机器学习学术速递[6.18]

AI 总结：提出高斯混合注意力（GMA），用K个高斯混合分量的潜在路由替代逐对查询-键比较，实现固定K的线性内存缩放，在长上下文分类任务中与注意力基线竞争。

链接：https://arxiv.org/abs/2606.18283

作者：Yongchao Huang, Hassan Raza

英文摘要：The dense token-to-token interaction pattern of standard dot-product attention remains a central bottleneck in scaling Transformer architectures to long contexts. We introduce \textbf{Gaussian Mixture Attention (GMA)}, a probabilistic attention-style sequence mixer that replaces explicit pairwise query--key comparison with routing through $K$ learned Gaussian mixture components. Queries and keys are mapped to posterior \textit{responsibility} vectors over a shared latent routing space; their overlap defines an implicit responsibility-space affinity, while values are written into and read from a $K$-slot latent memory. By exploiting the associativity of matrix multiplication, GMA avoids materializing the induced $N\times N$ affinity matrix and instead uses two responsibility matrices whose dominant activation storage scales as $\mathcal{O}(NK)$ rather than $\mathcal{O}(N^2)$ for fixed $K$. We formulate bidirectional and causal variants of GMA, provide an end-to-end differentiable parameterization of the Gaussian mixture components, and analyze its responsibility-modulated gradient structure, constrained non-negative low-rank affinity interpretation, and local routing stability. Empirically, GMA exhibits the intended fixed-$K$ linear memory scaling and is competitive with attention-style baselines on long-context classification, while causal GMA improves over tested linear/random-feature attention variants on WikiText-103 but remains behind optimized causal SDPA and Mamba in the current implementation. Analysis of learned responsibilities further shows broad component usage and moderate alignment with surface-form token categories, supporting GMA as a probabilistic, interpretable, fixed-$K$ linear-time attention-style alternative rather than a universal replacement for optimized softmax attention or state-space models.

AI 总结：提出鬼吸引子网络，一种理论推导的动力学解码器，通过构建盆地-吸引子结构实现高效闭环序列生成，在机器人动作解码任务中以2.3M参数匹配1.07B参数扩散变压器的离线精度，延迟降低32倍。

链接：https://arxiv.org/abs/2606.18315

机构：KTH Royal Institute of Technology（瑞典皇家理工学院）； Department of Production Engineering, KTH Royal Institute of Technology（瑞典皇家理工学院生产工程系）； Department of Decision and Control Systems, KTH Royal Institute of Technology（瑞典皇家理工学院决策与控制系统系）

作者：Tianyu Wang, Ying Wang, Zhihao Liu, Xi Vincent Wang, Lihui Wang

英文摘要：Sequential output generation with large-scale Transformer and diffusion decoders pays a memory cost that grows with sequence length, plus iterative per-step computation. Replacing them with small feed-forward decoders restores efficiency but produces unstructured latent representations that limit closed-loop control: phase-conditioned action generation and cross-step latent carry-over both require a latent geometry with stable basins. This article proposes Ghost Attractor Networks, a theoretically derived dynamical decoder whose latent evolves under a learned potential with drift and produces a basin-attractor structure by construction. Three desiderata (multi-modality, decoder-level single-pass switching, and constant memory) motivate the potential-drift form, and mode transitions arise as saddle-node bifurcations with ghost-attractor escape. A hierarchical phase-space decomposition separates first-order basin convergence from second-order proprioceptive refinement. Empirically, a Ghost trained end-to-end with a behavioral-cloning and contrastive objective exhibits the predicted gradient-flow contraction in its potential, with the gradient norm decaying by 67 percent across five integration steps on 1430 held-out samples. Ghost is evaluated as a robotic action decoder. A 2.3-million-parameter Ghost matches the offline accuracy of a 1.07-billion-parameter Diffusion Transformer at 462 times fewer parameters and 32 times lower latency, and beats five alternative 2M-parameter decoders (MLP, Neural ODE, CVAE, Transformer, 1-step Diffusion) on offline mean squared error by 5.9 to 29 percent. On the LIBERO-10 closed-loop benchmark, phase conditioning on Ghost's basin-structured latent yields a 13.5 percentage-point success-rate gain over a feed-forward MLP baseline, and persistent-latent ensembling reaches a 95.7 percent final success rate.

AI 总结：本文回顾了复数值循环语言模型SWAVE的演化过程，揭示了其设计假设的缺陷，并提出了cos-domination collapse等理论见解和工程原则。

链接：https://arxiv.org/abs/2606.18324

机构：EdgeVerve Systems Limited（EdgeVerve系统有限公司）

作者：Ramprasath Ganesaraja, Swathika N, Sahil Dilip Panse

英文摘要：SWave is a complex-valued recurrent language model (169.26M parameters, D=384, L=16, T=2048) trained on FineWeb-Edu using 2xH100 NVL. It was designed around three founding premises: that representing language as complex waves rather than real-valued numbers enables richer information encoding; that a Cayley-parameterised unitary transition provides a mathematical guarantee against state decay or explosion; and that a hidden state which rotates rather than shrinks preserves signal integrity over arbitrarily long contexts. The core of SWave evolved substantially across three development phases. The Resonance Head was found to structurally admit imaginary-channel collapse as a global loss minimum (a failure mode we term cos-domination collapse) and was superseded by an untied head with independent real and imaginary embedding tables from the Phase-Associative Memory (PAM) architecture. This resolved the degenerate minimum and enabled stable 200,000-step training (best-step PPL 22.0 at step 89,861). ComplexNorm and the Wave Propagation Scan proved load-bearing throughout all three phases and were retained to the final architecture. ProtectGatedScan was reframed as a structural prior rather than a learned behaviour. The four multi-scale retention concepts showed no measurable improvement under controlled evaluation and were found non-load-bearing. The ComplexGatedUnit was superseded by a real-valued squared-ReLU channel mixer with fewer parameters. The auxiliary training objectives showed no benefit once structural constraints were resolved. The investigation yields a formal characterisation of cos-domination collapse, a parallel scan with a log-space backward pass for numerical stability, six transferable engineering principles for complex-valued recurrent training, and a plan-to-code traceability methodology for catching structural divergences that conventional test suites miss.

AI 总结：提出RGNet，一种基于重正化群概念的神经网络架构，通过层次化粗粒化特征空间处理类别不平衡和多维噪声，在AI4I数据集上验证了其有效性。

链接：https://arxiv.org/abs/2606.18326

作者：Evgeny Nikulchev, Dmitry Ilin

英文摘要：The application of machine learning models in practical tasks faces challenges such as class imbalance and multidimensional noise. This paper proposes RGNet, a neural network architecture based on the concept of the renormalization group (RG), for hierarchical coarse-graining of the feature space. The model sequentially compresses the input dimensionality and concatenates all scales before classification, allowing it to capture both local details and global patterns. The notion of RG-flows is introduced - interpretable low-dimensional representations whose visualization via t-SNE reveals a discrete curvilinear structure confirming the effectiveness of coarse-graining. Experimental results are presented on the imbalanced AI4I dataset. The obtained results demonstrate that RGNet is a universal, interpretable, and competitive solution for fault prediction in applications with imbalanced classes.

AI 总结：提出LLMZero系统，利用LLM智能体通过树搜索发现多阶段RL后训练的自适应策略，揭示容量参数单调累积、正则化参数振荡的规律，在4个GRPO任务上相对基线提升9%-140%。

链接：https://arxiv.org/abs/2606.18388

机构：Amazon（亚马逊）

作者：Haoyang Fang, Wei Zhu, Boran Han, Alex Zhang, Zhenyu Pan, Shuo Yang, Shuai Zhang, Jiading Gai, Peng Tang, Cuixiong Hu, Xuan Zhu, Huzefa Rangwala, George Karypis, Bernie Wang

英文摘要：RL post-training strategies are dataset-dependent and reveal a recurring empirical pattern: capacity parameters accumulate monotonically across stages, while regularization parameters predominantly oscillate in response to shifting training dynamics. This distinction matters because fixed schedules commit all parameters to fixed trajectories and therefore cannot express the non-stationary exploration-exploitation tradeoffs that regularization must track; the principle provides actionable design rules for multi-stage training. We discover this through LLMZero, a system where LLM agents search over training trajectories via tree search, diagnosing pathologies at each checkpoint and proposing coordinated multi-parameter transitions. Across 4 diverse GRPO tasks, LLMZero discovers strategies that improve over the base model by 9% to 140% relative and over grid search by 6% to 15% relative, consistently outperforming random search and the skill-based agent. The structural principle transfers across tasks, providing an explanation for why discovered strategies take qualitatively different forms yet share similar parameter dynamics.

AI 总结：通过有序实Schur坐标分析单层tanh RNN，发现任务分布下循环矩阵存在功能冗余，特定非正常Schur耦合可被移除而不影响性能，揭示了任务限制的近似功能不变性。

链接：https://arxiv.org/abs/2606.18457

机构：Salk Institute for Biological Studies, La Jolla, CA, USA（索尔克生物研究所，拉霍亚，加利福尼亚州，美国）

作者：Simon Dräger

英文摘要： Recurrent networks can contain substantial functional redundancy in weight space: changing a recurrent matrix may leave the input-output rollout nearly unchanged on a task distribution, while similar-scale changes can destroy the same behavior. We study this redundancy in one-layer tanh RNNs using ordered real Schur coordinates. The Schur form separates spectral blocks from directed nonnormal couplings, giving a diagnostic basis for structured ablations that keep the input and readout maps fixed. In a fixed-length copy task, selected nonnormal Schur couplings can be removed with little loss in some trained solutions, whereas other couplings are necessary for accurate autonomous replay. Across flip-flop, sine generation, and context-dependent integration, the loss-preserving ablation profile varies across tasks and trained solutions. These results identify candidate approximate functional invariances, not universal symmetries of recurrent weight space. Schur-coordinate ablations provide a practical diagnostic for which structured perturbations preserve a trained recurrent solution and which ones disrupt its computation.

AI 总结：研究发现 SFT 过度训练导致 rollout 分布熵降低，使 GRPO 中优势信号消失，从而引发排名反转；提出基于熵的两阶段诊断方法可预警高风险检查点。

链接：https://arxiv.org/abs/2606.18487

机构：Stanford University（斯坦福大学）

作者：Siddharth Aphale, Kelly Liu

英文摘要：The standard heuristic of selecting the SFT checkpoint with the highest pass@1 for GRPO can fail when SFT compresses the rollout distribution. For binary rewards, the expected within group advantage variance is $p(1{-}p)(g{-}1)/g$; when early GRPO drives $p$ below $p^*(g)$, most groups have identical rewards and provide no group relative signal. We study SFT depth ladders for Qwen2.5-Coder-3B and DeepSeek-Coder-6.7B. We test Qwen2.5-Coder-3B across five depths and three seeds, and DeepSeek-Coder-6.7B across four matched depths and three seeds. On Qwen, pre RL pass@1 rises with SFT depth, but peak GRPO pass@10 falls from $0.806$ to $0.481$ (3 seed mean, $n{=}20$); pre RL entropy is positively associated with the GRPO outcome ($\rho{=}{+}0.69$). On DeepSeek, pass@1 remains far above $p^*(8){=}0.083$, and GRPO outcomes compress rather than invert. A two stage diagnostic, combining pre RL entropy triage with an early GRPO entropy monitor, flags high risk checkpoints and can stop failing runs early. Simple KL to reference regularisation and label smoothing variants do not rescue the collapsed Qwen checkpoint in our setting, suggesting the failure is not a trivial GRPO hyperparameter artefact.

AI 总结：本文发现RLVR模型的稀疏更新在参数空间中分散更远，形成近正交捷径导致合并脆弱，并提出SAR-Merging方法解决该问题。

链接：https://arxiv.org/abs/2606.18521

机构：Zhejiang University（浙江大学）； Simon Fraser University（西蒙菲莎大学）； The Chinese University of Hong Kong（香港中文大学）； Zhejiang Key Lab of Accessible Perception and Intelligent Systems（浙江省可感知智能系统重点实验室）

作者：Chenrui Wu, Zexi Li, Jiajun Bu, Jiangchuan Liu, Haishuai Wang

英文摘要：Reinforcement Learning with Verifiable Reward (RLVR) has emerged as a powerful post-training paradigm that surpasses Supervised Fine-Tuning (SFT) in eliciting reasoning intelligence and resisting catastrophic forgetting. Recent studies further reveal that RLVR induces highly sparse and off-principal parameter updates compared to SFT. This naturally raises the question: does such sparsity make RLVR models more amenable to model merging? If so, model merging would offer a scalable, training-free path to aggregate diverse reasoning capabilities from independently trained RLVR models. Surprisingly, we find the opposite, uncovering a sparsity curse: the sparse RLVR updates are spread farther apart in parameter space, forming near-orthogonal shortcuts that make aggregation inherently fragile. This is likely rooted in the stochasticity of RL optimization and the diversity of emergent reasoning patterns. Unlike SFT models that converge to shared, flat basins and merge naturally, RLVR models suffer severe degradation under standard merging methods. Through systematic empirical analysis of the update geometry, we characterize the mechanisms behind this failure and propose Sensitivity-aware Resolving Merging (SAR-Merging), a merging recipe tailored for the unique structure of RLVR parameter spaces. SAR-Merging resolves conflicts in overlapping update regions via Fisher Information-based sensitivity arbitration, followed by magnitude-aware sparsification and rescaling to preserve fragile reasoning pathways. Experiments on mathematical and coding benchmarks demonstrate that SAR-Merging substantially outperforms existing merging methods on RLVR models, enabling both single-task enhancement and multi-capability fusion.

AI 总结：针对循环Transformer，提出残差缩放因子应为1/N而非1/√L，并推导出多层的分解参数化，实现超参数从少循环到多循环的迁移。

链接：https://arxiv.org/abs/2606.18524

机构：Tsinghua University（清华大学）

作者：Shaowen Wang, Bingrui Li, Ge Zhang, Wenhao Huang, Shen Yan, Jian Li

英文摘要：Looped (weight-tied) Transformers apply a shared residual block $N$ times ($h \leftarrow h + \varepsilon\,f(h)$, same $f$ at each step), increasing effective depth without adding parameters. Prior depth-scaling analyses prescribe $\varepsilon = 1/\!\sqrt{L}$ for depth-$L$ residual networks. We show that this is insufficient for looped architectures: weight sharing makes residual updates correlated across iterations, requiring the stronger scaling $\varepsilon = 1/N$. For multi-layer blocks ($L$ unique layers looped $N$ times), we derive a factored parameterization $\varepsilon = \lambda/(N\!\sqrt{L})$ that separates the two sources of growth: $1/N$ controls the within-layer loop correlation, and $1/\!\sqrt{L}$ controls the across-layer variance. A key consequence is that the optimal learning rate depends only on the number of unique layers $L$, not on the loop count $N$, enabling direct hyperparameter transfer from small to large $N$ without retuning. Experiments on looped Transformers confirm that $1/N$ scaling improves trainability and yields better loss than $1/\!\sqrt{N}$ scaling across loop counts.

AI 总结：提出一种基于两水平重叠Schwarz区域分解的层次注意力机制，通过局部低秩注意力块与粗网格注意力块结合，在少参数下实现更快训练和更高精度。

链接：https://arxiv.org/abs/2606.18525

作者：Stephan Köhler, Oliver Rheinbach

英文摘要：We propose a hierarchical attention mechanism based on two-level overlapping Schwarz domain decomposition. The method is motivated by the observation that two-level Schwarz domain decomposition methods combine local subdomain corrections with a coarse level that communicates global, long-range information. We test its usefulness in the context of finite-dimensional operator learning using a simple, one-dimensional diffusion problem with homogeneous Dirichlet boundary conditions. Although elementary, this problem provides a controlled sequence-to-sequence setting in which the exact nonlocal solution operator is known. After discretization, learning the solution operator amounts to approximating the inverse of a symmetric positive definite matrix. As a baseline, we use a global softmax-free low-rank attention operator of the form $QK^T$. The proposed construction replaces this dense global factorization by a two-level additive structure: local low-rank attention blocks on overlapping subdomains are combined with a coarse attention block. The resulting operator has the form $$M_{\theta}^{-1} = \Phi Q_0 K_0^T \Phi^T + \sum_{i=1}^{N} R_i^T D_i^{1/2} Q_i K_i^T D_i^{1/2} R_i.$$ Here $R_i$ restricts to an overlapping subdomain, $D_i$ is a partition-of-unity weight, and $\Phi$ is a coarse interpolation (or prolongation) matrix. Numerical experiments for synthetic Fourier right-hand sides indicate that the domain-decomposition attention operator is able to train faster and can give more accurate approximations than a global low-rank attention baseline while using significantly fewer parameters.

AI 总结：提出PACT方法，通过识别并保留预训练权重中的承重墙维度，在任务向量中锚定任务特定核心，解决任务向量范式下任务冲突和性能下降问题，提升模型合并效果。

链接：https://arxiv.org/abs/2606.18627

机构：Shanghai Jiao Tong University（上海交通大学）； Nanyang Technological University（南洋理工大学）； The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））

作者：Ningyuan Shi, Zhipeng Zhou, Hao Wang, Chunyan Miao, Peilin Zhao

英文摘要：Model merging has emerged as a training-free alternative to multi-task learning, aiming to combine multiple task-specific fine-tuned models into a single multi-task model. Most existing model merging approaches follow the Task Arithmetic paradigm, which decomposes fine-tuned weights into pre-trained parameters and task vectors, and performs merging exclusively in the task-vector space. The effectiveness of this paradigm implicitly relies on the assumption that task-specific knowledge is encoded solely within task vectors. We argue that this assumption generally does not hold due to the intrinsic task preferences of pre-trained models. Specifically, we identify \textbf{Load-Bearing Wall (LBW) dimensions}, namely some task-critical knowledge that remains embedded in the pre-trained weights rather than being fully transferred into task vectors. We characterize LBW dimensions from both scalar-weight and subspace perspectives, thereby covering the major paradigms of existing model merging methods. Our analysis reveals that, by ignoring LBW dimensions, task-vector-based approaches fail to fully resolve task conflicts and may inadvertently damage task-specific knowledge encoded in the pre-trained model, leading to degradation. To address this issue, we propose PACT, which preserves the anchored task-specific cores (i.e., LBW dimensions) within task vectors by aligning their orthogonal complements with the subspace of the pre-trained weights. These aligned subspace components are then removed from the task vectors before applying existing model merging algorithms. Furthermore, we develop an efficient variant based on randomized SVD to improve scalability. PACT can be seamlessly integrated with existing methods. Extensive experiments across multiple benchmarks demonstrate that PACT consistently enhances mainstream model merging approaches and establishes new state-of-the-art performance.

AI 总结：提出统一理论代理InTrain，通过几何容量和优化韧性两个协同成分形式化架构的可训练性，在NAS基准上达到与集成方法相当的排序相关性。

链接：https://arxiv.org/abs/2606.18676

机构：School of Computer and Data Science, Fuzhou University（福州大学计算机与数据科学学院）； School of Computer and Data Science, Minjiang University（闽江学院计算机与数据科学学院）； School of Artificial Intelligence, Nanchang University（南昌大学人工智能学院）； Department of Computer Science, Hong Kong Baptist University（香港浸会大学计算机科学系）； School of Interdisciplinary Medicine and Engineering, Harbin Medical University（哈尔滨医科大学跨学科医学与工程学院）

作者：Qinqin Zhou, Fuhai Chen, Jipeng Wu, Zhiwei Chen, Zhikai Hu, Weiwei Cai

英文摘要：Training-free neural architecture search promises efficient discovery of high-performance networks without costly training. However, existing zero-cost proxies rely on fragmented heuristics that fail to capture the fundamental question: what makes an architecture trainable? This paper introduces Intrinsic Trainability (InTrain), a unified theoretical proxy that formalizes trainability as an architectural invariant emerging from two synergistic components: geometric capacity and optimization resilience. We operationalize intrinsic trainability through analysis of neural information processing. Geometric capacity is quantified via the participation ratio of activation covariance eigenspectrum, capturing the effective dimensionality of representation manifolds. Optimization resilience is measured through cumulative gradient health, assessing the robustness of backpropagation across network depth. InTrain synthesizes these dimensions through a scale-invariant multiplicative coupling, which we hypothesize is essential for capturing their synergistic, non-additive relationship. Extensive experiments on standard NAS benchmarks and search spaces demonstrate that InTrain achieves ranking correlations on par with state-of-the-art ensemble-based proxies and outperforms other single-metric methods.

AI 总结：提出受挫同步网络（FSN），通过复值耦合核和延迟项实现基于同步的注意力机制，在百万参数级字符级文本和代码任务上优于调优的RoPE-SwiGLU Transformer。

链接：https://arxiv.org/abs/2606.18694

作者：Joshua Nunley

英文摘要：A network of oscillators that synchronizes perfectly computes nothing further, so an attention architecture built from synchronization must locate its computation in structured departures from agreement. We introduce the Frustrated Synchronization Network (FSN), whose token states are phases on a torus and whose entire value pathway is one learned complex coupling kernel over harmonics and a one-step delay. Each component of the kernel is a frustration in the sense of the synchronization literature. The complex phases are static Kuramoto-Sakaguchi frustration angles, the signed harmonics are repulsive Daido components, and the delay term, which couples each token to the successors of the tokens it attends to, is algebraically identical to Kuramoto-Sakaguchi coupling whose frustration angle is the data's own transition, so next-token prediction is implemented as synchronization frustrated by the data. At matched one-million-parameter and training budgets on character-level text and code, the FSN's validation loss is below a tuned RoPE-SwiGLU transformer's at every epoch measured, and the comparison survives training the baseline to convergence: every thirty-epoch enwik8 seed finishes below the transformer's converged fifty-epoch loss of 1.611, and the FSN's completed fifty-epoch runs converge to 1.5953 +/- 0.0014. A variant with every feed-forward block replaced by mean-field coupling to learned collective modes, leaving no multilayer perceptron in the stack, tracks the transformer. On natural text the unfrustrated base layer falls behind the converged transformer at every copy depth, worst on long-range copy events; the kernel reverses the deficit at every depth of four and beyond. Headline comparisons are at the one-million-parameter scale; a scale ladder is complete through four million parameters with the advantage persisting, and remaining arms are marked as in progress.

AI 总结：提出TAPO方法，通过对比正确与错误轨迹构建微反思修正，实现从隐式分布对齐到显式轨迹构建的自蒸馏改进，在多个数学推理基准上优于GRPO。

链接：https://arxiv.org/abs/2606.18844

机构：Qwen Business Unit of Alibaba（阿里巴巴通义千问事业部）； Tsinghua University（清华大学）； Peking University（北京大学）

作者：Zhilin Huang, Hang Gao, Ziqiang Dong, Yuan Chen, Yifeng Luo, Chujun Qin, Jingyi Wang, Yang Yang, Guanjun Jiang

英文摘要：Self-distillation improves reasoning in large language models by using the model's own rollouts as training signal, typically through implicit logit-level alignment that minimizes KL divergence toward a privileged target distribution. However, because this supervision is generated via uncontrolled sampling, it provides no diagnostic insight into the model's specific errors or corrective guidance for its individual failure patterns. Consequently, the model learns to imitate a privileged distribution rather than receiving fine-grained corrections that pinpoint where and why its reasoning fails. In this paper, we propose Trajectory-Augmented Policy Optimization (TAPO), which advances self-distillation from implicit distributional alignment to explicit trajectory construction. During RL training, the model produces both correct and incorrect rollouts to the same query, and TAPO leverages this contrastive structure to construct micro-reflective corrections, new training trajectories that retain the model's erroneous reasoning up to the point of failure, then insert a natural-language diagnosis and corrected reasoning guided by a correct reference from the same sampling group. Since each trajectory is anchored in the learner's own prefix and solutions, the corrective signal preserves the model's on-policy distribution to a greater extent than the position-wise alignment imposed by KL-based methods. To integrate these trajectories, TAPO introduces difficulty-aware candidate selection at the model's capability boundary and decoupled advantage estimation to prevent gradient contamination. Experiments on AIME 2024, AIME 2025, and HMMT 2025 show that TAPO achieves consistent improvements over GRPO under the same number of training steps. Further analysis demonstrates that TAPO strengthens both first-pass reasoning and error-correction effectiveness.

AI 总结：提出GrapNet，一种将图作为可执行架构的神经基板，通过可编程接口支持结构编辑、冻结子图、局部审计等操作，在Split Fashion-MNIST和Split CIFAR-10上分别提升12.08和3.81个百分点的准确率。

链接：https://arxiv.org/abs/2606.18923

机构：Zirong Li（李子荣）

作者：Zirong Li

英文摘要：Programmability is a missing first-class interface in fixed-tensor neural networks: editing a relation, freezing a subgraph, auditing a local function, or changing the execution backend should be an operation on the neural program rather than ad-hoc parameter surgery. GrapNet studies this graph-as-network setting. The graph is the architecture and executable program, not an input data graph. Each compute node owns its next-layer child references and a trainable allocation vector aligned with those references; deleting a relation physically removes both the child reference and the corresponding allocation coordinate. Structural rules and execution policies live outside the node core, so the same child-owned graph can be grown, frozen, structurally edited, grouped into trainable family blocks, routed by attention over active relations, or lowered to dense snapshots after topology stabilizes. GrapNet composes with conventional modules through a vector-valued parent interface: dense layers, CNN encoders, ResNet feature extractors, attention blocks, and transformer representations can all feed one sensory GrapNode per coordinate. The evaluation is organized as a programmability stress suite rather than as a new replay benchmark. In a matched ten-seed Split Fashion-MNIST study, a plastic GrapNet+ER head reaches 63.16 percent seen-class accuracy versus 51.08 percent for a parameter-larger dense MLP+ER under the same seen-class loss and replay memory, with paired delta 12.08 points and p=1.3e-5. On Split CIFAR-10 with a frozen ImageNet ResNet-18 encoder, the same substrate improves the online head over MLP-256 by 3.81 points, with p=0.0026. These results support GrapNet as an editable neural graph substrate whose core value is structural programmability with faithful execution views.

AI 总结：提出ViGOS框架，通过解耦感知和推理，在MLLM后训练中避免文本捷径，提升图像依赖行为。

链接：https://arxiv.org/abs/2606.19120

机构：State Key Laboratory of Robotics and Intelligent Systems, Shenyang Institute of Automation, Chinese Academy of Sciences（中国科学院沈阳自动化研究所机器人学国家重点实验室）； University of Chinese Academy of Sciences（中国科学院大学）

作者：Sihan Wang, Xiyao Liu, Lianqing Liu, Zhi Han

英文摘要：On-policy self-distillation (OPSD) trains a model on its own rollouts and uses a frozen copy to provide dense token-level targets conditioned on a reference target. This works well for LLM reasoning, but a direct extension to multimodal large language models (MLLMs) can create a shortcut: the privileged target may guide tokens mainly based on the text reference target rather than the image. We propose ViGOS, a visually grounded OPSD framework for MLLM post-training. The student first writes a visual description and then reasons toward the final answer. For valid rollouts, an image-only perception teacher supervises the description, while a privileged reasoning teacher supervises the reasoning and final answer on the same student prefix. A reference teacher is used only for invalid rollouts to recover the output format. Across general vision-language, expert reasoning, visual math, spatial grounding, and visual-language-prior benchmarks, ViGOS keeps the main benefits of OPSD and improves image-grounded behavior in shortcut-prone settings.

AI 总结：提出INDEQS，一种基于图的NCDE预测方法，通过在不同架构位置注入有向图先验知识，结合内外混合机制和自适应图卷积，在合成和真实任务中优于无信息NCDE。

链接：https://arxiv.org/abs/2606.19138

机构：Fraunhofer Heinrich Hertz Institute（弗劳恩霍夫海因里希·赫兹研究所）； Technische Universität Berlin（柏林技术大学）

作者：Michael Detzel, Gabriel Nobis, Kristiyan Blagov, Juri Schubert, Jackie Ma, Wojciech Samek

英文摘要：Neural Controlled Differential Equations (NCDE) provide a powerful continuous-time framework for forecasting time series, but standard graph-based extensions typically learn spatial structure purely from data, even in settings where a directed graph structure is known a priori. We introduce Informed Neural controlled Differential EQuationS (INDEQS), a graph-based NCDE forecasting method that incorporates prior knowledge of a directed graph at distinct architectural positions. INDEQS separates inner mixing of hidden states across graph nodes from outer mixing between vector field and control, and offers both a lightweight graph-constrained variant and a more expressive variant, learning additional graph connections from data via adaptive graph convolutions. To systematically study when graph informedness is beneficial in forecasting, we devise a continuous advection simulation on directed graphs, yielding synthetic spatio-temporal datasets with known ground-truth flow structure. We then evaluate INDEQS on two real-world tasks: river discharge forecasting on a hydrological network and traffic flow prediction on PeMS08. Across these synthetic and real-world benchmarks, outer informedness consistently improves mean absolute error over an uninformed NCDE with comparable parameter count, particularly on larger graphs, while inner informedness offers a more parameter-efficient alternative when strict adherence to a known adjacency is desired. A comparison of discrete convolutional and continuous-time decoders further shows that continuous decoders yield better accuracy and greater temporal flexibility on real-world tasks. An implementation of INDEQS and the advection simulation is available at this https URL.

AI 总结：提出一种后验泛化框架，通过稀疏代理（SAE重建）认证语言模型，推导期望风险上界，并在GPT-2 Small等模型上验证非平凡界，揭示深层更易认证且特征分解区分语义对齐与统计稀疏性。

链接：https://arxiv.org/abs/2606.18383

机构：Department of Computer Science and Engineering, Indian Institute of Technology Patna（印度理工学院巴特那分校计算机科学与工程系）

作者：Dibyanayan Bandyopadhyay, Asif Ekbal

英文摘要：Sparse autoencoders (SAEs) are increasingly used to extract interpretable features from language models (LMs), yet a central question remains: when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM We study this through a post-hoc generalization framework that certifies the LM via a sparse proxy, obtained by replacing a native hidden activation with its pretrained SAE reconstruction. Our framework derives an upper bound on the base model's expected risk using four measurable quantities: proxy risk, SAE reconstruction gap, concept-pool mismatch, and sparse complexity. We interpret this certificate as an operational criterion for explanatory faithfulness. In particular, a non-vacuous bound indicates that the extracted sparse features retain meaningful predictive information, while small reconstruction and mismatch errors indicate that the proxy remains behaviorally close to the original model. Empirically, we show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes. A detailed layerwise analysis of Llama-3-8B reveals a strong depth dependence, with later layers becoming much easier to certify, associated with both stronger local fidelity and weaker downstream error amplification. Finally, through feature-shuffling ablations, we show that the decomposition distinguishes genuine semantic alignment from mere statistical sparsity, providing a useful diagnostic for when SAE-based explanations become less reliable.

AI 总结：提出MOLAR框架，通过分离干净属性推断与标签观测，利用图与文本模态的残差证据，从噪声标签中学习多模态分子表示，在自然噪声和标签翻转基准上优于基线方法。

链接：https://arxiv.org/abs/2606.18390

机构：Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）； Zhengzhou University（郑州大学）； The Education University of Hong Kong（香港教育大学）； The Chinese University of Hong Kong（香港中文大学）； Weizmann Institute of Science（魏茨曼科学研究所）

作者：Yingxu Wang, Kunyu Zhang, Nan Yin, Yu Li, Eran Segal

英文摘要：Motivation: Noisy labels are a common challenge in molecular property prediction because molecular annotations are often obtained from assays, curated databases, or weak annotation pipelines rather than directly observed clean biological states. Treating recorded labels as reliable supervision can cause models to memorize corrupted observations and learn misleading molecular evidence. In multimodal molecular representation learning, this issue can be amplified by graph-text fusion or alignment, which may propagate label-induced errors across modalities. Results: We propose MOLAR, a noise-aware framework for learning multimodal molecular representations from noisy labels. MOLAR separates latent clean-property inference from recorded-label observation: graph and text views contribute residual evidence to a clean-property distribution, and a categorical label-observation channel maps this distribution to recorded labels for training. This formulation derives posterior label reliability and modality-specific molecular evidence from the model. Experiments on naturally noisy molecular benchmarks and controlled label-flipping benchmarks show that MOLAR consistently outperforms representative baselines. Visualization analyses further show that MOLAR provides interpretable reliability and modality-evidence diagnostics.

AI 总结：提出双通道接地世界建模（DCGWM），通过分区潜空间和内向梯度流，结构性防止联合嵌入预测架构中多目标接地导致的目标干扰崩溃。

链接：https://arxiv.org/abs/2606.18688

作者：Akshay Hazare

英文摘要：Joint Embedding Predictive Architectures (JEPAs) are a leading approach to world model representation learning. We identify a failure mode in JEPA-based world models grounded against two qualitatively distinct external signals: physical dynamics (sparse, high-magnitude, constraint-satisfying gradient corrections) and social-behavioral dynamics (diffuse, distribution-matching corrections). We term this Objective Interference Collapse (OIC): we argue that joint learning in a shared latent space causes the dominant channel to systematically collapse the subordinate channel's representational subspace, in a manner not resolvable by loss weighting alone. We propose Dual-Channel Grounded World Modeling (DCGWM), designed to structurally prevent OIC through a partitioned latent space (physical subspace Z_p, behavioral subspace Z_b) with inward-only gradient flow. A Physical Grounding Channel updates only Z_p via VICReg-style alignment to physical measurements; a Social-Behavioral Grounding Channel updates only Z_b via alignment to trajectories from an emergent multi-agent simulation. An Inter-Channel Interface Module couples the subspaces at the task level without cross-subspace gradients. An Asymmetric Grounding Adherence Loss penalizes rollout drift with a hard hinge for physical violations and a soft KL for behavioral divergence. A Generative Rendering Layer is architecturally isolated from the latent world model. We present three theoretical results: the partition removes the gradient-interference pathway implicated in OIC; each grounded subspace inherits anti-collapse guarantees from its alignment objective; and generative isolation is necessary under a stated assumption on the generative objective's geometry. This manuscript establishes the problem formulation and architecture; experimental validation is ongoing and will be reported in a future revision.

AI 总结：提出LOGICA框架，在输出逻辑空间进行对比学习，通过门控跨模态适配器保留预训练似然接口，实现跨不同词汇表模型的上下文条件预测，在蛋白质-配体结合、TCR-肽活性和药物耐药性预测任务上超越现有方法。

链接：https://arxiv.org/abs/2606.18703

作者：Yanjun Shao, Yundi Chen, Yashvi Patel, Aurelien Pelissier, María Rodríguez Martínez

英文摘要：Pretrained biological language models expose per-token probability distributions through masked-token prediction, providing the likelihood interface central to sequence design, variant scoring, and mechanistic interpretation. Yet these distributions are learned from broad unlabeled corpora and are not naturally conditioned on task-specific biological contexts such as interaction partners, cellular environments, or therapeutic interventions. Existing contextual matching methods often distort this interface through pooled embeddings, contrastive latent spaces, or task-specific prediction heads. We introduce LOGICA (Logit-space Contrastive Alignment), a framework for context-conditioned prediction that performs contrastive learning directly in output-logit space. Using gated cross-modal adapters compatible with each model's native token head, LOGICA preserves the pretrained likelihood interface and converts contextualized token log-likelihoods into matching scores. Alignment is defined through context-sensitive token probabilities rather than proximity in a shared embedding space, enabling learning from sparse paired data across models with distinct vocabularies, without a shared tokenizer or decoder. LOGICA is particularly effective for mutation-local variant ranking, where comparisons reduce to context-conditioned likelihoods of mutant tokens at perturbed sites. Across protein--ligand binding, TCR--peptide activity, and drug-conditioned resistance prediction, LOGICA improves over prior state-of-the-art methods, including matched latent-contrastive and conditional MLM baselines, while retaining a token-level interface for interpretation and generation. On held-out-gene single-mutation drug-resistance prediction, LOGICA improves AUC from near-random latent-space baselines of $\sim$0.55 to $\sim$0.65.

AI 总结：提出无监督奖励优化框架，结合模型不确定性和语义一致性作为代理奖励，通过SRO和BRO算法优化PLMs，在无标签数据下实现可控蛋白质生成，性能接近有监督方法。

链接：https://arxiv.org/abs/2606.18961

机构：The Chinese University of Hong Kong（香港中文大学）； MBZUAI； Hong Kong University of Science and Technology（香港科学理工大学）

作者：Lanqing Li, Shentong Mo, Yang Yu, Pheng-Ann Heng

英文摘要：Protein language models (PLMs) have emerged as powerful tools for controllable biomolecular design, yet their post-training adaptation typically relies on costly wet-lab validation or curated preference datasets. To overcome this supervision bottleneck, we introduce unsupervised reward optimization of PLMs, a comprehensive framework for steerable protein generation without ground-truth labels. Our key insight is that task-agnostic rewards, which combine intrinsic model uncertainty with extrinsic semantic consistency informed by protein representation models, exhibit strong correlation with controllability measures across base models and temperature regimes. Building upon this discovery, we propose two offline algorithms: Soft Reward Optimization (SRO) and Binarized Reward Optimization (BRO), which effectively maximize the classical RLHF objective induced by these proxy rewards. Extensive experiments on compositional out-of-distribution prompts demonstrate that both methods significantly outperform competitive baselines (DPO, KTO), while approaching oracle performance across multiple sampling temperatures, model scales and protein families. Moreover, PLMs fine-tuned with unsupervised rewards can achieve consistently higher coverage compared to their base model in pass@k evaluations. By enabling self-improvement of PLMs through their own generated experience, our framework provides a scalable pathway toward controllable biomolecular design in settings where labeled preferences or experimental feedback are scarce or unavailable.

AI 总结：提出PROPEL框架，通过训练轻量级激活探针作为求解率代理，在无需重复求解器评估的情况下优化任务生成器，使生成任务集中在可学习前沿，提升数学、代码和软件工程任务的有效性。

链接：https://arxiv.org/abs/2606.18284

机构：Vmax； Goodfire AI

作者：Lorenz Wolf, Connor Watts, Roger Creus Castanyer, Geoffrey Bradway, Maxwill Lin, Augustine N. Mavor-Parker, Matthew Daborn-Sargent

英文摘要：The limiting resource for training agents via reinforcement learning (RL) is increasingly frontier task supply: valid, solvable tasks just difficult enough to train the current model. As reasoning and agentic models improve, fixed task distributions saturate, while naive synthetic generation yields tasks that are trivial, impossible, or ill-posed. Training a task generator with RL to optimize validity and learnability can address this bottleneck, but direct optimization requires repeated solver rollouts per candidate. For software-engineering (SWE) tasks, a single rollout can take tens of minutes; solver-in-the-loop generator training is intractable. We introduce PROPEL, a solver-amortized framework for training task generators at the targeted solve rate. PROPEL trains a lightweight activation probe on a one-time labeled corpus of generated tasks and solver outcomes. The probe predicts target-solver pass rate from a frozen generator reference model and serves as a proxy for solve rate during generator optimization, reducing generator evaluation to a single forward pass. Across math, code, and software-engineering at multiple model scales, PROPEL shifts generation toward the targeted solve rate: for coding, tasks generated at the learnable frontier increase from $10.1\% \rightarrow 20.0\%$ for a Qwen2.5-3B-Instruct solver and from $5.3\% \rightarrow 12.6\%$ for a Qwen2.5-7B-Instruct solver. For SWE, PROPEL increases the share of generations at the targeted solve rate from $9.8\% \rightarrow 19.6\%$ for Qwen3.5-27B on repositories not seen during training of probe and generator.

AI 总结：针对混合离散-连续动作、训练时安全约束和物理动力学形成的耦合问题，提出TRIDENT框架，通过Richardson-Romberg梯度校正、Lyapunov约束序列信任域更新和物理信息残差评论家，实现可证明的安全收敛，显著降低训练违规并提升奖励。

链接：https://arxiv.org/abs/2606.18308

机构：Peking University（北京大学）； Xiamen University（厦门大学）； National Taiwan University（国立台湾大学）； WHU（武汉大学）； THU / Jimei University（清华大学 / 集美大学）

作者：Zijie Meng, Ziwei Li, Yufei Liu, Zhiyu Li, Jiyuan Liu, Wenhua Nie, Bingcai Wei, Miao Zhang

英文摘要：Safe coordination in networked cyber-physical systems forces learning algorithms to simultaneously handle hybrid discrete-continuous actions, hard training-time safety constraints, and physics-governed dynamics. We show that these three features form a directed cycle of biases that defeats any naive composition of off-the-shelf modules, and formalize this as a three-way coupling lemma. We then introduce TRIDENT, the first MARL framework whose three components are co-designed to cancel each leak: a Richardson-Romberg gradient correction reducing Gumbel-Softmax bias from O(tau) to O(tau^2), a Lyapunov-constrained sequential trust-region update enforcing per-iterate feasibility, and a physics-informed residual critic that decomposes value rather than reward. We prove an O~(1/sqrt(K)) convergence rate to a constrained Nash equilibrium and an O(sqrt(K)) cumulative-violation bound. On multi-UAV mobile-edge computing, autonomous intersection management, and a hybrid SMAC variant, TRIDENT cuts training-time violations by 95.5% over MADDPG and 76.3% over MACPO, while improving reward by 13.5% over the strongest unconstrained baseline.

AI 总结： 提出Self-CTRL方法，通过强化学习优化语言模型自我解释与行为之间的一致性，在概率推理和宪法AI任务上显著提升一致性和安全性。

链接：https://arxiv.org/abs/2606.18327

机构：MIT CSAIL（麻省理工学院计算机科学与人工智能实验室）

作者：Itamar Pres, Laura Ruis, Melat Ghebreselassie, Belinda Z. Li, Jacob Andreas

英文摘要：Language models (LMs) that faithfully describe their own behavior can more easily be audited, understood, and trusted by users. This paper describes Self-Consistency Training with Reinforcement Learning (Self-CTRL), a method that optimizes for consistency between a LM's self-explanations and behavior on related inputs by updating explanations to better predict behavior or updating behavior to better match explanations. We apply our method in two domains. First, we study a formal probabilistic reasoning task in which LMs must learn to imitate a family of biased samplers and evaluated on their ability to report the associated biases. We find that consistency training improves the correlation between self-reported and behaviorally-measured latent biases from $R^2=0.24$ to $R^2=0.64$ on a set of held-out distributions, matching the generalization of direct ground-truth supervision. Second, we study a constitutional AI domain in which LMs must describe when they will refuse or comply with user requests. Here, Self-CTRL produces rules that faithfully describe the model's behavior on held-out requests, improving the refusal predictions of a third-party auditor model from $36\%$ to $92\%$. In the other direction, behavior updates improve alignment, reducing HarmBench failure rate from $15.0\%$ to $0.5\%$ without substantially increasing refusal on harmless prompts. By aligning explanations and behavior, our work provides a general recipe for training AI models to be safer, more transparent, and more controllable.

AI 总结：受神经科学启发，提出一种强化学习框架，利用局部线性嵌入捕捉状态局部结构，并通过注意力机制自适应融合动态与奖励特征，提升学习效率。

链接：https://arxiv.org/abs/2606.18469

机构：Mila – Quebec AI Institute（米拉-魁北克人工智能研究所）

作者：Somjit Nath, Jackson J Cone, Derek Nowrouzezahrai, Samira Ebrahimi Kahou

英文摘要：Neuroscientific research has revealed that the brain encodes complex behaviors by leveraging structured, low-dimensional manifolds and dynamically fusing multiple sources of information through adaptive gating mechanisms. Inspired by these principles, we propose a novel reinforcement learning (RL) framework that encourages the disentanglement of dynamics-specific and reward-specific features, drawing direct parallels to how neural circuits separate and integrate information for efficient decision-making. Our approach leverages locally linear embeddings (LLEs) to capture the intrinsic, locally linear structure inherent in many environments, mirroring the local smoothness observed in neural population activity, while concurrently deriving reward-specific features through the standard RL objective. An attention mechanism, analogous to cortical gating, adaptively fuses these complementary representations on a per-state basis. Experimental results on benchmark tasks demonstrate that our method, grounded in neuroscientific principles, improves learning efficiency and overall performance compared to conventional RL approaches, highlighting the benefits of explicitly modeling local state structures and adaptive feature selection as observed in biological systems.

AI 总结：提出量子退火增强Q学习框架，通过将Q值更新编码为QUBO问题并利用量子退火采样实现随机动作选择，解决高维非凸空间中的收敛问题，在C-MAPSS和工业数据集上显著优于基线方法。

链接：https://arxiv.org/abs/2606.18503

机构：Central University of Karnataka（卡纳塔克中央大学）； University College of Engineering, Anna University（安娜大学工程学院）； AIONOS India Pvt Ltd（AIONOS印度私人有限公司）； National Institute of Technology Tiruchirappalli（蒂鲁吉拉帕利国立理工学院）

作者：Manoranjan Gandhudi, Arunkumar V., G. R. Anil, Gangadharan G. R

英文摘要：Remaining useful life (RUL) estimation is central to predictive maintenance, where an unplanned failure can cost far more than the asset itself. Statistical degradation models miss the strong nonlinearity of real systems, and data-driven models often converge to suboptimal solutions in high-dimensional, non-convex search spaces. We propose a Quantum Annealing enhanced Q-Learning (QAQL) framework that couples the sampling behaviour of quantum annealing with the sequential decision making of Q-learning. Each Q-value update is encoded as a small quadratic unconstrained binary optimization (QUBO) whose ground state is the greedy action; rather than acting as a deterministic optimizer, the annealer returns a distribution over near-optimal actions across many reads, and this stochastic action selection supplies the exploration that curbs premature convergence on nonlinear degradation trajectories. The QUBO is solved on the D-Wave Advantage system using minor embedding, with the annealer woven into the reinforcement-learning loop rather than bolted on after training. We validate QAQL on two public benchmarks: the NASA C-MAPSS turbofan engine datasets and a device-fleet predictive maintenance dataset. Averaged over many independent runs and across six error metrics, QAQL outperforms the classical and quantum baselines considered in this study, with statistically significant improvements. The results indicate that quantum annealing is a usable, not merely theoretical, optimizer inside a reinforcement-learning loop for industrial predictive-maintenance applications.

AI 总结：提出GRID方法，从追求不同目标的异构示范者中提取通用奖励，训练通用智能体以学习环境通用能力，避免模式平均偏差，提升下游任务微调效率。

链接：https://arxiv.org/abs/2606.18537

机构：University of Washington（华盛顿大学）； NVIDIA（英伟达）

作者：Caleb Chang, Davin Win Kyi, Natasha Jaques, Karen Leung

英文摘要：Humans often acquire new skills by observing others, since observed behaviors implicitly reveal how to act in an environment. However, observations drawn from a heterogeneous population introduce conflicting behavioral signals, making it difficult to determine which behaviors are worth imitating. We address this challenge with General Reward Inference and Disentanglement (GRID), a social learning method that extracts universally useful behaviors from a heterogeneous population of demonstrators pursuing different goals. GRID decomposes per-agent reward functions into a general reward, capturing behaviors shared across all agents, and specific rewards, capturing individual preferences and objectives. Training exclusively on the general reward provides a new paradigm of generalist pretraining. It yields a generalist agent that internalizes universal environmental competencies, such as safety and basic task proficiency, without the mode-averaging bias that afflicts standard learning from demonstration techniques. This generalist serves as a superior prior for fine-tuning to downstream tasks, including preferences unseen during training. Experiments across a synthetic basis function decomposition, multi-agent Craftax, and a continuous autonomous driving simulator (Highway-Env) confirm that GRID successfully disentangles reward structure in a semantically meaningful way, outperforms standard learning from demonstration baselines, and enables more efficient and stable specialization.

AI 总结：提出首个任意时间多目标多臂老虎机算法Top-Two帕累托前沿汤普森采样(TTPFTS)，用于帕累托集识别，在合成环境和超大型分子库中验证有效性，并引入不确定性量化指标。

链接：https://arxiv.org/abs/2606.18785

机构：imec； Data Science Institute, Interuniversity Institute of Biostatistics and Statistical Bioinformatics, UHasselt（哈瑟尔特大学生物统计学与统计生物信息学跨大学研究所数据科学研究所）

作者：Lennert Saerens, Bram Silue, Eleni Litsa, Peter Vrancx, Pieter Libin

英文摘要：Identifying Pareto optimal solutions is critical to support multi-objective decision-making. We introduce the first anytime Multi-Objective Multi-Armed Bandit algorithm for the Pareto Set Identification problem, taking a Bayesian approach: Top-Two Pareto Front Thompson Sampling (TTPFTS). We benchmark TTPFTS against state-of-the-art fixed-budget Pareto Set Identification algorithms on synthetic environments. Next, we demonstrate its practical utility in a challenging multi-objective molecular discovery setting by efficiently exploring an ultra-large synthesis-on-demand molecular library. Furthermore, we introduce a novel uncertainty quantification metric that estimates our algorithm's confidence in the predicted Pareto set. We demonstrate that this metric effectively proxies true performance, yielding a robust methodology for monitoring learning progress in complex settings. Finally, we complement these empirical findings with a theoretical proof of the algorithm's asymptotic correctness.

AI 总结：提出SC-GRPO方法，利用自条件化分布间的KL散度作为GRPO梯度的乘性权重，实现细粒度信用分配，在数学、代码和智能体任务上平均提升8.1%。

链接：https://arxiv.org/abs/2606.18810

机构：Beijing Institute of Technology（北京理工大学）； Beihang University（北京航空航天大学）； Independent Researcher（独立研究者）

作者：Yingyu Shan, Yuhang Guo, Zihao Cheng, Zeming Liu, Xiangrong Zhu, Xinyi Wang, Jiashu Yao, Wei Lin, Hongru Wang, Heyan Huang

英文摘要：Reinforcement learning with verifiable rewards (RLVR) has driven substantial progress in training LLMs for reasoning tasks, but representative methods such as GRPO assign uniform credit across all tokens, wasting gradient on routine tokens while under-crediting pivotal reasoning steps. Existing token-level credit assignment methods require resources beyond the model's own rollouts. GRPO variants rely on process reward models or ground-truth answers. Knowledge distillation assigns credit through per-token divergence but requires external teachers (On-Policy Distillation) or privileged information (On-Policy Self Distillation). However, these dependencies limit applicability in the pure RLVR setting. We observe that conditioning the model on its own verified trajectories induces a measurable per-token KL divergence between the original and conditioned distributions, and prove that distilling from a self-teacher constructed by verified trajectories leads to infeasible weighted-average solutions when multiple verified trajectories exist. We propose SC-GRPO (Self-Conditioned GRPO), which uses KL divergence mentioned before as a multiplicative weight on GRPO gradients. Across five benchmarks spanning math, code, and agentic tasks, SC-GRPO consistently outperforms 8.1% over GRPO and 5.9% over DAPO with stronger OOD performance. Moreover, SC-GRPO achieves higher performance than OPD.

AI 总结：提出通过合成MDP构建强化学习基础模型，利用固定大小的充分统计量使注意力架构适用，在线和离线实验均优于传统算法。

链接：https://arxiv.org/abs/2606.18812

机构：École normale supérieure de Paris, PSL University, Paris, France（巴黎高等师范学院，PSL大学，法国巴黎）； Soda team, Inria Saclay, Palaiseau, France（Soda团队，法国国家信息与自动化研究所萨克雷中心，法国帕莱索）

作者：Abdelrahman Zighem, Jill-Jênn Vie

英文摘要： Foundation models for language and vision are powered by internet-scale data, while structured domains (tabular prediction, time-series forecasting, graph learning, reinforcement learning) are not. The substitute is synthetic data, which shifts the burden from collection to prior design. Such priors already exist for many structured tasks: TabPFN and its successors solve tabular classification with a transformer pretrained on a synthetic Bayesian prior. We make two points. \textbf{First}, reinforcement learning is the conspicuous gap: sampling a synthetic MDP is as feasible as sampling a synthetic tabular dataset, yet no in-context RL work treats prior design as a primary objective. \textbf{Second}, MDPs admit a fixed-size sufficient statistic, independent of the episodes observed and tabular in shape, which makes them directly amenable to the attention-based architectures used for tabular foundation models, with a policy head replacing the supervised target. Together these define the agenda for an RL foundation model. As a proof of concept, we train one model entirely on synthetic MDPs and show that, with no task-specific tuning, it solves held-out tabular benchmarks in context, both online and offline: online, in far fewer episodes than UCB-VI and tabular Q-learning, and offline, competitively with VI-LCB.

AI 总结：针对决策过程中信息增加与动作集缩小的不对称性，提出成熟马尔可夫决策过程（MMDP）框架，并基于过期动作优先级原则开发结构感知强化学习方法，实验证明其能提升学习效率。

链接：https://arxiv.org/abs/2606.18820

机构：Ant International（蚂蚁国际）； School of Economics, Sichuan University（四川大学经济学院）； School of Economics, Fudan University（复旦大学经济学院）

作者：Jiaxi Liu, Aiping Yang, Yuhang Yang, Shuqi Zhang, Zewei Dong, Jiangming Yang, Xuebin Chen

英文摘要：Sequential decision problems often exhibit an asymmetric evolution of information and decision flexibility: as a decision cycle unfolds, the agent receives richer information while feasible actions expire due to operational cutoffs, commitments, or resource constraints. Standard MDP formulations typically flatten this structure into stage-dependent state descriptions and action masks, thereby obscuring the nested information--action asymmetry that determines which decisions are urgent and which can be deferred. We introduce Maturing Markov Decision Processes (MMDPs), a formulation built around this information--action asymmetry. We characterize one of its key consequences through an expiring-action priority principle, which identifies the actions that must be resolved before the next stage. Motivated by this structure, we develop a structure-aware reinforcement learning framework with stage-aware policy design, expiring-action abstraction, and search-augmented learning with distillation. Experiments on a controlled multi-supplier replenishment problem, simplified cash-management environments of increasing complexity, and a production-scale simulator show that explicitly modeling this asymmetry improves learning efficiency and becomes increasingly valuable as decision problems scale.

AI 总结：提出REVES框架，通过将中间步骤的“接近正确”答案转化为解耦的修订和验证提示，实现高效的离策略数据生成，提升大语言模型的多步推理能力，在LiveCodeBench上比强化学习基线高6.5分。

链接：https://arxiv.org/abs/2606.18910

机构：Northwestern University（西北大学）； Amazon AGI（亚马逊人工智能实验室）； Qualcomm AI Research（高通人工智能研究）； University of Minnesota（明尼苏达大学）

作者：Yuanxin Liu, Ruida Zhou, Xinyan Zhao, Amr Sharaf, Hongzhou Lin, Arijit Biswas, Mohammad Ghavamzadeh, Zhaoran Wang, Mingyi Hong

英文摘要：Test-time scaling via sequential revision has emerged as a powerful paradigm for enhancing Large Language Model (LLM) reasoning. However, standard post-training methods primarily optimize single-shot objectives, creating a fundamental misalignment with multi-step inference dynamics. While recent work treats this as multi-turn reinforcement learning (RL), conventional approaches optimize over the multi-step trajectories directly, failing to further exploit the high-quality mistakes in intermediate steps that model can learn from correcting them. We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization. By converting the intermediate steps (``near-miss'' answers) in the successful recovery trajectories into decoupled revision and verification prompts, our approach concentrates training on both effective answer transformation and error identification. This approach enables efficient off-policy data generation and reduces the computational overhead of long-horizon sampling compared to standard multi-turn RL. On LiveCodeBench, using publicly available test cases as feedback, we observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training. Beyond coding, our approach matches the previously reported SOTA result on circle packing while using the smallest base model (4B) and far fewer rollouts than the much larger evolutionary search systems. Math results under ground-truth verification further confirm improved correction ability. It also generalizes to out-of-distribution constraint-satisfaction puzzles such as n\_queens and mini\_sudoku, where correctness is defined entirely by problem constraints. Code is available at this https URL.

AI 总结：提出OHIRL框架，在无标量奖励下通过固定通道感知流进行在线奖惩学习，利用内部轨迹评估器推断感知维度的效价，在XOR任务和CartPole等控制任务中达到高准确率。

链接：https://arxiv.org/abs/2606.18963

机构：Zirong Li（李 Cirong）

作者：Zirong Li

英文摘要： We study online reward-punishment learning when the environment provides no scalar reward or evaluative label. At each step the agent receives only a fixed-channel perceptual packet, and quantities such as pain, energy, contact, damage, or cognitive error are treated as perceptual dimensions whose valence must be inferred from transition consequences. OHIRL separates four roles: M_psi learns next-packet prediction, D_omega models residual dynamics, C_eta is a fixed internal post-transition trajectory evaluator, and B_xi learns to use the resulting value evidence for later policy updates and action scoring. C_eta uses a recovery-positive and persistence/growth-negative residual-regulation orientation; a coefficient-origin audit shows that equal-unit, raw-equal, and random monotone variants preserve more than 92% of the released top-action rankings, while sign inversion preserves 0%. The reward-free protocol exposes observation transitions while withholding environment rewards, delayed external evaluators, success labels, and action-goodness labels. A conditional error decomposition separates B_xi evidence-estimation error from residual policy-optimization error. In a 2x2-XOR packet task, medicine and chili acquire opposite value under visual XOR contexts, and the same pain or spice increase can be positive or negative depending on consequence structure; B_xi reaches 0.952 balanced reward-sign accuracy. In a full online-interleaved audit, M_psi reaches holdout R2=0.907, B_xi reaches 0.940 sign accuracy, and the policy reaches 0.979 optimal-action accuracy, while immediate packet scores, prediction-error rewards, shuffled targets, zero reward, and error-reduction controls collapse. Hidden-reward CartPole and Taxi controls, public-context no-leakage audits, and module-role ablations further test information boundaries and component necessity.

AI 总结：提出PQLRM算法，结合帕累托Q学习和奖励机，在多目标强化学习中高效逼近帕累托前沿，并处理非马尔可夫奖励。

链接：https://arxiv.org/abs/2606.19134

机构：Linköping University, Sweden（瑞典_linköping大学）； Univ. Lille, CNRS, Centrale Lille, UMR 9189 CRIStAL, F-59000 Lille, France（法国里尔大学、CNRS、中央里尔学院、UMR 9189 CRIStAL、法国里尔）； Univ. Toulouse, INRAE-MIAT, Toulouse, France（法国图卢兹大学、INRAE-MIAT、图卢兹）

作者：Arnaud Lequen, Clément Legrand-Lixon, Léo Saulières

英文摘要：We present Pareto Q-Learning with Reward Machines (PQLRM), a multi-objective reinforcement learning algorithm for tasks whose reward structure is specified by a set of reward machines (RMs). PQLRM combines Pareto Q-Learning (PQL), which maintains sets of vector-valued Q-estimates to approximate the Pareto front, with enhancements from Q-Learning with Reward Machines (QRM), which exploits the factored automaton structure of the reward signal. This yields a multi-policy algorithm that remains sample-efficient under non-Markovian, RM-encoded rewards. Experimental trials show that PQLRM converges faster than a naive PQL baseline applied to the cross-product MDP and can synthesize Pareto-optimal policies that QRM cannot.

AI 总结：针对电动汽车充电中离开时间未知导致强化学习策略效果差的问题，提出面向决策的强化学习框架，联合训练预测器与控制器，实现端到端优化，使总奖励提升14%，未供应能量减少55%。

链接：https://arxiv.org/abs/2606.19199

机构：Ghent University -- imec（根特大学 -- imec）

作者：Giuseppe Gabriele, Fabio Pavirani, Seyed Soroush Karimi Madahi, Chris Develder

英文摘要：The recent growth of EV adoption poses challenges for power systems, including increased peak demand and potential grid instability. Smart control of EV charging -- e.g., based on reinforcement learning (RL) -- can alleviate these issues by learning temporal and contextual patterns from historical data. Yet, in real-world scenarios, key features, such as departure time, often are unavailable. This, in turn, makes it harder for an RL agent to learn and execute an effective charging policy. To mitigate this uncertainty, a trained forecaster can approximate the unknown features from available data. However, since these forecasting models are typically trained for accuracy (rather than their impact on a downstream agent's decision quality), their errors may propagate and hinder the overall performance of a controller that is using the forecasts. To avoid this, we propose a decision-focused RL (DF-RL) framework in which the forecaster is trained end-to-end, i.e., with feedback from the charging policy actions taken by the RL agent. Such joint training of both the forecaster and controller ultimately results in higher-quality actions: our proposed DF-RL method yields superior charging decisions compared to other baselines, achieving up to a 14% improvement in total reward and a 55% reduction of unsupplied energy (i.e., charging that failed to happen because the EV already left), relative to the RL method without departure time forecasting.

AI 总结：针对GRPO等RL算法中策略熵崩溃问题，提出STARE方法，通过惊讶度分位数识别熵关键令牌并重加权其优势，结合目标熵闭环门控稳定熵，在1.5B-32B模型和多种任务上实现稳定训练，AIME24/25准确率提升4%-8%。

链接：https://arxiv.org/abs/2606.19236

机构：Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生院）； Tencent Hunyuan（腾讯混元）

作者：Haipeng Luo, Qingfeng Sun, Songli Wu, Can Xu, Wenfeng Deng, Han Hu, Yansong Tang

英文摘要：Reinforcement Learning with Verifiable Rewards algorithms like GRPO have emerged as the dominant post-training paradigm for complex reasoning in LLMs, yet commonly suffer from policy entropy collapse during training. We conduct a first-order gradient analysis of token-level entropy dynamics under GRPO and identify a token-level credit assignment mismatch: the per-token entropy variation decomposes into the product of the trajectory-level advantage and an entropy sensitivity function over the next-token distribution, yielding an advantage-surprisal four-quadrant structure and a near-criticality property. Motivated by it, we propose STARE (Surprisal-guided Token-level Advantage Reweighting for policy Entropy stability), which identifies entropy-critical token subsets via batch-internal surprisal quantiles, selectively reweights their effective advantages, and incorporates a target-entropy closed-loop gate for stable entropy regulation. Across model scales from 1.5B to 32B and three task families (Short CoT, Long CoT, and Multi-Turn Tool Use), STARE sustains stable RL training over thousands of steps while maintaining policy entropy within the target band. On AIME24 and AIME25, STARE outperforms DAPO and other competitive baselines by 4%-8% in average accuracy, with reflection tokens and response length growing in tandem, indicating sustained exploration-exploitation balance that further unlocks RL training this http URL is available at this https URL.

AI 总结：提出UBP2方法，通过联合推理奖励、动力学和值函数的不确定性来主动引导探索，在Meta-World基准上显著提高了样本效率。

链接：https://arxiv.org/abs/2606.19328

机构：Learning, Embodied Autonomy, and Forecasting (LEAF) Lab, University of Toronto（多伦多大学学习、具身自主与预测（LEAF）实验室）

作者：Mohamed Nabail, Leo Cheng, Jingmin Wang, Nicholas Rhinehart

英文摘要：Preference-based RL provides an approach to learning reward models from pairwise comparisons of behaviors, bypassing the need for explicit reward design. However, existing methods typically rely on passive data collection and suffer from poor sample efficiency, especially during the early stages of learning. We introduce a model-based approach that actively directs exploration by jointly reasoning over uncertainties in the reward, dynamics, and value functions. Our method, Uncertainty-Balanced Preference Planning (UBP2), uses ensembles of reward, dynamics, and value function models to evaluate candidate trajectories according to a unified score that combines expected reward, terminal value, and epistemic uncertainty. Planning under this objective yields an explicit tradeoff between exploitation and information acquisition without requiring ad hoc exploration heuristics. Under standard regularity assumptions, we establish sublinear regret guarantees for both finite-horizon and infinite-horizon settings. Empirically, experiments on the Meta-World benchmark show UBP2 achieves substantially higher sample efficiency than model-free preference-based methods and non-optimistic model-based baselines.

AI 总结：提出概念调制模型（CMMs），通过属性势统一条件潜变量模型的可识别性与外推分析，将基于转移的可识别性提升至条件设置，并导出代数外推准则。

链接：https://arxiv.org/abs/2606.18509

机构：Department of Statistics and Data Science, Carnegie Mellon University（卡内基梅隆大学统计与数据科学系）； Machine Learning Department, Carnegie Mellon University（卡内基梅隆大学机器学习系）

作者：Soheun Yi, Yizhou Lu, Chandler Squires, Pradeep Ravikumar

英文摘要：Reliable generalization in conditional latent variable models requires understanding both identifiability and extrapolation: how observed variation across attributes determines latent structure, and how that structure determines distributions at unseen attributes. However, existing identifiability and extrapolation guarantees are largely model-specific, with separate analyses in nonlinear ICA, causal representation learning, perturbation modeling, and related conditional latent variable models. We introduce concept modulation models (CMMs), an attribute-indexed class of conditional generative models with structure $A\to \Lambda \to C\to X$, where attributes select modulators, modulators induce latent concept laws, and concepts generate observed features. CMMs lift transition-based identifiability to conditional settings by showing that feature agreement on observed attributes induces a latent concept transition constrained by the CMM class. We express these constraints through attribute potentials, log-density ratios between attribute-conditioned concept laws, separating the generic lifting step from model-specific rigidity arguments. The same potentials control extrapolation: agreement at unseen attributes holds exactly when the transported attribute-potential identities extend to those attributes. This yields algebraic extrapolation criteria, identifies the common potential-based proof objects behind several existing identifiability and extrapolation results, and, when combined with the model-specific rigidity arguments in those works, recovers their stated conclusions.

AI 总结：针对现实世界中稀疏、不规则采样的多元时间序列，提出基于潜在随机微分方程的生成方法，将观测投影到连续时间随机动力系统，处理缺失和不规则采样，并捕获循环行为，在六个基准数据集上取得最优结果。

链接：https://arxiv.org/abs/2606.18898

机构：Josef Ressel Centre for Intelligent and Secure Industrial Automation, University of Applied Sciences, Salzburg, Austria（约瑟夫·雷塞尔智能与安全工业自动化中心，萨尔茨堡应用科学大学，奥地利）； University of Salzburg, Austria（萨尔茨堡大学，奥地利）

作者：Martin Uray, Dominik Geng, Florian Graf, Stefan Huber, Roland Kwitt

英文摘要：Multivariate time series anomaly detection (MTSAD) is critical for a wide range of application areas, such as industrial monitoring, cybersecurity, or healthcare. Real-world data is often sparse, irregularly sampled or partially observed, yet existing methods assume uniformly sampled time series. We propose a generative approach based on Latent SDEs that projects the observed time series on a continuous-time stochastic dynamical system, directly being able to handle missing observations and irregular sampling, while also naturally capturing possible cyclic behavior that many real-world use cases inherently possess. Experiments on six anomaly benchmark datasets show that our proposed method ranks first among state-of-the-art baselines. We further demonstrate that our method remains robust under severe data sparsity, while performance significantly degrades for the tested baseline methods. These results highlight latent SDEs as a natural inductive bias for anomaly detection in multivariate time series, especially in presence of real-world irregularities.

AI 总结：提出首个基于扩散模型的神经估计器 DIPHINE，用于计算连续非高斯动力系统的集成信息分解（$\Phi$ID），通过单个摊销网络联合估计所有互信息项，并利用 Möbius 逆变换恢复十六个原子。

链接： https://arxiv.org/abs/2606.18997

机构：KAUST（卡塔尔科学与技术部）； EURECOM（欧雷康）

作者：Simon Pedro Galeano Munoz, Mustapha Bounoua, Giulio Franzese, Pietro Michiardi, Maurizio Filippone

英文摘要：Uncovering the true informational architecture of real-world complex systems requires disentangling how their components uniquely store, redundantly share, and synergistically integrate information over time. Integrated Information Decomposition ($\Phi$ID) is a framework for decomposing the information dynamics of multivariate systems into sixteen non-overlapping atoms that characterize redundant, unique, and synergistic modes of information storage, transfer, and integration. Existing methods to compute $\Phi$ID are restricted to Gaussian or discrete systems, preventing its application to continuous non-Gaussian dynamical systems. We address this limitation by proposing DIPHINE (Diffusion-based $\Phi$-ID Neural Estimator), the first neural estimator that leverages score-based diffusion models to jointly estimate all the mutual information terms required by $\Phi$ID from a single amortized network, recovering the sixteen atoms through Möbius inversion. We provide a theoretical analysis of error propagation through the inversion, showing that the Jacobian of the mapping from mutual informations to atoms is integer-valued and that the synergy-to-synergy atom is provably the hardest to estimate. We demonstrate accurate recovery of ground-truth atoms on synthetic benchmarks, superior performance compared to established mutual information estimators, and the ability to extract physiologically interpretable information-dynamic structure on an application involving real data without any distributional assumptions.

AI 总结：针对流匹配模型因损失函数与样本质量不匹配导致的视觉缺陷，提出判别器引导的强化学习（DRL），利用预训练空间中判别器的logit作为奖励，显著提升无引导FID和语义FD，并改善偏好对齐。

链接：https://arxiv.org/abs/2606.19162

机构：FAIR at Meta（Meta FAIR）； Columbia University（哥伦比亚大学）； McGill University（麦吉尔大学）； Canada CIFAR AI Chair（加拿大CIFAR人工智能主席）

作者：Nicolas Beltran-Velez, Felix Friedrich, Zhang Xiaofeng, Reyhane Askari-Hemmat, Xiaochuang Han, Adriana Romero-Soriano, Michal Drozdzal

英文摘要：Score- and flow-matching models often rely on preference-based reinforcement learning for two purposes: aligning with subjective preferences and, surprisingly, recovering properties such as visual realism and coherent object structure that matching-based training is intended to learn from the data itself. We argue that this reflects a structural mismatch. Matching losses measure $\ell_2$ regression error on the velocity or score field under training-time marginals, a proxy poorly aligned with the visual and semantic properties that determine sample quality at inference. Given a reward aligned with these properties, RL sidesteps the mismatch by evaluating the model on its own samples and following the reward landscape directly. The challenge is to obtain such a reward without relying on human preferences, which are expensive and conflate data realism with annotator inclinations. We propose Discriminator-Guided RL (DRL). DRL trains a discriminator to separate data from base-model samples in a pretrained representation space and uses its logit as the reward in KL-regularized RL. The pretrained space restricts the discriminator to perceptually meaningful directions, and the logit estimates the log-likelihood ratio between data and model, which is the optimal reward for targeting the data distribution. Across SiT, JiT, REPA, and RAE, DRL reduces guidance-free FID (e.g., $9.38 \to 2.62$ on SiT) and semantic-space FD (e.g., $88.2 \to 19.3$ on DINOv3 for SiT), with consistent gains across all backbones, and improves human-preference rewards without training on them. It also yields a better Pareto frontier between preference reward and image fidelity under subsequent preference-based post-training, increasing alignment while reducing low-level artifacts such as oversaturation and excessive brightness.

AI 总结：提出大语言吉布斯方法，利用大语言模型的条件分布作为转移算子进行结构化概率推理，通过迭代重采样变量避免顺序偏差，在合成分布、一致性推理和贝叶斯结构学习中验证有效性。

链接：https://arxiv.org/abs/2606.19264

机构：University of Edinburgh, School of Informatics（爱丁堡大学信息学院）

作者：Sanghyeok Choi, Henry Gouk, Esmeralda S. Whitammer

英文摘要：The knowledge encoded in large language models (LLMs) can serve as a substrate for structured reasoning over variables describing a complex world, but accessing this knowledge in a probabilistically coherent manner poses a difficult inference problem. We propose Large Language Gibbs, a scheme for structured probabilistic inference that uses conditional distributions of an LLM as transition operators. Rather than sampling structured objects through single-pass autoregressive generation, we iteratively resample individual variables conditioned on others using an LLM's next-token conditionals. This approach avoids order-dependent biases and produces a stationary distribution that reflects a compromise between all local conditionals. We apply this approach to sampling from synthetic distributions, consistent reasoning tasks, and Bayesian structure learning. The results suggest that the use of LLM conditionals in MCMC is a practical alternative to one-pass generation for structured probabilistic inference under a world prior accessible through noisy LLM conditionals.

AI 总结：提出Diffusion-Proof框架，首次将扩散语言模型应用于形式定理证明，通过全证明生成和局部校正方法，在ProofNet和MiniF2F上分别提升1.61%和6.14%，并解决了一个DeepSeek-Prover-V2-7B无法解决的IMO问题。

链接：https://arxiv.org/abs/2606.19315

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

作者：Ruida Wang, Rui Pan, Pengcheng Wang, Shizhe Diao, Tong Zhang

英文摘要：Enhancing the formal math reasoning capabilities of Large Language Models (LLMs) has become a key focus in both mathematical and computer science communities in recent years. While significant progress has been made in using state-of-the-art Auto-Regressive (AR) LLMs for formal theorem proving, these models suffer from inherent limitations. Their next-token prediction generation methods may yield suboptimal performance due to the challenges of long-range coherence and the compounding of errors over long sequences. Recent advancements in diffusion LLMs (dLLMs), which generate text through iterative denoising of a multi-token block, offer a promising alternative. However, the application of dLLMs to formal mathematics, where maintaining long-range coherence is critical, remains largely understudied. To address the challenges above, we propose **Diffusion-Proof**, to the best of our knowledge, the first framework to train and apply dLLMs for formal theorem proving. Our frameworks contain training and inference methods for two models. The first one is *dLLM-Prover-7B*, which performs whole-proof writing with long-range coherent tactic usage. The second one is *dLLM-Corrector-7B*, which is a novel large block diffusion-based correction model. It leverages the in-filling capabilities of dLLMs to perform local proof correction using bi-directional information. Extensive experiments demonstrate that **Diffusion-Proof** relatively significantly outperforms the AR LLM baseline trained under the same dataset. **Diffusion-Proof** achieves an absolute improvement of **1.61%** on ProofNet-Test and **6.14%** on MiniF2F-Test benchmarks compare to the baseline. Notably, **Diffusion-Proof** successfully resolves one IMO problem that more advanced thinking model DeepSeek-Prover-V2-7B could not solve, showcasing the unique advantage of dLLMs in formal theorem proving.

AI 总结：本文通过微分几何、李群和流体力学，建立了冲击波理论与对称商化随机梯度下降学习动力学之间的显式数学联系，并应用于多种神经网络架构。

链接：https://arxiv.org/abs/2606.18303

机构：NEC Corporation（NEC公司）

作者：Taiki Miyagawa

英文摘要：We develop a mathematically explicit link between shock-wave theory and the symmetry-quotiented learning dynamics of stochastic gradient descent, drawing on differential geometry, Lie group theory, and fluid mechanics. Specifically, after quotienting parameter symmetries and applying local-entropy coarse-graining, the effective dynamics satisfy a viscous Hamilton--Jacobi equation on the quotient manifold. Moreover, under the assumption that the raw parameter dynamics can be summarized by a gradient field on the quotiented space, the gradient of the coarse-grained loss function obeys a Burgers-type equation, and shock formation can be established rigorously. We apply our theory to multilayer perceptrons, convolutional neural networks, Transformers, and mean-field networks, and show that they obey the Hamilton--Jacobi or Burgers-type equations. We conjecture that this framework also yields practical diagnostics for deep learning. In architectures such as Transformers, raw parameter norms are often distorted by symmetry redundancy and may therefore be misleading, whereas symmetry-corrected quotient observables provide a principled basis for monitoring, forecasting, and controlling training-phase transitions.

AI 总结：提出Fisher宽度作为统计流形上高斯宽度的类比，利用Fisher信息度量局部几何，并证明其保持高斯宽度的关键性质，应用于Fisher-Lipschitz假设类的泛化界。

链接：https://arxiv.org/abs/2606.18306

机构：Department of Mathematics, FPT University（FPT大学数学系）

作者：Vu Khac Ky

英文摘要：Gaussian width is a central geometric complexity measure in high-dimensional probability, compressed sensing, convex optimization, and learning theory. It quantifies the average extent of a set along random directions, thereby capturing the effective dimension of constraint sets, hypothesis classes, and descent cones. However, this notion is intrinsically Euclidean. Statistical models instead carry a natural Riemannian geometry induced by the Fisher information metric, where directions are scaled according to statistical distinguishability rather than ambient Euclidean length. We introduce Fisher width, a Fisher-geometric analogue of Gaussian width for statistical manifolds. At a parameter point $\theta$, Fisher width replaces the Euclidean identity by the local metric tensor $G(\theta)^{1/2}$, measuring the Gaussian width of the Fisher-rescaled set. This makes the resulting quantity sensitive to local statistical curvature and invariant under smooth reparameterizations. We develop the basic theory of Fisher width, showing that it retains key structural features of Gaussian width, including concentration, metric perturbation stability, and spectral comparison bounds with the Euclidean baseline, while also capturing anisotropic geometric effects invisible to Euclidean measures. As an application, we prove a generalization bound for Fisher-Lipschitz hypothesis classes and propose computable estimators, which we evaluate empirically on MNIST across three model classes. Fisher width is to statistical manifolds what Gaussian width is to Euclidean convex bodies. This work lays the foundation for studying complexity and learning on curved statistical manifolds.

AI 总结：本文指出，在生物医学表格数据中，测量噪声会削弱非线性结构，导致非线性模型与线性模型性能相当，并提出了一个精确的超额风险恒等式，揭示了测量可靠性、样本量和特征表示三个条件必须同时满足才能体现非线性优势。

链接：https://arxiv.org/abs/2606.18420

机构： Hertie Institute for AI in Brain Health, University of Tübingen（赫蒂人工智能脑健康研究所，图宾根大学）； Tübingen AI Center, University of Tübingen（图宾根人工智能中心，图宾根大学）； Department of Psychiatry and Neurosciences, Charité – Universitätsmedizin Berlin（精神病学与神经科学系，柏林夏里特医学院）； Bernstein Center for Computational Neuroscience, Berlin（伯恩斯坦计算神经科学中心，柏林）； German Center for Mental Health (DZPG), partner site Tübingen（德国心理健康中心（DZPG），图宾根合作站点）

作者：Marc-Andre Schulz, Kerstin Ritter

英文摘要：On biomedical tabular data, flexible models such as deep networks, gradient-boosted trees, and kernel methods are repeatedly matched or beaten by linear and logistic regression given the same features. The usual reaction is to treat this as a model-side shortfall, to be fixed with more data, a better architecture, or tuning, on the assumption that the nonlinear structure is there and the model has failed to capture it. We argue that these fixes cannot help when the binding limit is the measurement rather than the model, as it frequently is in biomedicine. Additive noise blurs the population-optimal predictor, and because blurring removes a function's fine, rapidly varying detail before its broad shape, it erases nonlinear structure faster than linear structure. A degree-$k$ interaction is attenuated by the $k$-th power of feature reliability, while the linear part is attenuated only once. At the reliabilities typical of biomedical measurement, the nonlinear advantage can vanish even when the underlying biology is strongly nonlinear, and what the noise removes cannot be recovered by a larger cohort or a more flexible model, only by better measurement. The nonlinearity is hidden, not absent, and a tie between linear and flexible models is not by itself a verdict on the biology. These pieces are classical, drawn from measurement-error statistics, psychometrics, and Gaussian analysis, and we assemble them into an exact excess-risk identity. Measurement reliability is one of three conditions, alongside sample size and feature representation, that must align for a flexible model to help, and together they leave only a narrow window that most biomedical tasks fall outside. Across 140 UK Biobank tasks, the gap between flexible and linear models, where it exists, carries the predicted noise signature, and the three conditions can be separated by intervention but not by a benchmark alone.

AI 总结：本文通过固定权重范数并改变输出温度，发现Grokking延迟主要由对数尺度（logit scale）决定，权重范数仅通过影响对数尺度间接起作用。

链接：https://arxiv.org/abs/2606.18465

机构：H&K Research Studio, Clevix LLC

作者：Truong Xuan Khanh

英文摘要：Grokking, the delayed jump from memorization to generalization, is usually tied to the weight norm: a smaller norm generalizes sooner. We ask what the norm actually controls. Holding the weight norm fixed by clamping and varying only an output temperature, we slide the grokking delay across its entire norm-induced range under cross-entropy; matching the effective logit scale back to baseline recovers about 85% of the delay at two moduli. Across a grid of norms and temperatures the delay collapses onto the logit scale alone (R2 = 0.97), with the norm adding 1-2% beyond it. The effect is loss-dependent: under mean-squared error the logit scale is pinned and the norm acts through a different route. A memorization control, a float64 softmax-collapse audit, and a no-LayerNorm transformer point to the same channel. Forking arms from one identical state, the delay follows the held norm value and not the clamp operation, which closes a rescaling-artifact concern. The proximal variable is the logit scale and the softmax saturation it drives; the weight norm is only an upstream handle. All numbers, tables, and figures reproduce from released code and data.

AI 总结：研究神经网络中多语义性源于叠加现象，通过数学分析稀疏输入下自编码器的L2重构损失上下界，验证并扩展了Elhage等人的实证结果。

链接：https://arxiv.org/abs/2606.18538

作者：Mriganka Basu Roy Chowdhury, Eric McLaughlin Weiner

英文摘要：One of the major difficulties in the mechanistic interpretability of neural networks is the occurrence of polysemanticity, which suggests that each neuron is typically responsible for multiple different tasks, impeding a clean interpretation of their function. The seminal paper of Elhage et al. (2022) argues that this occurs due to superposition, a phenomenon where the neural network represents distinct features as non-orthogonal directions in a lower-dimensional space, a strategy that allows much greater compression of the data without sacrificing fidelity due to the feature sparsity of input vectors. Elhage et al. (2022) empirically validates these hypotheses in a rather natural and simple autoencoder with sparse inputs. The contribution of the present work is to analyze the mathematical basis for the occurrence and optimality of superposition, while rigorously corroborating some of their findings. In particular, we provide upper and lower bounds for the L2 reconstruction loss, tight in the very sparse regime, for power activation functions. A short list of interesting open problems are also included at the end.

AI 总结：针对非平稳流中的在线分布预测问题，提出一种基于潜在簇几何的吉布斯准后验方法，通过可逆跳跃MCMC采样变维后验，并引入重启变体应对漂移，在亚线性腐败预算和运输代价下实现亚线性Wasserstein遗憾。

链接：https://arxiv.org/abs/2606.18778

作者：Navyansh Mahla, Prateek Chanda, Ganesh Ramakrishnan

英文摘要： Online learning in non-stationary streams is often formulated as tracking a point estimate, but many applications require predicting the full data-generating distribution. We study online distributional prediction under drift and adversarial corruption. Our approach represents each candidate law through a latent cluster geometry: a variable-size configuration of centers that organizes probability mass and induces a predictive distribution. A Gibbs quasi-posterior over these configurations yields an online predictor by posterior averaging, and the resulting variable-dimensional posterior can be sampled with reversible-jump MCMC. The method therefore avoids specifying a parametric streaming law while retaining a structured latent space for uncertainty, regularization, and comparison. We evaluate performance by cumulative Wasserstein-1 regret against the time-varying true law. The analysis separates two effects: corruption perturbs the loss-based posterior update, whereas drift makes long-horizon posterior memory stale. We address the latter with a restarted variant that temporally localizes the same quasi-Bayesian update. The resulting high-probability bounds decompose into a PAC-Bayesian complexity term, a corruption-sensitive posterior perturbation term, and a dynamic optimal-transport term driven by \(A_T^{\mathrm{OT}}=\sum_{t=2}^T W_2^2(p_{t-1}^*,p_t^*)\). Under bounded support, stable latent geometry, predictive-map regularity, oracle realizability, localized restart windows, sublinear transport action, and sublinear corruption budget, the restarted predictor achieves sublinear cumulative Wasserstein regret. These guarantees require no parametric model for the stream, drift mechanism, or corruption process.

AI 总结：提出利用环境间机制变化识别隐藏混淆和选择偏差，基于互信息构建可检验准则，并设计StruBI算法，在合成和真实数据上显著优于现有方法。

链接：https://arxiv.org/abs/2606.18834

作者：Praharsh Nanavati, Jilles Vreeken, David Kaltenpoth

英文摘要：Causal discovery methods commonly assume that all data is independently and identically distributed (i.i.d.) and that there are no unmeasured variables affecting the system. In practice, these assumptions are often violated, leading to inaccurate inference. In this paper, we study how to identify hidden confounding and selection biases from causal mechanism shifts. In particular, we show that structural biases lead to dependent mechanism shifts. That is, by considering for which variables the mechanisms change given data from different environments, we can tell which variables are unbiased, which are subject to hidden confounding, and which are undergoing selection bias. We formalize this into an empirically testable criterion based on mutual information, and show under which conditions it identifies structural biases. To tell which nodes are subject to what kind of bias, we introduce the StruBI algorithm. Experiments on synthetic and real-world data show that StruBI works well in practice, accurately recovering affected variable sets and types of biases, outperforming the state-of-the-art by a wide margin.

AI 总结：本文通过从布尔可满足性问题归约证明二值化神经网络的可满足性是NP完全的，并利用均匀遮挡导致的网络输出分段常数结构，提出多项式时间鲁棒性检查算法。

链接：https://arxiv.org/abs/2606.18918

机构：Indian Institute of Technology Goa（印度理工学院Goa）

作者：Harshit Goyal, Sudakshina Dutta

英文摘要：This paper studies the computational complexity of verification problems for Binarized Neural Networks (BNNs), where activations (and sometimes weights) are binary. We analyze two problems: satisfiability and robustness under uniform image occlusion. We show that BNN satisfiability is NP-complete via a reduction from Boolean satisfiability problem (SAT), and that uniform occlusion induces a piecewise-constant structure in the network output, enabling a polynomial-time robustness-checking algorithm.

AI 总结：本文对稀疏混合专家模型中的不连续性进行几何与随机分析，分类不连续阶数，建立渐近体积估计，证明随机路径几乎必然击中一阶不连续，并提出低开销平滑机制以提升性能。

链接：https://arxiv.org/abs/2606.19036

机构：Department of Mathematics, National University of Singapore, Singapore（新加坡国立大学数学系）； Faculty of Computer Science and Engineering, Ho Chi Minh City University of Technology (HCMUT), VNU-HCM, Ho Chi Minh City, Vietnam（胡志明市技术大学计算机科学与工程学院）

作者：Tho Tran Huu, Huu-Tuan Nguyen, Thien-Hai Nguyen, Nhat-Tri Ho, Viet-Hoang Tran, Tho Quan, Tan Minh Nguyen

英文摘要：Sparse Mixture-of-Experts (SMoE) architectures are now widely deployed in state-of-the-art language and vision models, where conditional routing allows scaling to very large networks. However, this very Top-$k$ expert selection that enables conditional routing also renders the SMoE map inherently discontinuous. In the vicinity of these discontinuity surfaces, even inputs that are arbitrarily close may activate substantially different sets of experts resulting in significantly different outputs. In this work we give a rigorous geometric and stochastic analysis of these discontinuities. We first classify them by order, determined by the number of tied experts at a switching event. Using measure-theoretic slicing arguments, we establish asymptotic volume estimates for the thickened discontinuity surfaces, showing that lower-order discontinuity sets dominate, whereas higher-order ones occupy a vanishingly small relative volume. Next, modeling random perturbations in the input space via a diffusion process, we prove that the path eventually encounter a discontinuity, and moreover that the first hit almost surely occurs on an order-1 discontinuity with explicit finite-time probability bounds. We further derive occupation-time bounds that quantify the duration the random path spend in the neighborhoods of each discontinuity order. These theoretical results imply that inputs are more likely to lie near lower order discontinuities. Motivated by this insight, we propose a simple smoothing mechanism that can be directly applied to existing SMoEs, softly incorporating experts near discontinuities; our analysis guarantees that the added computational overhead remains small while providing localized smoothing near discontinuities, and experiments across language and vision tasks show that smoothing not only enforces continuity of the SMoE map but also enhances empirical performance.

AI 总结： 利用损失和预测器的光滑性，将Gibbs预测器去随机化为后验均值处的确定性预测器，通过Jensen间隙类的Rademacher复杂度控制泛化界，并导出涉及参数雅可比和海森矩阵的正则化器。

链接：https://arxiv.org/abs/2606.19105

机构：Department of Computer Science and Software Engineering（计算机科学与软件工程系）； Université Laval（拉瓦尔大学）

作者：Alexandre Lemire Paquin, Brahim Chaib-Draa, Philippe Giguère

英文摘要：We study PAC-Bayes derandomization for smooth loss functions. Our goal is to obtain generalization bounds that hold with high probability for deterministic predictors by exploiting smoothness properties of both the loss and the predictor class. We show that passing from the Gibbs predictor to the deterministic predictor at the posterior mean has a precise cost, given by the generalization gap of the Jensen gap class. We control this class through its Rademacher complexity, leading to bounds for deterministic predictors that involve flatness quantities expressed in terms of parameter Jacobians and Hessians of the score map. The framework applies to both bounded and unbounded smooth loss functions, and we specialize the results to linear predictors and smooth neural networks. Finally, the Jacobian and Hessian quantities appearing in the theory motivate a practical regularizer. For BatchNorm networks, we compute this regularizer with respect to effective BatchNorm weights obtained by folding the BatchNorm transformation into the adjacent affine weights. Experiments on CIFAR-10 illustrate the behavior of this regularizer under different batch sizes.

AI 总结：针对混合建模中神经部分可能重复学习符号结构导致模型冗余的问题，提出正交正则化方法OrthoReg，直接惩罚符号与神经组件间的重叠，实现互补分解，提升符号恢复和分布外行为。

链接：https://arxiv.org/abs/2606.19145

机构：Technical University of Munich（慕尼黑工业大学）； Helmholtz Munich（亥姆霍兹慕尼黑中心）

作者：Till Richter, Niki Kilbertus

英文摘要：Dynamical systems are fundamental to modeling the natural world, yet modeling them involves a persistent trade-off: manually prescribed mechanistic models are interpretable by design but often overly simplistic and misspecified; in contrast, flexible data-driven neural methods lack physical insight. Hybrid modeling aims for the best of both worlds by combining a prescribed or symbolic, physics-based component with a flexible neural network. A critical challenge, however, is that the neural component may relearn mechanistic parts, yielding redundant and uninterpretable models, especially when the symbolic structure itself is discovered from data. Existing methods based on standard $L^2$ regularization rely on a projection argument that breaks when the symbolic component is learned through sparse discovery, allowing the neural augmentation to overlap with symbolic structure. We introduce \textbf{OrthoReg} (Orthogonal Regularization), which directly penalizes overlap between the symbolic and neural components, preventing symbolic structure from being absorbed by the neural residual. This yields a complementary decomposition: the symbolic part captures what the library can express, and the neural part captures what remains. On benchmark dynamical systems with partial library mismatch, OrthoReg improves symbolic recovery and out-of-distribution behavior.

AI 总结：研究随机动量方法（如重球法和加速SGD）在一致线性回归中的批次大小权衡，证明重球法不改善SGD的计算效率前沿但允许更大批次减少串行运行时间，而加速SGD的计算效率与串行运行时间权衡依赖于谱衰减。

链接：https://arxiv.org/abs/2606.19179

作者：Depen Morwani, Alexandru Meterez, Pranav Nair, Sham Kakade

英文摘要：Stochastic momentum methods such as heavy ball (HB), Nesterov momentum, and variants of Accelerated SGD (ASGD) [Kidambi et al., 2018] are widely used in modern training, but their stochastic benefits depend on two distinct quantities: serial runtime, the number of iterations needed to reach a target accuracy, and compute efficiency (CE), the inverse total gradient-query or FLOP cost. Larger batches reduce serial runtime without hurting CE only when the contraction gap grows linearly with batch size. We study stochastic HB and ASGD for consistent linear regression with Gaussian covariates and prove finite-dimensional, discrete-time lower bounds on their batch-size tradeoffs. Our first result shows that HB does not improve the CE frontier over SGD for arbitrary spectra; rather, it preserves SGD-level CE over a larger batch-size window, allowing larger batches to reduce serial runtime until HB reaches its deterministic accelerated scale. This window can be a factor $\sqrt{\kappa}$ larger than the SGD critical batch size. For ASGD, the picture is more spectrum-dependent: for rapidly decaying power-law spectra, ASGD improves small-batch CE over HB/SGD, but as batch size grows it trades this CE advantage for improved serial runtime. Synthetic linear-regression experiments verify these qualitative regimes, including near-overlap of ASGD and HB for slowly decaying spectra and the predicted CE--serial tradeoff for rapidly decaying spectra.

AI 总结：提出CodeBlock框架，通过选择结构完整的代码块而非孤立token进行稀疏监督，在仅使用1.9%监督token的情况下，在六个代码生成基准上取得优于全token微调的效果。

链接：https://arxiv.org/abs/2606.18286

机构：Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））； UC Santa Cruz（加州大学圣克鲁兹分校）； Ant Group（蚂蚁集团）； BAIA, ZJUT（浙江工业大学智能信息处理实验室）； D5Data.ai

作者：Zhijie Deng, Ling Li, Jinlong Pang, Kaiqin Hu, Qi Xuan, Zhaowei Zhu, Jiaheng Wei

英文摘要：Supervised fine-tuning of code LLMs typically applies uniform cross-entropy loss to all response tokens, implicitly assuming that every token provides equally useful learning signal. Recent token-level selection methods challenge this assumption in natural-language SFT by supervising only high-value tokens. However, directly transferring token-level masking to code can break syntactically and semantically coherent program units, because code depends on structural completeness and definition-use relations. We therefore propose CodeBlock, a structure-aware sparse supervision framework that selects structure-complete code evidence rather than isolated tokens. CodeBlock first selects high-quality instruction-response pairs, then partitions code responses into syntactically coherent coding items, estimates their utility by aggregating generalized cross-entropy over core logic tokens, and reranks them with data-flow reach and bridge signals to prioritize blocks that propagate or connect important program dependencies. During training, the full response remains available as context, while loss is applied only to selected code items and informative natural-language tokens. Experiments on six code-generation benchmarks show that CodeBlock achieves stronger average pass@1 than full-token SFT and competitive selection baselines, while using only 1.9% of supervised response tokens.

AI 总结：针对MoE模型专家级剪枝粒度粗、冗余识别不足的问题，提出基于归因引导和覆盖最大化的结构剪枝框架，将剪枝分配转化为通道分数覆盖优化问题，在50%剪枝率下结合4位量化保持精度，内存减少5.27倍。

链接：https://arxiv.org/abs/2606.18304

机构：School of Computer Science and Engineering, Beihang University（北京航空航天大学计算机科学与工程学院）； School of Artificial Intelligence, Beihang University（北京航空航天大学人工智能学院）； Nanyang Technological University（南洋理工大学）

作者：Yifu Ding, Jiacheng Wang, Ge Yang, Yongcheng Jing, Jinyang Guo, Xianglong Liu, Dacheng Tao

英文摘要：Mixture-of-Experts (MoE) models scale compute efficiently, yet remain expensive to deploy due to their substantial memory footprint and inference overhead. Prior compression methods mainly operate at the expert level, either removing entire experts or ranking experts by coarse-grained importance scores. However, such expert-wise decisions are often too coarse to capture fine-grained redundancy, leading to misallocated pruning budgets and limited compression. To address this problem, we observe that information within MoE experts is highly concentrated in a small subset of channels, leaving substantial redundancy even in experts deemed important. Based on this observation, we propose a structural pruning framework tailored for MoE models. Our method reformulates prune-ratio allocation as a channel-score coverage maximization problem and solves it efficiently using an attribution-based approximation. Experiments on DeepSeek and Qwen MoE models show that our method preserves model accuracy under 50% or 25% structured pruning when combined with 4-bit quantization. On Qwen3-30B-A3B, our approach reduces memory footprint by 5.27$\times$ and consistently outperforms state-of-the-art baselines across diverse benchmarks.

AI 总结：针对LLM推理中长度预测调度在分布偏移和尾延迟控制上的脆弱性，提出无预测的分布感知调度框架，通过轻量统计信号实现软优先级提升，结合缓存感知抢占，在多种工作负载下将P99 TTLT降低35-50%，TTFT降低34-47%。

链接：https://arxiv.org/abs/2606.18431

机构：Cornell University, Computer Science Department（康奈尔大学计算机科学系）； Cornell University, Electrical and Computer Engineering Department（康奈尔大学电气与计算机工程系）； Cornell University, Operations Research and Information Engineering Department（康奈尔大学运筹学与信息工程系）； Microsoft Azure System Research（微软Azure系统研究）； NVIDIA Corporation（英伟达公司）

作者：Yueying Li, Yuanfan Chen, Jiayang Chen, Esha Choukse, Haoran Qiu, G. Edward Suh, Rodrigo Fonseca, Ziv Scully, Udit Gupta

英文摘要：LLM serving exhibits extreme length variability, making size-based scheduling difficult in practice. Recent LLM schedulers approximate SJF/SRPT using predicted decode lengths or ranks and primarily report mean-centric metrics such as TTFT and TBT. We show that these prediction-driven policies can be fragile under distribution shifts, bursty arrivals, and GPU memory pressure, while offering limited control over the tail latency (P90-P99) that dominates user experience, even with perfect decode-length knowledge. We introduce a distribution-aware, prediction-free scheduling framework that replaces explicit length prediction with soft priority boosting driven by lightweight statistical signals. Our design co-optimizes scheduling and cache-aware preemption to account for memory-coupled decode dynamics across workload mixes. Evaluated on production and open-source traces, our method reduces P99 TTLT by up to 35-50% relative to SRPT with perfect length knowledge and reduces TTFT by 34-47% across workloads, including reasoning-heavy and chat-heavy tasks. These results demonstrate a robust alternative for optimizing tail latency in online LLM serving.

AI 总结：提出BLADE框架，通过拉格朗日乘子将双层优化转化为单层惩罚目标，避免逆Hessian计算，实现动态参考模型，理论保证一阶收敛，实验优于现有方法。

链接：https://arxiv.org/abs/2606.18650

机构：University of Oxford（牛津大学）； Renmin University of China（中国人民大学）； University of Chinese Academy of Sciences（中国科学院大学）

作者：Jiaxing Wang, Deping Xiang, Jin Xu, Zirui Liu, Zicheng Zhang, Guoqiang Gong, Jun Fang, Chao Liu, Pengzhang Liu, Tongxuan Liu, Ke Zhang, Qixia Jiang

英文摘要： As Large Language Model (LLM) datasets scale to trillions of tokens, data selection has emerged as a critical frontier to filter out uninformative noise and construct adaptive learning trajectories. Beyond static heuristic filtering, advanced data selection methods for LLM training largely follow two paradigms, each with fundamental limitations. Influence-based methods provide principled bi-level objectives but require intractable inverse-Hessian computations, while excess-loss methods are computationally efficient but rely on a static reference model that becomes misaligned with the evolving proxy model during training. We propose BLADE (Bi-Level Adaptive Data sElection), a Hessian-free framework for data selection. BLADE reformulates the bi-level optimization problem underlying influence-based methods as a penalized single-level objective via Lagrange multipliers, avoiding inverse-Hessian computation while revealing a principled connection to excess-loss based data selection. The resulting objective recovers an excess-loss form but replaces the static reference model with a dynamic one that stays synchronized with training. Theoretically, we prove that this penalized formulation guarantees first-order convergence. For efficient online batch selection, we instantiate BLADE as a memoryless randomized block-coordinate Frank-Wolfe algorithm. Extensive experiments show that BLADE consistently outperforms state-of-the-art data selection baselines, providing a practical recipe for LLM training.

AI 总结：提出稀疏性促进微调方法，利用E(3)等变材料基础模型的结构特性选择性更新参数，在能量和力预测任务中以约3%参数达到或超越全微调性能，并展示在磁矩预测等任务中的泛化性和可解释性。

链接：https://arxiv.org/abs/2606.18691

作者：Youngwoo Cho, Seunghoon Yi, Wooil Yang, Sungmo Kang, Young-woo Son, Jaegul Choo, Joonseok Lee, Soo Kyung Kim, Hongkee Yoon

英文摘要：Pre-trained materials foundation models, or machine learning interatomic potentials, leverage general physicochemical knowledge to effectively approximate potential energy surfaces. However, they often require domain-specific calibration due to physicochemical diversity as well as mismatches between practical computational settings and those used in constructing the pre-training data. To address this, we propose a sparsity-promoting fine-tuning method that selectively updates model parameters by exploiting the structural properties of E(3)-equivariant materials foundation models. On energy and force prediction tasks across molecular and crystalline benchmarks, our method matches or surpasses full fine-tuning and equivariant low-rank adaptation while updating only $\sim$3~\% of parameters, and in some cases as little as $\sim$0.5~\%. Beyond energy and force calibration, we further demonstrate task generalizability by applying our method to magnetic moment prediction and magnetism-aware total energy modeling. Finally, analysis of sparsity patterns reveals physically interpretable signatures, such as enhanced $d$-orbital contributions in transition metal systems. Overall, our results establish sparsity-promoting fine-tuning as a flexible and interpretable method for domain specialization of equivariant materials foundation models.

AI 总结：针对强化学习推演中自回归解码延迟瓶颈，提出感知系统的自推测解码框架，通过量化自推测解码器与感知系统的推测开关策略，在保持模型质量前提下降低推演和端到端延迟。

链接：https://arxiv.org/abs/2606.18967

机构：FuriosaAI； University of California, Berkeley（加州大学伯克利分校）

作者：Minseo Kim, Minjae Lee, Seunghyuk Oh, Kevin Galim, Donghoon Kim, Coleman Hooper, Harman Singh, Amir Gholami, Hyung Il Koo, Wonjun Kang

英文摘要：Reinforcement learning (RL) has become a representative post-training paradigm for LLMs, enabling strong reasoning and agentic capabilities. However, rollout generation remains a dominant latency bottleneck because autoregressive sampling decodes responses sequentially and a small number of long-tailed generations often determine completion time. Speculative decoding (SD) offers a natural way to address this bottleneck, as it is a well-established technique for serving fixed LLMs that reduces latency by rapidly drafting tokens and accepting them through parallel verification while preserving the target-model distribution. However, its practical speedups do not directly carry over to RL rollouts: (i) the evolving target policy makes any fixed drafter increasingly mismatched with the policy's output distribution; and (ii) active batch sizes shrink throughout rollout decoding, shifting decoding from compute-bound to memory-bound regimes where parallel verification can exploit underutilized compute. Therefore, accelerating RL rollouts requires both a drafter that remains effective under long, high-temperature generations from an evolving policy and system-aware use of SD that avoids compute-bound regimes. We present EfficientRollout, a system-aware self-SD framework designed to address this gap for RL rollouts. EfficientRollout induces a quantized drafter from the target model (i.e. self-speculative decoding), keeping it coupled to the evolving policy without separate drafter pretraining or online adaptation. It further coordinates a system-aware SD toggle policy with acceptance-aware draft-length adaptation, enabling speculation only in beneficial regimes while matching the drafting budget to evolving drafter quality. EfficientRollout reduces rollout and end-to-end latency by up to 19.6% and 12.7%, respectively, over an accelerated AR rollout baseline, while preserving final model quality.

AI 总结：提出FoMoE系统，通过跨工作节点分区专家层打破全副本范式，结合部分专家复制和跳跃令牌机制，显著降低通信开销并提升吞吐量。

链接：https://arxiv.org/abs/2606.19025

机构：DeepSeek-AI

作者：Lorenzo Sani, Zeyu Cao, Meghdad Kurmanji, Alex Iacob, Andrej Jovanovic, Yan Gao, Wanru Zhao, Nicholas D. Lane

英文摘要： Pre-training Large Language Models (LLMs) typically demands large-scale infrastructure with tightly coupled hardware accelerators. While increasing model and dataset scale remains the dominant driver of performance, Mixture-of-Experts (MoEs) architectures have recently achieved state-of-the-art results by decoupling parameter count from computational cost. This efficiency enables training massive models on constrained compute budgets, yet it typically requires the high-speed interconnects of a single datacenter. To overcome these physical limits, recent approaches such as DiLoCo and Photon use low-communication data-parallel methods to enable scaling across geographically distributed, weakly connected data centers. However, these methods suffer from a fundamental inefficiency: they require full model replicas at every site, which imposes prohibitive memory constraints and communication overheads. In this work, we introduce FoMoE, a system that breaks the full-replica paradigm by partitioning expert layers across workers. We demonstrate that FoMoE: (I) reduces communication costs by up to 1.42x over efficient baselines and 45.44x over DDP via partial expert replication in the studied regimes; (II) achieves empirical throughput speedups of up to 1.4x through a novel skip-token mechanism; and (III) shows stable routing in the trained proxy regimes and projects the communication/memory benefits to 100B-scale configurations through system modelling.

AI 总结：提出CAHP框架，将注意力头选择建模为全局图论问题，通过图聚类和信息论距离保留互补头，自动确定剪枝数量，在SST-5和MNLI上优于现有方法。

链接：https://arxiv.org/abs/2606.19150

机构：Bar-Ilan University（巴伊兰大学）

作者：Yaniv Livertovsky, Shahar Somin, Gonen Singer

英文摘要：The remarkable success of Transformer-based models in natural language processing stems from architectural scaling, which leads to a large number of parameters and hinders deployment in resource-constrained environments. While structured pruning offers a pathway to compression, existing state-of-the-art methods often rely on gradient-based importance ranking or stochastic gating, which suffer from instability, structural degeneration, and the need for extensive manual hyperparameter tuning. In this paper, we introduce CAHP (Complementary Attention Head Pruning), a novel post-hoc framework that redefines head selection as a global graph-theoretical problem. Rather than evaluating heads in isolation, CAHP utilizes graph-based clustering combined with information-theoretic distance measures to identify and preserve a topologically diverse subset of complementary attention heads. Without requiring a predefined sparsity level or pruning ratio, the framework automatically determines the number of selected attention heads across layers by identifying a diminishing marginal performance curve, where pruning additional heads leads to a sharp degradation in performance, as determined by the chosen polynomial degree. Extensive evaluations on the SST-5 and MNLI benchmarks, across different Transformer model scales, demonstrate that CAHP consistently outperforms competitive baselines, particularly in high-compression regimes. Furthermore, our structural analysis shows that CAHP avoids the "proximity bias" of gradient-based pruning methods, which tend to preserve heads mainly in layers close to the output, and instead retains a functionally critical set of attention heads in the model's intermediate layers.

AI 总结：提出SAGE方法，通过事后净化最终更新向量，在不重新运行原始遗忘流程的情况下，缓解大语言模型遗忘与保留能力之间的权衡。

链接：https://arxiv.org/abs/2606.18309

机构：Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University（上海交通大学图像处理与模式识别研究所）

作者：Jingyuan Zhang, Yucheng Bai, Peixi Wen, Zhehao Huang, Zhengbao He, Hanling Tian, Xinwen Cheng, Haiyin Ran, Xiaolin Huang

英文摘要：Large Language Model (LLM) unlearning aims to remove undesirable knowledge or behaviors while preserving retained capabilities. Current unlearning methods all involve a trade-off between unlearning and retention. We have found that the retention activation bias can also be used to quantify the damage an unlearning method inflicts on retention, without considering the specific implementation of the unlearning process. This allows us to restore retention performance for any unlearning method using a post-hoc approach. Therefore, we propose a complementary post-hoc setting to sanitize the final update vector without rerunning the original unlearning pipeline. In this setting, we design SAGE, Spectral Activation-GEometry Sanitization, a source-agnostic correction for final unlearning updates. SAGE collects real module inputs from a small retain proxy, extracts their dominant activation geometry, and solves a source-anchored optimization objective in closed form, which suppresses update components aligned with high-energy retained directions while preserving the source method's forgetting carrier. Across multiple unlearning methods, model scales, and benchmarks, SAGE consistently relieves the retain-forget trade-off, identifying post-hoc sanitization of final vectors as a practical and underexplored axis for machine unlearning.

AI 总结：针对分层联邦学习中客户端选择策略缺乏帕累托效率和策略证明性导致整体福利下降的问题，提出SCOPE-FL框架，采用顶级交易循环算法同时保证帕累托最优和策略证明性，并通过区块链智能合约实现奖励分配。

链接：https://arxiv.org/abs/2606.18384

机构：École de Technologie Supérieure (ÉTS)（高等技术学院）； Ferdowsi University of Mashhad（菲尔多西大学）； University of Toronto（多伦多大学）

作者：Seyed Salar Ghazi, Kaiwen Zhang, Mehdi feizi, Hans-Arno Jacobsen

英文摘要： Hierarchical Federated Learning (HFL) enables scalable collaborative model training across distributed devices while preserving data privacy. However, existing HFL client selection mechanisms suffer from a fundamental strategic inefficiency. By prioritizing stability over Pareto efficiency (PE), they produce suboptimal resource allocations, and without strategy proofness (SP), participants are incentivized to misrepresent their true preferences, both failures degrading system overall welfare in the Pareto sense in practice. To address it, we propose SCOPE-FL (Strategy-proof Chain-based Optimal pareto efficient Federated Learning), a synchronous HFL framework that formulates client selection as a two-sided school choice problem solved through the Top Trading Cycle (TTC) algorithm that simultaneously guarantees PE and SP. For reward distribution, SCOPE-FL employs a scalable Shapley value approximation based on One-Round Reconstruction (OR), ensuring compensation proportional to each client's contribution. The entire mechanism executes via blockchain smart contracts, providing the tamper-proof environment required for the SP guarantees to hold in practice. A comprehensive evaluation on MNIST, Fashion-MNIST, and CIFAR-10 demonstrates that SCOPE-FL outperforms state-of-the-art approaches, including DA, IAS, and other methods across model accuracy, convergence rate, and reward efficiency, while achieving communication latency comparable to DA and blockchain overhead significantly lower than DA at scale.

AI 总结：提出PSyGenTAB框架，将合成医疗数据生成建模为约束优化问题，通过增强拉格朗日方法嵌入可配置隐私约束，在保证隐私阈值的同时最大化临床数据效用，实验表明合成数据训练的模型性能与真实数据相当。

链接：https://arxiv.org/abs/2606.18518

机构：San Diego State University（圣地亚哥州立大学）； University of California, Irvine（加利福尼亚大学尔湾分校）

作者：Arshia Ilaty, Hossein Shirazi, Manasi Chitale, Kedar Hegde, Dhanalakshmi Ramesh, Rashmi S. Manjunath, Amir Rahmani, Hajar Homayouni

英文摘要：The development of medical AI is constrained by limited access to high-quality clinical data due to institutional silos and strict privacy regulations such as HIPAA and GDPR. Synthetic data generation offers a potential solution, but existing methods lack principled mechanisms to explicitly manage the privacy-utility trade-off, often degrading clinically meaningful patterns or risking patient re-identification. We present PSyGenTAB, a privacy-preserving generative framework that formulates synthetic healthcare data generation as a constrained optimization problem solved using the Augmented Lagrangian Method. By embedding configurable privacy constraints directly into model training, PSyGenTAB enforces minimum privacy thresholds while maximizing clinical data utility. Across multiple clinically motivated benchmarks, PSyGenTAB preserves inter-feature clinical relationships and minority-class diagnostic patterns essential for reliable health AI. Downstream evaluation using Train-on-Synthetic, Test-on-Real and Train-on-Real, Test-on-Synthetic protocols shows that models trained on synthetic data achieve performance comparable to those trained on real patient records. Privacy auditing further demonstrates reduced exact record reproduction and strong resilience to membership inference attacks. These results establish PSyGenTAB as a principled framework for balancing privacy protection and clinical utility in synthetic healthcare data, supporting secure cross-institutional AI development.

AI 总结：针对标签差分隐私回归问题，提出Cond-DP方法，利用公共特征矩阵的结构信息构造条件化矩阵以加速优化，在凸、强凸和非凸设置下提供收敛保证，并在线性回归中实现比DPSGD更快的收敛速度。

链接：https://arxiv.org/abs/2606.18773

机构：Microsoft（微软）； Google Research（谷歌研究院）

作者：Shuli Jiang, Walid Krichene, Nicolas Mayoraz

英文摘要：We study differentially private (DP) regression in settings where each data sample includes public, non-sensitive features -- common in applications such as recommendation and advertising systems. While such label-DP or semi-sensitive-feature settings have been primarily explored in the context of classification, effective approaches for regression remain underexplored. We introduce Cond-DP, a conditioned variant of DPSGD that leverages the structure of public feature matrices to improve optimization under privacy constraints. Motivated by the observation that these public features often exhibit rapidly decaying spectra, Cond-DP incorporates a data-driven conditioning matrix to reshape the optimization landscape and accelerate convergence. We provide convergence guarantees for convex, strongly convex, and non-convex settings, and recover standard DPSGD as a special case when the conditioning matrix is the identity. We show how to construct an effective conditioning matrix for Cond-DP directly from public features, enabling provably faster convergence than DPSGD in private linear regression without incurring additional privacy cost. Empirically, Cond-DP with this conditioning matrix consistently outperforms state-of-the-art baselines across a wide range of datasets and model architectures under label DP, demonstrating strong and robust performance in practice.

AI 总结：针对XGBoost模型提出XGBoost-Forget遗忘方法，在表格型网络入侵数据集上实现高效遗忘，保持模型性能的同时显著提升遗忘速度。

链接：https://arxiv.org/abs/2606.19220

机构：GECAD, ISEP, Polytechnic of Porto（波尔图理工学院工程学院GECAD研究所）

作者：Diana Magalhães, Eva Maia, João Vitorino, Isabel Praça

英文摘要：Machine Unlearning (MU) has emerged as an important technique for removing specific data points from trained models without requiring full retraining. However, most existing MU research focuses on deep learning and image data, leaving a gap in the domain of network intrusion detection, which relies heavily on tabular data. This work introduces XGBoost-Forget, an unlearning approach for the XGBoost model, to address this gap. The approach is evaluated on two tabular Network Intrusion (NI) datasets, IoT-23 and GeNIS, using multiple metrics to assess model performance, unlearning efficiency, and forgetting quality. The results show that XGBoost-Forget maintains predictive performance close to the original model while providing significantly faster unlearning, demonstrating its potential for MU in tabular NI settings.

AI 总结： 提出MAST方法，通过机制引导选择性更新参数，在遗忘RLVR诱导的推理行为时，显著降低对保留性能的附带损害。

链接：https://arxiv.org/abs/2606.19222

机构：School of Engineering, Institute of Science Tokyo, Japan（东京科学大学工学院）； College of Control Science and Engineering, Zhejiang University, China（浙江大学控制科学与工程学院）； Department of Electrical and Computer Engineering, National University of Singapore, Singapore（新加坡国立大学电气与计算机工程系）

作者：Chenyu Zhou, Qiliang Jiang, Shuning Wu, Xu Zhou

英文摘要：We propose MAST (Mechanism-Aligned Selective Targeting), a mechanism-guided method for unlearning RLVR-induced reasoning with substantially lower collateral damage than standard full-parameter updates. In matched SFT/RLVR checkpoints on Qwen2.5-Math-1.5B and Qwen3-1.7B-Base, the SFT-to-RLVR increment differs sharply from the SFT update in token-level delta-log-probability, and full-parameter gradient ascent forgets only by damaging retain MATH and GSM8K. MAST ranks attention-projection tensors by off-principal energy, update magnitude, and forget-gradient coupling magnitude, then updates only the top-ranked subset. On the primary model, MAST induces statistically significant target forgetting (MATH forget 45/150 to 37/150; McNemar p=0.0078) while preserving GSM8K (+0.8 pp) and MATH retain (-0.5 pp). The advantage reproduces across seeds, NPO/SimNPO objectives, and Qwen3, where MAST preserves GSM8K while full-parameter unlearning collapses it.

AI 总结：本文评估了仅使用零开销、隐私保护的NVML遥测（内容无关信号）对GPU工作负载分类的对抗鲁棒性，开发了一个分类器，在识别训练工作负载时达到98.2%的二元准确率，并对最具挑战性的意外工作负载达到43-87%的准确率。

链接：https://arxiv.org/abs/2606.19262

机构：Machine Intelligence Research Institute（机器智能研究所）； University of Virginia（弗吉尼亚大学）

作者：Robi Rahman, Sabiha Tajdari

英文摘要：Hardware-enabled monitoring of GPU workloads underpins many proposals for AI compute governance, but if developers can defeat monitoring mechanisms, such schemes are unworkable. We evaluate the adversarial robustness of GPU workload classification using only zero-overhead, privacy-preserving NVML telemetry: content-agnostic signals that observe physical effects of computation without accessing model weights, training data, or hyperparameters. Across 5 rounds of monitor-evader iteration, we evaluate 20 evasion strategy families on 9 GPU models spanning 4 architecture generations. We develop a classifier that achieves 98.2% binary accuracy at identifying training workloads across the whole corpus, and 43-87% accuracy against the most challenging unexpected workloads even when they are adversarially disguised.

AI 总结：研究发现稀疏自编码器（SAE）特征干预虽能抑制行为，但存在可恢复的失败模式，通过优化残差扰动可恢复原始行为，揭示特征级控制与行为完整性之间的差距。

链接：https://arxiv.org/abs/2606.18322

机构：The Hong Kong Polytechnic University（香港理工大学）

作者：Mingyue Cui, Linghui Shen, Xingyi Yang

英文摘要：Sparse Autoencoders (SAEs) decompose residual-stream activations into interpretable features. Recent latent-space defenses increasingly rely on these decompositions, assuming that identified "unsafe" SAE features serve as actionable handles for monitoring and intervention. In this paradigm, clamping a specific harmful feature is expected to reliably prevent model misbehavior. However, we show that this success may hide a recoverable failure mode: the clamp may block one visible route to a behavior without eliminating the behavior itself. We formulate this vulnerability as post-intervention recovery, a constrained residual-space optimization problem. Starting from the post-intervention residual state, we optimize residual perturbations to recover the pre-intervention behavior while preserving the post-intervention values of the targeted SAE features. Even under a strong threat model where the intervention remains active throughout optimization and generation, recovery remains possible. To rule out that recovery simply undoes the intervention, we use encoder-orthogonal updates for single-layer interventions and the corresponding feature-map Jacobian in the cross-layer setting. Across TPP, unlearning, IOI, and refusal steering experiments, this stress test reveals recoverable behavior despite successful feature-level intervention. Especially in the safety-critical refusal-steering setting, we achieve a 95.8% recovery rate on valid samples while keeping defended-feature relative drift to 0.131, substantially below suffix-based baselines. A recovery-path attribution analysis further localizes this recovery to the SAE reconstruction residual, the component left unexplained by the SAE. These results expose a gap between feature-level control and behavioral completeness: SAE features can support causal intervention, but controlling them does not guarantee control over the underlying behavior.

AI 总结：提出P$^2$CE算法，利用隔离森林异常检测和SHAP值，生成可行且帕累托最优的反事实解释，平衡可行性、合理性和计算效率。

链接：https://arxiv.org/abs/2606.18418

作者：Arthur Hendricks Mendes de Oliveira, Giovani Valdrighi, Marcos Medeiros Raimundo

英文摘要： The increasing use of machine learning algorithms in social applications has raised concerns about fairness and transparency, leading to the development of counterfactual explanations. These explanations supports individuals to understand and potentially alter unfavorable decisions in areas such as loan applications, job selections, and more, by providing actionable changes to input features that would lead to a desired outcome. Existing methods often struggle to balance feasibility, plausibility, and computational efficiency. To address this, we introduce P$^2$CE, an algorithm for generating plausible Pareto-optimal counterfactual explanations, offering users a diverse set of optimal trade-offs between different notions of feasibility. P$^2$CE employs an auxiliary isolation forest outlier detector to ensure that explanations are in accordance with the data distribution and leverages SHAP values to obtain optimal results with short computing times, regardless of the underlying model. Our algorithm was empirically evaluated on three datasets, demonstrating superior performance in terms of both solution quality and computational efficiency compared to related techniques.

AI 总结：提出签名过滤模块，通过移除干扰水印检测的签名令牌，在弱信号和低熵设置下将检测率从8-31%提升至78-99%，同时保持可控的假阳性率。

链接：https://arxiv.org/abs/2606.18430

机构：National Chengchi University（国立政治大学）

作者：Chih-Duo Hong, Yen-Pang Chen, Fang Yu

英文摘要：Statistical watermarks help organizations attribute large language model (LLM) outputs, yet existing detectors often struggle when watermark signals are weak, texts are repetitive, or watermarks are edited. We propose signature filtering, a detection-time module that enhances watermark detection without modifying watermark embedding and text generation. It learns a small set of ``signature'' tokens whose presence makes watermark tests unreliable, and removes these tokens before detection. The signatures are obtained by solving a mixed-integer linear program on a small training set, with constraints that maximize the true positive rate. We additionally derive finite-sample and asymptotic bounds under several attacker models (color-blind, color-adaptive, and distributionally correlated). On four well-known watermark families (Kgw, Sweet, Unigram, Exp), four benchmark corpora (C4, MBPP, HumanEval, Code-Search-Net), and six LLMs (Opt-1.3b, Opt-6.7b, Llama2-13b, Llama3.1-8b, Qwen2.5-14b, Phi-3-medium-14b), 2- and 3-gram signatures raise detection rates in weak-signal and low-entropy settings from 8~31% without filtering to 78~99% with filtering, while keeping false positives controllable and often negligible. In stress tests where we scramble sentences and perturb 25~50% of tokens by dilution, deletions, and substitutions, 2-gram filters for Kgw-style watermarks preserve most of the clean-text detection gains, often matching or outperforming the advanced WinMax watermark detector. Signature filtering thus provides a simple, scalable, and model-agnostic add-on to strengthen watermark-based provenance checks for LLM text in information processing workflows.

AI 总结：提出Veriphi系统，结合快速对抗攻击与α,β-CROWN形式化边界验证，实验表明训练方法有效性依赖数据集特性，IBP在MNIST上有效但在CIFAR-10上失效，PGD对抗训练在小扰动下达到94%认证准确率，并实现5倍验证加速。

链接：https://arxiv.org/abs/2606.18454

机构：TU Wien（维也纳工业大学）

作者：Pratik Deshmukh, Kartik Arya, Vasili Savin

英文摘要：We present Veriphi, a GPU-accelerated neural network verification system that combines fast adversarial attacks with formal bound certification using alpha,beta-CROWN methods. Through systematic experiments on MNIST and CIFAR-10 using three training methodologies (standard, adversarial, certified), we demonstrate that training method effectiveness is fundamentally dataset-dependent. Interval Bound Propagation (IBP) achieves 78% certified accuracy on simple MNIST (784 dimensions) but provides negligible certification performance on the more complex CIFAR-10 dataset, where PGD adversarial training dominates with 94% certification at small perturbations. We achieve 5x verification speedup through attack-guided falsification and scale our approach to production-size models (105.8M parameters) for real-world aerospace logistics optimization. Our results challenge the assumption that certified training universally outperforms adversarial training, showing context matters critically for verification strategy selection.

AI 总结：提出SWAAP框架，通过两阶段数据投毒（双层级优化寻找有害目标模型+梯度匹配隐蔽实现）操纵学习到的世界模型，导致规划性能显著下降，且能规避多种防御检测。

链接：https://arxiv.org/abs/2606.18697

作者：Yibin Hu, Xiaolin Sun, Zizhan Zheng

英文摘要：Model-based learning agents use learned world models to predict future states, plan actions, and adapt to new environments. However, the process of updating world models from collected experience creates a training-time attack surface: adversarially poisoned fine-tuning trajectories can manipulate the learned dynamics and thereby corrupt downstream planning. In this paper, we propose SWAAP, the first two-stage data poisoning framework for learned world models. In the first stage, SWAAP identifies a harmful target world model that induces low-return behavior under planning while remaining close to clean dynamics, using first-order bilevel optimization enabled by a transition-gradient theorem. In the second stage, SWAAP realizes this target through stealth-constrained gradient matching, modifying only a limited fraction of fine-tuning transition targets so that the induced training gradients steer the victim model toward the adversarial target, while a prediction-error regularizer encourages the poisoned targets to remain close to the world model's natural approximation error. To assess attack stealthiness, we evaluate defenses and detectability across three stages of the poisoning pipeline: pre-training detection of poisoned transitions, robust training during fine-tuning, and test-time monitoring of the resulting world model. Across diverse continuous-control tasks, SWAAP causes substantial performance degradation while keeping poisoned transitions close to clean data and evading the evaluated non-adaptive residual/CUSUM/TRIM-style defenses. These results reveal a practical vulnerability in world-model adaptation pipelines and highlight the need for robustness methods that protect both world-model training data and learned dynamics.

AI 总结：提出TRUST框架，通过概率Tsetlin机器和贝叶斯优化直接搜索满足用户指定置信度目标的最小输入变化，生成更稳健和可解释的反事实解释。

链接：https://arxiv.org/abs/2606.18832

机构：Group Research and Development Det Norske Veritas (DNV)（挪威船级社（DNV）集团研发部）

作者：K. Darshana Abeyrathna, Sara El Mekkaoui, Nils Enric Canut Taugbøl, Anuja Vats

英文摘要：Counterfactual explanations are widely used to provide algorithmic recourse in high-stakes decision-making systems. Most existing methods seek the smallest change to an input that flips a model's decision. However, decision-makers often rely not only on predicted labels but also on confidence thresholds and risk margins. Counterfactuals that barely cross a decision boundary can be fragile and unstable under noise or model variation. In this paper, we propose Target-confidence Recourse Using tSeTlin machines (TRUST), a framework in which users explicitly specify the desired prediction confidence for recourse. Rather than generating counterfactuals and evaluating confidence afterward, TRUST directly searches for minimal changes that satisfy a user-defined confidence target, enabling comparison of recourse options in terms of cost, confidence, and robustness. We instantiate TRUST using a Probabilistic Tsetlin Machine (PTM) combined with Bayesian optimization. The probabilistic clause-based structure of PTM links prediction confidence to the stability of decision rules. We show that counterfactuals satisfying the same rules can still differ substantially in reliability depending on how securely they satisfy those rules, revealing whether decisions are supported by robust or fragile clause activations. Experiments on synthetic and real-world datasets demonstrate that target-confidence counterfactuals produce more robust and interpretable recourse than conventional boundary-based approaches. Across multiple benchmarks, TRUST achieves perfect robustness while maintaining low recourse cost, including an L2 distance of 0.10 on the Haberman dataset at 0.92 confidence. By explicitly controlling confidence and exposing rule-level stability, TRUST provides actionable recourse for high-stakes decision support.

AI 总结：提出首个无需额外数据即可认证视觉语言模型在语义层面（如形状、大小、风格）鲁棒性的框架，通过文本提示作为语义代理并量化决策边界，确保预测类别在语义变换下不变。

链接：https://arxiv.org/abs/2606.18839

机构：School of Computing \& Information Systems, University of Melbourne, Australia

作者：Peiyu Yang, Paul Montague, Feng Liu, Andrew C. Cullen, Amardeep Kaur, Christopher Leckie, Sarah M. Erfani

英文摘要：Vision-language models (VLMs) are now widely used in downstream tasks. However, real-world applications often expose VLMs to distribution shifts induced by semantic variation (e.g., shape, size, and style). Robustness certification determines if a model's prediction changes when transformations are applied to its input. While most certification frameworks study geometric or pixel-level transformations over inputs, this work proposes a novel framework that enables certifying VLM robustness under semantic-level transformations. Leveraging the open-vocabulary capability of VLMs, we use text prompts as semantic proxies to construct transformations parameterized by an extent that controls the degree of semantic variation. By characterizing the VLM decision boundary in closed form, our framework quantitatively certifies extent intervals for which the predicted class remains unchanged under the semantic transformation. Our framework is the first to certify VLM robustness under semantic-level variations without requiring additional data for each variation, making it practical to apply. Experiments on both synthetic and real-world data show that our framework enables certifying robustness under diverse semantic variations across scenarios.

AI 总结：研究通过特征选择和岭正则化应对战略操纵的分类问题，发现仅基于可操纵性排除特征通常次优，提出联合优化特征集与正则化水平的算法，并在医疗支付基准上验证。

链接：https://arxiv.org/abs/2606.18867

机构：University of California, Berkeley（加州大学伯克利分校）； University of Texas, Austin（德克萨斯大学奥斯汀分校）； Cornell Tech（康奈尔科技校区）； Stanford University（斯坦福大学）； University of Pennsylvania（宾夕法尼亚大学）； Harvard University（哈佛大学）； Inria, Paris（法国国家信息与自动化研究所巴黎分部）

作者：Jivat Neet Kaur, Pratik Patil, Divya Shanmugam, Emma Pierson, Michael I. Jordan, Nika Haghtalab, Meena Jagadeesan, Ahmed Alaa, Serena Wang

英文摘要：When algorithmic predictors inform resource allocation in high-stakes domains such as healthcare, these predictors must account for strategic manipulation of input features. The typical solution is to redesign the predictor itself to explicitly account for strategic interactions. In practice, however, decision makers are often constrained to adjusting coarser levers within existing prediction pipelines. For example, healthcare organizations often select which features to exclude based on perceived manipulability, while using standard regularization procedures to shrink the coefficients of retained features. In this work, we initiate a formal study of strategic classification through feature selection and its interaction with ridge regularization. Our main finding is that excluding individual features based on their manipulability alone is generally suboptimal. We provide a fine-grained characterization of the performance of a feature subset under optimal regularization, yielding new insights for policy design. Motivated by this characterization, we develop a practical algorithm for jointly choosing the feature set and the level of ridge regularization. Through a real-world case study on a healthcare payments benchmark, we illustrate how our algorithm can guide the design of coarse policy levers in practice. Our results provide a principled, practical framework for mitigating the effects of strategic behavior in algorithmic decision-making systems.

AI 总结：提出K跳高斯扩散核作为预处理模块，通过多跳扩散和高斯权重平衡局部与全局信息，在噪声或结构复杂图中优于传统消息传递和现有扩散方法。

链接：https://arxiv.org/abs/2606.18317

机构：Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences（中国科学院深圳先进技术研究院）； Southern University of Science and Technology（南方科技大学）

作者：Xuling Zhang, Peng Wang, Daiyan Li, Aoran Huang, Zeiwei Chen, Yongkui Yang

英文摘要：Most graph neural network (GNN) cores rely on graph convolutions, typically implemented as message passing between direct (single-hop) neighbors. In many real-world graphs, edges can be noisy or poorly defined, limiting information propagation to local neighborhoods. Existing diffusion kernels, such as Personalized PageRank (PPR) and Heat Kernel, alleviate this issue through global propagation, but still struggle with complex local structures and distant node noise. To address these limitations, we propose a K-Hop Gaussian (KHG) diffusion kernel as a preprocessing module for graph data. KHG introduces multi-hop diffusion with Gaussian weighting for remote nodes, balancing local and global information propagation before applying standard GNNs. Experiments on multiple benchmark datasets demonstrate that KHG significantly outperforms traditional message-passing GNNs, as well as PPR and Heat Kernel diffusion, particularly in noisy or structurally complex graphs.

AI 总结：提出TMR-GGNN框架，通过时间窗口内异构实体交互建模、动态多关系图构建、时间感知注意力机制和对比学习解码器，结合InfoNCE与Focal Loss复合损失函数，解决数据不平衡和欺诈模式演化问题。

链接：https://arxiv.org/abs/2606.18444

作者：Rohit Tewari, Shubhankar Shilpi, Navin Chhibber, Devendra Singh Parmar, Sunil Khemka, Piyush Ranjan

英文摘要：In recent years, credit card fraud detection has faced significant challenges due to highly imbalanced data, evolving fraud patterns, and complex relational structures among transaction entities. To address these issues, this research proposes a novel framework called Timeaware Multi Relational Guided Graph Neural Network (TMR GGNN). Particularly, the proposed TMR GGNN extends the encoder decoder Graph Neural Network GNN architecture by modeling heterogeneous interactions across customers, merchants, devices, and IPs over temporal windows. Subsequently, the proposed TMR GGNN approach constructs a dynamic, multi relational graph and incorporates a time aware relational attention mechanism within the encoder to adaptively weigh the transaction relevance based on temporal proximity and semantic context. Consequently, the decoder employs a contrastive learning module to distinguish between real and synthesized transaction patterns, while improving the models generalization of rare fraud cases. Additionally, to effectively manage severe class imbalances and emphasize discriminative learning, a composite loss function combining Information Noise Contrastive Estimation (InfoNCE) based contrastive loss with Focal Loss is introduced. This integration assists in improving fraud identification while mitigating false negatives.

AI 总结：提出RelAD框架，通过条件稀疏门控属性重建和双视图多关系边重建，有效检测关系数据中的属性异常和连接模式异常，在6个基准数据集上优于现有方法。

链接：https://arxiv.org/abs/2606.18621

机构：Griffith University（格里菲斯大学）； Guangxi University（广西大学）

作者：Shiyuan Li, Yunfeng Zhao, Yue Tan, Qingfeng Chen, Yixin Liu, Shirui Pan

英文摘要：Relational databases are widely used for managing structured data in real-world systems. Detecting anomalies from such relational data is crucial for identifying fraud, risks, and abnormal behaviors, yet remains under-explored. The key challenges lie in the intrinsic complexity of relational data: multi-table attributes are high-dimensional and heterogeneous, making sparse abnormal clues easy to overwhelm by normal or irrelevant information; and anomalies may further manifest as abnormal connection patterns across different foreign-key relations, which existing tabular and graph anomaly detection methods are ill-suited to capture. To address them, we propose RelAD, a reconstruction-based framework that captures anomalies from both attribute and relational edge reconstruction. RelAD contains two core modules: conditional sparse-gated attribute reconstruction, which suppresses redundant multi-table attributes and emphasizes abnormal semantic blocks, and dual-view multi-relational edge reconstruction, which detects relation-specific abnormal connections from both intrinsic and behavioral entity profiles. The resulting attribute and relational signals are integrated through a lightweight fusion module to produce the final anomaly score. We further construct 6 benchmark datasets with systematic anomalies, on which extensive experiments show that RelAD consistently outperforms other baselines while achieving competitive efficiency.

AI 总结：提出各向异性图扩散网络（AGDN），通过MixScore转移矩阵和各向异性扩散策略，有效利用图结构信息求解旅行商问题，在多种实例规模和分布上优于现有方法。

链接：https://arxiv.org/abs/2606.19185

机构：Florida State University（佛罗里达州立大学）； Singapore Management University（新加坡管理大学）

作者： Bolin Shen, Ziwei Huang, Zhiguang Cao, Yushun Dong

英文摘要：The Traveling Salesman Problem (TSP) is a cornerstone of combinatorial optimization and arises in many practical scenarios. Although graph-based learning approaches have been explored for TSP, the question of how to exploit graph structure more effectively remains open. We present the Anisotropic Graph Diffusion Network (AGDN), a new Graph Neural Network framework designed to solve TSP. Our method tackles two central difficulties: (1) the lack of informative topological prior in fully connected TSP graphs, and (2) losing connected nodes in the optimal solution after the commonly used graph sparsification techniques. To overcome these issues, we construct a MixScore transition matrix that merges node similarity with pairwise distance, and we develop an anisotropic graph diffusion strategy that supports efficient information exchange across multiple hops. Comprehensive experiments spanning diverse instance sizes and node distributions show that AGDN consistently outperforms existing methods while keeping computation time competitive. Furthermore, AGDN generalizes well to problem sizes and distributions beyond those seen during training. The implementation is publicly available at: this https URL.

AI 总结：提出P-K-GCN，结合样条GCN和Koopman算子理论，在非规则几何上实现时空超分辨率，并通过物理损失和理论分析保证误差降低。

链接：https://arxiv.org/abs/2606.19303

机构：Department of Industrial & Systems Engineering, The University of Tennessee, Knoxville（田纳西大学诺克斯维尔分校工业与系统工程系）； Charles F. Dolan School of Business, Fairfield University（费尔菲尔德大学查尔斯·F·多兰商学院）； Department of Electrical Engineering & Computer Science, The University of Tennessee, Knoxville（田纳西大学诺克斯维尔分校电气工程与计算机科学系）

作者：Xizhuo (Cici) Zhang, Zekai Wang, Fei Liu, Bing Yao

英文摘要：High-fidelity simulation of spatiotemporal dynamics is computationally prohibitive, necessitating efficient super-resolution techniques to reconstruct high-resolution data from coarse-grained inputs. Traditional data-driven methods often lack physical constraints, and simple physics-informed learning struggles with irregular spatial geometries and intricately evolving temporal dynamics. To tackle these challenges, we propose a Physics-augmented Koopman-enhanced Graph Convolutional Network (P-K-GCN) for spatiotemporal super-resolution on irregular geometries. Specifically, a continuous spline-based GCN is first designed to extract spatial dependencies directly from coarse graph, and Koopman operator theory is incorporated to project the nonlinear dynamics into a compact latent space where temporal progression is linearized. Second, we augment the optimization objective with a physics-based loss to force the data-driven reconstructions to adhere to physical laws for improving predictive fidelity and robustness. Finally, we provide a rigorous theoretical analysis, establishing that the physics augmentation and Koopman regularization mathematically guarantees a reduction in super-resolution error by diminishing Rademacher complexity and tightening generalization bounds. We evaluate our framework on reconstructing spatially high-resolution cardiac electrodynamics across a 3D heart geometry from sparse low-resolution measurements. Numerical experiments demonstrate that our method achieves superior accuracy compared to baseline models.

AI 总结：提出本质子空间分解（ESD）和合并（ESM/ESM++）方法，通过正交化任务更新的主成分来减少多任务合并中的干扰，无需训练即可实现高效多任务学习。

链接：https://arxiv.org/abs/2606.19164

机构：School of Computer Science and Engineering, Southeast University（东南大学计算机科学与工程学院）； Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education（教育部新一代人工智能技术及其跨学科应用重点实验室（东南大学））； Huawei Inc.（华为公司）

作者：Longhua Li, Lei Qi, Xin Geng, Qi Tian

英文摘要：Model merging aims to enable multi-task learning by integrating the capabilities of multiple models fine-tuned from the same pre-trained checkpoint into a single model. Its core challenge is inter-task interference among task-specific parameter updates. In this paper, we analyze the output shifts induced by task updates and observe that their energy is concentrated in a small number of principal directions. We call the subspace spanned by these directions the essential subspace. In contrast, most remaining directions carry little task-relevant energy, but their accumulation across multiple task updates can cause severe interference during merging. Motivated by this observation, we propose Essential Subspace Decomposition (ESD), which decomposes each task update according to the principal components of its activation shift. Based on ESD, we introduce Essential Subspace Merging (ESM), a training-free static merging method that orthogonalizes and fuses essential components into one compact multi-task model. We further extend ESM to ESM++, a training-free dynamic merging method that decomposes task-specific residuals into low-rank experts and selects the most relevant expert through prototype-based routing during forward inference. Extensive experiments across multiple task sets and model scales demonstrate that ESM and ESM++ effectively preserves task knowledge while reducing inter-task interference.

AI 总结：提出DRIFT方法，利用在线策略影响函数解决标准影响函数在指令微调数据归因中的近邻偏差和梯度范数偏差问题，通过模型自身生成作为验证目标，提升7B模型性能上限。

链接：https://arxiv.org/abs/2606.18307

机构：Tsinghua University（清华大学）

作者：Zefan Wang, Lincheng Li, Tianyu Yu, Yuan Yao

英文摘要：Optimizing the training data distribution for Supervised Fine-Tuning (SFT) dictates the capability of Large Language Models (LLMs). While existing data curation methods excel at accelerating training under constrained budgets, they are less suited to elevating the capability upper bound. The challenge here is no longer to identify a smaller subset that preserves performance, but to refine the data distribution toward instances most capable of improving the final model. To address this problem, we explore instance-level data attribution using Influence Functions (IF). We identify that standard IF formulations struggle in this setting due to two structural limitations: a proximity gap caused by off-policy validation targets, and a severe bias towards gradient norm. We propose DRIFT (Data Refinement via On-Policy Influence Functions for Supervised Fine-Tuning). Instead of relying on external reference data, DRIFT utilizes the model's on-policy rollouts as validation targets, which empirically minimizes the parameter proximity gap and better aligns with the local neighborhood assumption of IF. It further applies signed weighting based on trajectory correctness and debiases influence scores against the gradient hacking issue, allowing a small set of validation queries to act as reliable anchors for attributing the full dataset. Experiments on 7B-parameter instruction and reasoning models show that DRIFT consistently raises the performance ceiling on both, outperforming existing data curation baselines.

AI 总结：为加速系外行星气候模拟，提出ThousandWorlds基准数据集，包含五个全球气候模型的约1800次模拟，用于评估机器学习模拟器在低数据、多模拟器参数到场回归任务中的性能。

链接：https://arxiv.org/abs/2606.18338

机构：University of Cambridge（剑桥大学）； University of Oxford（牛津大学）； University of Colorado Boulder（科罗拉多大学博尔德分校）； University of Bristol（布里斯托大学）； Purdue University（普渡大学）； University of Exeter（埃克塞特大学）

作者：Edward T. Stevenson, Mei Ting Mak, Eric Wolf, Denis E. Sergeev, Tobi Hammond, N. J. Mayne, Miles Cranmer

英文摘要：The search for life beyond Earth will depend on detecting faint signatures in the atmospheres of potentially habitable exoplanets. Interpreting those signatures requires understanding the host planet's climate: the same molecule may signal life on one planet and abiotic chemistry on another. Global climate models (GCMs) provide this understanding, but individual runs can require up to millions of core-hours and substantial domain expert time. Machine-learning emulators could remove this bottleneck, but progress has been limited by the absence of a curated, multi-model exoclimate dataset. We introduce ThousandWorlds, an ML-ready benchmark for exoclimate emulation and for the broader regime of low-data, multi-simulator, parameter-to-field regression. The dataset contains approximately 1800 simulations from five GCMs, mapping eight planet parameters to 3D atmospheric fields including temperature, humidity, winds, clouds, and radiation. Three nested subsets define progressively harder challenges: single-simulator regression, multi-simulator regression with complete observations, and multi-simulator regression with structured missingness. We propose two evaluation protocols: one for ranking methods, and one that measures performance relative to the disagreement between GCMs themselves. We evaluate seven baselines spanning simple methods, deep learning, and Gaussian processes. GP-based methods perform best, suggesting that ThousandWorlds exposes a regime where off-the-shelf deep learning does not yet succeed. Data: this https URL. Code: this https URL.

AI 总结：本文提出状态分层评估方法，发现时间序列基础模型在交通状态转换时准确率和预测区间覆盖率显著下降，并提出了双峰混合增强方法以改善转换状态覆盖。

链接：https://arxiv.org/abs/2606.18367

机构：University of California, Berkeley（加州大学伯克利分校）； Duke University（杜克大学）； National University of Singapore（新加坡国立大学）； Northeastern University（东北大学）； University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）； Southern Methodist University（南卫理公会大学）

作者：Yingshuo Wang, Xian Sun, Lingdong Kong, Wei Gao, Yanhang Li, Zhichao Fan, Zexin Zhuang

英文摘要：Standard benchmarks evaluate time series foundation models (TSFMs) using aggregate metrics, but these can mask severe failures in critical operating regimes. We introduce regime-stratified evaluation and apply it to three TSFMs on two standard traffic speed benchmarks. Traffic exhibits abrupt regime switching between free-flow and congested states, producing bimodal speed distributions during transitions. When we stratify by traffic regime, both accuracy and prediction-interval coverage degrade sharply during transitions: transition-regime MAE reaches 11 mph (versus 3 mph overall), and empirical coverage of 90% prediction intervals drops as low as 55%. These failures are invisible in aggregate metrics because free-flow observations dominate the sample. A simple historical conditional baseline (sampling from per-sensor training distributions) achieves better transition coverage than any TSFM, but has far worse overall accuracy. We propose bimodal mixture augmentation (BMA), a post-hoc method that combines TSFM forecasts with historical distributional knowledge, approaching the historical baseline's transition coverage while preserving the TSFM's accuracy. Our results suggest that TSFM benchmarks should incorporate regime-aware evaluation to surface failures that aggregate metrics hide.

AI 总结：提出可重复的VLM评判协议评估单图3D网格质量，发现几何有效性和渲染CLIP等廉价代理方法无法替代VLM评判。

链接：https://arxiv.org/abs/2606.18451

机构：Transformer Lab

作者：Ali Asaria, Tony Salomone, Deep Gandhi

英文摘要：Single-image-to-3D generators are improving quickly, but there is no agreed, human-free way to tell whether one generated mesh is better than another. Practitioners commonly rely on cheap automatic proxies (render-space CLIP similarity and mesh geometry-validity statistics), yet how well these track perceived quality is unestablished. We make two contributions. First, we propose and validate a reproducible VLM-judge evaluation protocol: a fixed 24-view headless render rig, two independent vision-language judge families, and a mandatory position-bias correction that queries both presentation orders and keeps only order-consistent verdicts. The two judge families agree substantially with each other (Cohen's kappa = 0.66), well above the chance-agreement floor. Second, using this protocol as the reference, we show the cheap proxies do not substitute for it. Geometry validity is only a weak signal on average (because, as we show, it is bimodal) and stays below our pre-registered target, while render-CLIP is at chance. A learned Bradley-Terry head collapses onto a single manifoldness statistic (giving render-CLIP a negative weight) and matches geometry-only exactly, so learning the feature weights buys nothing. The proxy is also bimodal: it is significantly above chance on contrasts with visible geometric defects but at chance on ambiguous contrasts, consistent with geometry validity tracking the judge only when the defect is visually salient. We therefore recommend the VLM-judge protocol as a reliable, reproducible evaluator under the conditions tested (two feed-forward generators on Google Scanned Objects, with a face-drop degradation regime) and advise against geometry/CLIP proxies as optimization targets.

AI 总结：提出TS-Fault基准，通过参数化故障场景（沿观测/机制、单变量/多变量两轴）评估时间序列预测模型鲁棒性，发现干净数据准确性与鲁棒性负相关、机制级故障重排排名、基础模型最脆弱。

链接：https://arxiv.org/abs/2606.18539

作者：Yuyang Zhao, Lian Xu, Hao Miao, Chenxi Liu, Hao Xue

英文摘要：Time series forecasting (TSF) underpins consequential decisions in energy, transportation, finance, and healthcare, yet TSF models are almost universally ranked by a single number (e.g., average error) on clean held-out data, under the implicit assumption that it predicts deployed reliability. However, real faults are not i.i.d noise but structured events with temporal shape, broken cross-variable dependencies, regime change coupled with missingness, and causal propagation across a sensing pipeline. Treating TSF robustness as a data-quality problem, we present TS-Fault, a benchmark that evaluates forecasting models under explicit, parameterized fault scenarios with controllable semantic difficulty. TS-Fault organizes recurring failures into four modes along two orthogonal axes (observation- vs mechanism-level; univariate vs multivariate) and injects each fault into the most prediction-critical window via a unified importance score. This design enables robustness to be tested against the structures models actually rely on, rather than reduced to generic noise sensitivity. We evaluate 21 models across 6 datasets, 4 modes, and 5 difficulty levels under a paired clean/corrupt protocol. The results reveal three findings that contradict common leaderboard intuition: (i) clean-data accuracy anti-correlates with robustness; (ii) clean rankings are preserved under observation-level faults but reshuffled under mechanism-level faults; and (iii) all catastrophic failures occur under mechanism-level faults, with foundation models achieving the highest clean-data accuracy yet exhibiting the greatest fragility. The code is publicly available at this https URL.

AI 总结：针对1型糖尿病血糖预测算法缺乏标准化评估基准的问题，提出MetaboNet-Bench多模态基准，集成血糖、胰岛素和碳水化合物数据，通过多个模型对比验证多模态数据对模型性能的影响。

链接：https://arxiv.org/abs/2606.18640

机构：Department of Genetics, Stanford University School of Medicine（斯坦福大学医学院遗传学系）； Replica Health； Boston Children’s Hospital, Harvard Medical School（哈佛医学院波士顿儿童医院）； Diabetes Research Institute, Mills-Peninsula Medical Center（米尔斯半岛医学中心糖尿病研究所）

作者：Nathaniel Jeffries, Miriam Wolff, Sam Royston, Elizabeth Healey, Caleb Mayer, David Klonoff, Michael Snyder, Tao Wang

英文摘要：Glucose forecasting algorithms are an important aspect of glycemic control management in type 1 diabetes. So far, the research community has developed numerous algorithms and models for forecasting. However, it is well-recognized that the lack of standardized model performance evaluation benchmarks makes fair comparison difficult and hinders further innovation, and thus benchmark standardization is in urgent need. Furthermore, many published glucose forecasting algorithms are limited to CGM data alone, ignoring other multimodal signals such as insulin dosing and carbohydrate intake. Here, we introduce MetaboNet-Bench, a benchmark for multimodal glucose forecasting for patients with type 1 diabetes that provides an extensible open-source evaluation framework for comparison of glucose forecasting algorithms that leverage glucose, insulin, and carbohydrate data. We then demonstrate its utility by benchmarking several recently published glucose forecasting models and a custom multimodal time-series model, representing different model architectures. The results show that the benefit of adding data modalities is conditioned on the complexity of the model and that incorporating more clinical metrics helps identify meaningful gaps to fill for future research.

AI 总结：针对表格流学习中分布漂移问题，提出上下文管理策略CURE，通过不确定性门控准入和冗余感知驱逐管理上下文，在七个流上相对提升最高27.0%。

链接：https://arxiv.org/abs/2606.18677

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

作者：Jinmo Lee, Doyun Choi, Moongi Choi, Jaemin Yoo

英文摘要： Tabular stream learning requires predictions on sequentially arriving examples under distribution shift. While standard methods adapt by updating model states, tabular foundation models (TFMs) make predictions conditioned on a labeled context in an in-context manner, making them a natural alternative for stream learning. This shifts the challenge from how to update the model to how to manage the context. We propose a future information view that yields three practical requirements for context management: preserve recent examples, retain uncertain examples, and remove redundant examples. We instantiate these requirements as CURE (Context management via Uncertainty-aware admission and Redundancy aware Eviction), a context-managing policy with entropy-gated admission and redundancy-aware eviction. Across seven streams, CURE shows up to 27.0% relative improvement over classical stream learners, remains robust across multiple TFM backbones, and ranks first among other policy variants. Code and datasets are available at this https URL.

AI 总结：提出RouteJudge平台，通过匿名成对比较评估LLM路由策略的决策质量，并发布ORBIT工具箱标准化路由工作流，支持可复现和偏好感知的路由评估。

链接：https://arxiv.org/abs/2606.18774

机构：School of Artificial Intelligence, Nanjing University（南京大学人工智能学院）； National Key Laboratory for Novel Software Technology, Nanjing University（南京大学计算机软件新技术国家重点实验室）； SinapisAI

作者：Guannan Lai, Haoran Hu, Han-Jia Ye

英文摘要：We present RouteJudge, an online pairwise preference evaluation framework for LLM routing systems, with a public platform available at this https URL. Different from model-level response evaluation, RouteJudge focuses on router-level decision quality. For each user query, multiple routing strategies independently recommend candidate models under the same model pool and budget constraints. The selected model responses are then presented to users through anonymous pairwise comparisons, and the resulting user preferences are attributed back to the routing strategies behind the compared responses. Each evaluation record stores the query, routing decisions, model responses, preference labels, cost, latency, and task metadata, enabling preference-aware, cost-aware, and task-conditioned analysis of LLM routers. To support the continuous expansion of routing methods in RouteJudge, we further release ORBIT (Optimal Routing and Budgeted Inference Toolbox), a modular and extensible toolbox that standardizes the end-to-end workflow of LLM routing. ORBIT provides unified interfaces for benchmark loading, query representation, router implementation, budget-aware evaluation, and method comparison, allowing researchers to develop and evaluate routing algorithms under consistent protocols. It also serves as the submission and integration layer for RouteJudge: researchers can implement routing methods within ORBIT, validate them on existing routing benchmarks, and submit compatible routers for online preference-based evaluation. The code of ORBIT is available at this https URL.

AI 总结：提出GateMem基准，评估多主体共享内存代理在效用、访问控制和遗忘三方面的治理能力，发现现有方法无法同时满足三者。

链接：https://arxiv.org/abs/2606.18829

作者：Zhe Ren, Yibo Yang, Yimeng Chen, Zijun Zhao, Benshuo Fu, Zhihao Shu, Bingjie Zhang, Yangyang Xu, Dandan Guo, Shuicheng Yan

英文摘要：Memory benchmarks for LLM agents largely assume single-user settings, leaving shared assistants for hospitals, workplaces, campuses, and households understudied. In these deployments, multiple principals write to a common memory pool and query it under different roles, scopes, and relationships, so memory quality requires governance as well as recall. We introduce GateMem, a benchmark for multi-principal shared-memory agents. GateMem jointly evaluates utility for legitimate long-horizon requests with state updates, access control across contextual authorization boundaries, and agent-facing active forgetting after explicit deletion requests. It spans medical, office, education, and household domains, with long-form multi-party episodes, incremental memory injection, hidden checkpoints, structured judging, and leak-target annotations. Across diverse baselines and backbone models, no method simultaneously achieves strong utility, robust access control, and reliable forgetting. Long-context prompting often yields the best governance score at high token cost, while retrieval-based and external-memory methods reduce cost yet still leak unauthorized or deleted information. These results show current memory agents remain far from reliable shared institutional deployment.

AI 总结：提出一种基于统计对偶性的半监督聚类框架，通过a-contrario推理和感知算法，利用种子标签初始化并迭代排除异常点，实现鲁棒聚类，在少量种子下达到强性能。

链接：https://arxiv.org/abs/2606.18833

机构：Cyber Innovation Lab, Airbus, Newport, UK（空中客车公司网络创新实验室（英国纽波特））

作者：Nassir Mohammad

英文摘要： This paper introduces a semi-supervised clustering framework grounded in the statistical duality between grouping principles and anomaly detection. We address the challenge of robust cluster definition in noisy environments -- a task where partitioning algorithms often over-assign outliers and density-based methods remain sensitive to heuristic global parameters. Drawing on \textit{a-contrario} statistical reasoning and Gestalt proximity principles, we define a cluster as a maximal subset of data points containing no anomalies relative to a null hypothesis of uniform randomness. Central to this approach is the Perception algorithm, which utilises a principled expectation-based threshold ($\mathbb{E} < 1$) to identify outliers without manual parameter tuning. By treating clustering as the dual of anomaly detection, we employ an iterative ``clustering-by-exclusion'' mechanism. The algorithm is seed-guided, leveraging minimal user-provided labels to initialise robust cluster medians and form initial groups, which are subsequently expanded by admitting non-anomalous points. This approach naturally isolates fringe points, isolated noise, and emerging unknown clusters. We evaluate the method on synthetic and real-world benchmarks, including image and text datasets represented through raw, linear-reduced, and neighbourhood-preserving embeddings. Results demonstrate that with as few as 10--30 seeds per cluster, the proposed method achieves competitive and often very strong performance under a practical low-tuning benchmarking protocol, while maintaining linear scalability with respect to both observations and dimensionality for a fixed number of seeded clusters and iterations.

AI 总结：通过受控基准测试，比较量子与经典生成器在脑MRI数据增强中的性能，发现两者均未显著优于仅用真实数据训练，且量子生成器无额外优势。

链接：https://arxiv.org/abs/2606.18970

机构：Department of Mathematics（数学系）； Department of Political and Social Sciences（政治与社会科学系）

作者：Syed Mujtaba Haider, Silvia Figini

英文摘要：Medical image classification is often constrained by limited labeled data, motivating generative augmentation; recently, quantum generative models have been proposed for this purpose, frequently reporting accuracy gains. However, such claims are typically based on single training runs, do not match the parameter budgets of the quantum and classical generators, and do not characterize the data regime in which any benefit appears. We present a controlled benchmark that isolates the contribution of a quantum generator to brain-MRI augmentation. Images are encoded into a KL-regularized latent space in which a conditional Wasserstein GAN with gradient penalty is trained using either a variational quantum generator or a classical generator of near-identical parameter count (1648 vs. 1632). Synthetic samples are decoded and used to augment a pretrained classifier across labeled data fractions from 5% to 100%, evaluated over eight random seeds with paired significance testing (with multiple-comparison correction) and with intraset diversity and latent-distribution analyses. Across all fractions, no augmentation variant significantly outperforms real-data-only training, and the quantum and classical generators are statistically indistinguishable. Any low-data benefit behaves as regularization rather than faithful data expansion:synthetic samples are off distribution and severely mode collapsed precisely where data is scarce, and the quantum generator is no more diverse thanits classical counterpart. We release the protocol as a testbed for rigorous evaluation of quantum generative augmentation in medical imaging.

AI 总结：提出 Act2Answer 协议，通过动作回答评估 VLA 模型的知识保留，发现模型在简单概念上表现良好，但在丰富语义类别上存在差距，且 VQA 联合训练有助于知识保留。

链接：https://arxiv.org/abs/2606.19297

机构：CogAI Lab（CogAI实验室）； FusionBrain Lab（FusionBrain实验室）； IAI MSU（莫斯科大学人工智能研究所）； Lomonosov MSU（莫斯科国立罗蒙诺索夫大学）； NUST MISIS（国立研究型技术大学MISIS）； Applied AI Institute（应用人工智能研究所）； HSE University（高等经济大学）； Generalizable AI Systems（通用人工智能系统实验室）； ISP RAS（俄罗斯科学院系统编程研究所）； MIRAI； Domain-specific NLP Group（领域特定自然语言处理组）

作者：Nikita Kachaev, Andrey Moskalenko, Matvey Skripkin, Nikita Kurlaev, Daria Pugacheva, Albina Burlova, Mikhail Kolosov, Denis Shepelev, Andrey Kuznetsov, Elena Tutubalina, Aleksandr I. Panov, Alexey K. Kovalev, Vlad Shakhuro

英文摘要：Embodied Vision-Language-Action (VLA) models are typically obtained by fine-tuning powerful pretrained VLMs on robotics data, yet it is unclear how much commonsense and factual knowledge they retain after adaptation. Failures on knowledge-sensitive tasks are ambiguous, conflating missing knowledge with poor generalization of low-level control. We introduce Act2Answer, a lightweight protocol that adapts VLM knowledge benchmarks to VLA evaluation by requiring agents to answer through action. Each question becomes a short tabletop episode where the agent performs a single object-placement action to select among candidate answers, yielding an action-grounded success rate with reduced control confounds. We curate a test suite of such environments across diverse commonsense and world-knowledge categories and introduce layerwise intent probing to localize answer-relevant information across the VLM backbone and action head. In a large-scale study of 7 VLA models and 9 VLM baselines, we systematically rank models across categories, finding that VLAs show solid performance on simple concepts while exhibiting larger gaps on richer semantic categories relative to their source VLMs, that VQA co-training is associated with better knowledge retention, and that answer-relevant signals peak in middle VLA layers but attenuate in upper layers. Act2Answer is available at this https URL.

AI 总结：提出Artemis框架，通过区域级因果干预学习特定脑区的混杂因素表示，消除fMRI和DTI多模态神经影像中人口统计学混杂因素对GNN的影响，在三个基准上提升性能。

链接：https://arxiv.org/abs/2606.18287

机构：University of Pittsburgh（匹兹堡大学）； University of Maryland（马里兰大学）； University of Southern California（南加州大学）； Binghamton University（宾汉姆顿大学）； University of Texas Rio Grande Valley（德克萨斯大学里奥格兰德河谷分校）

作者：Siyuan Dai, Yang Du, Kun Zhao, Zhusuyi Chen, Heng Huang, Paul Thompson, Chao Shi, Haoteng Tang, Liang Zhan

英文摘要： Multimodal neuroimaging, integrating functional connectivity from fMRI and structural connectivity from DTI, enables non-invasive analysis of brain networks using graph neural networks. However, demographic factors such as age and sex systematically confound the relationship between brain connectivity and clinical outcomes, causing GNNs to exploit spurious shortcuts rather than learning causally invariant representations. While recent causal GNN methods introduce causality at the graph-modeling level, their causal mechanisms remain domain-agnostic without accounting for the real-world confounders inherent in clinical neuroimaging data. Moreover, brain networks are constructed from atlas-based parcellations where each region exhibits distinct sensitivity to demographic factors, necessitating region-aware adjustment. We propose Artemis, a region-level causal framework that bridges this gap with causal intervention at each brain region independently by learning region-specific confounder representations with lightweight parameters. Our adjustment comprehensively utilized the multimodal functional and structural features for graph reasoning as a plug-in module compatible with arbitrary GNN backbones. Experiments on three benchmarks, ADNI for disease diagnosis, OASIS for dementia staging, and HCP for sex classification, demonstrate consistent improvements over representative GNN-based baselines. Multiple supporting experiments further demonstrate statistical significance and neuroscientific interpretability.

AI 总结：综述了基于AI的土壤湿度建模方法，分为五类：统计时间序列、地统计、经典机器学习、深度学习和概率/贝叶斯方法，利用多源数据实现回归或分类。

链接：https://arxiv.org/abs/2606.18316

机构：Electrical Engineering\ University of Technology\ , Sweden

作者：Ilektra Tsimpidi, George Georgoulas, Vidya Sumathy, George Nikolakopoulos

英文摘要：Soil Moisture (SM) modelling constitutes a complex spatiotemporal learning problem characterised by nonlinear environmental interactions, heterogeneous data sources, and limited ground observations. Physics-based approaches, such as water balance models, rely on explicit hydrological equations and high-quality inputs, but their computational cost and scalability limitations restrict large-scale deployment. Data-driven artificial intelligence (AI) methods have emerged as flexible alternatives, enabling the extraction of empirical relationships between soil moisture and environmental variables with reduced modelling assumptions. This work presents a structured survey of AI-based models for soil moisture estimation and classification. Existing approaches are organized into five categories: (a) statistical time-series models, (b) geostatistical methods (c) classical machine learning (ML) models, (d) Deep Learning (DL) models and (e) Probabilistic/Bayesian methods. These models leverage historical soil moisture records, meteorological variables, vegetation indices, topography, soil characteristics, and geolocation data to perform regression or classification tasks.

AI 总结：提出ASTRA模拟器，通过微调ASR将词错误率降至23.45%，并集成AI评估框架，实现可扩展的标准化ATCO训练。

链接：https://arxiv.org/abs/2606.18319

机构：Air Emerging Technologies High-Speed Experimentations and Research (AETHER), RSAF Agile Innovation Digital (RAiD), Republic of Singapore Air Force（新加坡共和国空军敏捷创新数字实验室空中新兴技术高速实验与研究）

作者：Ethan Chew, Enjia Wu, Iruss Eng Wei Yeow, Ian Weiqin Lim, Ranen Sim, Brandon Koh Ziheng, Kaleb Nim, Caden Toh Jun Yi, Wei Dong Soin, Darius Kai Keat Koh, Galen King Yu Tay, Prannaya Gupta, Jonathan Ee Fang Koong, Yong Zhi Lim

英文摘要：Air Traffic Control Operators (ATCOs) are vital in ensuring the safe, orderly, and efficient flow of air traffic, yet training capacity is constrained by reliance on specialized human trainers known as simpilots, who must role-play both pilots and ATCOs in a simulated airspace. Existing automated solutions rely on Western-centric speech models that perform poorly in Singaporean operational contexts, with off-the-shelf systems exhibiting Word Error Rates (WER) of up to 107.80% on Singaporean-accented aviation speech. We introduce ASTRA, an end-to-end training simulator that automates these simpilot roles through a pipeline that transcribes ATCO speech, interprets instructions, and generates appropriate pilot and ATCO responses using locally adapted voice models. Our fine-tuned Automatic Speech Recognition (ASR) pipeline reduces WER to 23.45%, substantially outperforming existing approaches in this domain. Beyond traffic simulation, ASTRA incorporates an AI-assisted performance evaluation framework that assesses trainee radiotelephony communications across accuracy, brevity, and completeness, achieving post-optimization scores of 91.7%, 88.2%, and 86.9%, respectively. Built on open-source foundations such as DSPy and Unsloth, this approach enables scalable, standardized ATCO assessment while reducing instructor workload.

AI 总结：研究揭示拒绝推断方法在信用评分中因反馈循环导致评估指标误导，提出通过少量探索打破循环并诊断问题。

链接：https://arxiv.org/abs/2606.18479

机构：Northeastern University（东北大学）； KTH Royal Institute of Technology（瑞典皇家理工学院）

作者：Bruno Scarone, Ricardo Baeza-Yates

英文摘要：Reject inference methods are widely used to mitigate survival bias in credit scoring, yet their effectiveness remains poorly understood. We systematically evaluate several such methods and uncover a structural failure mode: in a natural retraining cycle, models whose accuracy improves while recall collapses create an illusion of improvement that leads practitioners to believe the system is getting better when, in fact, its rejection quality -- the ability to correctly screen out defaulters -- is deteriorating. We then propose a controlled exploration strategy that breaks the feedback loop without statistical assumptions: the lender deliberately approves a fraction of rejected applicants and observes their true outcomes. We show that accuracy and rejection quality give opposite recommendations on whether to explore: accuracy favors no exploration, while rejection quality improves with it, confirming that standard evaluation metrics are misleading under selection bias. Even minimal exploration rates (2--5\%) prove sufficient in our experiments to diagnose the severity of the feedback loop at near-zero cost. Our findings are consistent across two machine learning methods and three real-world datasets, and suggest that standard evaluation protocols are inadequate for assessing models trained under survival bias.

AI 总结：提出一种可解释的因果发现引导框架，从多模态PSG中推导层次化睡眠恢复评分(SRS)，在两大队列中SRS与感知恢复的关联强度是AHI的2.5倍。

链接：https://arxiv.org/abs/2606.18506

机构：University of California, Irvine（加州大学尔湾分校）

作者：Saba A. Farahani, Elahe Khatibi, Manoj Vishwanath, Amir M. Rahmani, Hung Cao

英文摘要：Objective sleep assessment relies on polysomnography (PSG), yet clinical impact is often better reflected in patient-reported outcomes (PROs) such as sleepiness and fatigue. Existing summary indices, including the Apnea-Hypopnea Index (AHI), provide limited insight into the multidomain physiology underlying functional recovery. We propose an interpretable, causal-discovery--guided framework for deriving a hierarchical Sleep Recovery Score (SRS) from multimodal PSG. Using two large population cohorts (MESA: n=1540; MrOS: n=825), we apply directed acyclic graph (DAG) learning to identify candidate physiological drivers spanning respiratory burden, hypoxic burden, sleep fragmentation, sleep architecture, and autonomic regulation. Although derived from clinical PSG, these domains map naturally to sensing streams increasingly available in connected health technologies, including wearable ECG, oximetry, and sleep-stage estimation devices. To preserve mechanistic plausibility, we introduce a two-stage screening process that combines physiology-based constraints with constrained LLM-assisted auditing to identify and remove structural confounders and construct-overlapping variables. Across cohorts, these five domains emerge as recurrent physiological domains associated with recovery, and the resulting SRS shows up to 2.5$\times$ stronger alignment with perceived recovery than AHI. By linking multimodal sleep physiology to patient-centered outcomes through an interpretable, bias-aware, and domain structured framework, this work provides a practical foundation for recovery modeling across both clinical sleep studies and emerging smart and connected health settings.

AI 总结：提出WGAN方法，通过可学习的校准变换将变化检测器响应分布映射回参考分布，在探测器模型和模拟量能器数据上验证了恢复老化系数和改善能量分布一致性的能力。

链接：https://arxiv.org/abs/2606.18561

机构：Laboratory of Methods for Big Data Analysis, HSE University（大数据分析方法实验室，高等经济大学）

作者：Saraa Ali, Vladimir Bocharnikov, Fedor Ratnikov, Mikhail Hushchyn, Artem Ryzhikov, Denis Derkach

英文摘要：The quality of recorded data depends on the stability of the sensor system that acquires it. Sensor motion and aging can degrade the performance and stability of downstream data-driven methods. We present a Wasserstein-GAN-inspired approach for unsupervised inference of physically interpretable transformation parameters that map a changed detector response distribution back to a nominal reference distribution. In contrast to standard generative modeling, the generator is used as a learnable calibration transformation whose trainable weights represent the sought parameters, while the critic provides a distributional distance signal via the Wasserstein objective. We validate the approach on a tracking-detector toy model with controlled layer shifts and demonstrate its application on high-granularity Geant4-simulated calorimeter data with cell-wise aging effects. The method recovers aging coefficients for individual cells with correlation to ground truth and improves agreement between calibrated and reference energy-sum distributions, while exhibiting the expected degradation at increasing channel-to-channel noise levels. These results indicate that adversarial distribution matching can serve as a data-driven component of calibration strategies in settings where direct labels for degradation parameters are unavailable.

AI 总结：提出一种多模态框架，结合跨模态融合和梯度反转去学习，减少人口统计信息对轻度认知障碍检测的偏见，在跨语言数据集上缩小性能差距。

链接：https://arxiv.org/abs/2606.18571

机构：University of Massachusetts Lowell, USA（马萨诸塞大学洛厄尔分校）

作者：William Nguyen, Jiali Cheng, Hadi Amiri

英文摘要：Mild Cognitive Impairment (MCI) is a medical condition characterized by a noticeable decline in memory, language, or thinking abilities. MCI detection from spontaneous speech is promising for scalable screening. However, learned models often exploit demographic cues correlated with labels, resulting in a large performance gap across subgroups. We present a multimodal framework that combines (i) cross-model fusion between modalities (speech, text, and image), and (ii) unlearning using gradient reversal that discourages the shared embedding from encoding task-irrelevant demographic attributes. Evaluated on the multilingual benchmarks TAUKADIAL and PREPARE, our method outperforms the state-of-the-art multilingual and multimodal baseline in MCI classification while substantially reducing the performance gap across patient subgroups (sex and language). We further analyze transfer across datasets, showing that demographic unlearning helps learn more robust representations for MCI detection.

AI 总结：提出scGTN框架，通过孪生图变换网络整合基因表达与细胞间结构信息，利用最优传输策略进行自监督聚类，在多个数据集上优于现有方法。

链接：https://arxiv.org/abs/2606.18672

机构：Sichuan University（四川大学）； University of International Business and Economics（对外经济贸易大学）； Great Bay University（大湾区大学）； The Education University of Hong Kong（香港教育大学）

作者： Jinke Wu, Yifan Wang, Siyu Yi, Caiyang Yu, Ziyue Qiao, Nan Yin, Jiancheng Lv, Wei Ju

英文摘要：Single-cell RNA sequencing (scRNA-seq) serves a pivotal role in characterizing gene expression at the cellular level, enabling the identification of cell types and advancing the understanding of cellular heterogeneity. Despite the significant progress in scRNA-seq data clustering, we argue that current methods always ignore the sparsity and noise, as well as the complex intercellular structural information inherent in scRNA-seq data. Toward this end, in this paper, we propose a novel single-cell RNA-seq clustering framework via deep Siamese Graph Transformer Network (termed scGTN), which explicitly integrates gene expression profile and intercellular structural dependencies for cell clustering. In particular, we formulate scRNA-seq data as a graph and construct two augmented graph views that serve as dual views to capture complementary intercellular information. Then, a Siamese graph transformer network is employed to explicitly incorporate shortest-path information and node-wise distances for capturing richer structural relationships between cells. Finally, we employ an optimal transport strategy to guide the cell clustering in a self-supervised manner. Extensive experiments on multiple benchmark scRNA-seq datasets demonstrate that our scGTN consistently outperforms existing methods. Our code is available at this https URL.

AI 总结：提出一种光子量子神经场，将坐标编码为可训练光学相位，通过多光子Fock空间干涉混合并从光子数测量解码，作为物理信息残差最小化的可训练表示，在七种PDE基准上展示相位复杂度转变，在困难区域误差低一个数量级且参数少约四分之一。

链接：https://arxiv.org/abs/2606.18713

机构：Xidian University（西安电子科技大学）； National University of Singapore（新加坡国立大学）

作者：Jiale Linghu, Hao Dong, Yangshuai Wang

英文摘要：Photonic quantum machine learning offers a route to trainable physical representations built from phase, interference and measurement. However, its role in scientific machine learning remains largely unexplored. Physics-informed neural fields provide a natural setting, because differential equations require trial spaces that preserve phase, frequency and derivative structure. Here we introduce a photonic quantum neural field in which coordinates become trainable optical phases, are mixed by multi-photon Fock-space interference and are decoded from photon-number measurements. The photonic circuit is optimized as the neural-field representation itself, not as a fixed feature map or hardware accelerator. Photonic measurement is therefore a trainable representation on which the physics-informed residual is minimized. Across seven elliptic, wave, nonlinear dispersive and inverse PDE benchmarks, we observe a phase-complexity transition: classical coordinate and Fourier-feature networks suffice in smooth regimes, whereas the photonic field is most accurate when residual derivatives amplify phase mismatch. In the hardest regimes it gives the lowest errors, with margins reaching an order of magnitude and about one quarter of the trainable parameters of classical baselines. Frozen and shuffled controls, together with noise stress tests, attribute this gain to learned interference and stable Fock-probability readout under compound perturbations. These results identify photonic quantum measurement as a representation-learning principle for scientific machine learning.

AI 总结：提出图锚定交叉注意力Transformer（GGATN），通过全局过程图作为结构化记忆、Transformer自注意力编码序列位置、图锚定交叉注意力注入过程拓扑，结合维特比式图约束解码，一次性生成完整事件序列，在六个基准日志上优于LLM基线。

链接：https://arxiv.org/abs/2606.18726

机构：Department of Computer Science, University of Milan（米兰大学计算机科学系）

作者：Fang Wang, Ernesto Damiani

英文摘要：Structurally constrained event sequence generation remains challenging because generated paths must preserve transition feasibility, temporal order, termination, and attribute consistency. In predictive process monitoring (PPM), this challenge appears as full event sequence generation, whereas existing work mainly addresses component tasks such as next activity, remaining time, outcome, and attribute prediction. This paper proposes the Graph Grounded Cross Attention Transformer Neural Network (GGATN) for this unified PPM task. GGATN uses a global process graph as structured activity memory, contextualizes sequence positions through Transformer self attention, and injects process topology through graph grounded cross attention. Unlike autoregressive decoding, GGATN generates activities, timestamps, length, and event level and sequence level attributes in a single pass, followed by Viterbi style graph constrained decoding for feasible paths and explicit termination. Experiments on six benchmark event logs show more reliable generation quality than local instruction prompted LLM baselines. GGATN achieves strong performance on sequence similarity, Damerau Levenshtein similarity, bigram based control flow similarity, and duration distribution, while maintaining zero hallucinated activities and zero sequence level attribute inconsistency. Ablation analyses confirm the global graph encoder as a stable structural prior. Interpretability analyses show how graph structure, sequence context, feedback refinement, and constrained decoding shape generation.

AI 总结：提出混合SNN-CNN模型，从智能手机视频合成事件相机数据，实现高效准确的跌倒检测。

链接：https://arxiv.org/abs/2606.18732

机构：School of Electrical Engineering Pontificia Universidad Católica de Valparaíso, Chile

作者：Guillermo Rojas, Gonzalo Soto, Daniel Yunge

英文摘要： This work presents the development of hybrid models that integrate spiking neural networks (SNNs) with components of convolutional neural networks (CNNs) to learn from simulated event-based camera data (Dynamic Vision Sensor, DVS) generated from conventional smartphone videos. Aimed primarily at human fall detection, the approach leverages the energy efficiency and spatio-temporal processing capabilities of SNNs by converting video frames into event-based data. The proposed models are evaluated through simulations on multiple datasets, comparing their performance to that of traditional machine learning models. Results demonstrate significant gains in efficiency without sacrificing accuracy, underscoring the potential of combining SNNs and DVS technology for complex tasks in real-world environments.

AI 总结：测试不同架构的归纳偏差对模拟平流层突然增温动力学的影响，发现三维垂直耦合是关键，但低预测误差不保证物理一致性。

链接：https://arxiv.org/abs/2606.18857

机构：Technical University of Denmark（丹麦技术大学）； University of Cambridge（剑桥大学）； University of Exeter（埃克塞特大学）

作者：Oskar Bohn Lassen, Simon Driscoll, Stephen I. Thomson, Sebastian Schemm, Francisco C. Pereira

英文摘要：Machine-learning emulators are increasingly used for weather prediction and have the potential to extend skill on subseasonal-to-seasonal timescales by learning dynamically important sources of predictability. A key challenge is whether the models can exploit predictability anchors, such as stratospheric variability, that influence tropospheric circulation beyond short lead times. We test how architectural inductive bias affects emulation of sudden stratospheric warming (SSW) dynamics using paired idealised Isca simulations that differ only in an imposed wave-2 heating perturbation. Across convolutional, transformer, and graph-based architectures trained for one-step prediction, model differences are modest when the stratosphere is dynamically quiet but widen substantially when SSW-like variability is active. Our results identify explicit three-dimensional vertical coupling as a key inductive bias for machine-learning emulation of stratospheric dynamics. However, Eliassen-Palm flux diagnostics show that low forecast error does not guarantee physically faithful wave-mean-flow interaction, with coherent errors remaining in stratospheric wave-driving structure.

AI 总结：提出稳定元反馈半监督学习框架，通过噪声感知解耦和运动学门控伪标签，利用大规模无标签数据提升自动紧急制动性能，实现超100:1正误触发比和35%无事故里程提升。

链接：https://arxiv.org/abs/2606.18864

机构：Li Auto（理想汽车）

作者：Xiangyu Wang, Yang Zhan, Mengxiang Hao, Chuanchuan Zhong, Yansong Jia, Junjie Zhang, Yu Han, Xin Jiang, Zhen Cao, Ying Wang, Yulun Song, Zhitao Xu

英文摘要：This paper studies how to scale learning-based automatic emergency braking (AEB) with massive unlabeled fleet data under production constraints. Our approach is based on meta-feedback semi-supervised learning (MF-SSL), where a teacher generates pseudo labels for unlabeled driving data and is updated using a small labeled anchor set as safety-critical feedback. In production, anchor ambiguity and labeled-unlabeled mismatch can amplify systematic pseudo-label errors, leading to spurious triggers. We propose a stabilized MF-SSL framework with (i) Noise-Aware Decoupling, which removes ambiguity-prone anchors from the teacher's supervised update path, and (ii) kinematics-gated pseudo-labeling with a teacher conflict penalty to suppress mismatch-induced risk hallucinations on unlabeled data while maintaining broad coverage. Extensive experiments show consistent gains as unlabeled data scale from 1M to 1B windows, improving safety while keeping comfort stable. The 1B-trained student model is deployed to hundreds of thousands of vehicles and validated over \$10^9$ km of driving, achieving a positive-to-false activation ratio exceeding 100:1 and a 35% improvement in accident-free driving mileage over a production rule-only baseline.

AI 总结：提出域偏移感知神经网络，通过最大均值差异策略对齐源域与目标域特征，解决变工况下旋转轴不平衡质量估计的回归问题，实验证明该方法在域偏移未知时显著提升预测精度。

链接：https://arxiv.org/abs/2606.18882

机构：Rio de Janeiro State University (UERJ)（里约热内卢州立大学）； University of Campinas (UNICAMP)（坎皮纳斯大学）； CPQD； São Francisco University (USF)（圣弗朗西斯科大学）； Federal University of Rio de Janeiro (UFRJ)（里约热内卢联邦大学）

作者：Bernardo Feijó Junqueira, Claudio Kiyoshi Umezu, Bruno Bilhar Karaziack, Tomaz Junior, Daniel Alves Castello

英文摘要：This work investigates the application of a domain-shift aware neural network for regression tasks aimed at estimating unbalance masses in rotating shafts under varying operating conditions. Experimental data were collected from a test rig in which a primary shaft, equipped with a flange carrying unbalanced masses, was driven at different rotational speeds, while a secondary shaft could be optionally activated to introduce domain discrepancy. The unbalance masses were positioned at a fixed radial distance, and the dynamic response of the system was recorded using triaxial accelerometers. The inverse problem of mass estimation is formulated within a domain adaptation framework, where the network is trained with a maximum mean discrepancy strategy to align feature representations across source and target distributions. The results demonstrate the effectiveness of explicitly addressing domain shift in improving prediction accuracy, especially when the system's physical behavior and sources of domain discrepancy are not fully known and fall outside the training conditions. These findings highlight the potential of domain-shift aware models for regression tasks in Structural Health Monitoring.

AI 总结： 提出通过LLM启发式获取马尔可夫随机场充分统计量的零样本主动特征获取框架，解决数据标注不足问题，在IBD患者诊断中优于现有方法。

链接：https://arxiv.org/abs/2606.18933

机构：Faculty of EE, Technion（技术学院电子工程系）； Faculty of Medicine, Technion（技术学院医学院）； CytoReason； NVIDIA

作者：Binyamin Perets, Natalie Mendelson, Shiran Vainberg, Yehuda Chowers, Shai Shen-Orr, Shie Mannor

英文摘要：Active feature acquisition (AFA) sequentially selects which features to observe to reach a classification or ranking decision. Its central limitation is reliance on large amount of labeled data to fit probabilistic models guiding acquisition. Large language models (LLMs) supply unsupervised domain knowledge, but are poor sequential planners. Asking one to both know and decide conflates capabilities best kept separate. Here, we develop a framework for zero-shot AFA through disciplined elicitation: asking the LLM only for what it can be trusted to return, the unary deviations and pairwise co-variations that are the sufficient statistics of a Markov random field (MRF). We apply our framework to two settings: binary classification and top-$k$ identification. In practice, the LLM reliably returns only discriminative statistics, what distinguishes the classes rather than each class in isolation, which precludes classical AFA. We apply a maximum-entropy closure that resolves this gauge ambiguity. We evaluate on a cohort of Inflammatory Bowel Disease (IBD) patients, an active clinical setting where diagnostic ambiguity and patient heterogeneity obstruct stable treatment strategies. Our framework outperforms the LLM both on real labels and on its own extracted beliefs. Where it matters most, on the hardest patients, our top-$k$ acquisition policy markedly outperforms all existing methods.

AI 总结：提出LSTM-ViT混合框架，结合地表观测时序与大气廓线，预测HRRR降水、风速和温度预报误差，相比基线LSTM性能提升，尤其降水误差预测技能提高约两倍。

链接：https://arxiv.org/abs/2606.19026

机构：Atmospheric Sciences Research Center, University at Albany, SUNY（纽约州立大学奥尔巴尼分校大气科学研究中心）； University of Oklahoma（俄克拉荷马大学）； State Weather Risk Communication Center, University at Albany, SUNY（纽约州立大学奥尔巴尼分校州天气风险沟通中心）

作者：David Aaron Evans, Jay C. Rothenberger, Kara J. Sulia, Nick P. Bassill, Chris D. Thorncroft

英文摘要：Forecast errors in high-resolution numerical weather prediction (NWP) systems are often linked to unresolved planetary boundary layer (PBL) processes, convection, terrain-induced circulations, and other vertically structured atmospheric phenomena. Previous work demonstrated that Long Short-Term Memory (LSTM) networks can successfully predict forecast errors in the High-Resolution Rapid Refresh (HRRR) model using mesonet observations, but we believe performance degradation is linked to periods of complex vertical atmospheric evolution. To address this limitation, we develop a hybrid LSTM-Vision Transformer (LSTM-ViT) framework that combines temporal sequence learning from surface observations with atmospheric profiles from the New York State Mesonet profiler network. The LSTM-ViT framework is trained to predict HRRR hourly precipitation, 10 m wind speed, and 2 m temperature forecast errors at individual mesonet stations. Across all three predictors, incorporation of profiler-derived atmospheric structure improves forecast error prediction skill relative to the baseline LSTM architecture, with the largest gains occurring at shorter forecast lead times and during periods of enhanced PBL activity. Improvements are particularly pronounced for precipitation forecast error, where the LSTM-ViT framework achieves approximately a twofold increase in predictive skill relative to the baseline LSTM while better capturing convectively driven error evolution and reducing degradation associated with PBL processes. These results demonstrate that combining temporal sequence learning with vertically informed attention mechanisms provides a physically meaningful pathway for improving forecast error prediction in operational NWP systems. Our research offers forecasters enhanced guidance regarding model bias and forecast confidence.

AI 总结：针对Airbnb中客人序列长、探索性强且标签稀疏的问题，提出JourneyFormer序列建模解决方案，通过优化数据选择、ID嵌入、模型架构和标签归因，并在两个生产面上通过在线A/B测试验证了其有效性。

链接：https://arxiv.org/abs/2606.19108

机构：Airbnb

作者：Daochen Zha, Chun How Tan, Xin Liu, Bin Xu, Han Zhao, Xiaowei Liu, Tracy Yu, Hui Gao, Huiji Gao, Liwei He, Stephanie Moyerman, Sanjeev Katariya

英文摘要：Sequence modeling has become increasingly popular in recommendation and ranking algorithms, owing to its capacity to model users' historical behaviors and infer user intentions. Despite its theoretical simplicity, the practical deployment of a sequence model in production is non-trivial due to complexity of the sequence and sparse labels. For example, in Airbnb, guest sequences are often long, exploratory and complex, and we focus on booking labels, which are sparse. As such, we are often required to make various design decisions regarding data and modeling to strike a balance between effectiveness and scalability. This work delved into these production challenges and deployed JourneyFormer, a sequence modeling solution for search ranking at Airbnb. We detail crucial design considerations, covering aspects such as guest event selection, ID embeddings, model architecture, and label attribution. Additionally, we describe several tailored strategies to accelerate model training and inference. JourneyFormer has been successfully deployed within Airbnb's production, where its effectiveness and impact have been evidenced not only by improved offline ranking metrics but also by significant gains in key business metrics through online A/B testing across 2 production surfaces.

AI 总结：提出ChronoSurv，一种基于有向图的多模态生存分析框架，通过层次化拓扑和异质消息传递建模临床轨迹，在头颈癌数据集上取得最优判别性能与可靠校准。

链接：https://arxiv.org/abs/2606.19140

机构：Université Paris-Saclay, CentraleSupélec, MICS, France（巴黎-萨克雷大学，中央理工-高等电力学院，MICS，法国）； University of Lyon, INSA Lyon, CREATIS, France（里昂大学，INSA里昂，CREATIS，法国）

作者：Hugo Miccinilli, Theo Di Piazza

英文摘要：Accurate survival prediction is essential for personalized treatment planning in head and neck cancer, yet remains challenging due to the heterogeneous and high-dimensional nature of multimodal clinical data. While deep survival models have improved predictive performance over classical statistical approaches, existing methods typically rely on static fusion strategies or temporally agnostic modeling, limiting their ability to capture structured clinical workflows. In this work, we propose ChronoSurv, a heterogeneous hierarchical directed graph framework for multimodal survival analysis. ChronoSurv represents patient care as a progression-aware clinical trajectory using directed graphs aligned with key diagnostic steps. A hierarchical topology incorporates fine-grained, coarse, and global representations, further supporting flexible adaptation to missing modalities, while heterogeneous message passing models complex and asymmetric relationships across modalities and clinical steps. Experimental results on two public datasets demonstrate that ChronoSurv achieves state-of-the-art discriminative performance while maintaining statistically reliable calibration. Comprehensive ablation studies further confirm the contribution of each architectural component, highlighting the potential of trajectory-aware graph modeling for multimodal survival prediction.

AI 总结：提出一种扩展的帕累托前沿引导采样框架，通过将高斯过程代理的约束满足概率和鲁棒性作为多目标优化目标，结合交互式仪表盘实现人机协同的约束感知生物过程优化。

链接：https://arxiv.org/abs/2606.19230

机构：Imperial College London（伦敦帝国理工学院）； DataHow AG； ETH Zurich（苏黎世联邦理工学院）

作者：Samuel Stricker, Claus Wirnsperger, Alessandro Butté, Laura Helleckes, Gonzalo Guillén Gosálbez, Antonio del Rio Chanona, Mehmet Mercangöz

英文摘要：This work presents an extension to Pareto Front Guided Sampling (PFGS), a Human-in-the-Loop (HitL) Bayesian Optimization (BO) framework in which Gaussian process (GP) surrogate-derived quantities are reformulated as objectives of a multi-objective optimization problem, and the resulting Pareto front is exposed to a domain expert for interactive candidate selection rather than returning a single automated recommendation. The framework is extended in two directions: constrained optimization is addressed by incorporating the posterior probability of satisfying output specification limits as an explicit Pareto objective, computed analytically from the GP posterior distribution; robust optimization is addressed by a Monte Carlo sampling strategy that estimates expected lower-confidence performance over a user-defined variability of input perturbations, capturing performance degradation under likely implementation deviations. The resulting multi-dimensional Pareto representation renders trade-offs between predicted performance, model uncertainty, probabilistic constraint satisfaction, and input robustness simultaneously visible through pairwise two-dimensional projections on an interactive dashboard, enabling selection criteria to be iteratively refined as the surrogate model improves and development objectives evolve. The framework is showcased on an eight-dimensional fed-batch Chinese Hamster Ovary (CHO) cell culture simulator demonstrating systematic identification of high-performing, feasibility-compliant, and perturbation-resilient operating conditions, and illustrating how expert-defined requirements provide a principled stopping criterion and support informed allocation of experimental resources.

AI 总结：提出SCAN方法，通过多尺度聚类增强重建型异常检测，在表示层集成正常模式聚类中心约束重建，在异常判据层结合聚类概率与重建误差，并利用邻域中心表示改进聚类性能，在多个真实数据集上达到最优。

链接：https://arxiv.org/abs/2606.19255

机构：East China Normal University（华东师范大学）； APPLab, Huawei（华为2012应用实验室）； Huawei（华为）

作者：Xingze Zheng, Hanyin Cheng, Siyuan Wang, Yiting Hao, Peng Chen, Yuan Jun, Yang Shu

英文摘要：Time series anomaly detection plays a crucial role in a wide range of real-world applications. Reconstruction-based methods have become the mainstream paradigm, but they suffer from over-generalization and under-generalization problems, which are challenging to balance. To address this, we introduce multi-scale clustering to enhance reconstruction-based methods. At the representation level, we integrate the cluster center representations of normal patterns to constrain the model to target representative normal patterns for reconstruction, preventing dominance of powerful capacity and representation capability. At the anomaly criterion level, we derive anomaly confidence score based on cluster membership probability and combine it with reconstruction error, providing dual criteria for detection. Furthermore, the effectiveness of the cluster center representations and anomaly confidence score depends on the clustering performance. Accordingly, we extract neighborhood-centered representations for multi-view clustering to improve clustering performance. Extensive experiments on multiple real-world datasets from diverse application domains demonstrate the state-of-the-art performance of SCAN.

AI 总结：本研究利用环境声音和光照强度数据，通过高效序列神经网络模型预测ICU患者谵妄风险，发现声音是主要预测因子，结合光照可改善短期预测，AUC达0.80。

链接：https://arxiv.org/abs/2606.19292

机构：University of Florida（佛罗里达大学）； Stanford University（斯坦福大学）

作者： Jiaqing Zhang, Sabyasachi Bandyopadhyay, Miguel Contreras, Jessica Sena, Yuanfang Ren, Andrea Davidson, Ziyuan Guan, Tezcan Ozrazgat-Baslanti, Subhash Nerella, Azra Bihorac, Parisa Rashidi

英文摘要：Delirium is a common and serious complication in the Intensive Care Unit (ICU), associated with increased morbidity, prolonged hospital stays, and higher healthcare costs. Despite its prevalence, early prediction and prevention remain challenging. Environmental factors such as ambient sound and light may influence the onset of delirium, yet they are often overlooked in risk assessments. In this study, we examined whether light intensity and sound pressure levels can independently predict delirium across multiple prediction horizons. We evaluated four efficient sequential neural network models on data collected from 9 ICUs across 309 patients to predict delirium for 10 prediction-window sizes. We reported feature importance and direction of influence using Shapley Additive Explanations analysis. The convolutional model achieved the strongest discrimination, with AUC = 0.80 on sound data and on combined data. Sound features were the dominant predictors overall. Integrating sound with light improved short-term ($<1$ week) prediction, with the combined model assigning the highest risk immediately after the sensing period. These findings suggest that passive ambient sensing, especially sound, can add a clinically meaningful, interpretable signal for delirium risk estimation and offer a practical pathway to enrich multimodal ICU prediction and prevention strategies.

AI 总结：提出用可执行程序近似深度网络组件行为的方法，针对Transformer注意力头，通过生成Python程序再现注意力模式，实现可解释性。

链接：https://arxiv.org/abs/2606.19317

机构：NJIT（新泽西理工学院）； MIT EECS（麻省理工学院电气工程与计算机科学系）； MIT CSAIL（麻省理工学院计算机科学与人工智能实验室）

作者：Amiri Hayes, Belinda Li, Jacob Andreas

英文摘要：A longstanding goal of research on interpretable deep learning is to replace opaque neural computations with human-meaningful symbolic descriptions. In this paper, we propose an approach for approximating the behavior of components of deep networks with executable programs. We focus on attention heads in transformer language models. For a given head, we first compute its associated attention matrices on a collection of randomly selected training examples. Next, we prompt a pre-trained language model with a summary of these matrices, and instruct it to generate a set of Python programs that can reproduce the associated attention patterns given only text from the input sentence. Finally, we re-rank programs according to how well our final set of programs predict behavior on held-out inputs. We demonstrate that a set of fewer than 1,000 such generated programs can reproduce the attention patterns of heads in GPT-2, TinyLlama-1.1B, and Llama-3B, achieving an average Intersection-over-Union similarity above 75% on TinyStories. Moreover, the best-fit programs can replace neural attention heads without substantially affecting model behavior: replacing 25% of attention heads with programmatic surrogates across the three models incurs only a 16% average perplexity increase, while maintaining performance on a variety of downstream question answering benchmarks. This work contributes a scalable pipeline for reverse-engineering attention heads in transformer models using human-readable, executable code, advancing a path toward symbolic transparency in neural models.