our model successfully extrapolates across scales, predicting highly effective mixtures for large, unseen scales using parameters fitted on smaller ones. In addition, our model achieves these results using significantly fewer parameters compared to previous empirical laws. Our code is available at this https URL.

AI 总结：本文研究了深度高斯过程先验在深度增长时的极限行为，识别出RBF核带宽的尖锐阈值，低于该阈值时先验收敛到非退化非高斯分布，具有非零坐标依赖。

链接：https://arxiv.org/abs/2606.08218

机构：Technion, IIT（以色列理工学院）； NVIDIA（英伟达）

作者：Mark Kozdoba, Shie Mannor

英文摘要：Compositional priors describe the generic properties of layered functions in deep Bayesian models, where deep neural networks with random weights are a canonical this http URL the wide-network limit, the prior is a Gaussian process with a depth-dependent kernel, and its behaviour as depth grows has been extensively studied through this kernel. Here, we study another case, where each layer itself is a vector valued Gaussian process, and our aim is similarly to understand the limiting behaviour of the prior as depth grows. Previous GP work has established that for the RBF kernel and a certain range of bandwidths $r$, the prior degenerates in the limit, converging to the set of constant functions -- which is not useful as a probabilistic model. In this paper we establish several new results. First, we identify a sharp bandwidth threshold $r_c(d) = \Theta(\sqrt{d})$ above which the limit is degenerate, strengthening the earlier bounds. Second, and more importantly, we show that for $r$ below the threshold $r_c(d)$ the prior converges to a limit distribution $\pi_{\bar{Z}}$. We also prove that these distributions are non-degenerate and non-Gaussian, with non-vanishing dependence between coordinates. In contrast to the previously known degenerate regime, deep Gaussian process priors can therefore admit non-trivial limits. Empirically, we verify the threshold across a range of dimensions $d$, and demonstrate a complex multimodal behaviour of the limit distributions $\pi_{\bar{Z}}$ -- a regime that becomes increasingly narrow with $d$ and would be hard to identify without knowing the threshold.

AI 总结：研究对称多类型正交非负矩阵三因子分解问题，提出基于KKT条件的定点法和基于ADAM的三阶段算法，在合成数据和引文网络上验证了分解质量与聚类、链接预测等任务中的竞争力。

链接：https://arxiv.org/abs/2606.08291

作者：Rok Hribar, Gregor Papa, Janez Povh, Andrej Kastrin

英文摘要：We study the symmetric multi-type orthogonal non-negative matrix tri-factorization problem, where several symmetric non-negative matrices are simultaneously approximated by factors of the form $GS_{i}G^{\top}$, with a shared non-negative and orthogonal factor $G$. This model is motivated by clustering and network analysis, where non-negativity improves interpretability and orthogonality gives a natural assignment-type structure to the latent factor. Since the resulting optimization problem is highly non-convex, we develop two heuristic algorithms for computing high-quality local solutions. The first one is a fixed point method derived from the Karush-Kuhn-Tucker conditions after adding a penalty term for the orthogonality constraint. The second one is a three-stage ADAM-based method that combines non-negativity-preserving optimization, orthogonalization, and restricted ADAM refinement on the feasible set. We evaluate both methods on synthetic data, including noisy instances, and on citation network benchmarks. The synthetic experiments show that both algorithms recover factorizations close to the optimum and remain stable under noise. On real networks, the learned embeddings are competitive with or better than standard baselines such as SVD, node2vec, and classical link prediction heuristics in link prediction, node clustering, and node classification tasks.

AI 总结：提出基于权重变化的傅里叶分形维数作为泛化度量，并设计傅里叶优化器正则化该维数，在CIFAR-10等数据集上实现与泛化差距的高相关性。

链接：https://arxiv.org/abs/2606.08308

机构：Institute of Mathematics, Statistics and Scientific Computing - University of Campinas（坎皮纳斯大学数学、统计与科学计算研究所）

作者：Joao B. Florindo, Davi Wanderley Misturini

英文摘要： Predicting the generalization performance of deep neural networks without relying on hold-out validation data is a fundamental challenge in machine learning. While Stochastic Gradient Descent (SGD) drives the optimization of these highly parameterized models, its heavy-tailed, non-Gaussian dynamics induce complex, scale-invariant trajectories in the parameter space. In this paper, we propose a novel generalization measure based on the Fourier fractal dimension of the network's weight variations. By analyzing the characteristic function of the Lévy-driven stochastic differential equations in the frequency domain, we extract a metric that robustly captures the geometric complexity of the learning process. Furthermore, we introduce a customized Fourier-based optimizer designed to actively regularize this fractal dimension during training. Extensive empirical evaluations on the CIFAR-10, SVHN, and MNIST datasets demonstrate that our proposed Fourier generalization measure exhibits a strong correlation with the actual generalization gap. Our method achieves state-of-the-art Kendall rank correlation coefficients, outperforming a wide array of existing norm-based, margin-based, and PAC-Bayesian measures. Ultimately, this work highlights the potential of frequency-domain fractal analysis as both a powerful predictor for model generalizability and a principled foundation for developing more stable optimization algorithms.

AI 总结：研究Muon优化器通过极分解替换矩阵梯度，证明其偏置为平坦谱，在欠定回归中导出奇异值动力学，实验表明其效果依赖于谱方向活跃度。

链接：https://arxiv.org/abs/2606.08388

机构：Massachusetts Institute of Technology（麻省理工学院）

作者：Pierfrancesco Beneventano, Mahmoud Abdelmoneum, Tomaso Poggio

英文摘要：Muon replaces a matrix gradient $G=U\Sigma V^\top$ by its polar factor $UV^\top$. This keeps the singular directions selected by the gradient, but makes the update spectrum flat. We study the optimization bias created by this operation. Under explicit alignment assumptions, we prove that the polar update is the one-step entropy-maximizing choice among bounded updates that use the gradient singular directions and do not adapt to the current weight spectrum. In an underdetermined regression model, we derive exact singular-value dynamics for continuous-time Muon and identify a measurement-dependent condition under which the normalized spectrum moves toward equal nonzero singular values. This geometry also rules out a common low-rank interpretation: at fixed Frobenius norm, Muon's distinguished state has a flat spectrum, whereas nuclear-norm minimization favors spectral concentration. Controlled matrix-sensing experiments separate the effect from simple gradient rescaling, show that norm-matched gradient descent does not reproduce Muon, and recover the predicted flattening trend across broad ablations. In small NanoGPT pretraining, Muon preserves stable rank, has a broad learning-rate plateau, and improves validation loss relative to AdamW; in a matched small-ViT control, the ranking reverses. The resulting picture is regime-dependent: Muon is not universally superior, but its flat-spectrum bias can help when many spectral directions need to remain active.

AI 总结：研究神经时间序列模型中交互发现的真实性问题，提出基于输入支持几何的可辨识性理论，并给出有效秩作为拟合前诊断工具。

链接：https://arxiv.org/abs/2606.08390

作者：Valentina Kuskova, Dmitry Zaytsev, Michael Coppedge

英文摘要：When a neural time-series model reports that one variable modulates another's effect on a target, is the discovered interaction a property of the data or an artifact of model flexibility? We argue that this is fundamentally a question of identifiability, governed by the geometry of the observed input support rather than by the specific neural architecture. We study the problem in a multiplicative-gating extension of neural additive vector autoregression (GNAVAR), in which source contributions are modulated by other lagged variables. We show that representational capacity is not identifiability: dependent inputs induce leakage between edge-specific interaction terms, and low-dimensional support permits distinct interaction decompositions that agree on the observed data while differing elsewhere. We then prove a population identifiability theorem for normalized minimal GNAVAR decompositions under explicit support conditions, including settings with shared modulators. The theory yields a simple practitioner-facing diagnostic: the effective rank of the joint lag-block covariance predicts, before fitting, whether interaction recovery is feasible for a given candidate set. When the candidate set is unknown, a two-seed stability check provides a practical operational test. The same support condition organizes empirical outcomes into the three states predicted by the theory. Our results show that interaction recoverability depends on support geometry, that effective rank provides a practical pre-fit diagnostic, and that instability across independent fits is a characteristic signature of non-identifiable interaction discovery. The identifiability phenomenon, the support condition, and the instability signature are model-agnostic; GNAVAR is the vehicle that makes them provable.

AI 总结：提出方向线性可分性度量（LSM），通过搜索包含目标类所有样本的仿射半空间并测量最小竞争样本入侵量，为神经表征的类间几何提供不对称、类级、目标归一化的诊断工具。

链接：https://arxiv.org/abs/2606.08721

机构：State Key Laboratory of Novel Software Technology, School of Intelligence Science and Technology, Nanjing University（南京大学智能科学与技术学院软件新技术国家重点实验室）； AI for Good (AIGO), Istituto Italiano di Tecnologia（意大利技术研究院AI for Good (AIGO)）； DITEN, University of Genoa（热那亚大学DITEN）； State Key Laboratory of Novel Software Technology, School of Artificial Intelligence, Nanjing University（南京大学人工智能学院软件新技术国家重点实验室）

作者：Yi Wei, Xuan Qi, Furao Shen

英文摘要： Modern neural classifiers commonly rely on linear readouts, yet predictive metrics alone do not characterize the class-wise geometry of the representations on which such readouts operate. We introduce the directional linear separability measure (LSM), a finite-sample diagnostic for one-sided affine separability. For a target class A and a competing set B, LSM searches over affine halfspaces that contain all samples in A and measures the smallest competing-sample intrusion that must remain on the target side, normalized by |A|. The resulting quantity is asymmetric, class-wise, target-normalized, and applicable to finite representations extracted from neural networks. We establish its supporting-hyperplane characterization, relate it to optimal affine classification accuracy, and prove invariance under full-rank linear embeddings. These results separate changes caused by linear reparameterization from those caused by information loss or nonlinear geometric transformations. We also give a penalty-based affine search for estimating class-wise LSM in high-dimensional features, with reported values computed from the original discrete preservation and violation criterion. Finally, we analyze coordinatewise gated nonlinearities as finite-sample geometric operators and empirically use LSM to diagnose class-wise intrusion across common deep-learning components and architectures.

AI 总结：针对Transformer无法学习某些简单布尔函数（如奇偶函数）的问题，通过分析参数空间几何，证明敏感函数在参数空间中占据极小区域，随机初始化几乎必然错过，从而解释了可表达但不可学习的现象。

链接：https://arxiv.org/abs/2606.08768

机构：Machine Learning, ICML（机器学习，ICML）

作者：Blanka Köver, Alexandra Butoi, Anej Svete, Michael Hahn, Ryan Cotterell

英文摘要：Transformers consistently fail to learn certain simple functions that are provably expressible with specific parameter settings. This gap between learnability and expressivity is particularly prominent for sensitive functions -- functions whose output is likely to change if a single bit of the input is flipped -- for example, PARITY. While prior work has established that transformers exhibit a bias toward functions with low average sensitivity, the precise mechanism underlying this bias remains poorly understood. To shed light on this phenomenon, we study the geometry of transformers' parameter space. We show that sensitive functions -- even when representable -- occupy a vanishingly small region that random initialization is very likely to miss. Specifically, we shift the focus from average sensitivity to the full sensitivity profile -- the distribution of sensitivity values across all inputs -- and prove that randomly initialized transformers almost surely compute functions which have low-sensitivity strings. Consequently, any function that lacks such strings is provably unlearnable.

AI 总结：提出结合拉格朗日分解的决策聚焦学习框架，通过新代理目标和两种损失函数，在保持可并行化的同时，有效处理大规模约束优化问题，实验表明在变量数多八倍的实例上优于传统方法。

链接：https://arxiv.org/abs/2606.08797

机构：Polytechnique Montréal（蒙特利尔综合理工学院）； Ecole Polytechnique（巴黎综合理工学院）； UCLouvain（鲁汶大学）； Mila - Québec AI Institute（魁北克人工智能研究所）； KU Leuven（荷语鲁汶大学）

作者：Stéphane Eilles-Chan Way, Hugo Percot, Quentin Cappart, Tias Guns, Louis-Martin Rousseau

英文摘要：Decision-focused learning has shown great promise for addressing predict-then-optimize problems, particularly in the presence of under-specified models. However, its practical deployment is often hindered by high computational costs and limited scalability, as it requires solving a constrained optimization problem for each training instance at every iteration. To address these challenges, we propose a novel framework that incorporates Lagrangian decomposition into the decision-focused learning paradigm. Specifically, we introduce a new surrogate objective along with two loss functions for evaluating and training the underlying prediction model. We further propose two variants of our approach, which offer different trade-offs between computational efficiency and solution quality. Our framework can be seamlessly integrated with standard decision-focused learning methods, including Smart Predict-then-Optimize (SPO+) and Implicit Maximum Likelihood Estimation (IMLE). Through experiments on two standard benchmarks, the multi-dimensional knapsack problem and quadratic portfolio optimization, we demonstrate that our approach achieves competitive performance while remaining amenable to parallelization. In particular, it consistently outperforms traditional decision-focused learning methods on large-scale instances, involving up to eight times more variables than those typically considered in related work. The implementation is available at this https URL.

AI 总结：提出LEAF框架，通过输入凸神经网络学习Moreau包络来加速凸优化，降低模型复杂度并保持收敛性，实验显示比最先进求解器快一个数量级。

链接：https://arxiv.org/abs/2606.08993

机构：University of Central Florida（中佛罗里达大学）

作者：Binh Nguyen, Trinh Tran, Truong X. Nghiem

英文摘要：We propose LEAF, a learning-enabled ADMM framework for accelerated convex optimization. The key idea is to approximate the Moreau envelope of the objective function using an Input Convex Neural Network (ICNN), resulting in a learned model that preserves convexity and smoothness. This leads to the proposed Moreau Envelope Learning ADMM (MEL-ADMM) and its splitting variant sMEL-ADMM. Unlike existing approaches that learn high-dimensional operators directly, LEAF learns a scalar-valued Moreau envelope, significantly reducing model complexity and improving data efficiency. The framework accommodates a broad class of convex problems with smooth and non-smooth objectives. By embedding convexity explicitly through the ICNN architecture, the proposed approach maintains high approximation accuracy while preserving key structural properties of the optimization problem. Both MEL-ADMM and sMEL-ADMM are developed with theoretical guarantees of convergence and feasibility under the learned model. Rigorous analysis shows that the proposed methods achieve convergence rates comparable to classical ADMM while reducing per-iteration computational cost. Numerical experiments demonstrate up to an order-of-magnitude speedup over state-of-the-art solvers while maintaining low optimality gaps

AI 总结： 提出更紧的收敛分析，揭示混合矩阵所有特征值影响收敛速率，并通过实验验证比仅用谱间隙的分析更准确。

链接：https://arxiv.org/abs/2606.09154

作者：Yuki Takezawa, Anastasia Koloskova, Sebastian U. Stich

英文摘要：Decentralized SGD is a fundamental algorithm in decentralized learning, although the influence of an underlying network topology on its convergence behavior is not yet fully understood. Existing convergence analyses have shown that topologies with a small spectral gap significantly deteriorate the convergence rate of Decentralized SGD in both homogeneous and heterogeneous cases. However, many prior papers have reported that indeed the choice of the topology has a significant experimental impact in the heterogeneous case, but has little experimental impact on training behavior in the homogeneous case. In this paper, we present a tighter convergence analysis of Decentralized SGD, offering a more precise understanding of how topologies affect the convergence rate than the prior analysis. Specifically, unlike existing convergence analyses that used only the spectral gap as a property of the topology, our novel analysis shows that all eigenvalues of the mixing matrix affect the convergence rate. Throughout the experiments, we carefully evaluated the convergence behavior of Decentralized SGD and demonstrated that our novel convergence analysis can more accurately describe the effect of topology on the convergence rate.

AI 总结：本文刻画了深度L、总参数W的Transformer的VC维，并建立了思维链学习的样本复杂度上下界，揭示了参数与序列长度对学习所需样本量的影响。

链接：https://arxiv.org/abs/2606.09731

作者：Chenxiao Yang, Nathan Srebro, Zhiyuan Li

英文摘要：We tightly characterize the VC dimension of depth-$L$ Transformers with a total of $W$ parameters, mapping an input sequence of length $T$ to a single output, establishing an upper bound of $O(L W \log (T W))$ and a nearly matching lower bound of $\Omega(L W \log (T W / L))$. We further tightly characterize the sample complexity of chain-of-thought learning using such a Transformer, showing teacher forcing (i.e. selecting a predictor consistent with the entire chain-of-thought on training data) learns with sample complexity $O\left(L W \log \left(\left(T+T^{\prime}\right) W\right)\right)$ and that any learning rule that uses chain-of-thought data requires at least $\Omega\left(L W \log \left(\left(T+T^{\prime}\right) W / L\right)\right)$ examples, where $T$ is the input length and $T^{\prime}$ is the number of autoregressive steps.

AI 总结：本文研究前馈ReLU网络在固定读出和二次损失下的梯度下降动力学，将其重写为训练集空间上的集体动力学，并揭示深度网络中权重诱导的Gram算子层次结构。

链接：https://arxiv.org/abs/2606.09744

作者：Claudio Nordio

英文摘要：We study feed-forward ReLU networks with fixed readout and quadratic loss. The aim is to rewrite gradient descent not primarily as a dynamics in weight space, but as a collective dynamics closed in terms of fields defined on the training-set space. For a single hidden layer, the weight variables can be eliminated from the activation dynamics, yielding a closed equation for the residuals governed by a collective kernel that factorizes into an input-geometric matrix and a dynamical co-activation matrix. For deeper networks, the residual dynamics retains a clean layer-wise kernel structure. However, from depth three onward, closure requires a hierarchy of weight-induced Gram operators that mediate information transport across layers.

AI 总结：针对扩散语言模型中双向注意力导致共享前缀KV不稳定的问题，提出双向前缀缓存（bicache），通过动态识别安全层深度重用KV，避免精度崩溃，提升吞吐量36.3%-98.3%。

链接：https://arxiv.org/abs/2606.07571

机构：Korea University（高丽大学）

作者：Younghun Go, Jaehoon Han, Changyong Shin, Chuk Yoo, Gyeongsik Yang

英文摘要：Key-value (KV) caching for shared prefixes is essential for high-throughput large language model (LLM) serving, but it faces critical challenges in emerging diffusion language models (DLMs). In DLMs, bidirectional attention means that updating any token dynamically alters the entire context and its corresponding KVs. Thus, existing caching techniques developed for LLMs, which assume that KVs remain invariant once computed, corrupt the shared prefix KVs. Our experiments show that applying these techniques to DLMs causes model accuracy to collapse to near zero. To unlock high-throughput DLM serving, we propose bidirectional prefix caching, bicache, the first KV caching technique for shared prefixes in DLMs. bicache is designed based on key observations from our comprehensive analysis: shared prefix KVs remain stable and reusable in shallow layers, while the depth of shallow layers depends on the fraction of shared prefix tokens in each request. Thus, bicache dynamically identifies a safe layer depth for reusing shared prefix KVs and eliminates redundant computation. Evaluations demonstrate that bicache significantly improves serving throughput by 36.3%-98.3% compared to existing techniques without accuracy collapse (only 0-1.8% difference).

AI 总结：提出两阶段方法，从人类引导的智能体辅助部署到自主技能系统，在AMD XDNA 2 NPU上实现8种LLM的端到端自动部署，性能超越或持平人工优化基线。

链接：https://arxiv.org/abs/2606.07586

机构：AMD Research and Advanced Development（AMD研究与高级开发）

作者：Jiajie Li, Erwei Wang, Zhiru Zhang, Samuel Bayliss

英文摘要：Spatial neural processing units (NPUs) provide an energy-efficient platform for edge LLM inference, but efficiently deploying an LLM end-to-end on such hardware remains labor-intensive. Although AI coding agents have begun to lower this cost, existing studies have largely focused on single-kernel optimization rather than end-to-end LLM deployment on resource-constrained spatial NPUs. We present a two-stage methodology, instantiated on the AMD XDNA 2 NPU, that progresses from human-guided development to agent autonomy. In the first stage, we develop a reference deployment of Llama-3.2-1B through human-guided agent assistance. The resulting implementation achieves a speedup of 2.2x on prefill and 4.0x on decode over the hand-optimized baseline, with the optimization trajectory and its lessons recorded as structured documentation throughout. In the second stage, we distill the documentation into an agent skill system consisting of eight phases, orchestrating the optimization and debugging skill sets, with numerical correctness strictly enforced at each phase. Using our agent skill system, we autonomously deploy eight additional decoder-only LLMs (Llama-3.2-3B, SmolLM2-1.7B, Qwen2.5-{0.5B, 1.5B, 3B}, Qwen3-{0.6B, 1.7B, 4B}) end-to-end on the AMD XDNA 2 NPU using the open-source compiler stack. To our knowledge, these models have not previously been deployed on AMD NPUs via any open-source software stack. Each deployment completes in 0.5-4 hours of agent wall time with almost no human guidance, and passes the numerical-correctness gates, demonstrating functional generalization to previously unencountered LLMs. Three of the eight match or exceed the sustained performance of our Llama-3.2-1B reference deployment, suggesting that the resulting implementations can be competitive without additional model-specific human engineering.

AI 总结：提出基于多臂赌博机算法的结构化剪枝框架，通过将每个神经元视为臂并评估移除奖励，在表格分类、回归及深度网络任务上验证了UCB1和汤普森采样等策略的有效性。

链接：https://arxiv.org/abs/2606.07615

作者：Salem Ameen, Sunil Vadera

英文摘要：Deep neural networks often contain redundant hidden units. Removing individual weights can reduce parameter count, but unstructured sparsity is not always easy to exploit in standard dense implementations. This paper develops a structured pruning framework in which complete neurons are removed using multi-armed bandit (MAB) algorithms. Each candidate neuron is treated as an arm; pulling an arm temporarily masks that neuron, measures the change in loss on a sampled mini-batch, restores the neuron, and updates an estimate of its safe-removal reward. The framework supports stochastic policies, including Epsilon-Greedy, Softmax, UCB1 and Thompson Sampling, and multiplicative-weight policies, including Hedge-style multiplicative weights and EXP3. We evaluate the method on tabular classification, tabular regression and deep neural-network benchmarks covering image, text and reasoning tasks. Statistical comparisons using the Friedman test followed by the Nemenyi post-hoc test show significant differences between methods. On tabular classification tasks, UCB1 obtains the highest mean rank among pruning policies and improves on the unpruned neural network. On regression tasks, UCB1 obtains the highest mean rank and is statistically competitive with, or superior to, several standard regression models according to R^2. On deep-learning tasks, UCB1 and Thompson Sampling obtain the strongest ranks, and several MAB policies significantly outperform the unpruned model, magnitude-based neuron pruning and greedy activation-variation pruning. The results show that MAB-based neuron pruning is an effective and computationally practical approach for structured model reduction.

AI 总结：提出ScaleSweep方法，通过扫描可行块尺度候选并选择最小化目标函数的候选，优化NVFP4量化中的尺度初始化，理论推导扫描范围边界，在Llama和Qwen模型上提升量化性能，缩小与全精度的差距。

链接：https://arxiv.org/abs/2606.07618

机构：Wangxuan Institute of Computer Technology, Peking University（北京大学王选计算机技术研究所）

作者：Li Lin, Xiaojun Wan

英文摘要：NVFP4 is a recently introduced hardware-supported FP4 format that improves the fidelity of 4-bit quantization through fine-grained block scales. However, existing NVFP4 scale initialization methods still primarily rely on AbsMax initialization, which leaves a noticeable gap to the optimal solution. To address this, we propose ScaleSweep, a simple and efficient scale optimization method that sweeps over feasible block scale candidates and selects the candidate that minimizes a target objective. We further provide a theoretical analysis of NVFP4 quantization and derive both lower and upper bounds for the required sweep range under mean square error (MSE) and weighted mean square error (WMSE) between the original tensor and the quantized reconstructed tensor. The proposed bounds substantially reduce the sweep space while preserving the optimal candidate, enabling negligible overhead compared with the baseline quantization operators. Experiments on Llama and Qwen models demonstrate that ScaleSweep consistently improves quantization performance over existing initialization methods and further narrows the gap to full precision. In particular, under aggressive end-to-end quantization of weights, activations, KV cache, and query states, ScaleSweep preserves more than 93% of the full-precision performance.

AI 总结：针对大语言模型推理中KV缓存传输的通信瓶颈和跨模型重用时的语义错位问题，提出语义缓存蒸馏（SCD）框架，通过低秩子空间重建和稀疏过渡层归一化输入预测，实现高达2.65倍的首令牌时间加速，且生成质量接近理想情况。

链接：https://arxiv.org/abs/2606.07684

作者：Qianli Ma, Zhiqing Tang, Hanshuai Cui, Zhi Yao, Weijia Jia

英文摘要：Disaggregated serving alleviates memory bottlenecks in Large Language Model (LLM) inference but creates a severe communication bottleneck: transmitting high-dimensional Key-Value (KV) caches often dominates time-to-first-token (TTFT). Moreover, reusing caches across heterogeneous models (e.g., base and fine-tuned variants) causes semantic misalignment that accumulates over layers, degrading generation quality. We propose Semantic Cache Distillation (SCD), a loss-constrained framework that replaces raw KV transmission with compact semantic codes. SCD addresses these challenges via two mechanisms: (1) Reuse, which reconstructs most layers from low-rank subspaces to minimize transfer cost, and (2) Patch, which predicts normalized inputs at sparse transition layers to truncate error propagation. Empirically, SCD delivers up to 2.65 $\times$ TTFT speedup over the oracle consumer prefill and dominates quantization and selective recomputation baselines on the quality--latency Pareto frontier in bandwidth-constrained regimes, while keeping generation quality within 5\% F1 of the oracle.

AI 总结：研究在混合长上下文模型中，通过Oracle引导的稀疏预填充减少密集注意力计算，在保持任务性能的同时实现加速，并验证了可行性、索引器质量和运行时加速潜力。

链接：https://arxiv.org/abs/2606.07703

作者：Hongxing Wang, Harenome Razanajato, Zhen Zhang, Yujie Yuan, Hongsheng Liu

英文摘要：Long-context prefill remains expensive because full/GQA layers still score the historical sequence, even in hybrid models with local, sparse, linear, or recurrent components. We study how much dense attention is needed to preserve task-level behavior under explicit support granularity and top-k budgets. We introduce an attention-mass top-k oracle for existing GQA checkpoints: for each layer and query position, it computes dense attention, selects head-averaged token support, and recomputes attention only on that support. The oracle is a diagnostic reference, not a deployable accelerator, and separates sparse-budget feasibility from indexer error and runtime realization effects. On Qwen-family retrieval-heavy evaluations, the longest per-query oracle rows stay within 1 point of dense, and a Qwen3.5-9B RULER-style sweep from 4K to 100K stays within 0.48 points. Guided by the oracle, we derive a head-collapsed auxiliary indexer trained by KL distillation from dense attention-mass distributions while keeping the backbone frozen. With separately distilled Qwen3.5-0.8B and Qwen3.5-9B indexers, the reported 16K/32K validation macro gaps are +2.04 and +1.13 points, treated as quality preservation rather than improvement; fused selection-block-shared support can introduce a larger realization gap. Preliminary single-card TTFT measurements show distilled-indexer sparse serving speedups of 1.71x for Qwen3.5-0.8B on NPU and 1.93x for Qwen3.5-9B on GPU against its dense FlashAttention-2 baseline. Additional random-init stress rows reach 3.44x, indicating sparse-runtime headroom but not validated output quality. This first release separates oracle feasibility, distilled-indexer quality, and runtime headroom, leaving a fully matched quality-latency frontier to future work.

AI 总结：提出基于数组数学（MoA）的缩放点积注意力重表述，通过代数构造消除所有中间数组，实现O(n dk + n dv)数据移动，相比标准实现O(n^2 + n dk + n dv)显著降低内存流量，并验证了数值精度。

链接：https://arxiv.org/abs/2606.07713

机构：University at Albany（奥尔巴尼大学）； Université Paris-Est Créteil（巴黎东大学克雷泰伊分校）

作者：Lenore Mullin, Gaetan Hains

英文摘要：The attention mechanism is the dominant computational bottleneck in modern transformer-based AI. Its standard implementation incurs quadratic memory traffic in the sequence length~$n$, and DRAM accesses cost 100--1000$\times$ more energy than arithmetic operations on contemporary hardware, so any analysis focused solely on FLOP counts fundamentally mischaracterises the bottleneck. We present a Mathematics of Arrays (MoA) reformulation of scaled dot-product attention and its numerically stable softmax, deriving a Denotational Normal Form (DNF) that eliminates all intermediate arrays -- including the implicit transposed-key buffer and every softmax temporary -- by algebraic construction rather than empirical tuning. The DNF achieves $O(n_{dk} + n{_{dv}})$ data movement versus $O(n^2 + n_{dk} + n_{dv})$ for the standard implementation, where $n$ is the sequence length, $dk$ is the key dimensionality and $dv$ the value dimensionality, and is verified numerically against PyTorch at full double-precision floating-point on concrete inputs. Unlike hardware-specific accelerators or empirical tiling schemes such as FlashAttention, MoA simultaneously provides array fusion, shape-transformation correctness, and predictive cost models from a single algebraic framework. Memory minimality is a theorem established before any code is written. A predictive performance model projects $2$--$100\times$ speedup and $2$--$50\times$ energy reduction, with the advantage widening at exascale. The derivation establishes a formally verified pipeline from Python specification through (ONF) Operational Normal Form, and dimension-lifted hardware mapping, providing performance-portable AI kernels of direct relevance to DARPA edge-deployment and DOE exascale priorities.

AI 总结：提出Still方法，通过单次前向传递的轻量级Perceiver层实现KV缓存压缩，在8×至200×压缩比和8k至128k上下文长度下兼顾速度与质量，长上下文任务超越最强基线8-22分。

链接：https://arxiv.org/abs/2606.07878

机构：Baseten

作者：Charles O'Neill, Alex Sandomirsky, Harry Partridge, Mudith Jayasekara, Max Kirkby

英文摘要： The KV cache is the memory bottleneck of long-horizon language model deployment. Practically, a deployable compactor must be lightweight enough to call during inference, expressive enough to preserve context under constraint, and reusable across a trajectory. Existing compaction methods satisfy only part of this requirement: selection methods are lightweight but subset-bound, while synthesis methods are expressive but rely on per-context optimization. Here we introduce Still, a small per-layer Perceiver trained once against a frozen base model that produces compact keys and values in a single forward pass. On Qwen and Gemma models, Still occupies the favorable side of the speed--quality frontier across compression ratios from $8\times$ to $200\times$ and context lengths from $8$k to $128$k. On the long-context RULER grid, Still exceeds the strongest baseline by 8--22 points. The same compact cache also supports free-form summarization, preserving most of the full-context gain on HELMET and winning a pairwise LongBench summarization comparison against KV-Distill. Because compaction is a forward pass, Still can be applied iteratively, entering a long-horizon regime unavailable to per-context methods. We show that amortization makes long-context cache compaction tractable, and synthesis makes its compact state useful at extreme compression.

AI 总结：提出PartitionSel方法，通过划分拟阵约束下的梯度匹配效用最大化，实现跨域小批量选择，减少冗余并提升训练兼容性，在LLM微调中取得鲁棒性提升。

链接：https://arxiv.org/abs/2606.07954

作者：Prayas Agrawal, Prateek Chanda, Ishita Khatri, Ganesh Ramakrishnan, Bamdev Mishra, Pratik Jawanpuria

英文摘要：Training large language models (LLMs) on heterogeneous data requires selecting minibatches that balance convergence speed with coverage across domains. Existing methods either select samples independently within each domain or rely on computationally expensive proxy models to learn continuous domain weights. We propose PartitionSel, a cross-domain minibatch selection approach that maximizes a validation-guided gradient-matching utility under per-domain budgets encoded as a partition-matroid constraint. By coupling the per-domain budgets through a single utility, PartitionSel is designed to reduce redundancy in selections across domains. The proposed objective is weakly submodular and admits an orthogonal matching pursuit algorithm with provable approximation guarantees. Empirically, we evaluate PartitionSel for minibatch selection during the fine-tuning of Qwen2.5 and Llama-3 on MetaMathQA and Mol-Instructions. PartitionSel achieves robust gains over per-domain and domain-agnostic baselines on both benchmarks. It also reduces the number of conflicting gradient pairs within each batch, indicating that the cross-domain coupling translates into more compatible training updates.

AI 总结：提出STAR-KV框架，通过可微阈值机制实现注意力头和块级别的自适应秩选择，结合混合分解和低秩感知混合精度量化，在多种LLM上达到75%的KV缓存压缩，结合量化可减少20倍，并实现6.9倍注意力模块加速和3.1倍端到端生成吞吐提升。

链接：https://arxiv.org/abs/2606.08382

作者：Priyansh Bhatnagar, Ashkan Moradifirouzabadi, Se-Hyun Yang, SeungJae Lee, Jungwook Choi, Mingu Kang

英文摘要：Low-rank projection has emerged as a promising approach for compressing the KV cache by exploiting hidden-dimension redundancy. However, prior methods rely on fixed or heuristic rank selection and struggle to achieve aggressive compression with minimal accuracy degradation. We propose STAR-KV, an adaptive low-rank KV cache compression framework with fine-grained rank control. STAR-KV encompasses 1) a differentiable thresholding mechanism that enables optimal rank selection at both attention-head and block levels, 2) a hybrid decomposition strategy that applies different low-rank factorizations according to the sensitivity of key and value projections, and 3) a low-rank-aware mixed precision quantization that leverages data statistics for near lossless low-bit quantization. Evaluated across multiple LLMs and benchmarks, STAR-KV achieves up to 75% KV cache compression and up to 20x overall KV cache reduction when combined with quantization. Enabled by custom Triton-based GPU kernels, STAR-KV delivers up to 6.9x speedup for the attention module and 3.1x end-to-end generation throughput. Our code is publicly available at: this https URL.

AI 总结：针对RLVR中长上下文rollout计算昂贵的问题，提出Sparrow方法，通过动态稀疏度调度保持token级策略失配的下尾统计量稳定，在Qwen3系列模型上实现2.0-2.4倍加速，并推广到更大模型和编程领域。

链接：https://arxiv.org/abs/2606.08446

机构：Carnegie Mellon University（卡内基梅隆大学）； Cornell University（康奈尔大学）； Intel（英特尔）； Amazon AGI（亚马逊AGI）

作者：Yang Zhou, Ranajoy Sadhukhan, Zhaofeng Sun, Zhuoming Chen, Souvik Kundu, Saket Dingliwal, Sai Muralidhar Jayanthi, Aram Galstyan, Haizhong Zheng, Beidi Chen

英文摘要：Despite being powerful, reinforcement learning with verifiable rewards (RLVR) induces extremely long COT, making it computationally expensive. Since RLVR per-step cost is dominated by long-context rollout generation, sparse attention offers a promising way to accelerate dense rollout. However, sparse rollouts require a delicate stability-efficiency tradeoff: overly aggressive sparsity causes collapse, while overly lenient sparsity gives insufficient speedup. In this work, we study this tradeoff through sparse-to-dense actor-policy mismatch. We first observe that sparse rollout collapse is not driven by uniform degradation across tokens: most sparse tokens align perfectly with dense even under aggressive sparsity. Motivated by this, we hypothesize that sparse rollout training remains stable if the lower tail of per-token actor-policy mismatch stays above a critical threshold throughout the trajectory. We introduce a dynamic sparsity schedule that keeps this tail statistic constant during generation and validate our hypothesis. Across Qwen3 thinking-family models, keeping the tail mismatch statistic near a consistent threshold generally enables stable training. We then use a cost model to find the sparsity schedule for maximum speedup under this mismatch threshold, achieving 2.2x, 2.4x, and 2.0x rollout speedups when training Qwen3-1.7B, Qwen3-4B, and Qwen3-8B. Empirically, we show the thresholds generalize to a larger model (Qwen3-14B) and another RL domain (coding). Finally, our analysis naturally motivates DistillSparse: lightweight LoRA-based distillation on sparse rollout lets more aggressive sparsity reach the same sparse-to-dense mismatch threshold, yielding higher speedup.

AI 总结：提出EinSort方法，通过索引排序发现张量中的低秩结构，实现大语言模型权重和KV缓存的张量化压缩，相比基线方法提升了重构质量。

链接：https://arxiv.org/abs/2606.08565

作者：Toshiaki Koike-Akino, Jing Liu, Ye Wang

英文摘要：Tensor networks provide efficient representations for compressing large neural networks. By carefully designing shapes and topologies, they can significantly reduce memory and computational costs. However, identifying implicit low-rank structures in large foundation models remains challenging due to their enormous scale and un-structured weight distributions. We propose an adaptive tensorization method that discovers inherent low-rank structure in a target tensor by index ordering. Experiments on weight and KV-cache compression demonstrate improved reconstruction quality compared to baselines.

AI 总结：提出OrderDP框架，通过随机子集选取与top-q样本选择实现无偏梯度估计，提供收敛性和泛化性理论保证，在CIFAR和ImageNet上降低40%训练成本且保持精度。

链接：https://arxiv.org/abs/2606.08574

作者：Chenhan Jin, Shengze Xu, Qingsong Wang, Fan Jia, Dingshuo Chen, Tieyong Zeng

英文摘要：Data pruning (DP), as an oft-stated strategy to alleviate heavy training burdens, reduces the volume of training samples according to a well-defined pruning method while striving for near-lossless performance. However, existing approaches, which commonly select highly informative samples, can lead to biased gradient estimation compared to full-dataset training. Furthermore, the analysis of this bias and its impact on final performance remains ambiguous. To address these challenges, we propose OrderDP, a plug-and-play framework that aims to obtain stable, unbiased, and near-lossless training acceleration with theoretical guarantees. Specifically, OrderDP first randomly selects a subset and then chooses the top-$q$ samples, where unbiasedness is established with respect to a surrogate loss. This ensures that OrderDP conducts unbiased training in terms of the surrogate objective. We further establish convergence and generalization analyses, elucidating how OrderDP affects optimal performance and enables well-controlled acceleration while ensuring guaranteed final performance. Empirically, we evaluate OrderDP against comprehensive baselines on CIFAR-10, CIFAR-100, and ImageNet-1K, demonstrating competitive accuracy, stable convergence, and exact control -- all with a simpler design and faster runtime, while reducing training cost by over 40%. Delivering both strong performance and computational efficiency, our method serves as a robust and easily adaptable tool for data-efficient learning. The code is publicly available at this https URL.

AI 总结：本文在Loihi 2神经形态芯片上实现了卷积稀疏编码的局部竞争算法，并与GPU基线对比，展示了其在结构化稀疏推理中的可行性和优势。

链接：https://arxiv.org/abs/2606.08584

机构：Mercedes-Benz AG（梅赛德斯-奔驰集团）； Institut für Robotik und Kognitive Systeme, Universität zu Lübeck（吕贝克大学机器人与认知系统研究所）

作者：Geoffrey Kasenbacher, Daniel Ruepp, Gerrit A. Ecke

英文摘要：Sparse coding provides a principled framework for signal representation by expressing an input as a linear combination of only a small number of basis functions. The Locally Competitive Algorithm (LCA) is particularly attractive in the context of neuromorphic computing because its dynamics, leaky integration, thresholding, and lateral inhibition map naturally to neuromorphic hardware. While prior work has studied non-convolutional LCA on Loihi 2, the convolutional setting is of particular interest because it introduces spatial structure, weight sharing, overlapping receptive fields, and scaling behavior that are more representative of practical sparse inference workloads. In this work, we present a Loihi 2 implementation of convolutional sparse coding via the LCA and evaluate it against a conventional GPU baseline on the same inference problems. The implementation follows a one-layer recurrent LCA formulation and extends it to convolutional feature maps with local inhibitory kernels derived from pairwise filter interactions. To the best of our knowledge, this is the first implementation and benchmark of convolutional LCA on Loihi 2. Our goal is not only to demonstrate feasibility, but also to clarify in which operating regimes convolutional sparse inference becomes attractive on neuromorphic hardware. The resulting study positions convolutional LCA as a useful benchmark for structured sparse inference on emerging neuromorphic systems.

AI 总结：针对预填充-解码分离架构中KV缓存传输开销大的问题，提出SpectrumKV，通过为每个令牌分配不同精度（FP16/INT8/INT4）实现混合精度传输，并设计轻量部署探测自适应选择精度策略，在相同传输预算下显著提升模型质量并降低TTFT。

链接：https://arxiv.org/abs/2606.08635

作者：Yang Pengju

英文摘要： Prefill-decode (PD) disaggregation decouples prompt processing from token generation, but it also turns the key-value (KV) cache into a network payload. Existing PD-side KV reduction methods are mostly binary: selected tokens are transmitted at full precision and the rest are not transmitted. This paper argues that binary selection leaves a useful design space unused. SpectrumKV assigns a precision level to each token instead: attention sinks and other high-importance tokens are protected at FP16, medium-importance tokens are sent at INT8, and low-importance tokens are sent at INT4 when the model can tolerate it. The main practical complication is that INT4 tolerance is model-dependent. Qwen2.5-7B catastrophically fails under INT4 KV quantization, while Mistral-7B and Gemma-2-9B remain stable. SpectrumKV therefore runs a lightweight deployment-time probe: three aggressive NIAH trials under a 3-tier policy. Models that pass use FP16+INT8+INT4; models that fail fall back to FP16+INT8. Across Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3, and Gemma-2-9B-it, SpectrumKV improves quality at the same transfer budget. At a 50% normalized KV budget on WikiText-2, SpectrumKV changes perplexity by +1.97%,-0.06%, and-0.44%, respectively, compared with PDTrim's +25.85%, +22.07%, and +35.63%. On NIAH retrieval at 4096 tokens, the adaptive policy reaches 52.6% on Qwen at the aggressive b=0.3 budget versus 26.3% for PDTrim, and reaches 100% by b=0.5; Mistral and Gemma preserve retrieval under the 3-tier policy. End-to-end GPU timing of the transfer path shows 50-62% TTFT reductions at b=0.5. These results suggest that PD KV transfer should be treated as a precision-allocation problem, not only as token pruning.

AI 总结：提出C$^3$ache方法，通过跨推理块缓存和重用去噪残差，加速世界动作模型推理，实现高达2.5倍加速且任务成功率几乎无损。

链接：https://arxiv.org/abs/2606.08962

机构：George Mason University（乔治梅森大学）； University of Central Florida（中佛罗里达大学）

作者：Weisen Zhao, Lam Nguyen, Zhicong Lu, Yuzhang Shang

英文摘要：World Action Models (WAMs) generalize better than standard Vision-Language-Action (VLA) policies to novel motions and environments, because a video-modeling objective lets them learn from abundant unlabeled video rather than scarce labeled robot demonstrations. This generalization is computationally expensive. To complete a task, a WAM runs over multiple inference chunks, and each chunk requires a costly denoising process. Existing acceleration methods reduce this cost by caching and reusing computation within a single chunk's denoising trajectory. Our empirical analysis reveals a substantial source of redundancy they overlook: redundancy across chunks. When a robot executes a smooth behavior, the residuals computed at a given denoising step are strongly correlated from one chunk to the next. We introduce C$^3$ache, a training-free method that caches and reuses these residuals across inference chunks at the same denoising step. Experiments on benchmarks with a Fast-WAM backbone show that C$^3$ache achieves up to a $2.5\times$ speedup in total wall-clock inference time, with negligible degradation in task success rate.

AI 总结：提出统一几何框架解释后训练量化失败与量化感知训练恢复机制，揭示量化感知训练通过梯度感知谷壁使量化点返回低损失盆地。

链接：https://arxiv.org/abs/2606.09012

机构：University of California, Berkeley（加州大学伯克利分校）； University of Pennsylvania（宾夕法尼亚大学）

作者：Hanyang Li, Jianhao Ma, Ying Cui

英文摘要：Post-training quantization (PTQ) converts a trained full-precision model into low-bit weights without task-level retraining, while quantization-aware training (QAT) incorporates quantization into the training loop. Although PTQ is efficient and often accurate at moderate bitwidths, it can fail sharply at aggressive bitwidths; QAT is more expensive but can often recover the lost accuracy. We propose a unified geometric framework that explains both PTQ failure and QAT recovery. We model full-precision training as following a low-loss \emph{river} inside a wider \emph{valley}: a normal neighborhood of the river forms a nearly flat \emph{basin}, while leaving this basin incurs a sharp loss increase. When the quantization grid is comparable to the basin width, local PTQ objectives, including rounding and Hessian-based second-order reconstruction, can select a high-loss deployed quantized point outside the basin even when nearby low-loss quantized points exist. In this regime, straight-through-estimator-based QAT has a useful bias: it evaluates gradients at the deployed quantized weights while updating latent full-precision weights, causing the gradient to sense the valley wall and acquire an inward component that steers subsequent quantized iterates back into the basin. We formalize this mechanism through a local landscape model, construct a geometric PTQ failure mode, and prove finite-time QAT recovery under local quantizer-compatibility assumptions. Experiments across vision and language models under multiple neural-network quantization schemes corroborate the predicted basin-crossing failure of PTQ and the corresponding recovery mechanism of QAT.

AI 总结：提出前瞻稀疏注意力（LSA），基于DeepSeek-V4架构的神经记忆索引器，通过预测未来上下文需求仅保留关键KV块，在超长上下文场景下将物理KV缓存压缩至全上下文的13.5%，同时保持或略微提升下游准确率。

链接：https://arxiv.org/abs/2606.09079

机构：Independent Researchers（独立研究者）； Tencent（腾讯）； The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））； Tsinghua University（清华大学）

作者：Yan Wang, Qifan Zhang, Jiachen Yu, Tian Liang, Dongyang Ma, Xiang Hu, Zibo Lin, Chunyang Li, Zhichao Wang, Jia Li, Yujiu Yang, Haitao Mi, Dong Yu

英文摘要： Conventional LLMs keep the full KV cache loaded during decoding, causing a severe GPU memory bottleneck for ultra-long context serving. In this report, we propose Lookahead Sparse Attention (LSA), a novel inference paradigm powered by a Neural Memory Indexer built upon the DeepSeek-V4 architecture. Rather than passively attending to all historical tokens, LSA proactively predicts future context demands and preserves only the query-critical KV chunks in the GPU memory. Crucially, we instantiate this architecture via a backbone-free decoupled training strategy. By formulating the indexer as a standard dual-encoder architecture, we train it independently using standard retrieval training frameworks without ever loading the massive backbone model into GPU memory. We demonstrate that this "less is more" paradigm significantly maximizes serving efficiency while acting as an effective attention denoiser in tasks that rely on long-term global memory. Across primary long-context evaluation suites (e.g., LongBench-v2, LongMemEval, and RULER), FM-DS-V4 compresses the average physical KV cache footprint down to merely 13.5% of the full-context baseline, while consistently preserving or slightly elevating downstream accuracy (+0.6% absolute margin on average). Crucially, at extreme 500K scales, FlashMemory suppresses the physical KV cache overhead by over 90% without destabilizing the backbone's core reasoning capacities.

AI 总结：提出CANS框架，利用FedLinUCB-DW算法让异构设备自适应学习最优DNN分区，通过共享在线推理反馈和离线经验加速多用户边缘协同推理，显著降低延迟。

链接：https://arxiv.org/abs/2606.09175

机构：Harbin Institute of Technology Shenzhen（哈尔滨工业大学（深圳））； Politecnico di Milano（米兰理工大学）； Harbin Institute of Technology（哈尔滨工业大学）； Fudan University（复旦大学）； Shanghai Academy of Artificial Intelligence for Science（上海人工智能科学研究院）

作者：Zheshun Wu, Ziyang Zhang, Changyao Lin, Zenglin Xu, Jie Liu

英文摘要：Recently, mobile edge computing (MEC)-enabled collaborative deep neural network (DNN) inference has emerged as a promising approach for delivering intelligent services to resource-constrained mobile devices. A representative scenario is multi-user collaborative edge inference, where distinct devices independently partition their DNN models and offload backend computation to a common edge server over wireless networks. However, determining the optimal DNN partition for each device is challenging due to unknown and time-varying system conditions, including fluctuating wireless links and diverse device capabilities. To address this problem, we propose Cooperative Autodidactic NeuroSurgeon (CANS), a collaborative edge inference framework that enables devices to adaptively learn optimal DNN partitions by sharing informative feedback during online inference. To handle the challenge of device heterogeneity and better leverage offline inference experience, we integrate a novel FedLinUCB-DW algorithm that groups devices of the same type and warm-starts online exploration using local offline early-exit inference experience. Furthermore, we provide theoretical guarantees for FedLinUCB-DW by deriving the regret upper bound. We also validate our method on both a simulated environment and a hardware prototype system. Empirical evaluations demonstrate that CANS achieves lower inference latency compared to state-of-the-art baselines. Especially, in prototype experiments on two edge devices, the proposed CANS reduced average inference latency by up to 50% compared to the non-cooperative baseline.

AI 总结：提出一种受世界模型启发的评估器，通过轻量级过渡模型在连续潜在空间中展开调度动作，避免昂贵AST变异和重复编码，在TVM AutoScheduler中实现比Ansor更优的延迟和测量效率。

链接：https://arxiv.org/abs/2606.09312

机构：Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences（中国科学院大学杭州高等研究院）； Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所）； University of Chinese Academy of Sciences（中国科学院大学）

作者：Haolin Pan, Lianghong Huang, Xvlin Zhou, Mingjie Xing, Yanjun Wu

英文摘要：Tensor program optimization is essential for modern machine learning systems, but its search space is enormous. Existing auto-schedulers reduce measurement cost with learned cost models, yet they usually evaluate each candidate as a static code snapshot, ignoring the schedule trajectory that produced it. This makes them insensitive to action dependencies and vulnerable to superficial code variations. We propose a \emph{world-model-inspired} evaluator that models schedule evaluation as action-conditioned latent dynamics over program states. Starting from the initial program, it rolls out scheduling actions in a continuous latent space with a lightweight transition model, avoiding expensive AST mutation and repeated code encoding. The final dynamic representation is combined with action and hardware features to rank candidates. Implemented in TVM AutoScheduler, our method improves representative-subgraph latency over Ansor by 1.37$\times$ on GPU and 1.54$\times$ on CPU under the same 64-trial budget. It also matches Ansor-10K within 2.2% geometric mean using 10$\times$ fewer measurements, and accelerates full-model inference over PyTorch/PyTorch-opt(cuDNN) by 4.61$\times$/3.67$\times$ geometric mean.

AI 总结：针对神经网络形式化验证中GPU内存瓶颈，将张量并行（TP）和全分片数据并行（FSDP）适配到auto_LiRPA/α,β-CROWN框架，TP实现约2倍峰值内存降低但边界紧度下降，FSDP实现80-90%基础内存降低且边界与单GPU逐位一致，并支持完整验证和卷积层。

链接：https://arxiv.org/abs/2606.09377

机构：Lomonosov Moscow State University（莫斯科国立大学）； Central University（中央大学）

作者：Sergei Vorobyov, Eugene Ilyushin

英文摘要：Formal neural network verification -- proving that a network satisfies safety properties for \emph{all} inputs in a specified domain -- is bounded in practice by GPU memory: standard implementations of bound-propagation algorithms (IBP, CROWN, $\alpha$-CROWN) require weight and relaxation-coefficient matrices to reside entirely on one accelerator. We adapt two parallelism techniques originally developed for large-scale model training to the \texttt{auto\_LiRPA}\,/\,$\alpha,\beta$-CROWN verification framework. \textbf{Tensor Parallelism (TP)} shards both weight and $A$-matrices across GPUs, achieving ${\approx}2\times$ peak-memory reduction at $P{=}2$; soundness is confirmed on VNN-COMP 2022 MNIST-FC benchmarks, though bound tightness degrades with the number of sharded zones due to forced IBP substitution for intermediate bounds inside sharded zones. \textbf{Fully Sharded Data Parallelism (FSDP)} shards only weight matrices with a per-layer \texttt{AllGather}, producing bounds that are \emph{bitwise identical} to the single-GPU baseline: baseline memory drops by 80--90\%, peak memory by 34--39\% on wide MLPs. FSDP integrates cleanly with complete verification ($\beta$-CROWN + Branch-and-Bound) and with convolutional layers (\texttt{BoundConv}); a complete \emph{unsat} result is obtained for CIFAR-100 ResNet-large (VNN-COMP 2024) under FSDP. Across all experiments the memory bottleneck in $\alpha$-CROWN+BaB mode proves to be per-neuron alpha tensors, not weight matrices, pointing to the key direction for future work.

AI 总结：针对资源受限设备上部署安全大语言模型（LLM）的挑战，提出基于软提示与蒸馏训练的安全对齐方法，在最小化额外计算开销的同时实现优越的安全-有用性权衡。

链接：https://arxiv.org/abs/2606.09388

机构：Qualcomm AI Research（高通人工智能研究院）

作者：Motasem Alfarra, Cristina Pinneri, Dana Kianfar, Mohammed Almousa, Christos Louizos

英文摘要：Deploying safe large language models (LLMs) on resource-constrained edge devices presents a critical challenge: while dual-model systems combining LLMs with guard models provide effective safety guarantees, their substantial memory and computational demands make them prohibitively expensive for on-device deployment. This paper presents a comprehensive study of parameter-efficient safety alignment methods for resource-constrained settings. Through systematic evaluation across multiple LLM architectures, training objectives, and parameter-efficient fine-tuning approaches, we identify that soft prompts combined with distillation-based training consistently outperform alternative methods. We introduce distillation frameworks based on total variation and KL divergence that effectively transfer safety behaviors from guard models into learned soft prompts. Our evaluations on various benchmarks demonstrate that this combination achieves superior safety-usefulness trade-offs compared to LoRA adapters, steering vectors, and direct optimization methods, while requiring minimal additional memory and compute at inference time. These findings establish soft prompt distillation as the preferred approach for safety alignment in on-device LLM deployment.

AI 总结：提出跨分词器在线策略蒸馏方法，通过精确的token映射算法使教师模型概率分布信号能跨不同分词器传播，显著提升计算效率。

链接：https://arxiv.org/abs/2606.09456

机构：The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））； Tencent（腾讯）； The Hong Kong University of Science and Technology（香港科技大学）

作者：Yifan Niu, Han Xiao, Dongyi Liu, Zelong Wang, Dihong Gong, Yasheng Wang, Jia Li

英文摘要：On-Policy Distillation (OPD) has become a core technique in the post-training of Large Language Models (LLMs) for transferring knowledge from domain experts to student models. However, existing OPD distillation methods require teacher and student models to share the same tokenizer, restricting the applicability of OPD within the model series. Current mainstream practice typically employs Supervised Fine-Tuning (SFT) on teacher-generated responses for cross-tokenizer distillation, which fails to capture the rich knowledge embedded in the teacher's probability distribution. In this work, we enable the standard on-policy distillation method to operate across model families, ensuring that high-fidelity token-level signals can propagate across different tokenizers with a precise token-mapping algorithm. Extensive experiments show that cross-tokenizer OPD is significantly more compute-efficient than baselines on various benchmarks. Our results unlock a broader range of teacher-student pairs for OPD, opening up new avenues for adapting and enhancing interactions between LLMs.

AI 总结：针对在线策略蒸馏中学生陷入低KL一致陷阱导致训练信号弱的问题，提出KAT动态终止规则，过滤弱监督，在数学基准上提升avg@k 2.66%和pass@k 3.43%，同时减少59.73%的rollout长度。

链接：https://arxiv.org/abs/2606.09471

机构：The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））； The Hong Kong University of Science and Technology（香港科技大学）； The Hong Kong Polytechnic University（香港理工大学）； Eastern Institute of Technology, Ningbo（宁波东方理工大学）

作者：Haoran Xin, Anhao Zhao, Ying Sun, Jin Li, Xiaoyu Shen, Hui Xiong

英文摘要：On-policy distillation (OPD) provides dense token-level supervision by asking a teacher to score student-generated rollouts. However, when the student drifts into an unrecoverable prefix, the teacher may locally agree with the degraded state, producing low reverse KL but little corrective training signal. We identify this persistent regime as a low-KL agreement trap. Further analyses show that tokens during and after such traps produce less useful supervision signals. We propose KAT (KL Agreement Trap Termination), an online OPD termination rule that detects persistent low-KL agreement with a dynamic training-adaptive threshold. By filtering weak supervision from degenerate agreement, KAT improves avg@k accuracy by 2.66% and pass@k by 3.43% across four mathematical benchmarks, while reducing average rollout length by 59.73%.

AI 总结：提出BUDDY框架，通过轻量决策模块根据输入动态选择top-k层，并复用KV缓存支持解码时自适应路由，在严格预算控制下提升精度-计算权衡。

链接：https://arxiv.org/abs/2606.09514

作者：Yuhua Zhou, Shaoqi Yu, Shichao Weng, Changhai Zhou, Mingze Yin, Fei Yang, Aimin Pan

英文摘要： Large language models (LLMs) incur high inference cost due to their depth and parameter scale. Depth pruning can reduce latency by skipping redundant Transformer blocks, but existing methods (i) provide limited control under user-specific compute budgets and (ii) typically fix the routing path, failing to adapt as the context grows during decoding. We propose Buddy, a budget-driven dynamic depth routing framework. Buddy uses a lightweight Decision Module to score intermediate layers conditioned on the input and deterministically executes the top-k layers to satisfy a given budget. To support decode-time adaptation, Buddy reuses the first-layer KV cache as a low-overhead global context source and pools it together with the newest token representation before each routing decision. When no explicit budget is provided, an optional Budget Predictor estimates an input-dependent compute level to balance quality and efficiency. Experiments on Llama-family and Qwen models show that Buddy is competitive with strong static pruning baselines and often improves the accuracy-compute trade-off, while uniquely supporting strict budget control, decode-time rerouting, and multiple budgets within a single trained model.

AI 总结：提出AutoMegaKernel系统，将Llama模型编译为单个持久CUDA内核，通过静态调度验证器确保无死锁和无竞争，自动生成10种模型正确超内核，并在NVIDIA推理卡上以W8A16精度超越cuBLAS bf16。

链接：https://arxiv.org/abs/2606.09682

机构：RightNow AI

作者：Jaber Jaber, Osama Jaber

英文摘要：AutoMegaKernel (AMK) compiles a HuggingFace Llama-family model into a single persistent cooperative CUDA kernel that runs the whole forward pass in one launch, with no per-model hand-written CUDA. The contribution is the system, not raw speed. A frozen schedule-IR validator statically certifies deadlock-freedom and race-freedom via static graph checks (not a mechanized proof), so an unsafe agent-proposed schedule is rejected before launch: across 7,160 adversarial schedules (6,091 unsafe) it had zero false-accepts and accepted all 360 real lowerings. The same source retargets sm_80/sm_90/sm_120 from one codebase, auto-generates correct megakernels for 10 of 10 supported models, and on a real SmolLM2-135M checkpoint reproduces HuggingFace greedy decode token-for-token (perplexity match 2.5e-7). An unattended, agent-drivable autoresearch loop self-improves the megakernel over its own baseline (1.25-1.72x). A search-found int8 (W8A16) megakernel beats CUDA-graphed cuBLAS bf16 at batch-1 decode across NVIDIA's datacenter inference fleet: L4 up to 1.33x, the current-gen L40S 1.25-1.27x, A10G up to 1.08x at scale, and the consumer RTX 5090 1.19-1.23x. The ordering is not a clean function of bandwidth (the 864 GB/s L40S beats the 600 GB/s A10G); the divide is inference-class vs training-class. AMK trails cuBLAS on the high-bandwidth training-class A100/H100, where the harness localizes the cross-SM-sync bottleneck; we report the gap plainly. This is a precision-asymmetric (W8A16 vs bf16) comparison at decode position 0; the largest real checkpoint is TinyLlama-1.1B. Code and the harness: this https URL

AI 总结：提出BrainSurgery工具，通过声明式YAML计划实现神经网络检查点的鲁棒可复现张量操作，支持结构修改、数学变换和张量重塑，内置断言验证防止静默错误。

链接：https://arxiv.org/abs/2606.09707

机构：University of Southern Denmark（南丹麦大学）

作者：Gianluca Barmina, Annemette Broch Pirchert, Andrea Blasi Núñez, Lukas Galke Poech, Peter Schneider-Kamp

英文摘要：As deep learning models scale, managing, inspecting, and modifying large checkpoints has become increasingly challenging. Researchers often need to alter model weights for layer restructuring, precision casting, low-rank factorization, and architectural debugging, yet these workflows often rely on fragile ad-hoc Python scripts. Here, we introduce BrainSurgery, a tool for robust and reproducible "tensor surgery" on neural network checkpoints, and provide a system demonstration covering four examples and three case studies from model upcycling to LoRA extraction. By abstracting storage formats and memory management, BrainSurgery executes complex transformations through declarative YAML plans. It supports structural modifications, mathematical transformations, and tensor reshaping through expressive regex and structural targeting, while built-in assertions validate tensor shapes, data types, and values to prevent silent errors. We envision that BrainSurgery will provide a strong foundation for future research through its reproducible and validated operations.

AI 总结：提出HASA方法，根据客户端异构性分数分配子网宽度，在固定计算预算下提升平均和最差客户端准确率。

链接：https://arxiv.org/abs/2606.07621

机构：University of Naples "Federico II"（那不勒斯腓特烈二世大学）； Queen Mary University of London（伦敦玛丽女王大学）； Brandenburg University of Technology Cottbus-Senftenberg（勃兰登堡工业大学）； Tallinn University of Technology（塔林理工大学）； University of Zanjan（赞詹大学）

作者：Amir Hossein Shahdadian, Ahmed M. Abdelmoniem, Mahdi Taheri, Samira Nazari, Christian Herglotz

英文摘要： Edge services increasingly use federated learning to personalize on-device models while keeping sensitive data local. In practice, deployments must handle heterogeneity in both client resources and local data distributions. Model-heterogeneous federated learning lowers client cost by allowing each client to train a subnet of a shared supernet, but most subnet-allocation policies are driven by device constraints and do not explicitly account for statistical heterogeneity. This paper proposes Heterogeneity-Aware Subnet Allocation (HASA), a train-only rule that assigns subnet widths based on client heterogeneity scores computed from local training data while enforcing a fixed size-weighted compute budget. This design enables budget-matched comparisons with alternative allocation policies. On an article-title next-word prediction benchmark with seven clients, HASA improves unweighted mean client test accuracy over uniform allocation across 10 matched seeds, increasing mean client test accuracy from 13.82 percent to 14.32 percent, and improves worst-client accuracy on average. In a matched-budget comparison with representative partial-training baselines, HASA achieves the strongest worst-client and tail-client accuracy on this benchmark. A directionality ablation shows that assigning smaller subnets to more heterogeneous clients degrades both mean and tail performance. A cross-domain image-classification study further shows that the effectiveness of heterogeneity-aware allocation depends on how well the heterogeneity score reflects clients' need for additional model width.

AI 总结：针对联邦学习中客户端数据与系统异构性导致收敛慢、鲁棒性差的问题，提出代理辅助的进化客户端选择框架，将选择问题建模为组合优化，用代理模型加速进化搜索，实验表明收敛更快、能耗更低、鲁棒性更强。

链接：https://arxiv.org/abs/2606.07702

机构：Jiangnan University（江南大学）； The Hong Kong Polytechnic University（香港理工大学）

作者：Lin Qiang, Sun Xiaoyan, Hu Yao, Fang Wei

英文摘要：The heterogeneity of client data and systems makes it difficult to achieve satisfactory convergence speed and robustness in federated learning with random client selection. To address this issue, this paper proposes a surrogate-assisted client evolutionary selection framework for federated learning. In this framework, some typical client selection strategies are first used to generate candidate sets, and a metric function that integrates model performance, communication latency, and energy consumption is developed to formulate the client selection problem as a combinatorial optimization one. Subsequently, a surrogate model is constructed using the candidate selections and metric to efficiently approximate the performance of selected client subsets. An evolutionary algorithm is employed to search the combinatorial space of client selections, guided by the surrogate model to accelerate convergence. Experiments on MNIST, CIFAR10, CINIC10, and TinyImageNet demonstrate that the proposed algorithm achieves faster convergence, lower energy consumption, and improved robustness compared to existing methods.

AI 总结：针对纵向联邦学习中样本重建攻击的防御问题，提出基于因果表示学习的CausShield方法，将共享表示分解为任务相关与无关部分，实现全周期隐私保护，理论证明收敛性，实验优于七种最新方法。

链接：https://arxiv.org/abs/2606.08027

机构：Nanjing University of Science and Technology（南京理工大学）； University of Western Australia（西澳大学）； Hohai University（河海大学）； Nanjing University（南京大学）

作者：Yongqi Jiang, Yansong Gao, Siguang Chen, Anmin Fu

英文摘要：Vertical federated learning (VFL) is a distributed learning paradigm that leverages vertically partitioned features across isolated parties without sharing raw samples; however, it remains vulnerable to active sample reconstruction attacks. Existing defenses fail to achieve a satisfactory trade-off between model utility and privacy protection, due to either suppressing task-relevant information alongside privacy-sensitive features or relying on end-to-end supervised training to converge the defense module, which exposes the model to early-epoch vulnerability. To address this challenge, we adopt a structural causal model (SCM) insight and construct CausShield. From a task-learning standpoint, causal features within a raw sample are those that are directly relevant and contributory to the learning objective, whereas non-causal features are task-irrelevant but often encode sample-specific private information, thereby facilitating reconstruction. Importantly, we lay a theoretical foundation to prove this insight. CausShield thus decomposes the shared representations between the client and the coordinating server in VFL into task-relevant and task-irrelevant components to ensure full-cycle privacy protection. Nonetheless, the decomposition is inherently challenging due to the dual objectives of preserving model utility while mitigating privacy leakage. We address this via a carefully formulated optimization problem, which is solved through unsupervised representation learning. We further theoretically prove that CausShield preserves the convergence behavior of standard VFL. Extensive experiments compare CausShield against seven SOTAs, including InvL (USENIX Security'25), and evaluate robustness against advanced reconstruction attacks such as URVFL (NDSS'25). Results demonstrate that CausShield consistently outperforms in privacy protection, model utility, and computational efficiency.

AI 总结：提出物理一致零空间对齐（PCNSA）框架，通过伪零空间守恒预处理保持物理零空间与测量伪零空间的几何对应，从而检测低幅值但高影响的隐蔽虚假数据注入攻击。

链接：https://arxiv.org/abs/2606.08473

机构：Ben-Gurion-University（本-古里安大学）

作者：Xin Li, Chenhan Xiao, Jonathan Cohen, Aviad Elyashar, Yang Weng, Rami Puzis

英文摘要：False data injection attacks (FDIAs) introducing small measurement perturbations can still cause large deviations in power system state estimation when the injected signals align with the pseudo-null space of the system model. Existing model- and data-driven detectors may fail to identify such low-magnitude but high-impact attacks because residual tests ignore changes hidden in the pseudo-null space, while subspace learning methods capture correlation patterns without enforcing physical consistency. This paper proposes Physically Consistent Null Space Alignment (PCNSA), a framework that detects stealthy FDIAs by preserving, through preprocessing, the geometric correspondence between the physical null space and the measurement-derived pseudo-null space. The key point is a Pseudo-null Space Conserved data Preprocessing (PSCP) step that re-expresses measurements in the physical coordinate frame before subspace extraction. We prove that PSCP preserves the separation between row space and its orthogonal complement, a property that conventional per-feature standardization violates. This keeps the singular value decomposition (SVD)-derived pseudo-null subspace aligned with the physical residual space without explicit knowledge of H. Experiments on IEEE 14-, 30-, 57-, and 118-bus systems confirm this principle in practice: stealthy attacks that evade XTM, LSTM, AE and Isolation Forest baselines appear as clear deviations in the aligned subspace, yielding higher F1-score and detection accuracy while remaining robust under partial observability and realistic PMU noise.

AI 总结：针对联邦图学习中客户端级模态缺失问题，提出拓扑感知跨模态插补框架PRISM，通过联邦检索缺失模态语义并利用拓扑控制注入局部图传播，在六个多模态图数据集上平均提升4.48%。

链接：https://arxiv.org/abs/2606.09301

机构：Beijing Institute of Technology（北京理工大学）

作者：Zekai Chen, Miao Zhang, Jiayang Xing, Xunkai Li, Xun Wu, Rong-Hua Li, Guoren Wang

英文摘要：Multimodal federated graph learning (MM-FGL) aims to collaboratively learn from decentralized graphs with text and images. However, real-world clients may not share a common modality basis: a visual-search client may contain image--interaction graphs but no seller descriptions, while a catalog client may provide text but no product images. We refer to this practical setting as client-level modality deficiency. Unlike random instance-wise missingness, a deficient client lacks the local semantic basis needed to reconstruct the absent modality. More importantly, in graph learning, incomplete representations initialize message passing, so imputation errors can be filtered, mixed, and amplified by the receiving topology. To address this gap, we propose \textbf{PRISM} (\textbf{P}roactive \textbf{R}etrieval and \textbf{I}mputation via \textbf{S}tructural \textbf{M}eta-prompting), a topology-aware federated cross-modal imputation framework. Rather than reconstructing the missing modality solely from local observations, PRISM recovers missing-modality semantics from the federation and introduces them into local graph propagation under topology-aware control. Experiments on six multimodal graph datasets across graph-centric and modality-centric tasks show that PRISM consistently improves modality-deficient clients, outperforming state-of-the-art baselines by \textbf{4.48}\% on average.

AI 总结：通过系统变化适配数据分布，使用鲁棒成员推断和金丝雀数据提取攻击，评估差分隐私下大语言模型的实际隐私风险，发现分布偏移显著影响隐私脆弱性，LoRA等参数高效微调方法对分布外数据提供最佳实证保护。

链接：https://arxiv.org/abs/2606.09401

作者：Bartłomiej Marek, Lorenzo Rossi, Vincent Hanke, Xun Wang, Michael Backes, Franziska Boenisch, Adam Dziedzic

英文摘要：Recent work has applied differential privacy (DP) to adapt large language models (LLMs) for sensitive applications, offering theoretical guarantees. However, its practical effectiveness remains unclear, partly due to LLM pretraining, where overlaps and interdependencies with adaptation data can undermine privacy despite DP efforts. To analyze this issue in practice, we investigate privacy risks under DP adaptations in LLMs using state-of-the-art attacks such as robust membership inference and canary data extraction. We benchmark these risks by systematically varying the adaptation data distribution, from exact overlaps with pretraining data, through in-distribution (IID) cases, to entirely out-of-distribution (OOD) examples. Additionally, we evaluate how different adaptation methods and different privacy regimes impact the vulnerability. Our results show that distribution shifts strongly influence privacy vulnerability: the closer the adaptation data is to the pretraining distribution, the higher the practical privacy risk at the same theoretical guarantee, even without direct data overlap. We find that parameter-efficient fine-tuning methods, such as LoRA, achieve the highest empirical privacy protection for OOD data. Our benchmark identifies key factors for achieving practical privacy in DP LLM adaptation, providing actionable insights for deploying customized models in sensitive settings. Looking forward, we propose a structured framework for holistic privacy assessment beyond adaptation privacy, to identify and evaluate risks across the full pretrain-adapt pipeline of LLMs.

AI 总结：本文通过匹配强对手成员推理攻击的最坏情况成功度，提供从纯-DP ε到GDP μ的原则性映射，并推荐 μ≈ε/5 作为保守通用转换。

链接：https://arxiv.org/abs/2606.09582

作者：Bogdan Kulynych, Antti Honkela

英文摘要：Recent work argues for using Gaussian differential privacy (GDP) to report the privacy guarantees in privacy-preserving machine learning. We provide principled mappings from pure-DP $\varepsilon$ to GDP $\mu$ by matching the worst-case success of a strong-adversary membership inference attack in terms of three metrics: multiplicative advantage at fixed FPR, precision at fixed recall, and the standard privacy profile. We tabulate $\mu$ values across a useful range of parameters and recommend $\mu \approx \varepsilon/5$ as a conservative general-purpose conversion.

AI 总结：提出核契约框架，通过数值、统计、运行时和可观测性条款约束训练核与推理核之间的分布偏差，并推导偏差界以保障策略梯度无偏性。

链接：https://arxiv.org/abs/2606.07581

作者：Bruce Changlong Xu, Lan Wu

英文摘要： A modern post-training pipeline often writes one symbol for its policy, pi_theta, while evaluating it through two different programs: a training kernel optimized for autograd and an inference kernel optimized for low-precision, fused, dynamically batched serving. In finite precision, these kernels can induce different distributions at identical weights, with the gap concentrated on slices that aggregate benchmarks under-represent. This paper proposes kernel contracts: a contract-first framework for specifying acceptable divergence between K_train and K_inf. A contract C = (N, S, R, O, Pi) combines numerical, statistical, runtime, and observability clauses with an escalation policy from violations to routing actions. We derive a chain of bounds from logit drift to total-variation distance to bounded reward drift, and specialize it to RL post-training, where per-token importance-ratio drift yields a bound on policy-gradient bias under explicit support and norm assumptions. We also describe a four-stage promotion pipeline, online routing loop, and minimal YAML DSL for contract artifacts. This is a framework and vocabulary paper; we do not report production-scale empirical validation.

AI 总结：提出对微调权重更新进行SVD截断尾部，无需重训练或组标签即可减少虚假关联，在多个模型和基准上以<2%的准确率损失将差距降低最多5倍。

链接：https://arxiv.org/abs/2606.07596

机构：UCLA（加州大学洛杉矶分校）； Northeastern University（东北大学）

作者：Edward Sun, Dmitrii Troitskii

AI 总结：本文提出将序贯统计推断应用于大语言模型可信赖性，围绕表示、有效性和监控三个任务展开，将LLM交互视为依赖随机过程，提供不确定性保证并检测行为变化。

链接：https://arxiv.org/abs/2606.07624

机构：H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology（佐治亚理工学院工业与系统工程系）

作者：Yao Xie

AI 总结：提出利用激活空间中的性状方向监测监督微调中的涌现失调，通过低维几何特征实现高效检测，在7-9B模型上达到0.990 AUROC。

链接：https://arxiv.org/abs/2606.07631

机构：University of Maryland（马里兰大学）

作者：Huy Nghiem, Sy-Tuyen Ho, Sarah Wiegreffe, Hal Daumé III

AI 总结：研究激活引导在对抗性文本扰动下的鲁棒性，发现所有方法、模型和设置中方向鲁棒性下降高达64%，置信度崩溃，层选择脆弱，揭示其结构性脆弱性。

链接：https://arxiv.org/abs/2606.07696

机构：Independent Researcher（独立研究员）； Indiana University（印第安纳大学）

作者：Kien Le, Thai Le

AI 总结：研究多智能体LLM在协调博弈中面对拜占庭攻击和通信拓扑限制的脆弱性，发现智能体无法集体适应背叛，且显式限制拓扑会破坏合作，而隐式限制则不影响。

链接：https://arxiv.org/abs/2606.07790

机构：Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)（穆罕默德·本·扎耶德人工智能大学）

作者：Aya El Mir, Martin Takáč, Salem Lahlou

AI 总结：提出“应变连贯性”模式，即编码代理识别到问题但仍按原计划行动，通过构建Claude Sonnet 4.6检测器在44条轨迹上实现94%故障预测精度，优于基线方法。

链接：https://arxiv.org/abs/2606.07889

作者：Marut Pandya, Kasey Zhang, Baiqing Lyu

AI 总结：提出语义法定数保证（SQA），一种通过多样化验证者群体和风险自适应法定数谓词，将非确定性LLM代理的不安全操作批准率从18.5%降至0.3%的控制平面原语。

链接：https://arxiv.org/abs/2606.08021

机构：OpenKedge.io

作者：Jun He, Deying Yu

AI 总结：本文提出行为安全与干预鲁棒性之间的“审计差距”，通过构建解离模型和引入潜在脆弱性评分（LVS），证明行为安全指标不足以衡量表征层面的鲁棒性。

链接：https://arxiv.org/abs/2606.08044

机构：Stanford University（斯坦福大学）； University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）； Technical University of Denmark（丹麦技术大学）

作者：Enyi Jiang, Anders Gjølbye, Yibo Jacky Zhang, Sanmi Koyejo

AI 总结：提出Causal Agent Replay (CAR)方法，通过结构因果模型和干预操作，对LLM智能体失败步骤进行反事实归因，解决现有方法无法定位决策步骤的问题。

链接：https://arxiv.org/abs/2606.08275

机构：Carnegie Mellon University（卡内基梅隆大学）

作者：Jaineet Shah

AI 总结：提出一种干预前筛选框架，利用特征统计预测SAE引导的副作用（效果不稳定和附带扩散），在多个模型和字典上验证了解码器几何等信号优于基线，但预测效果因模型而异。

链接：https://arxiv.org/abs/2606.08365

机构：University of Michigan（密歇根大学）

作者：Evan Duan

AI 总结：提出统一图校准攻击（UGCA）框架，通过KL散度损失、重排序机制和混合损失等策略，在保持分类精度下显著提高期望校准误差，揭示高精度或多类模型更易受攻击。

链接：https://arxiv.org/abs/2606.08467

机构：Virginia Polytechnic Institute and State University（弗吉尼亚理工学院暨州立大学）； The Pennsylvania State University（宾夕法尼亚州立大学）； VinUniversity； University of Illinois at Chicago（伊利诺伊大学芝加哥分校）

作者：Cuong Dang, Jiahao Zhang, Hieu Ta Quang, Dung Le, Lu Cheng, Suhang Wang

AI 总结：提出一种联合有限样本证书，同时上界选择性风险、下界接受概率和部署效用，适用于自适应阈值选择，通过比率风险的经验伯恩斯坦界等方法，在ImageNet和COCO上比Hoeffding-CRC提升22个百分点接受前沿，且紧致约10倍。

链接：https://arxiv.org/abs/2606.08517

机构：Chongqing University of Posts and Telecommunications（重庆邮电大学）； Army Medical University (Third Military Medical University)（陆军军医大学（第三军医大学））

作者：Xiaoli Yu, Jiamiao Liu

AI 总结：针对数据稀缺下算子学习的不确定性量化，提出基于扰动的共形预测框架，在二维Navier-Stokes基准上比现有方法生成更窄的共形带，同时保持目标覆盖。

链接：https://arxiv.org/abs/2606.08654

机构： University of Oklahoma（俄克拉荷马大学）； Fudan University（复旦大学）

作者：Weinan Wang, Bowen Gang, Hao Deng

AI 总结：研究激活引导是否引发突现失调，通过扩展评估范围，发现激活引导可导致广泛失调，且比微调产生更连贯的有害响应，并分析了关键因素。

链接：https://arxiv.org/abs/2606.08682

机构：Nanyang Technological University（南洋理工大学）； Sun Yat-sen University（中山大学）； University of Science and Technology of China（中国科学技术大学）； National University of Singapore（新加坡国立大学）

作者：Qi Cao, Jian Lou, Meiting Liu, Wenjie Feng, Dan Li, See-Kiong Ng, Anh Tuan Luu

AI 总结：本文通过定义基于对数概率差异的因果影响度量，并利用电路发现技术，研究视觉语言模型幻觉输出的反事实鲁棒性，推导出检测不稳定所需的最小反事实样本数。

链接：https://arxiv.org/abs/2606.08777

作者：Abhivansh Gupta, Simardeep Singh, Advika Sinha, Shreyansh Modi, Akshat Tomar

AI 总结：针对AI在模糊任务上的长期扩散威胁，提出蓝队与红队对抗框架，通过弱模型评分训练强模型，并发现红队可利用多目标进化提示优化找到评分高但性能差的子版本行为，蓝队则通过对抗优化提升鲁棒性。

链接：https://arxiv.org/abs/2606.08892

机构：Anthropic Fellows Program (via MATS)（Anthropic 研究员计划（通过 MATS））； EPFL（洛桑联邦理工学院）； Redwood Research（红木研究）； Anthropic

作者：Mikhail Terekhov, Caglar Gulcehre, Vivek Hebbar, Joe Benton

AI 总结：提出用小Transformer编码器将轨迹映射到单位球面，使嵌入距离近似奖励与元数据的L1距离，线性探针检测奖励黑客，AUC达0.9467，成本比LLM-as-judge低四个数量级。

链接：https://arxiv.org/abs/2606.08893

机构：Tamarillo

作者：Iván Belenky, Joaquín Itria, Steven Johns

AI 总结：提出DynaCF框架，通过在线测量反事实扰动下的边际变化和偏好翻转来动态降低捷径敏感样本的权重，从而缓解奖励模型中的捷径学习问题。

链接：https://arxiv.org/abs/2606.09043

机构：The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））； New Jersey Institute of Technology（新泽西理工学院）； Institute of Computing Technology, CAS（中国科学院计算技术研究所）

作者：Fengyuan Liu, Yongliang Miao, Zirui He, Yanguang Liu, Fei Sun, Mengnan Du

AI 总结：本文提出使用分布奖励模型统一RLHF中的悲观主义方法，通过闭式有效奖励公式连接现有启发式方法，并揭示其隐含假设。

链接：https://arxiv.org/abs/2606.09073

作者：Ely Hahami, Yoel Zimmermann, Ray Zhou, Jack Benarroch Jedlicki

AI 总结：研究发现在基于RAG的LLM推荐中，安全训练会导致注入提示反而压制目标品牌推荐率，揭示了安全机制可能被逆向利用的风险。

链接：https://arxiv.org/abs/2606.09204

作者：Hyunseok Paeng

AI 总结：针对离线安全强化学习易受数据投毒攻击的问题，提出Safe-RULE框架，通过反学习移除恶意样本影响，无需从头训练或访问原始环境，实验证明能有效提升安全性。

链接：https://arxiv.org/abs/2606.09559

机构：University of Notre Dame（圣母大学）

作者：Shixiong Jiang, Taozheng Zhu, Fanxin Kong

AI 总结：提出将消息传递神经网络诱导的随机块模型映射到单位n-球面的拓扑框架，用于比较训练后的图神经网络，并实现无需重新训练的迁移学习候选检索。

链接：https://arxiv.org/abs/2606.07598

机构：KnotTheory.ai Inc.（KnotTheory.ai 公司）； Dept. of Mathematics, Emporia State University（恩波利亚州立大学数学系）

作者：Gopal Anantharaman

AI 总结：提出图神经网络框架，利用Cayley图等图表示，仅通过结构信息区分可解群与不可解群，探索图神经网络学习群论代数性质的能力。

链接：https://arxiv.org/abs/2606.07619

作者：Tal Weissblat

AI 总结：针对图神经网络可能泄露训练图邻接信息的问题，提出基于马尔可夫链近似的攻击方法MC-GRA(+)和防御方法MC-GPB(+)，在异质图上实现高保真重构攻击并有效防御。

链接：https://arxiv.org/abs/2606.08067

机构：Hong Kong Baptist University（香港浸会大学）； Shanghai Jiao Tong University（上海交通大学）； Stanford University（斯坦福大学）

作者：Zhanke Zhou, Bo Han, Xuan Li, Jiangchao Yao, Sanmi Koyejo, Michael K. Ng

AI 总结：提出网格图网络（MGN）预测任意孔洞几何2D结构的von Mises应力场，通过编码节点类型、相对边特征和全局特征实现平移和旋转不变性，在未见几何和载荷下R²≥0.97，优于传统模型。

链接：https://arxiv.org/abs/2606.08287

作者：Josiah D. Kunz, Kamal Choudhary

AI 总结：提出GeoGNN双塔架构，利用地理邻接图学习空间嵌入，结合时间序列表示，通过点积匹配实现时间序列地理定位，在电力消费数据集上平均提升约27%的定位精度。

链接：https://arxiv.org/abs/2606.08303

机构：Emory University（埃默里大学）； Oak Ridge National Laboratory（橡树岭国家实验室）； University of Southern California（南加州大学）

作者：Toan Tran, Waqwoya Abebe, Abhishek Potnis, Supriya Chinthavali, Cyrus Shahabi, Li Xiong, Dalton Lunga

AI 总结：本文提出图基础模型在动力学中需具备归纳跨网络泛化能力，通过仅基于合成多层网络训练的ts-net模型，在真实多层网络上实现零样本泛化，并优于传统方法。

链接：https://arxiv.org/abs/2606.08306

作者：Michał Czuba, Mateusz Stolarski, Adam Piróg, Piotr Bielak, Piotr Bródka

AI 总结：针对超图神经网络在异质性节点上性能下降的问题，提出异质性感知的自适应蒸馏方法HADES，通过量化节点异质性调节教师知识迁移，使学生模型性能超越教师并实现最高12.3倍加速。

链接：https://arxiv.org/abs/2606.08978

机构：Chung-Ang University（中央大学）； Chungbuk National University（忠北国立大学）

作者：Joohee Cho, David Yoon Suk Kang, Yunyong Ko

AI 总结：提出并行层次池化和反池化算子，构建首个超图U-Net架构，在分类、重构和异常检测任务上超越现有方法。

链接：https://arxiv.org/abs/2606.09051

机构：Institute for Financial Services Analytics, University of Delaware（特拉华大学金融服务分析研究所）； Department of Applied Economics and Statistics, University of Delaware（特拉华大学应用经济学与统计学系）； Department of Electrical and Computer Engineering, University of Kentucky（肯塔基大学电气与计算机工程系）； Department of Electrical and Computer Engineering, University of Delaware（特拉华大学电气与计算机工程系）

作者：Fuli Wang, Wei Qian, Daniel L. Lau, Gonzalo R. Arce

AI 总结：提出TL-HFD算法，通过局部活动区域和阈值化边界激活实现超图种子聚类的局部扩散，保证与全局更新等价并给出有限时间对偶次优性界。

链接：https://arxiv.org/abs/2606.09340

机构：KTH Royal Institute of Technology（瑞典皇家理工学院）； Johns Hopkins University（约翰霍普金斯大学）

作者：Meher Chaitanya, Sebastian Dalleiger, Luana Ruiz

AI 总结：提出Graph Mamba Operator (GraMO)，通过将状态空间模型与图交互学习集成到单一循环中，实现长期时空依赖的联合建模，在N体系统、运动捕捉和机器人数据集上取得最低误差。

链接：https://arxiv.org/abs/2606.09432

机构：Indian Institute of Science, Bangalore（印度科学研究所，班加罗尔）； Indian Institute of Technology, Delhi（印度理工学院，德里）

作者：Karn Tiwari, Niladri Dutta, N M Anoop Krishnan, Prathosh A P

AI 总结：提出MetaEvo两阶段框架，通过偏好优化增强模型从任务经验中抽象原则的能力，并在模块化架构中积累复用，持续提升推理性能。

链接：https://arxiv.org/abs/2606.07603

机构：School of Computer Science and Technology, Beijing Institute of Technology（北京理工大学计算机科学与技术学院）； Beijing Institute of Technology Southeast Academy of Information Technology（北京理工大学东南信息技术研究院）

作者：Bowen Ren, Heyan Huang, Yinghao Li, Yang Gao

AI 总结：提出用范畴论中的Kan扩展形式化迁移学习中的结构不变量，定义传递差异度量，并在链复形和持久模块中给出有限余核公式，通过瓶颈距离计算持久值不变量，实验验证了该方法能识别正确的任务函子并检测破坏迁移相关拓扑的表征坍塌。

链接：https://arxiv.org/abs/2606.07627

机构：Friedrich-Alexander Universität Erlangen-Nürnberg（埃尔朗根-纽伦堡大学）

作者：Luciano Melodia

AI 总结：提出记忆中心式LLM自适应方法，通过双轮廓条件算子与最小增益采样课程，解决上游记忆激活下游LLM的跨模型适应问题，在多项QA任务中优于基线。

链接：https://arxiv.org/abs/2606.07711

机构：Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学高瓴人工智能学院）； Weixin, Tencent（腾讯微信）； Institute for Artificial Intelligence, Peking University（北京大学人工智能研究院）

作者：Hao Yang, Shiqi Shen, Haoxuan Li, Zhipeng Wang, Zhi Gong, Xu Chen

AI 总结：研究任务粒度顺序对持续学习中灾难性遗忘的影响，通过CIFAR-100上的粗到细、细到粗和平坦三种训练策略，结合弹性权重巩固（EWC）方法，发现先学习一般类别可减少遗忘。

链接：https://arxiv.org/abs/2606.08013

机构：College of Engineering（工程学院）； College of Natural Science（自然科学学院）

作者：Emre Alyamac, Himanshu Janmeda, Shashwat Krishna, Yash Vijay

AI 总结：提出一种基于检索相似分割问题的进化学习方法，通过重用已有管道避免从头训练模型，降低开发成本，并分析跨域迁移的可行性。

链接：https://arxiv.org/abs/2606.08155

机构：University of Augsburg（奥格斯堡大学）

作者：Andreas Margraf, Henning Cui, Jörg Hähner

AI 总结：提出受睡眠启发的无监督回放机制，在多个新任务顺序训练后应用，以部分恢复所有先前学习任务的性能，防止灾难性遗忘。

链接：https://arxiv.org/abs/2606.08447

作者：Anthony Bazhenov, Jean Erik Delanois, Giri P. Krishnan

AI 总结：提出COLD框架，利用漂移加惩罚原理调节稳定性-可塑性权衡，通过虚拟队列控制遗忘，理论保证收敛性，实验优于现有方法。

链接：https://arxiv.org/abs/2606.08452

机构：IIIT Delhi（德里印度理工学院）； IIT Dharwad（达尔瓦德印度理工学院）

作者：Nazreen Shah, Govinda Arya, Bharath B.N., Ranjitha Prasad

AI 总结：提出ProjectionTL框架，通过分层贝叶斯建模与自适应投影实现源选择与特征选择，缓解负迁移，提升跨域学习的准确性、稳定性和可解释性。

链接：https://arxiv.org/abs/2606.08691

机构：Vanderbilt University Medical Center（范德比尔特大学医学中心）； Columbia University（哥伦比亚大学）

作者：Samhita Pal, Tian Gu

AI 总结：提出INFUSER框架，通过生成器与求解器的协同进化，利用影响力分数和DuGRPO优化，从文档池中自适应生成训练数据，显著提升模型推理性能。

链接： https://arxiv.org/abs/2606.09052

作者：Siyu Chen, Miao Lu, Beining Wu, Heejune Sheen, Fengzhuo Zhang, Shuangning Li, Zhiyuan Li, Jose Blanchet, Tianhao Wang, Zhuoran Yang

AI 总结：提出LargeMonitor框架，利用大型预训练模型（LVM和LMM）解耦检测与诊断，实现无任务持续学习中的零样本漂移检测和语义病因诊断，提升现有算法性能。

链接：https://arxiv.org/abs/2606.09430

机构：HKU（香港大学）； Qicore Tech（启科科技）

作者：Mingqi Yuan, Xiaoquan Sun, Shihao Luo, Jiayu Chen

AI 总结：本文通过动态等距机制保持深度神经网络在持续学习中的可塑性，提出等距正则化方法和AdamO优化器，在多个基准上匹配或超越现有方法。

链接：https://arxiv.org/abs/2606.09762

作者：Andries Rosseau, Robert Müller, Ann Nowé

AI 总结：提出RL4F基准，基于DIII-D托卡马克历史数据构建评估环境，比较多种离线RL方法在等离子体控制任务上的性能，发现基于模型的离线RL方法平均表现最佳。

链接：https://arxiv.org/abs/2606.07550

机构：Central South University（中南大学）； Chongqing University（重庆大学）； Carnegie Mellon University（卡内基梅隆大学）； The University of Hong Kong（香港大学）

作者：Yang Fu, Haomin Bao, Rohit Sonker, Xiaoyan Hu, Aravind Venugopal, Jeff Schneider, Jiayu Chen

AI 总结：提出基于Transformer的医疗推荐系统MedicalRec，利用从3000篇论文中构建的MedicalRec-Bench数据集（含5000+记录），无需重新训练即可为医疗图像分类任务推荐最优模型，最高HitRate@100达75.5%。

链接：https://arxiv.org/abs/2606.07553

作者：Roghayeh Taghavi, Aysa Hasanazde Bashkandi, Amir Ali Bengari, Mohammad Amin Raji, Mohammad Salahi Ardekani, Parisa Mardukhian, Parvaneh Rezaei, Ramin Mousa

AI 总结：研究发现多种LLM路由方法存在“路由平台”现象，即准确性趋同且远低于理想路由器，主要原因是可预测性瓶颈；通过增大训练数据、更强编码器和端到端微调可突破平台。

链接：https://arxiv.org/abs/2606.07587

机构：Rice University（莱斯大学）； Amazon（亚马逊）

作者：Yifan Lu, Qiyue Zhang, Shenrun Zhang, Zhibo Yu, Zhuang Wang, Hanjie Chen, Jiarong Xing

AI 总结：提出ResearchClawBench基准，包含10个领域40个任务，通过多模态评分标准评估自主科研能力，最强智能体仅得21.5分，揭示当前系统在实验协议、证据匹配和科学核心方面的不足。

链接：https://arxiv.org/abs/2606.07591

机构：Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）

作者： Wanghan Xu, Shuo Li, Tianlin Ye, Qinglong Cao, Yixin Chen, Hengjian Gao, Yiheng Wang, Qi Li, Kun Li, Sheng Xu, Shengdu Chai, Fangchen Yu, Xiangyu Zhao, Zhangrui Zhao, Weijie Ma, Zijie Guo, Haoyu Zhou, Haoxiang Yin, Lixue Cheng, Chaofan Hu, Haoxuan Li, Lu Mi, Xuxuan Xie, Yifan Zhou, Ruizhe Chen, Zhiwang Zhou, Xingjian Guo, Yuhao Zhou, Xuming He, Shengyuan Xu, Xinyu Gu, Jiamin Wu, Mianxin Liu, Chunfeng Song, Fenghua Ling, Dongzhan Zhou, Shixiang Tang, Yuqiang Li, Mao Su, Peng Ye, Siqi Sun, Bin Wang, Xue Yang, Zhenfei Yin, Tianfan Fu, Guangtao Zhai, Wanli Ouyang, Bo Zhang, Lei Bai, Wenlong Zhang

AI 总结：针对预训练数据混合中因高质量数据重复率变化导致的小规模实验外推失败问题，提出重复控制子采样方法，在1/16目标token预算下实现接近最优混合，揭示了重复动态而非规模决定实验泛化性。

链接：https://arxiv.org/abs/2606.07597

机构：Imperial College London（帝国理工学院）； Cohere

作者：Kevin Zhou, Lisa Alazraki, Kris Cao, Marek Rei

AI 总结：本文通过转录因子结合基准测试，揭示不同可解释性方法常产生矛盾解释、无法定位已知调控基序且不能忠实反映模型决策，主张采用类似临床试验的系统验证框架。

链接：https://arxiv.org/abs/2606.07607

作者：Shasha Zhou, Mingyu Huang, Ke Li

AI 总结：提出项目反应缩放定律（IRSL），将项目反应理论融入缩放定律框架，通过Beta-IRT模型利用语言模型的概率响应，将参数复杂度从O(M×N)降至O(M+N)，在预训练和测试时缩放场景中仅用50个问题即可实现可靠估计。

链接：https://arxiv.org/abs/2606.07616

作者：Sang Truong, Yuheng Tu, Rylan Schaeffer, Sanmi Koyejo

AI 总结：针对现实数据中的类别不平衡和噪声标注问题，提出一种利用基础模型先验的主动学习框架，通过不平衡感知的协同决策选择信息量最大的样本，在图像和文本数据集上实现超过50%的标注节省。

链接：https://arxiv.org/abs/2606.07630

机构：University of California, Riverside（加州大学河滨分校）； Carnegie Mellon University（卡内基梅隆大学）； Worcester Polytechnic Institute（伍斯特理工学院）

作者：Jiancheng Zhang, Meiqing Li, Qi Zhang, Yinglun Zhu

AI 总结：本文提出应用生命周期评估方法全面核算AI系统从硬件制造到训练推理的全链条资源消耗与环境影响，以弥补传统单一训练或推理成本评估的不足。

链接：https://arxiv.org/abs/2606.07632

作者：Jared Fernandez, Clara Na, Yonatan Bisk, Constantine Samaras, Emma Strubell

AI 总结：提出层次主动区域剪枝（HARP），一种高效的基于训练的数据选择方法，通过层次结构和经验贝叶斯推断降低选择成本，同时保持下游对齐，在多个基准上优于最强基线最多8.9分，且训练样本减少约7倍。

链接：https://arxiv.org/abs/2606.07690

机构：Cornell University（康奈尔大学）； The Hong Kong University of Science and Technology（香港科技大学）

作者：Ning Wang, Zhengxin Zhang, Maosen Tang, Yitang Gao, Claire Cardie, Sainyam Galhotra

AI 总结：提出SySRs算法，通过配对比较和自适应分配评估预算，利用模型相似性降低LLM评估成本，在15个基准上平均错误率最低。

链接：https://arxiv.org/abs/2606.07726

机构：ETH Zurich（苏黎世联邦理工学院）； Centrale Supélec（中央理工-高等电力学院）； ENS de Cachan（卡尚高等师范学校）； MPI for Intelligent Systems, Tübingen（马克斯·普朗克智能系统研究所，图宾根）

作者：Zifan Lyu, Chahine Nejma, Tobias Wegel, Fanny Yang, Florian E. Dorner

AI 总结：提出一个包含漂移模拟、时序感知评估和超参数优化协议的基准测试框架，在7个真实数据集上评估14种漂移检测方法，揭示其优劣并建立基线性能。

链接：https://arxiv.org/abs/2606.07789

机构：University of Coimbra（科英布拉大学）； Victoria University of Wellington（惠灵顿维多利亚大学）； Commerzbank（德国商业银行）； University of Waikato（怀卡托大学）； AI Institute, University of Waikato（怀卡托大学人工智能研究所）

作者：Vitor Cerqueira, Heitor Murilo Gomes, Marco Heyden, Bernhard Pfahringer, Albert Bifet

AI 总结：提出仪器化数据作为观测数据和模板合成数据之外的第三种选择，每个数据点携带产生它的机制模型、显式不确定性及可执行的反事实族，通过V&V仪器化图像到模拟管道实现，支持因果干预。

链接：https://arxiv.org/abs/2606.07865

机构：University of the Witwatersrand（威特沃特斯兰德大学）

作者：Daniel N. Wilke

AI 总结：针对机器学习气候降尺度中高分辨率模拟资源有限的问题，提出通过时间分布采样而非连续块状采样来分配训练年份，以更好地捕捉强迫气候响应和内部变率，实验表明时间分布采样在固定预算下性能最优。

链接：https://arxiv.org/abs/2606.07898

机构：University of California, Los Angeles（加州大学洛杉矶分校）； University of Wyoming（怀俄明大学）

作者：Karandeep Singh, Stefan Rahimi, Chad W. Thackeray, Stephen Cropper, Alex Hall

AI 总结：提出Tab-PE算法，将Private Evolution框架扩展至表格数据，通过启发式算子迭代优化候选数据集，在保持差分隐私的同时高效处理高阶相关性，相比基线AIM分类准确率提升最高10%，速度提升28倍。

链接：https://arxiv.org/abs/2606.08259

机构：Microsoft（微软）

作者：Toan Tran, Arturs Backurs, Zinan Lin, Victor Reis, Li Xiong, Sergey Yekhanin

AI 总结：本文复现了Renold等人对1995-2020年美国航空公司利润周期的聚类实验，通过PCA和核PCA分析，发现六聚类分类在原始7维和3维PC空间中具有几何鲁棒性，并验证了数据的内在线性流形结构。

链接：https://arxiv.org/abs/2606.08322

机构：Swiss Federal Railways (SBB)（瑞士联邦铁路（SBB））； University of Berne（伯尔尼大学）

作者：Andreas Schlapbach

AI 总结：提出RiskNet，一个从多语言新闻构建的大规模AI风险事件数据集，通过结构化流水线进行事件识别、对齐和多维分类，支持AI安全、治理和风险分析研究。

链接：https://arxiv.org/abs/2606.08376

机构：Beijing University of Posts and Telecommunications（北京邮电大学）； Beijing Key Laboratory of Multimodal Data Intelligent Perception and Governance（多模态数据智能感知与治理北京市重点实验室）

作者：Leihan Zhang, Wecheng Ye, Xianlong Ma, Haochuan Liu, Yang Li, Qianyu Zhang, Jinliang Chen, Qiang Yan

AI 总结：提出PIPE-Cypher流水线，利用本地大模型从企业属性图自动生成平衡的NL-to-Cypher基准，通过模式分析、逆向查询约束生成和执行验证等步骤，实现可重复的基准构建。

链接：https://arxiv.org/abs/2606.08481

机构：Halıcıoğlu School of Data Science and Computing, University of California, San Diego（加利福尼亚大学圣迭戈分校哈勒乔卢数据科学与计算学院）； Independent Researcher（独立研究员）

作者：Suraj Ranganath, Anish Raghavendra

AI 总结：提出频谱审计框架，结合非周期/周期分解、相位保持傅里叶干预等，发现深度学习模型对非周期成分的依赖是任务依赖且架构通用的，在睡眠-觉醒分类中影响显著，临床异常检测中中等，运动想象中最小，并扩展到ECG。

链接：https://arxiv.org/abs/2606.08583

机构：Indian Knowledge Systems and Mental Health Applications (IKSMHA) Center, Indian Institute of Technology Mandi（印度理工学院曼迪分校印度知识体系与心理健康应用中心）； School of Computing and Electrical Engineering, Indian Institute of Technology Mandi（印度理工学院曼迪分校计算与电气工程学院）

作者：Jasmeet Singh Bindra, Siddharth Panwar, Shubhajit Roy Chowdhury

AI 总结：通过固定架构仅改变通道宽度（1.05K-40.26K参数），发现EEG去噪重建性能在3-6.5K参数时饱和，且重建度量不预测下游BCI效用，超紧凑模型（33-46KB）适用于边缘部署。

链接：https://arxiv.org/abs/2606.08594

机构：Indian Knowledge Systems and Mental Health Applications (IKSMHA) Center, Indian Institute of Technology Mandi（印度理工学院曼迪分校印度知识体系与心理健康应用中心）； School of Computing and Electrical Engineering, Indian Institute of Technology Mandi（印度理工学院曼迪分校计算与电气工程学院）

作者：Jasmeet Singh Bindra, Siddharth Panwar, Shubhajit Roy Chowdhury

AI 总结：针对深度主动学习中人工标注噪声导致性能下降的问题，提出一种通过分配部分标注预算重新标注已标注数据来去噪的框架，实验表明在相同预算下更高效且最终数据集噪声较少。

链接：https://arxiv.org/abs/2606.08718

作者：Md Abdullah Al Forhad, Weishi Shi

AI 总结：针对无源数据下精确满足声明性分析结果的需求，提出结果一致性合成任务，通过闭式条件伽马抽样实现精确聚合，并构建SpecBench基准，证明一致性保真度正交。

链接：https://arxiv.org/abs/2606.08736

机构：Independent Researcher（独立研究员）

作者：Muhammed Rasin

AI 总结：针对合成电子病历评估过度依赖统计相似性而忽视临床有效性的问题，提出基于流行病学的多维度评估框架，发现当前生成模型虽能复现边缘分布，但无法同时保持亚组结构、效应估计和依赖关系，导致评估高估数据质量。

链接：https://arxiv.org/abs/2606.08903

机构：Centre for Big Data Research in Health, the University of New South Wales（新南威尔士大学健康大数据研究中心）

作者：Nicholas I-Hsien Kuo, Blanca Gallego, Louisa Jorm

AI 总结：针对现有评估指标忽视预测锐度与流行偏差鲁棒性的问题，提出广义评估框架PROBE，通过排序变换器和排序聚合器实现更全面、灵活且一致的模型评估。

链接：https://arxiv.org/abs/2606.08921

机构：Chung-Ang University（中央大学）； Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）

作者：Sooho Moon, Jian Kang, Yunyong Ko

AI 总结：提出PROBE-Web交互系统，通过调整预测锐度和流行度偏差鲁棒性两个视角，灵活评估KGC模型，并提供四种关键功能。

链接：https://arxiv.org/abs/2606.08926

机构：Chung-Ang University（中央大学）

作者：Sooho Moon, Yunyong Ko

AI 总结：通过条件互信息检验，发现中间网格状态的结构属性可预测符号ARC-AGI求解器在相同任务内的成败，主要预测信息沿单一网格复杂度轴分布，且跨求解器架构泛化。

链接：https://arxiv.org/abs/2606.09026

作者：Ayan Pendharkar

AI 总结：提出Janus程序，通过诱饵校准和留出数据验证，判断语言模型错误解释的可信度，避免选择偏差。

链接：https://arxiv.org/abs/2606.09046

机构：Meta Platforms（Meta平台）

作者：Vyzantinos Repantis, Ameya Gawde, Harshvardhan Singh

AI 总结：提出一种基于GEMM维度的剪枝方法分类法，通过统一基准框架系统评估不同剪枝方法在加速-质量帕累托前沿上的表现，发现静态深度剪枝在低质量损失下最优，为LLM剪枝加速提供统一视角。

链接：https://arxiv.org/abs/2606.09080

机构：Ningbo Institute of Digital Twin, Eastern Institute of Technology, Ningbo（宁波数字孪生研究院，东方理工大学（宁波））； Department of Computing, The Hong Kong Polytechnic University（香港理工大学计算学系）； Munich Center for Machine Learning, LMU Munich（慕尼黑大学机器学习慕尼黑中心）

作者：Haozhe Hu, Hao Wu, Anhao Zhao, Longwei Ding, Peiran Yin, Yunpu Ma, Xiaoyu Shen

AI 总结：提出Orange Lab，一种基于Web的可视化数据分析环境，通过组件展示范式将机器学习工作流嵌入任意网页，实现动态交互与数据驱动叙事，降低数据科学使用门槛。

链接：https://arxiv.org/abs/2606.09239

作者：Matej Bevec, Aleš Erjavec, Vesna Tanko, Lena Trnovec, Lan Žagar, Ana Farič, Janez Demšar, Blaž Zupan

AI 总结：提出ERBench基准，通过方程恢复任务评估符号回归算法，强调在变化维度、采样大小、分布和域下的鲁棒性，填补现有基准的空白。

链接：https://arxiv.org/abs/2606.09276

机构：University of Jena（耶拿大学）

作者：Paul Kahlmeyer, Henrik Voigt, Michael Habeck, Joachim Giesen

AI 总结：本文指出当前概率电价预测中评分规则偏向锐度而忽视校准，导致过自信估计，呼吁未来研究转向校准感知的目标和架构。

链接：https://arxiv.org/abs/2606.09517

机构：Institute for Automation and Applied Informatics（自动化与应用信息学研究所）； Karlsruhe Institute of Technology（卡尔斯鲁厄理工学院）

作者：Jan Niklas Lettner, Hadeer El Ashhab, Benjamin Schäfer

AI 总结：针对条件生成在组合偏移下的评估难题，提出一种基于训练分布的后验信任分数，结合全局真实性和属性忠实度，实现样本过滤、排序和弃权，提升生成质量。

链接：https://arxiv.org/abs/2606.09601

机构：Institute of Science and Technology Austria (ISTA)（奥地利科学技术学院）； Pyramidal Inc（Pyramidal公司）； Achira Inc（Achira公司）

作者：Berker Demirel, Valentino Maiorca, Marco Fumero, Theofanis Karaletsos, Francesco Locatello

AI 总结：提出首个基于持久用户身份的交互式原生iOS模拟器基准iOSWorld，包含26个新应用和133个任务，评估代理在单应用、多应用及记忆个性化任务上的表现，最佳配置整体准确率52%，多应用任务仅37%。

链接：https://arxiv.org/abs/2606.09764

机构：Carnegie Mellon University（卡内基梅隆大学）

作者：Lawrence Keunho Jang, Mareks Woodside, Geronimo Carom, Andrew Keunwoo Jang, Jing Yu Koh, Ruslan Salakhutdinov

AI 总结：提出STARIXNet轻量神经网络，通过捕获多系统指标的时空关系进行多变量资源分配，优先服务稳定性再考虑成本效率，在沃尔玛生产环境中节省10%-50%成本。

链接：https://arxiv.org/abs/2606.07565

机构：Walmart Global Tech（沃尔玛全球科技）

作者：Ahmed Abdulaal, Maruf Aytekin, Thilaga kumaran Srinivasan, Tomer Lancewicki

AI 总结：提出MST-Direct扩展方法，通过稀疏Sinkhorn匹配器、多变量元组匹配和克里金条件化，实现大规模、多变量、条件地质统计模拟，精确保持联合分布。

链接：https://arxiv.org/abs/2606.07578

作者：Tcharlies Bachmann Schmitz

AI 总结：提出一种结合FT-Transformer与XGBoost的混合架构，通过校准感知堆叠集成处理类别不平衡和特征交互，在银行客户流失数据集上F1达62.10%，AUC-ROC为0.861。

链接：https://arxiv.org/abs/2606.07582

机构：Independent Researcher, Austin, TX, USA（独立研究员，美国德克萨斯州奥斯汀）； Independent Researcher, Leander, TX（独立研究员，美国德克萨斯州利安德）； Texas A & M University-Victoria, Victoria, TX（德克萨斯农工大学维多利亚分校）

作者：Joyjit Roy, Samaresh Kumar Singh, Laxmi Shaw

AI 总结：提出QDSP框架，集成配额引导子空间采样和可微决策结构感知，在极低出生体重婴儿队列中实现高精度死亡/脑瘫预测，并提供可解释的临床决策路径。

链接：https://arxiv.org/abs/2606.07606

机构：College of Computer Science, Sichuan Normal University（四川师范大学计算机科学学院）； West China Second University Hospital, Sichuan University（四川大学华西第二医院）

作者：Ling Wang, Xiaolong Li, Hui Zhou, Jing Shi, Fuhao Zhang, Dapeng Chen, Nan Mu

AI 总结：本文利用随机森林递归特征消除法分析尼日利亚调查数据，发现少量预测因子即可高精度识别贫困状态和不平等线位置，表明机器学习可优化调查设计并降低数据需求。

链接：https://arxiv.org/abs/2606.07614

机构：Cantabria University（坎塔布里亚大学）； SOAS University of London（伦敦大学亚非学院）； United Nations University World Institute for Development Economics Research (UNU-WIDER)（联合国大学世界发展经济学研究所）

作者：Vanesa Jordá, Miguel Niño-Zarazúa

AI 总结：提出基于Transformer的框架，利用历史队列长度、等待时间和旅客吞吐量数据，预测登机口和安检点未来两小时的队列长度与等待时间，支持主动排队管理。

链接：https://arxiv.org/abs/2606.07622

机构：Korea Aerospace University（韩国航空大学）； Korea Airports Corporation（韩国机场公社）

作者：Juhwan Lee, Seokbin Yoon, Keumjin Lee, Hojong Baik, Seyeon Jung

AI 总结：提出三模态假新闻检测框架KITE，联合建模文本、视觉和知识表示，利用跨模态注意力整合特征，在基准数据集上显著优于单双模态基线。

链接：https://arxiv.org/abs/2606.07651

作者：Kevin Patel, Shashi Bhushan Jha

AI 总结：针对物联网环境中MLaaS组合因动态性而失效的问题，提出一种测试时自适应（TTA）组合框架，通过TTA感知可组合性模型和服务级自适应模型，在推理时调整服务并保持组合性能，显著降低计算时间。

链接：https://arxiv.org/abs/2606.07685

作者：Deepak Kanneganti, Sajib Mistry, Sheik Mohammad Mostakim Fattah, Aneesh Krishna

AI 总结：提出一种知识包容的自适应PINN框架，通过整合文本和网络结构知识改进微生物群落建模，在真实和模拟数据集上性能提升最高53%。

链接：https://arxiv.org/abs/2606.07686

机构：University of Melbourne（墨尔本大学）； Academia Sinica（中央研究院）

作者：Ravisha Rupasinghe, Rajith Vidanaarachchi, Asela Hevapathige, Sachith Seneviratne, Sen-Lin Tang, Saman Halgamuge

AI 总结：提出首个环境机械生物信号基础模型BCG-FM，利用床垫压电传感器无感采集心冲击图，通过14.6万人的275万小时数据预训练，在生物年龄估计上达到3.26年MAE，并实现15种健康状态的临床相关判别。

链接：https://arxiv.org/abs/2606.07692

作者：Magnus Ruud Kjaer, Haejun Han, Ashish Neupane, David Q. Sun

AI 总结：针对船舶交通流数据高度稀疏且间歇性爆发的问题，提出一种模型无关的可学习Tweedie头作为即插即用输出模块，通过优化闭合形式的Tweedie单元偏差并预测均值，同时学习节点级方差幂以捕获港口区域异质性，在真实AIS数据上显著提升RMSE。

链接：https://arxiv.org/abs/2606.07694

机构：Korea Advanced Institute of Science and Technology (KAIST)（韩国科学技术院）

作者：Kyeongjun Lee, Heeyoung Kim

AI 总结：本研究通过整合PharmGKB的药理基因组学先验知识（CYP酶注释）作为特征向量，增强图神经网络在药物相互作用预测中的性能，在配对数据划分下显著提升DDI类型分类，但未能突破信息天花板。

链接：https://arxiv.org/abs/2606.07698

机构：AI Solutions Berlin

作者：Juergen Dietrich

AI 总结：提出基于图同构神经网络（GIN）的EssentialGIN模型，整合PPI网络拓扑与基因表达、直系同源、亚细胞定位等多源生物数据，在人类等复杂生物中显著优于现有方法。

链接：https://arxiv.org/abs/2606.07700

作者：Sahar Mansouri-Rad, Zahra Narimani, Parvin Razzaghi, Nazanin Hosseinkhan

AI 总结：提出FunctionEvolve框架，利用表达式树组织符号回归搜索，通过结构摘要、局部树编辑和结构感知系数拟合，在LLM-SRBench合成子集上以Claude Opus 4.6实现82.9%的SA@50，较同基线提升4.5倍。

链接：https://arxiv.org/abs/2606.07704

作者：Zeyu Xia, Jun Zhu, Dong Yan

AI 总结：提出多目标回归框架，利用LLM从自然叙事中提取连续情感特征，结合动态功能连接和机器学习算法，实现从fMRI数据中解码连续情感轨迹，并揭示可解释的情感特异性脑网络拓扑。

链接：https://arxiv.org/abs/2606.07707

机构：NTNU（挪威科技大学）； Kavli Institute for Systems Neuroscience, NTNU（挪威科技大学卡弗里系统神经科学研究所）； Microsoft（微软）

作者：Lemei Zhang, Peng Liu, Hans Dahle Kvadsheim, August Sætre Aasvær, Shuer Ye, Reza Bonyadi, Maryam Ziaei, Jon Atle Gulla

AI 总结：本研究通过可视化与几何分析，探究自杀意念检测模型内部如何编码心理风险因素，发现主题增强能提升低表征风险因素表示的清晰度与可解释性。

链接：https://arxiv.org/abs/2606.07714

机构：University of Ottawa（渥太华大学）； National Research Council Canada（加拿大国家研究委员会）

作者：Hamideh Ghanadian, Isar Nejadgholi, Hussein Al Osman

AI 总结：提出几何感知三平面场网络(GTF-Net)，通过双流骨干网络结合自适应傅里叶神经算子与CNN，实现车辆气动压力和壁面剪切应力的高效预测，在精度上超越现有方法。

链接：https://arxiv.org/abs/2606.07724

机构：Southern University of Science and Technology（南方科技大学）； Shenzhen Tenfong Technology Co., Ltd.（深圳腾风科技有限公司）； Eastern Institute of Technology（东方理工高等研究院）

作者：Kangkang Qi, Huiyu Yang, Keqi Ding, Yunpeng Wang, Yuntian Chen, Yuanwei Bin, Rikui Zhang, Jianchun Wang

AI 总结：针对高维含噪热扩散问题，提出物理信息神经网络（PINN）框架，在噪声和维度较高时显著优于有限差分法（FDM），实现精度与效率的权衡。

链接：https://arxiv.org/abs/2606.07982

机构：Kathmandu University（加德满都大学）

作者：Shreesh Bhattarai, Harish Chandra Bhandari

AI 总结：提出SafeECGMatch框架，通过双分支架构提取时频特征，结合自适应标签平滑和温度缩放校准模型，在标签分布不匹配下实现可靠的开集分类和OOD检测。

链接：https://arxiv.org/abs/2606.08037

机构：Hankuk University of Foreign Studies（韩国外国语大学）

作者：Hongkyu Koh, Ikbeom Jang

AI 总结：提出约束感知优化框架，结合平衡均方误差、孪生反对称正则化器和OOD边缘一致性损失，在不改变SPURS架构下提升蛋白质稳定性预测的鲁棒性，在多个基准上取得显著改进。

链接：https://arxiv.org/abs/2606.08100

机构：Birla Institute of Technology and Science Pilani, Hyderabad Campus（比拉理工学院海得拉巴校区）

作者：A Shivram, Aneesh S. Chivukula, Manik Gupta, Sourav Chowdhury

AI 总结：提出TRUST-SCF框架，利用Transformer对交易序列建模，通过金融对齐的注意力偏置、连续延迟预测和标签高效评分管道，实现动态信用评分，实验表明优于基线。

链接：https://arxiv.org/abs/2606.08140

机构：Department of Growth Barook Co.（Growth Barook公司）

作者：Mohammadamin Davoodabadi, Amirabbas Shakeri

AI 总结：提出LogNEO，利用GPT-Neo模型和基于位置感知奖励的PPO微调，在HDFS、BGL和Thunderbird基准上达到F1分数0.927、0.913和0.984，召回率比LogGPT提升6%，并在生产部署中实现45ms端到端延迟。

链接：https://arxiv.org/abs/2606.08153

作者：David Eje, Tanmay Sharma, Khush Patel, Manuel Mazzara, Leonard Johard

AI 总结：提出AttentionCap，一种定制化Transformer，结合Gram表示、对称注意力输出层和归一化拉普拉斯损失，实现多层多节点下的高精度电容矩阵预测，速度提升192倍。

链接：https://arxiv.org/abs/2606.08161

机构：Dept. Computer Science & Tech., BNRist, Tsinghua Univ., Beijing, China（清华大学计算机科学与技术系，北京信息科学与技术国家研究中心）； School of IC, BNRist, Tsinghua Univ., Beijing, China（清华大学集成电路学院，北京信息科学与技术国家研究中心）； School of IC, Peking Univ., Beijing, China（北京大学集成电路学院）

作者：Jiechen Huang, Hector R. Rodriguez, Dingcheng Yang, Zuochang Ye, Yibo Lin, Wenjian Yu

AI 总结：提出一个结合专家知识和迁移学习的半自动化平台，为非专家推荐完整的机器学习流水线，并自动提取数据特征，通过一阶逻辑推理提供排名算法。

链接： https://arxiv.org/abs/2606.08212

机构：LATECE Lab, Université du Québec à Montréal（LATECE实验室，魁北克大学蒙特利尔分校）

作者：Lokman Saleh, Hafedh Mili, Mounir Boukadoum

AI 总结：提出GPT-Micro范式，结合大语言模型、热力学约束和稀疏数据，实现自主发现本构模型，在印刷电子测试中数据量减少70%、发现时间缩短400倍。

链接：https://arxiv.org/abs/2606.08238

作者：Soumik Dutta, Kiarash Naghavi Khanghah, Sania Shree, Logan McNeil, Thomas Feldhausen, Hongyi Xu, Rajiv Malhotra

AI 总结：提出将自然语言查询转换为结构化图并通过确定性规划器执行的系统，利用深度优先搜索解决跨工具依赖，实现高可靠性查询。

链接：https://arxiv.org/abs/2606.08300

作者：Aishwarya Chakravarthy, Vidhi Kulkarni, Duen Horng Chau

AI 总结：提出利用Kolmogorov-Arnold网络从气压测量数据重建鸟类呼吸动力学方程，揭示隐藏的两相肌肉激活模式，并通过肌电图验证。

链接：https://arxiv.org/abs/2606.08479

作者：Julian Szereszewski, Facundo Fainstein, Leandro E. Fernandez, Gabriel B. Mindlin

AI 总结：针对生成式推荐中奖励模型因曝光偏差导致噪声的问题，提出AdaGRPO框架，通过策略难度和奖励可区分性诊断动态切换GRPO与监督学习，在电商数据集上提升召回率并抑制幻觉。

链接：https://arxiv.org/abs/2606.08480

机构：JD.com（京东）； Waseda University（早稻田大学）； University of Electronic Science and Technology of China（电子科技大学）

作者：Kewei Xu, Junbo Qi, Yanyan Zou, Pengfei Zhang, Xingzhi Yao, Shengjie Li

AI 总结：提出STELLAR框架，通过图-时间编码器、上下文锚定潜在对齐和不平衡感知解码模块，联合优化动态栖息地上下文和群落结构，有效解决物种分布建模中的时空耦合与长尾不平衡问题。

链接：https://arxiv.org/abs/2606.08484

机构：Sun Yat-sen University（中山大学）； Cornell University（康奈尔大学）； Foshan University（佛山大学）； Cornell Lab of Ornithology（康奈尔鸟类学实验室）

作者：Shufeng Kong, Tao Yu, Yuanyuan Wei, Caihua Liu, Junwen Bai, Yingheng Wang, Marc Grimson, Daniel Fink, Carla P. Gomes

AI 总结：利用Transformer分析癌症患者常规实验室检测的纵向轨迹，预测162种治疗相关并发症，性能优于单时间点方法，验证了轨迹数据对器官功能恶化的早期编码能力。

链接：https://arxiv.org/abs/2606.08538

机构：Technical University of Munich（慕尼黑工业大学）； Charité - Universitätsmedizin Berlin（柏林夏里特医学院）； German Heart Center（德国心脏中心）

作者：Jannik Lübberstedt, Krischan Braitsch, Jacqueline Lammert, Christof Winter, Florian Gabriel, Tristan Lemke, Christopher Zirn, Markus Graf, Friedrich Puttkammer, Hartmut Häntze, Johannes Moll, Anirudh Narayanan, Andrei Zhukov, Fabian Drexel, Zeineb Ben Chaaben, Sebastian Ziegelmayer, Su Hwan Kim, Marion Högner, Jan Kirschke, Florian Bassermann, Marcus Makowski, Christian Wachinger, Lisa Adams, Keno Bressem

AI 总结：针对三维水凝物预测中零膨胀长尾分布导致的过度平滑问题，提出物理引导的双解码框架PredHydro-Net，通过解耦架构、小波频率解耦和对抗训练，在极端事件检测和光谱表示上优于现有模型。

链接：https://arxiv.org/abs/2606.08563

机构：Chinese Academy of Meteorological Sciences（中国气象科学研究院）； Xiong’an Institute of Meteorological Artificial Intelligence（雄安气象人工智能研究院）

作者：Dandan Chen, Yaqiang Wang

AI 总结：提出Memory-as-a-Layer (MAL)适配器，利用测试时神经记忆为对话语音情感识别提供上下文，在不修改大型音频语言模型的前提下提升性能。

链接：https://arxiv.org/abs/2606.08573

作者：Daniel Chen, Qicong Hu, Yang Xiao, Ting Dang, Hong Jia

AI 总结：提出一种全局结构的量子神经网络，减少量子电路中酉矩阵数量，训练时间降低97%，训练完成率提升25%，实现100%训练成功率，纠错性能超越以往研究。

链接：https://arxiv.org/abs/2606.08592

机构：Meiji University（明治大学）

作者：Shun Ryuzaki, Hideo Mukai

AI 总结：提出首个风电基础模型Tyan-WP，通过静态站点嵌入和功率感知气象融合模块，在零样本场景下实现超短期概率预测，显著优于传统模型。

链接：https://arxiv.org/abs/2606.08630

机构：School of Information Science and Technology, University of Science and Technology of China（中国科学技术大学信息科学技术学院）； China Electric Power Research Institute（中国电力科学研究院）

作者：Jiahui Huang, Ao Luo, Lei Liu, Hongwei Zhao, Tengyuan Liu, Ruibo Guo, Bo Wang, Zhao Wang, Bin Li

AI 总结：提出Comp-MCTS框架，在固定LLM调用预算下，通过树搜索最大化生成唯一且经oracle验证的反事实，平衡数量与质量。

链接：https://arxiv.org/abs/2606.08696

作者：Yasuo Tabei

AI 总结：针对空间转录组数据噪声大、分辨率低的问题，提出SNR-ST-Mix数据增强框架，通过空间邻域约束和表达相似性加权混合生成生物合理的合成样本，提升深度神经网络插补性能。

链接：https://arxiv.org/abs/2606.08712

机构：Northwestern University（西北大学）； Yale University（耶鲁大学）

作者：Hongyi Yu, Yaoyu Fang, Jiahe Qian, Xinkun Wang, Lee A. Cooper, Bo Zhou

AI 总结：利用知识图谱的K近邻方法在基因敲除扰动预测中表现优异，结合强化学习优化的LLM可达到最先进性能。

链接：https://arxiv.org/abs/2606.08816

机构：University College London（伦敦大学学院）； University of Manchester（曼彻斯特大学）； Valence Labs（Valence实验室）； Recursion（Recursion公司）

作者：Jake Fawkes, Liam Hodgson, Jason Hartford

AI 总结：针对现有基于表示的时间序列异常检测方法忽略振幅信息导致性能下降的问题，提出PAI方案，通过诊断模块和分数增强函数融合振幅相关分数，在TSB-AD-U-Eva和TAB UV数据集上平均VUS-PR提升98.4%和36.8%。

链接：https://arxiv.org/abs/2606.08935

机构：HUAWEI（华为）； KAIST（韩国科学技术院）

作者：Kang Zhang, Wei Jian Lau, Shoushou Ren, Dong Lin, Joon Son Chung, Chuanhao Sun

AI 总结：提出将Cox比例风险模型的时间事件风险信息迁移到大语言模型中的方法，通过文本提示微调Qwen模型，在三个数据集上取得有竞争力的区分度和校准性，并发现隐藏状态呈现连续风险梯度。

链接：https://arxiv.org/abs/2606.08945

机构：Centre for Big Data Research in Health, the University of New South Wales（新南威尔士大学健康大数据研究中心）

作者：Nicholas I-Hsien Kuo, Blanca Gallego, Louisa Jorm

AI 总结：提出TRIAGE框架，利用大语言模型对竞争性临床结果生成辩证推理，缓解风险极化，实现连续风险评分与可解释推理，在三个基准上AUPRC提升3.3%，校准误差降低81%。

链接：https://arxiv.org/abs/2606.09030

机构：KAIST（韩国科学技术院）； AITRICS； University of Wisconsin-Madison（威斯康星大学麦迪逊分校）

作者：Hyeongwon Jang, Gyouk Chu, Changhun Kim, Joonhyung Park, Hangyul Yoon, Eunho Yang

AI 总结：提出一种将粒子系统状态视为希尔伯特空间中的函数、用学习到的神经基函数线性子空间近似状态空间的降阶建模框架，实现大规模拉格朗日模拟的高效表示与预测，在百万粒子SPH模拟中R²>0.99。

链接：https://arxiv.org/abs/2606.09065

作者：Tu Do, Shannon Ryan, Santu Rana

AI 总结：针对心理理论后训练中的捷径问题，提出Thinking-RFT方法，结合可验证奖励和显式推理链，在多个无捷径数据集上显著提升推理能力，尤其在复杂高阶推理和多模态场景中表现优异。

链接：https://arxiv.org/abs/2606.09092

作者：Jike Zhong, Yuxiang Lai, Ming Li, Yuheng Li, Wuao Liu, Behzad Dariush, Konstantinos Psounis, Shao-Yuan Lo

AI 总结：提出BAVAR-BLED算法，结合贝叶斯平均向量自回归和椭圆分布Black-Litterman模型，在TD3架构下自适应分配资产，在道琼斯工业平均指数成分股上实现夏普比率1.72和总收益57.26%。

链接：https://arxiv.org/abs/2606.09104

机构：University of Liverpool（利物浦大学）； Xi'an Jiaotong-Liverpool University（西交利物浦大学）

作者：Daniil Mikriukov (1 and 2), Ruoyu Sun (2), Angelos Stefanidis (2), Jionglong Su (2), Zhengyong Jiang (2) ((1) University of Liverpool, (2) Xi'an Jiaotong-Liverpool University)

AI 总结：提出融合时空图神经网络（STGCN）与检索增强生成（RAG）的精准农业系统，实现30天天气预报、作物推荐及农业问答，在尼泊尔1359个地点数据上STGCN预测MSE达0.011。

链接：https://arxiv.org/abs/2606.09160

作者：Prajwal Thapa, Yagya Raj Pandeya

AI 总结：提出基于时序图注意力网络（T-GAN）的框架，从时空追踪数据中识别足球比赛控球阶段，实现战术意图（入侵空间、保持控球、得分）和六个子阶段的分类，F1分数达0.87（意图级）和0.79（得分阶段）。

链接：https://arxiv.org/abs/2606.09289

机构：Technical University of Munich（慕尼黑工业大学）

作者：Yuesen Li, Daniel Link

AI 总结：研究机器学习仿真卫星温室气体反演算法的时间稳定性，发现预测精度随时间下降，加入时间特征可改善Lasso和神经网络模型的XCH4预测，简单Lasso模型表现优于复杂方法且更稳定。

链接：https://arxiv.org/abs/2606.09313

机构：EURECOM； National Institute for Environmental Studies（国立环境研究所）

作者：Nugzar Gognadze, Motonobu Kanagawa, Yu Someya, Hisashi Yashiro

AI 总结：提出基于TabTransformer的模型，通过学习分类特征的嵌入向量，生成密集的足球事件表示，在下游任务中优于基线方法。

链接：https://arxiv.org/abs/2606.09327

机构：Institute of Exercise Training and Sport Informatics, German Sport University Cologne（科隆德国体育大学运动训练与体育信息学研究所）

作者：Weiran Yang, Daniel Memmert, Maximilian Klemp-Weins

AI 总结：提出基于条件归一化流的物理信息神经网络框架，利用Chapman-Kolmogorov方程和线性化SDE基分布，高效求解多种初始条件下FPE的算子，引入时间加权损失函数解决小时间不稳定性。

链接：https://arxiv.org/abs/2606.09434

机构：Fuzhou University（福州大学）； Louisiana State University（路易斯安那州立大学）； Beijing Normal-Hong Kong Baptist University（北京师范大学-香港浸会大学联合国际学院）； École Polytechnique Fédérale de Lausanne（洛桑联邦理工学院）； Chinese Academy of Sciences（中国科学院）

作者：Li Zeng, Xiaoliang Wan, Yaobin Wang, Fabio Nobile, Tao Zhou

AI 总结：提出损失引导的自适应尺度细化框架，通过插值、路由和尺度池更新自动发现任务有效尺度，在NaCl水溶液体系中降低力预测误差。

链接：https://arxiv.org/abs/2606.09480

作者：Limin Yu

AI 总结：提出自动定向方法，通过约束谱双聚类、贪心局部搜索和多臂老虎机框架联合选择用户和物品构建多个不重叠营销活动，在合成数据、Amazon评论和商业数据上优于模拟退火。

链接：https://arxiv.org/abs/2606.09623

机构：Umeå University（于默奥大学）； KTH Royal Institute of Technology（皇家理工学院）； University of Würzburg（维尔茨堡大学）

作者：Maja Lindström, Natalija Glisovic, Jan von Pichowski, Tommy Löfstedt, Martin Rosvall

AI 总结：本文提出问题导向视角，通过二维相图组织方程可发现性，并引入表示-评估-优化（REO）框架抽象发现过程，旨在从数据中推断物理定律，推动理论修正与新概念形成。

链接：https://arxiv.org/abs/2606.09638

机构：School of Computer Science, Shanghai Jiao Tong University（上海交通大学计算机科学与工程学院）； Ningbo Key Laboratory of Advanced Manufacturing Simulation, Eastern Institute of Technology（东部理工学院宁波先进制造仿真重点实验室）； The State Key Lab of Brain-Machine Intelligence, Zhejiang University（浙江大学脑机智能全国重点实验室）； Department of Statistics and Data Science, Yale University（耶鲁大学统计与数据科学系）； Department of Chemical and Environmental Engineering, Yale University（耶鲁大学化学与环境工程系）； Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学高瓴人工智能学院）； School of Engineering Sciences, University of Chinese Academy of Sciences（中国科学院大学工程科学学院）； DP Technology

作者：Siyu Lou, Hao Xu, Wenguan Wang, Lu Lu, Hao Sun, Yang Liu, Linfeng Zhang, Dongxiao Zhang, Yuntian Chen

AI 总结：针对阿尔茨海默病进展异质性和数据稀疏问题，提出结合局部转换建模与序列建模的数字孪生框架，利用多模态纵向数据预测认知状态并量化不确定性，在ADNI数据上表现优异。

链接：https://arxiv.org/abs/2606.09671

机构：University of Southampton（南安普顿大学）； University Hospital Southampton NHS Foundation Trust（南安普顿大学医院NHS基金会信托）； Faculty of Medicine, University of Southampton（南安普顿大学医学院）

作者：Yinyu Huang, Yilin Zhang, Sofia Michopoulou, Christopher Kipps, Rahman Attar

AI 总结：针对云边连续体中节点冷启动问题，提出一种结合数据混合与神经架构搜索的自动化时间序列预测架构，有效提升预测精度并加速收敛。

链接：https://arxiv.org/abs/2606.09787

机构：Eurecom； OpenAirInterface

作者：Abd Elghani Meliani, Arora Sagar, Adlen Ksentini, Raymond Knopp

AI 总结：提出层次涌现框架(HEF)，将涌现建模为机制景观中的相变，证明在结构假设下物理可行且收敛到唯一不动点，并在111个模算术变换器实验中验证了相变指纹。

链接：https://arxiv.org/abs/2606.07563

机构：H&K Research Studio（H&K 研究工作室）； Clevix LLC（Clevix 有限责任公司）

作者：Truong Xuan Khanh

AI 总结：提出CARTOGRAPH验证层，通过未解析子空间引导、模糊闭合和残差库检测，在多个测试中优于原始投影，并能识别和撤销库外机制。

链接：https://arxiv.org/abs/2606.07576

作者：Neel Tushar Shah, Manglam Kartik

AI 总结：本文主张大型语言模型应学习个性化偏好而非聚合偏好，分析聚合偏好的理论局限与实证问题，提出通过有界个性化框架兼顾个体自主与集体安全。

链接：https://arxiv.org/abs/2606.07629

作者：Cristina Garbacea

AI 总结：提出基于比特等价的信息论定义开放环境，证明经典赌博机非开放，设计算法实现开放学习。

链接：https://arxiv.org/abs/2606.08369

机构：Stanford University（斯坦福大学）

作者：Wanqiao Xu, Yifan Zhu, Benjamin Van Roy

AI 总结：本文从哲学视角统一反问题、稀疏辨识、神经常微分方程和神经算子等数据驱动建模策略，指出它们仅在输入-输出关系的模型类假设上不同，并论证只有某些模型能发现机制并实现泛化。

链接：https://arxiv.org/abs/2606.08956

机构：University of Colorado Boulder（科罗拉多大学博尔德分校）

作者：Conor Rowan

AI 总结：通过计算轨迹长度、曲率等几何指标，发现语义相关提示在中间层收敛、推理任务曲率更大、歧义token轨迹分叉，并揭示三层结构。

链接：https://arxiv.org/abs/2606.09287

机构：MetriQual； London, UK（英国伦敦）； Athens, GR（希腊雅典）

作者：Vishal Pandey, Gopal Singh