机器学习学术速递[2.4]

点击阅读原文访问arxivdaily.com，涵盖CS|物理|数学|经济|统计|金融|生物|电气领域，更有搜索、收藏等功能！

cs.LG 方向，今日共计359篇

大模型相关(60篇)

【1】Fast-Slow Efficient Training for Multimodal Large Language Models via Visual Token Pruning
标题：通过视觉标记修剪对多模式大型语言模型进行快速-缓慢高效训练
链接：https://arxiv.org/abs/2602.03815

作者：Dingkun Zhang,Shuhan Qi,Yulin Wu,Xinyu Xiao,Xuan Wang,Long Chen
摘要：Multimodal Large Language Models (MLLMs) suffer from severe training inefficiency issue, which is associated with their massive model sizes and visual token numbers. Existing efforts in efficient training focus on reducing model sizes or trainable parameters. Inspired by the success of Visual Token Pruning (VTP) in improving inference efficiency, we are exploring another substantial research direction for efficient training by reducing visual tokens. However, applying VTP at the training stage results in a training-inference mismatch: pruning-trained models perform poorly when inferring on non-pruned full visual token sequences. To close this gap, we propose DualSpeed, a fast-slow framework for efficient training of MLLMs. The fast-mode is the primary mode, which incorporates existing VTP methods as plugins to reduce visual tokens, along with a mode isolator to isolate the model's behaviors. The slow-mode is the auxiliary mode, where the model is trained on full visual sequences to retain training-inference consistency. To boost its training, it further leverages self-distillation to learn from the sufficiently trained fast-mode. Together, DualSpeed can achieve both training efficiency and non-degraded performance. Experiments show DualSpeed accelerates the training of LLaVA-1.5 by 2.1$\times$ and LLaVA-NeXT by 4.0$\times$, retaining over 99% performance. Code: https://github.com/dingkun-zhang/DualSpeed

【2】Understanding Agent Scaling in LLM-Based Multi-Agent Systems via Diversity
标题：通过多样性理解基于LLM的多代理系统中的代理扩展
链接：https://arxiv.org/abs/2602.03794

作者：Yingxuan Yang,Chengrui Qu,Muning Wen,Laixi Shi,Ying Wen,Weinan Zhang,Adam Wierman,Shangding Gu
摘要：LLM-based multi-agent systems (MAS) have emerged as a promising approach to tackle complex tasks that are difficult for individual LLMs. A natural strategy is to scale performance by increasing the number of agents; however, we find that such scaling exhibits strong diminishing returns in homogeneous settings, while introducing heterogeneity (e.g., different models, prompts, or tools) continues to yield substantial gains. This raises a fundamental question: what limits scaling, and why does diversity help? We present an information-theoretic framework showing that MAS performance is bounded by the intrinsic task uncertainty, not by agent count. We derive architecture-agnostic bounds demonstrating that improvements depend on how many effective channels the system accesses. Homogeneous agents saturate early because their outputs are strongly correlated, whereas heterogeneous agents contribute complementary evidence. We further introduce $K^*$, an effective channel count that quantifies the number of effective channels without ground-truth labels. Empirically, we show that heterogeneous configurations consistently outperform homogeneous scaling: 2 diverse agents can match or exceed the performance of 16 homogeneous agents. Our results provide principled guidelines for building efficient and robust MAS through diversity-aware design. Code and Dataset are available at the link: https://github.com/SafeRL-Lab/Agent-Scaling.

【3】Reasoning with Latent Tokens in Diffusion Language Models
标题：扩散语言模型中的潜在符号推理
链接：https://arxiv.org/abs/2602.03769

作者：Andre He,Sean Welleck,Daniel Fried
摘要：Discrete diffusion models have recently become competitive with autoregressive models for language modeling, even outperforming them on reasoning tasks requiring planning and global coherence, but they require more computation at inference time. We trace this trade-off to a key mechanism: diffusion models are trained to jointly predict a distribution over all unknown tokens, including those that will not actually be decoded in the current step. Ablating this joint prediction yields faster inference but degrades performance, revealing that accurate prediction at the decoded position relies on joint reasoning about the distribution of undecoded tokens. We interpret these as latent tokens and introduce a method for modulating their number, demonstrating empirically that this enables a smooth tradeoff between inference speed and sample quality. Furthermore, we demonstrate that latent tokens can be introduced into autoregressive models through an auxiliary multi-token prediction objective, yielding substantial improvements on the same reasoning tasks where they have traditionally struggled. Our results suggest that latent tokens, while arising naturally in diffusion, represent a general mechanism for improving performance on tasks requiring global coherence or lookahead.

【4】LLM-Inspired Pretrain-Then-Finetune for Small-Data, Large-Scale Optimization
标题：受法学硕士启发的预训练然后微调，用于小数据、大规模优化
链接：https://arxiv.org/abs/2602.03690

作者：Zishi Zhang,Jinhui Han,Ming Hu,Yijie Peng
摘要：We consider small-data, large-scale decision problems in which a firm must make many operational decisions simultaneously (e.g., across a large product portfolio) while observing only a few, potentially noisy, data points per instance. Inspired by the success of large language models (LLMs), we propose a pretrain-then-finetune approach built on a designed Transformer model to address this challenge. The model is first pretrained on large-scale, domain-informed synthetic data that encode managerial knowledge and structural features of the decision environment, and is then fine-tuned on real observations. This new pipeline offers two complementary advantages: pretraining injects domain knowledge into the learning process and enables the training of high-capacity models using abundant synthetic data, while finetuning adapts the pretrained model to the operational environment and improves alignment with the true data-generating regime. While we have leveraged the Transformer's state-of-the-art representational capacity, particularly its attention mechanism, to efficiently extract cross-task structure, our approach is not an off-the-shelf application. Instead, it relies on problem-specific architectural design and a tailored training procedure to match the decision setting. Theoretically, we develop the first comprehensive error analysis regarding Transformer learning in relevant contexts, establishing nonasymptotic guarantees that validate the method's effectiveness. Critically, our analysis reveals how pretraining and fine-tuning jointly determine performance, with the dominant contribution governed by whichever is more favorable. In particular, finetuning exhibits an economies-of-scale effect, whereby transfer learning becomes increasingly effective as the number of instances grows.

【5】When Single Answer Is Not Enough: Rethinking Single-Step Retrosynthesis Benchmarks for LLMs
标题：当单一答案还不够时：重新思考LLM的一步回归合成基准
链接：https://arxiv.org/abs/2602.03554

作者：Bogdan Zagribelnyy,Ivan Ilin,Maksim Kuznetsov,Nikita Bondarev,Roman Schutski,Thomas MacDougall,Rim Shayakhmetov,Zulfat Miftakhutdinov,Mikolaj Mizera,Vladimir Aladinskiy,Alex Aliper,Alex Zhavoronkov
摘要：Recent progress has expanded the use of large language models (LLMs) in drug discovery, including synthesis planning. However, objective evaluation of retrosynthesis performance remains limited. Existing benchmarks and metrics typically rely on published synthetic procedures and Top-K accuracy based on single ground-truth, which does not capture the open-ended nature of real-world synthesis planning. We propose a new benchmarking framework for single-step retrosynthesis that evaluates both general-purpose and chemistry-specialized LLMs using ChemCensor, a novel metric for chemical plausibility. By emphasizing plausibility over exact match, this approach better aligns with human synthesis planning practices. We also introduce CREED, a novel dataset comprising millions of ChemCensor-validated reaction records for LLM training, and use it to train a model that improves over the LLM baselines under this benchmark.

【6】Can Large Language Models Generalize Procedures Across Representations?
标题：大型语言模型可以跨表示概括过程吗？
链接：https://arxiv.org/abs/2602.03542

作者：Fangru Lin,Valentin Hofmann,Xingchen Wan,Weixing Wang,Zifeng Ding,Anthony G. Cohn,Janet B. Pierrehumbert
摘要：Large language models (LLMs) are trained and tested extensively on symbolic representations such as code and graphs, yet real-world user tasks are often specified in natural language. To what extent can LLMs generalize across these representations? Here, we approach this question by studying isomorphic tasks involving procedures represented in code, graphs, and natural language (e.g., scheduling steps in planning). We find that training LLMs with popular post-training methods on graphs or code data alone does not reliably generalize to corresponding natural language tasks, while training solely on natural language can lead to inefficient performance gains. To address this gap, we propose a two-stage data curriculum that first trains on symbolic, then natural language data. The curriculum substantially improves model performance across model families and tasks. Remarkably, a 1.5B Qwen model trained by our method can closely match zero-shot GPT-4o in naturalistic planning. Finally, our analysis suggests that successful cross-representation generalization can be interpreted as a form of generative analogy, which our curriculum effectively encourages.

【7】Not All Negative Samples Are Equal: LLMs Learn Better from Plausible Reasoning
标题：并非所有的负样本都是相等的：LLM从合理的推理中学习得更好
链接：https://arxiv.org/abs/2602.03516

作者：Zixiang Di,Jinyi Han,Shuo Zhang,Ying Liao,Zhi Li,Xiaofeng Ji,Yongqi Wang,Zheming Yang,Ming Gao,Bingdong Li,Jie Wang
摘要：Learning from negative samples holds great promise for improving Large Language Model (LLM) reasoning capability, yet existing methods treat all incorrect responses as equally informative, overlooking the crucial role of sample quality. To address this, we propose Plausible Negative Samples (PNS), a method that synthesizes high-quality negative samples exhibiting expected format and structural coherence while ultimately yielding incorrect answers. PNS trains a dedicated model via reverse reinforcement learning (RL) guided by a composite reward combining format compliance, accuracy inversion, reward model assessment, and chain-of-thought evaluation, generating responses nearly indistinguishable from correct solutions. We further validate PNS as a plug-and-play data source for preference optimization across three backbone models on seven mathematical reasoning benchmarks. Results demonstrate that PNS consistently outperforms other negative sample synthesis methods, achieving an average improvement of 2.03% over RL-trained models.

【8】Lookahead Path Likelihood Optimization for Diffusion LLMs
标题：扩散LLC的前瞻路径可能性优化
链接：https://arxiv.org/abs/2602.03496

作者：Xuejie Liu,Yap Vit Chun,Yitao Liang,Anji Liu
摘要：Diffusion Large Language Models (dLLMs) support arbitrary-order generation, yet their inference performance critically depends on the unmasking order. Existing strategies rely on heuristics that greedily optimize local confidence, offering limited guidance for identifying unmasking paths that are globally consistent and accurate. To bridge this gap, we introduce path log-likelihood (Path LL), a trajectory-conditioned objective that strongly correlates with downstream accuracy and enables principled selection of unmasking paths. To optimize Path LL at inference time, we propose POKE, an efficient value estimator that predicts the expected future Path LL of a partial decoding trajectory. We then integrate this lookahead signal into POKE-SMC, a Sequential Monte Carlo-based search framework for dynamically identifying optimal unmasking paths. Extensive experiments across 6 reasoning tasks show that POKE-SMC consistently improves accuracy, achieving 2%--3% average gains over strong decoding-time scaling baselines at comparable inference overhead on LLaDA models and advancing the accuracy--compute Pareto frontier.

【9】Self-Verification Dilemma: Experience-Driven Suppression of Overused Checking in LLM Reasoning
标题：自我验证困境：经验驱动抑制LLM推理中过度使用检查
链接：https://arxiv.org/abs/2602.03485

作者：Quanyu Long,Kai Jie Jiang,Jianda Chen,Xu Guo,Leilei Gan,Wenya Wang
备注：19 pages, 8 figures
摘要：Large Reasoning Models (LRMs) achieve strong performance by generating long reasoning traces with reflection. Through a large-scale empirical analysis, we find that a substantial fraction of reflective steps consist of self-verification (recheck) that repeatedly confirm intermediate results. These rechecks occur frequently across models and benchmarks, yet the vast majority are confirmatory rather than corrective, rarely identifying errors and altering reasoning outcomes. This reveals a mismatch between how often self-verification is activated and how often it is actually useful. Motivated by this, we propose a novel, experience-driven test-time framework that reduces the overused verification. Our method detects the activation of recheck behavior, consults an offline experience pool of past verification outcomes, and estimates whether a recheck is likely unnecessary via efficient retrieval. When historical experience suggests unnecessary, a suppression signal redirects the model to proceed. Across multiple model and benchmarks, our approach reduces token usage up to 20.3% while maintaining the accuracy, and in some datasets even yields accuracy improvements.

【10】Risk Awareness Injection: Calibrating Vision-Language Models for Safety without Compromising Utility
标题：风险意识注入：在不损害实用性的情况下校准视觉语言模型以实现安全
链接：https://arxiv.org/abs/2602.03402

作者：Mengxuan Wang,Yuxin Chen,Gang Xu,Tao He,Hongjie Jiang,Ming Li
摘要：Vision language models (VLMs) extend the reasoning capabilities of large language models (LLMs) to cross-modal settings, yet remain highly vulnerable to multimodal jailbreak attacks. Existing defenses predominantly rely on safety fine-tuning or aggressive token manipulations, incurring substantial training costs or significantly degrading utility. Recent research shows that LLMs inherently recognize unsafe content in text, and the incorporation of visual inputs in VLMs frequently dilutes risk-related signals. Motivated by this, we propose Risk Awareness Injection (RAI), a lightweight and training-free framework for safety calibration that restores LLM-like risk recognition by amplifying unsafe signals in VLMs. Specifically, RAI constructs an Unsafe Prototype Subspace from language embeddings and performs targeted modulation on selected high-risk visual tokens, explicitly activating safety-critical signals within the cross-modal feature space. This modulation restores the model's LLM-like ability to detect unsafe content from visual inputs, while preserving the semantic integrity of original tokens for cross-modal reasoning. Extensive experiments across multiple jailbreak and utility benchmarks demonstrate that RAI substantially reduces attack success rate without compromising task performance.

【11】On the Entropy Dynamics in Reinforcement Fine-Tuning of Large Language Models
标题：大型语言模型强化微调中的熵动力学
链接：https://arxiv.org/abs/2602.03392

作者：Shumin Wang,Yuexiang Xie,Wenhao Zhang,Yuchang Sun,Yanxi Chen,Yaliang Li,Yanyong Zhang
摘要：Entropy serves as a critical metric for measuring the diversity of outputs generated by large language models (LLMs), providing valuable insights into their exploration capabilities. While recent studies increasingly focus on monitoring and adjusting entropy to better balance exploration and exploitation in reinforcement fine-tuning (RFT), a principled understanding of entropy dynamics during this process is yet to be thoroughly investigated. In this paper, we establish a theoretical framework for analyzing the entropy dynamics during the RFT process, which begins with a discriminant expression that quantifies entropy change under a single logit update. This foundation enables the derivation of a first-order expression for entropy change, which can be further extended to the update formula of Group Relative Policy Optimization (GRPO). The corollaries and insights drawn from the theoretical analysis inspire the design of entropy control methods, and also offer a unified lens for interpreting various entropy-based methods in existing studies. We provide empirical evidence to support the main conclusions of our analysis and demonstrate the effectiveness of the derived entropy-discriminator clipping methods. This study yields novel insights into RFT training dynamics, providing theoretical support and practical strategies for optimizing the exploration-exploitation balance during LLM fine-tuning.

【12】MeKi: Memory-based Expert Knowledge Injection for Efficient LLM Scaling
标题：MeKi：基于内存的专家知识注入，实现高效的LLM扩展
链接：https://arxiv.org/abs/2602.03359

作者：Ning Ding,Fangcheng Liu,Kyungrae Kim,Linji Hao,Kyeng-Hun Lee,Hyeonmok Ko,Yehui Tang
摘要：Scaling Large Language Models (LLMs) typically relies on increasing the number of parameters or test-time computations to boost performance. However, these strategies are impractical for edge device deployment due to limited RAM and NPU resources. Despite hardware constraints, deploying performant LLM on edge devices such as smartphone remains crucial for user experience. To address this, we propose MeKi (Memory-based Expert Knowledge Injection), a novel system that scales LLM capacity via storage space rather than FLOPs. MeKi equips each Transformer layer with token-level memory experts that injects pre-stored semantic knowledge into the generation process. To bridge the gap between training capacity and inference efficiency, we employ a re-parameterization strategy to fold parameter matrices used during training into a compact static lookup table. By offloading the knowledge to ROM, MeKi decouples model capacity from computational cost, introducing zero inference latency overhead. Extensive experiments demonstrate that MeKi significantly outperforms dense LLM baselines with identical inference speed, validating the effectiveness of memory-based scaling paradigm for on-device LLMs. Project homepage is at https://github.com/ningding-o/MeKi.

【13】GFlowPO: Generative Flow Network as a Language Model Prompt Optimizer
标题：GFlowPO：生成流网络作为语言模型提示优化器
链接：https://arxiv.org/abs/2602.03358

作者：Junmo Cho,Suhan Kim,Sangjune An,Minsu Kim,Dong Bok Lee,Heejun Lee,Sung Ju Hwang,Hae Beom Lee
摘要：Finding effective prompts for language models (LMs) is critical yet notoriously difficult: the prompt space is combinatorially large, rewards are sparse due to expensive target-LM evaluation. Yet, existing RL-based prompt optimizers often rely on on-policy updates and a meta-prompt sampled from a fixed distribution, leading to poor sample efficiency. We propose GFlowPO, a probabilistic prompt optimization framework that casts prompt search as a posterior inference problem over latent prompts regularized by a meta-prompted reference-LM prior. In the first step, we fine-tune a lightweight prompt-LM with an off-policy Generative Flow Network (GFlowNet) objective, using a replay-based training policy that reuses past prompt evaluations to enable sample-efficient exploration. In the second step, we introduce Dynamic Memory Update (DMU), a training-free mechanism that updates the meta-prompt by injecting both (i) diverse prompts from a replay buffer and (ii) top-performing prompts from a small priority queue, thereby progressively concentrating the search process on high-reward regions. Across few-shot text classification, instruction induction benchmarks, and question answering tasks, GFlowPO consistently outperforms recent discrete prompt optimization baselines.

【14】Entropy-Gated Selective Policy Optimization:Token-Level Gradient Allocation for Hybrid Training of Large Language Models
标题：熵门控选择性策略优化：用于大型语言模型混合训练的令牌级梯度分配
链接：https://arxiv.org/abs/2602.03309

作者：Yuelin Hu,Zhengxue Cheng,Wei Liu,Li Song
备注：accepted by cscwd2026
摘要：Hybrid training methods for large language models combine supervised fine tuning (SFT) on expert demonstrations with reinforcement learning (RL) on model rollouts, typically at the sample level. We propose Entropy Gated Selective Policy Optimization (EGSPO), a three stage framework that extends sample level mixing with token level gradient modulation. Stage 1, SFT expert learning, establishes a reliable warm up policy using expert demonstrations with a pure SFT loss. Stage 2, RL rollout generation, samples trajectories from the current policy and computes per token predictive entropy. Stage 3, the EGSPO mechanism, applies entropy gated gradient allocation: a predictive entropy module routes high entropy tokens to full PPO updates to encourage exploration, and low entropy tokens to attenuated PPO updates to reduce variance and preserve knowledge. Critically, both branches incorporate the advantage function A_t, ensuring that incorrect trajectories receive consistent negative learning signals and preventing reinforcement of confident errors. EGSPO achieves consistent improvements on mathematical reasoning benchmarks, with gains of 3.8 percent on AIME and 2.9 percent on MATH over the CHORD phi baseline, while incurring only 3.4 percent additional computational overhead.

【15】R1-SyntheticVL: Is Synthetic Data from Generative Models Ready for Multimodal Large Language Model?
标题：R1-SyntheticDL：生成模型中的合成数据准备好适用于多模式大型语言模型了吗？
链接：https://arxiv.org/abs/2602.03300

作者：Jingyi Zhang,Tianyi Lin,Huanjin Yao,Xiang Lan,Shunyu Liu,Jiaxing Huang
摘要：In this work, we aim to develop effective data synthesis techniques that autonomously synthesize multimodal training data for enhancing MLLMs in solving complex real-world tasks. To this end, we propose Collective Adversarial Data Synthesis (CADS), a novel and general approach to synthesize high-quality, diverse and challenging multimodal data for MLLMs. The core idea of CADS is to leverage collective intelligence to ensure high-quality and diverse generation, while exploring adversarial learning to synthesize challenging samples for effectively driving model improvement. Specifically, CADS operates with two cyclic phases, i.e., Collective Adversarial Data Generation (CAD-Generate) and Collective Adversarial Data Judgment (CAD-Judge). CAD-Generate leverages collective knowledge to jointly generate new and diverse multimodal data, while CAD-Judge collaboratively assesses the quality of synthesized data. In addition, CADS introduces an Adversarial Context Optimization mechanism to optimize the generation context to encourage challenging and high-value data generation. With CADS, we construct MMSynthetic-20K and train our model R1-SyntheticVL, which demonstrates superior performance on various benchmarks.

【16】Agentic Proposing: Enhancing Large Language Model Reasoning via Compositional Skill Synthesis
标题：抽象提案：通过组合技能合成增强大型语言模型推理
链接：https://arxiv.org/abs/2602.03279

作者：Zhengbo Jiao,Shaobo Wang,Zifan Zhang,Xuan Ren,Wei Wang,Bing Zhao,Hu Wei,Linfeng Zhang
备注：23page4
摘要：Advancing complex reasoning in large language models relies on high-quality, verifiable datasets, yet human annotation remains cost-prohibitive and difficult to scale. Current synthesis paradigms often face a recurring trade-off: maintaining structural validity typically restricts problem complexity, while relaxing constraints to increase difficulty frequently leads to inconsistent or unsolvable instances. To address this, we propose Agentic Proposing, a framework that models problem synthesis as a goal-driven sequential decision process where a specialized agent dynamically selects and composes modular reasoning skills. Through an iterative workflow of internal reflection and tool-use, we develop the Agentic-Proposer-4B using Multi-Granularity Policy Optimization (MGPO) to generate high-precision, verifiable training trajectories across mathematics, coding, and science. Empirical results demonstrate that downstream solvers trained on agent-synthesized data significantly outperform leading baselines and exhibit robust cross-domain generalization. Notably, a 30B solver trained on only 11,000 synthesized trajectories achieves a state-of-the-art 91.6% accuracy on AIME25, rivaling frontier-scale proprietary models such as GPT-5 and proving that a small volume of high-quality synthetic signals can effectively substitute for massive human-curated datasets.

【17】Beyond Suffixes: Token Position in GCG Adversarial Attacks on Large Language Models
标题：超越后缀：GCG对大型语言模型的对抗性攻击中的代币位置
链接：https://arxiv.org/abs/2602.03265

作者：Hicham Eddoubi,Umar Faruk Abdullahi,Fadi Hassan
备注：12 pages, 10 figures
摘要：Large Language Models (LLMs) have seen widespread adoption across multiple domains, creating an urgent need for robust safety alignment mechanisms. However, robustness remains challenging due to jailbreak attacks that bypass alignment via adversarial prompts. In this work, we focus on the prevalent Greedy Coordinate Gradient (GCG) attack and identify a previously underexplored attack axis in jailbreak attacks typically framed as suffix-based: the placement of adversarial tokens within the prompt. Using GCG as a case study, we show that both optimizing attacks to generate prefixes instead of suffixes and varying adversarial token position during evaluation substantially influence attack success rates. Our findings highlight a critical blind spot in current safety evaluations and underline the need to account for the position of adversarial tokens in the adversarial robustness evaluation of LLMs.

【18】Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoning
标题：Accordion思维：高效且可读的LLM推理的自我调节步骤总结
链接：https://arxiv.org/abs/2602.03249

作者：Zhicheng Yang,Zhijiang Guo,Yinya Huang,Yongxin Wang,Wenlei Shi,Yiwei Wang,Xiaodan Liang,Jing Tang
摘要：Scaling test-time compute via long Chain-ofThought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity. In this paper, we introduce Accordion-Thinking, an end-to-end framework where LLMs learn to self-regulate the granularity of the reasoning steps through dynamic summarization. This mechanism enables a Fold inference mode, where the model periodically summarizes its thought process and discards former thoughts to reduce dependency on historical tokens. We apply reinforcement learning to incentivize this capability further, uncovering a critical insight: the accuracy gap between the highly efficient Fold mode and the exhaustive Unfold mode progressively narrows and eventually vanishes over the course of training. This phenomenon demonstrates that the model learns to encode essential reasoning information into compact summaries, achieving effective compression of the reasoning context. Our Accordion-Thinker demonstrates that with learned self-compression, LLMs can tackle complex reasoning tasks with minimal dependency token overhead without compromising solution quality, and it achieves a 3x throughput while maintaining accuracy on a 48GB GPU memory configuration, while the structured step summaries provide a human-readable account of the reasoning process.

【19】Merging Beyond: Streaming LLM Updates via Activation-Guided Rotations
标题：超越合并：通过激活引导轮换流媒体LLM更新
链接：https://arxiv.org/abs/2602.03237

作者：Yuxuan Yao,Haonan Sheng,Qingsong Lv,Han Wu,Shuqi Liu,Zehua Liu,Zengyan Liu,Jiahui Gao,Haochen Tan,Xiaojin Fu,Haoli Bai,Hing Cheung So,Zhijiang Guo,Linqi Song
摘要：The escalating scale of Large Language Models (LLMs) necessitates efficient adaptation techniques. Model merging has gained prominence for its efficiency and controllability. However, existing merging techniques typically serve as post-hoc refinements or focus on mitigating task interference, often failing to capture the dynamic optimization benefits of supervised fine-tuning (SFT). In this work, we propose Streaming Merging, an innovative model updating paradigm that conceptualizes merging as an iterative optimization process. Central to this paradigm is \textbf{ARM} (\textbf{A}ctivation-guided \textbf{R}otation-aware \textbf{M}erging), a strategy designed to approximate gradient descent dynamics. By treating merging coefficients as learning rates and deriving rotation vectors from activation subspaces, ARM effectively steers parameter updates along data-driven trajectories. Unlike conventional linear interpolation, ARM aligns semantic subspaces to preserve the geometric structure of high-dimensional parameter evolution. Remarkably, ARM requires only early SFT checkpoints and, through iterative merging, surpasses the fully converged SFT model. Experimental results across model scales (1.7B to 14B) and diverse domains (e.g., math, code) demonstrate that ARM can transcend converged checkpoints. Extensive experiments show that ARM provides a scalable and lightweight framework for efficient model adaptation.

【20】Reinforcement Learning with Promising Tokens for Large Language Models
标题：具有大型语言模型有希望的令牌的强化学习
链接：https://arxiv.org/abs/2602.03195

作者：Jing-Cheng Pang,Liang Lu,Xian Tang,Kun Jiang,Sijie Wu,Kai Zhang,Xubin Li

【21】DynSplit-KV: Dynamic Semantic Splitting for KVCache Compression in Efficient Long-Context LLM Inference
标题：DynSplit-KV：高效长上下文LLM推理中KVache压缩的动态语义拆分
链接：https://arxiv.org/abs/2602.03184

作者：Jiancai Ye,Jun Liu,Qingchen Li,Tianlang Zhao,Hanbin Zhang,Jiayi Pan,Ningyi Xu,Guohao Dai

【22】Self-Hinting Language Models Enhance Reinforcement Learning
标题：自我暗示语言模型增强强化学习
链接：https://arxiv.org/abs/2602.03143

作者：Baohao Liao,Hanze Dong,Xinxing Xu,Christof Monz,Jiang Bian

【23】Contrastive Concept-Tree Search for LLM-Assisted Algorithm Discovery
标题：LLM辅助算法发现的对比概念树搜索
链接：https://arxiv.org/abs/2602.03132

作者：Timothee Leleu,Sudeera Gunathilaka,Federico Ghimenti,Surya Ganguli

【24】Quantized Evolution Strategies: High-precision Fine-tuning of Quantized LLMs at Low-precision Cost
标题：量化进化策略：以低精度成本对量化LLM进行高精度微调
链接：https://arxiv.org/abs/2602.03120

作者：Yinggan Xu,Risto Miikkulainen,Xin Qiu
备注：Preprint version

【25】Function-Space Empirical Bayes Regularisation with Large Vision-Language Model Priors
标题：具有大视觉语言模型先验的功能空间经验Bayes正规化
链接：https://arxiv.org/abs/2602.03119

作者：Pengcheng Hao,Huaze Tang,Ercan Engin Kuruoglu,Wenbo Ding

【26】Consensus Group Relative Policy Optimization for Text Generation
标题：共识小组文本生成的相对策略优化
链接：https://arxiv.org/abs/2602.03102

作者：Yuki Ichihara,Yuu Jinnai,Kaito Ariu,Eiji Uchibe

【27】Evaluating LLMs When They Do Not Know the Answer: Statistical Evaluation of Mathematical Reasoning via Comparative Signals
标题：当LLM不知道答案时评估它们：通过比较信号对数学推理进行统计评估
链接：https://arxiv.org/abs/2602.03061

作者：Zihan Dong,Zhixian Zhang,Yang Zhou,Can Jin,Ruijia Wu,Linjun Zhang

【28】CoBA-RL: Capability-Oriented Budget Allocation for Reinforcement Learning in LLMs
标题：CoBA-RL：LLM中强化学习的以能力为导向的预算分配
链接：https://arxiv.org/abs/2602.03048

作者：Zhiyuan Yao,Yi-Kai Zhang,Yuxin Chen,Yueqing Sun,Zishan Xu,Yu Yang,Tianhao Hu,Qi Gu,Hui Su,Xunliang Cai

【29】Clarify Before You Draw: Proactive Agents for Robust Text-to-CAD Generation
标题：绘制前澄清：用于稳健的文本到CAD生成的主动代理
链接：https://arxiv.org/abs/2602.03045

作者：Bo Yuan,Zelin Zhao,Petr Molodyk,Bin Hu,Yongxin Chen
备注：In Review

【30】Generalizable and Interpretable RF Fingerprinting with Shapelet-Enhanced Large Language Models
标题：使用Shapelet增强型大型语言模型的可推广和可解释的RF指纹识别
链接：https://arxiv.org/abs/2602.03035

作者：Tianya Zhao,Junqing Zhang,Haowen Xu,Xiaoyan Sun,Jun Dai,Xuyu Wang
备注：12 pages, 7 figures, IMWUT submission

【31】Rethinking Music Captioning with Music Metadata LLMs
标题：用音乐元数据LLM重新思考音乐字幕
链接：https://arxiv.org/abs/2602.03023

作者：Irmak Bukey,Zhepei Wang,Chris Donahue,Nicholas J. Bryan
备注：Accepted to ICASSP 2026

【32】FedKRSO: Communication and Memory Efficient Federated Fine-Tuning of Large Language Models
标题：FedKRSO：大型语言模型的通信和内存高效联合微调
链接：https://arxiv.org/abs/2602.03019

作者：Guohao Yang,Tongle Wu,Yuanxiong Guo,Ying Sun,Yanmin Gong
备注：Accepted by INFOCOM 2026

【33】NLI:Non-uniform Linear Interpolation Approximation of Nonlinear Operations for Efficient LLMs Inference
标题：NLI：用于高效LLM推理的非线性运算的非均匀线性内插逼近
链接：https://arxiv.org/abs/2602.02988

作者：Jiangyong Yu,Xiaomeng Han,Xing Hu,Chen Xu,Zhe Jiang,Dawei Yang
备注：Admitted to ICLR 18pages 5 figures

【34】Where Norms and References Collide: Evaluating LLMs on Normative Reasoning
标题：规范和参考文献发生冲突的地方：评估规范性推理的LLM
链接：https://arxiv.org/abs/2602.02975

作者：Mitchell Abrams,Kaveh Eskandari Miandoab,Felix Gervits,Vasanth Sarathy,Matthias Scheutz
备注：Accepted to the 40th AAAI Conference on Artificial Intelligence (AAAI-26)

【35】Reasoning about Reasoning: BAPO Bounds on Chain-of-Thought Token Complexity in LLMs
标题：关于推理的推理：BAPO与LLM中思想链代币复杂性的界限
链接：https://arxiv.org/abs/2602.02909

作者：Kiran Tomlinson,Tobias Schnabel,Adith Swaminathan,Jennifer Neville
备注：28 pages

【36】TraceNAS: Zero-shot LLM Pruning via Gradient Trace Correlation
标题：TraceNAS：通过梯度轨迹相关进行Zero-ShotLLM修剪
链接：https://arxiv.org/abs/2602.02891

作者：Prajna G. Malettira,Manish Nagaraj,Arjun Roy,Shubham Negi,Kaushik Roy
备注：Preprint

【37】"I May Not Have Articulated Myself Clearly": Diagnosing Dynamic Instability in LLM Reasoning at Inference Time
链接：https://arxiv.org/abs/2602.02863

作者：Jinkun Chen,Fengxiang Cheng,Sijia Han,Vlado Keselj
备注：21 pages, 12 figures, 15 tables

【38】Zero Sum SVD: Balancing Loss Sensitivity for Low Rank LLM Compression
标题：零和MVD：平衡低等级LLM压缩的损失敏感性
链接：https://arxiv.org/abs/2602.02848

作者：Ali Abbasi,Chayne Thrash,Haoran Qin,Shansita Sharma,Sepehr Seifi,Soheil Kolouri

【39】Chain of Simulation: A Dual-Mode Reasoning Framework for Large Language Models with Dynamic Problem Routing
标题：模拟链：具有动态问题路由的大型语言模型的双模式推理框架
链接：https://arxiv.org/abs/2602.02842

作者：Saeid Sheikhi

【40】A Single Revision Step Improves Token-Efficient LLM Reasoning
标题：单一修订步骤改进了令牌高效的LLM推理
链接：https://arxiv.org/abs/2602.02828

作者：Yingchuan Zhang,Terry Ma,Wenxuan Zhong,Ping Ma

【41】Scaling-Aware Adapter for Structure-Grounded LLM Reasoning
标题：用于基于结构的LLM推理的可扩展感知适配器
链接：https://arxiv.org/abs/2602.02780

作者：Zihao Jing,Qiuhao Zeng,Ruiyi Fang,Yan Yi Li,Yan Sun,Boyu Wang,Pingzhao Hu
备注：Under review at ICML 2026

【42】Privately Fine-Tuned LLMs Preserve Temporal Dynamics in Tabular Data
标题：私人微调的LLM保留表格数据中的时间动态
链接：https://arxiv.org/abs/2602.02766

作者：Lucas Rosenblatt,Peihan Liu,Ryan McKenna,Natalia Ponomareva

【43】From Task Solving to Robust Real-World Adaptation in LLM Agents
标题：从任务求解到LLM代理中的稳健现实世界适应
链接：https://arxiv.org/abs/2602.02760

作者：Pouya Pezeshkpour,Estevam Hruschka

【44】Entropy-Guided Dynamic Tokens for Graph-LLM Alignment in Molecular Understanding
标题：分子理解中Graph-LLM比对的信息引导动态标记
链接：https://arxiv.org/abs/2602.02742

作者：Zihao Jing,Qiuhao Zeng,Ruiyi Fang,Yan Sun,Boyu Wang,Pingzhao Hu
备注：Accepted by ICLR 2026

【45】Monotonicity as an Architectural Bias for Robust Language Models
标题：单调性作为稳健语言模型的架构偏见
链接：https://arxiv.org/abs/2602.02686

作者：Patrick Cooper,Alireza Nadali,Ashutosh Trivedi,Alvaro Velasquez
备注：12 pages, 1 figure

【46】A Positive Case for Faithfulness: LLM Self-Explanations Help Predict Model Behavior
标题：忠诚的积极案例：LLM自我安慰有助于预测模型行为
链接：https://arxiv.org/abs/2602.02639

作者：Harry Mayne,Justin Singh Kang,Dewi Gould,Kannan Ramchandran,Adam Mahdi,Noah Y. Siegel

【47】Performance of Small Language Model Pretraining on FABRIC: An Empirical Study
标题：FABRIC上小语言模型预训练的性能：实证研究
链接：https://arxiv.org/abs/2602.02632

作者：Praveen Rao

【48】Step-Wise Refusal Dynamics in Autoregressive and Diffusion Language Models
标题：自回归和扩散语言模型中的逐步拒绝动力学
链接：https://arxiv.org/abs/2602.02600

作者：Eliron Rahimi,Elad Hirshel,Rom Himelstein,Amit LeVi,Avi Mendelson,Chaim Baskin

【49】PeerRank: Autonomous LLM Evaluation Through Web-Grounded, Bias-Controlled Peer Review
标题：PeerRank：通过网络支持、偏见控制同行评审进行自主LLM评估
链接：https://arxiv.org/abs/2602.02589

作者：Yanki Margalit,Erni Avram,Ran Taig,Oded Margalit,Nurit Cohen-Inger

【50】Uncertainty and Fairness Awareness in LLM-Based Recommendation Systems
标题：基于LLM的推荐系统中的不确定性和公平性意识
链接：https://arxiv.org/abs/2602.02582

作者：Chandan Kumar Sah,Xiaoli Lian,Li Zhang,Tony Xu,Syed Shazaib Shah
备注：Accepted at the Second Conference of the International Association for Safe and Ethical Artificial Intelligence, IASEAI26, 14 pages

【51】Beyond Experience Retrieval: Learning to Generate Utility-Optimized Structured Experience for Frozen LLMs
标题：超越经验检索：学习为冻结的LLM生成效用优化的结构化体验
链接：https://arxiv.org/abs/2602.02556

作者：Xuancheng Li,Haitao Li,Yujia Zhou,Yiqun Liu,Qingyao Ai

【52】HyPAC: Cost-Efficient LLMs-Human Hybrid Annotation with PAC Error Guarantees
标题：HyPAC：具有PAC错误保证的经济高效的LLC-人类混合注释
链接：https://arxiv.org/abs/2602.02550

作者：Hao Zeng,Huipeng Huang,Xinhao Qu,Jianguo Huang,Bingyi Jing,Hongxin Wei

【53】D$^2$Quant: Accurate Low-bit Post-Training Weight Quantization for LLMs
标题：D $'#39; 2$Quant：LLM的准确低位训练后体重量化
链接：https://arxiv.org/abs/2602.02546

作者：Xianglong Yan,ChengZhu Bao,Zhiteng Li,Tianao Zhang,Shaoqiu Zhang,Ruobing Xie,Samm Sun,Yulun Zhang

【54】SPA-Cache: Singular Proxies for Adaptive Caching in Diffusion Language Models
标题：SPA-缓存：扩散语言模型中用于自适应缓存的奇异代理
链接：https://arxiv.org/abs/2602.02544

作者：Wenhao Sun,Rong-Cheng Tu,Yifu Ding,Zhao Jin,Jingyi Liao,Yongcheng Jing,Dacheng Tao
备注：18 pages, 6 figures.The code repository is available at https://github.com/wenhao728/spa-cache

【55】WorldVQA: Measuring Atomic World Knowledge in Multimodal Large Language Models
标题：WorldVQA：在多模式大型语言模型中测量原子世界知识
链接：https://arxiv.org/abs/2602.02537

作者：Runjie Zhou,Youbo Shao,Haoyu Lu,Bowei Xing,Tongtong Bai,Yujie Chen,Jie Zhao,Lin Sui,Haotian Yao,Zijia Zhao,Hao Yang,Haoning Wu,Zaida Zhou,Jinguo Zhu,Zhiqi Huang,Yiping Bao,Yangyang Liu,Y. Charles,Xinyu Zhou

【56】HMVLA: Hyperbolic Multimodal Fusion for Vision-Language-Action Models
标题：HMVLA：视觉-语言-动作模型的双曲多模式融合
链接：https://arxiv.org/abs/2602.02533

作者：Kun Wang,Xiao Feng,Mingcheng Qu,Tonghua Su
备注：5 pages,5 figures,ICASSP

【57】IMU-1: Sample-Efficient Pre-training of Small Language Models
标题：IMU-1：小语言模型的样本高效预训练
链接：https://arxiv.org/abs/2602.02522

作者：George Grigorev
备注：16 pages

【58】GraphDancer: Training LLMs to Explore and Reason over Graphs via Curriculum Reinforcement Learning
标题：GraphDancer：训练法学硕士通过课程强化学习探索和推理图形
链接：https://arxiv.org/abs/2602.02518

作者：Yuyang Bai,Zhuofeng Li,Ping Nie,Jianwen Xie,Yu Zhang
备注：15 pages, Project website: https://yuyangbai.com/graphdancer/

【59】CreditAudit: 2D Auditing for LLM Evaluation and Selection
标题：CreditAudit：LLM评估和选择的2D审计
链接：https://arxiv.org/abs/2602.02515

作者：Yiliang Song,Hongjun An,Jiangong Xiao,Haofei Zhao,Jiawei Shao,Xuelong Li
备注：First update

【60】Augmenting Parameter-Efficient Pre-trained Language Models with Large Language Models
标题：用大型语言模型增强参数高效的预训练语言模型
链接：https://arxiv.org/abs/2602.02501

作者：Saurabh Anand,Shubham Malaviya,Manish Shukla,Sachin Lodha
备注：22 pages, 9 figures, 11 tables, short paper was accepted in ACM SAC 2024

Graph相关(图学习|图神经网络|图优化等)(11篇)

【1】Data-Driven Graph Filters via Adaptive Spectral Shaping
标题：通过自适应光谱整形的数据驱动图形过滤器
链接：https://arxiv.org/abs/2602.03698

作者：Dylan Sandfelder,Mihai Cucuringu,Xiaowen Dong

【2】Causal Graph Learning via Distributional Invariance of Cause-Effect Relationship
标题：通过因果关系分布不变性进行因果图学习
链接：https://arxiv.org/abs/2602.03353

作者：Nang Hung Nguyen,Phi Le Nguyen,Thao Nguyen Truong,Trong Nghia Hoang,Masashi Sugiyama

【3】Unveiling Covert Toxicity in Multimodal Data via Toxicity Association Graphs: A Graph-Based Metric and Interpretable Detection Framework
标题：通过毒性关联图揭示多峰数据中的隐性毒性：基于图的度量和可解释检测框架
链接：https://arxiv.org/abs/2602.03268

作者：Guanzong Wu,Zihao Zhu,Siwei Lyu,Baoyuan Wu

【4】GraDE: A Graph Diffusion Estimator for Frequent Subgraph Discovery in Neural Architectures
标题：Grade：神经架构中频繁子图发现的图扩散估计
链接：https://arxiv.org/abs/2602.03257

作者：Yikang Yang,Zhengxin Yang,Minghao Luo,Luzhou Peng,Hongxiao Li,Wanling Gao,Lei Wang,Jianfeng Zhan

【5】Topology Matters: A Cautionary Case Study of Graph SSL on Neuro-Inspired Benchmarks
标题：布局很重要：基于神经启发基准的图SSL警示案例研究
链接：https://arxiv.org/abs/2602.03217

作者：May Kristine Jonson Carlon,Su Myat Noe,Haojiong Wang,Yasuo Kuniyoshi

【6】Causal Graph Spatial-Temporal Autoencoder for Reliable and Interpretable Process Monitoring
标题：因果图时空自动编码器，用于可靠且可解释的过程监控
链接：https://arxiv.org/abs/2602.03004

作者：Xiangrui Zhang,Chunyue Song,Wei Dai,Zheng Zhang,Kaihua Gao,Furong Gao

【7】RPG-AE: Neuro-Symbolic Graph Autoencoders with Rare Pattern Mining for Provenance-Based Anomaly Detection
标题：RPG-AE：具有稀有模式挖掘的神经符号图自动编码器，用于基于源的异常检测
链接：https://arxiv.org/abs/2602.02929

作者：Asif Tauhid,Sidahmed Benabderrahmane,Mohamad Altrabulsi,Ahamed Foisal,Talal Rahwan

【8】SC3D: Dynamic and Differentiable Causal Discovery for Temporal and Instantaneous Graphs
标题：SC 3D：时间和瞬时图的动态和可区分的因果发现
链接：https://arxiv.org/abs/2602.02830

作者：Sourajit Das,Dibyajyoti Chakraborthy,Romit Maulik
备注：8 pages

【9】Eidolon: A Practical Post-Quantum Signature Scheme Based on k-Colorability in the Age of Graph Neural Networks
标题：Eidolon：图神经网络时代基于k-可着色性的实用后量子签名方案
链接：https://arxiv.org/abs/2602.02689

作者：Asmaa Cherkaoui,Ramon Flores,Delaram Kahrobaei,Richard Wilson
备注：23 pages, 5 figures

【10】GASTON: Graph-Aware Social Transformer for Online Networks
标题：GASTON：在线网络的图形感知社交Transformer
链接：https://arxiv.org/abs/2602.02524

作者：Olha Wloch,Liam Hebert,Robin Cohen,Lukasz Golab
备注：Submitted to ICWSM

【11】Generator-based Graph Generation via Heat Diffusion
标题：基于生成器的热扩散图生成
链接：https://arxiv.org/abs/2602.03612

作者：Anthony Stephenson,Ian Gallagher,Christopher Nemeth
备注：Submitted to ICML; 8+15 pages; 20 figures

Transformer(9篇)

【1】SymPlex: A Structure-Aware Transformer for Symbolic PDE Solving
标题：Symbolism：用于符号性偏脱方程求解的结构感知Transformer
链接：https://arxiv.org/abs/2602.03816

作者：Yesom Park,Annie C. Lu,Shao-Ching Huang,Qiyang Hu,Y. Sungtaek Ju,Stanley Osher
备注：27 pages

【2】Explaining the Explainer: Understanding the Inner Workings of Transformer-based Symbolic Regression Models
标题：解释解释者：了解基于转换器的符号回归模型的内部工作
链接：https://arxiv.org/abs/2602.03506

作者：Arco van Breda,Erman Acar
备注：8 pages, 5 figures

【3】Spatiotemporal Decision Transformer for Traffic Coordination
标题：交通协调的时空决策Transformer
链接：https://arxiv.org/abs/2602.02903

作者：Haoran Su,Yandong Sun,Hanxiao Deng

【4】Tabula RASA: Exposing and Breaking the Relational Bottleneck in Transformers
标题：Tabula RASA：揭露并打破《Transformer》中的关系瓶颈
链接：https://arxiv.org/abs/2602.02834

作者：Jonas Petersen,Camilla Mazzoleni,Riccardo Maggioni
备注：16 pages, 4 figures, 8 tables

【5】LmPT: Conditional Point Transformer for Anatomical Landmark Detection on 3D Point Clouds
标题：LmPT：用于3D点云解剖地标检测的条件点Transformer
链接：https://arxiv.org/abs/2602.02808

作者：Matteo Bastico,Pierre Onghena,David Ryckelynck,Beatriz Marcotegui,Santiago Velasco-Forero,Laurent Corté,Caroline Robine--Decourcelle,Etienne Decencière
备注：This paper has been accepted at International Symposium on Biomedical Imaging (ISBI) 2026

【6】CAPS: Unifying Attention, Recurrence, and Alignment in Transformer-based Time Series Forecasting
标题：CAPS：在基于转换器的时间序列预测中统一注意力、重现性和一致性
链接：https://arxiv.org/abs/2602.02729

作者：Viresh Pati,Yubin Kim,Vinh Pham,Jevon Twitty,Shihao Yang,Jiecheng Lu

【7】Every Bit Counts: A Theoretical Study of Precision-Expressivity Tradeoffs in Quantized Transformers
标题：每一位都很重要：量化Transformer中精确性-表现性权衡的理论研究
链接：https://arxiv.org/abs/2602.02707

作者：Sayak Chakrabarti,Toniann Pitassi,Josh Alman

【8】Learnable Koopman-Enhanced Transformer-Based Time Series Forecasting with Spectral Control
标题：具有谱控制的可学习Koopman增强型基于变换器的时间序列预测
链接：https://arxiv.org/abs/2602.02592

作者：Ali Forootani,Raffaele Iervolino

【9】Physics-inspired transformer quantum states via latent imaginary-time evolution
标题：物理启发的Transformer量子状态通过潜在的时间进化
链接：https://arxiv.org/abs/2602.03031

作者：Kimihiro Yamazaki,Itsushi Sakata,Takuya Konishi,Yoshinobu Kawahara

GAN|对抗|攻击|生成相关(17篇)

【1】Bridging Online and Offline RL: Contextual Bandit Learning for Multi-Turn Code Generation
标题：连接在线和线下RL：用于多回合代码生成的上下文强盗学习
链接：https://arxiv.org/abs/2602.03806

作者：Ziru Chen,Dongdong Chen,Ruinan Jin,Yingbin Liang,Yujia Xie,Huan Sun

【2】CTTVAE: Latent Space Structuring for Conditional Tabular Data Generation on Imbalanced Datasets
标题：CTTVAE：不平衡数据集中条件表格数据生成的潜在空间结构
链接：https://arxiv.org/abs/2602.03641

作者：Milosh Devic,Jordan Gierschendorf,David Garson

【3】Optimization and Generation in Aerodynamics Inverse Design
标题：空气动力反设计中的优化与生成
链接：https://arxiv.org/abs/2602.03582

作者：Huaguan Chen,Ning Lin,Luxi Chen,Rui Zhang,Wenbing Huang,Chongxuan Li,Hao Sun

【4】Most Convolutional Networks Suffer from Small Adversarial Perturbations
标题：大多数卷积网络都会遭受小的对抗性扰动
链接：https://arxiv.org/abs/2602.03415

作者：Amit Daniely,Idan Mehalel

【5】Spectral Evolution Search: Efficient Inference-Time Scaling for Reward-Aligned Image Generation
标题：光谱进化搜索：用于奖励对齐图像生成的高效推断时间缩放
链接：https://arxiv.org/abs/2602.03208

作者：Jinyan Ye,Zhongjie Duan,Zhiwen Li,Cen Chen,Daoyuan Chen,Yaliang Li,Yingda Chen

【6】Adversarial construction as a potential solution to the experiment design problem in large task spaces
标题：对抗性构建作为大任务空间中实验设计问题的潜在解决方案
链接：https://arxiv.org/abs/2602.03172

作者：Prakhar Godara,Frederick Callaway,Marcelo G. Mattar
备注：7 pages, 7 figures

【7】Quant VideoGen: Auto-Regressive Long Video Generation via 2-Bit KV-Cache Quantization
标题：Quant VideoGen：通过2位KV缓存量化自动回归长视频生成
链接：https://arxiv.org/abs/2602.02958

作者：Haocheng Xi,Shuo Yang,Yilong Zhao,Muyang Li,Han Cai,Xingyang Li,Yujun Lin,Zhuoyang Zhang,Jintao Zhang,Xiuyu Li,Zhiying Xu,Jun Wu,Chenfeng Xu,Ion Stoica,Song Han,Kurt Keutzer
备注：11 pages, 7 figures

【8】From Tokens to Numbers: Continuous Number Modeling for SVG Generation
标题：从令牌到数字：用于VG生成的连续数字建模
链接：https://arxiv.org/abs/2602.02820

作者：Michael Ogezi,Martin Bell,Freda Shi,Ethan Smith

【9】Membership Inference Attacks from Causal Principles
标题：来自因果原则的会员推理攻击
链接：https://arxiv.org/abs/2602.02819

作者：Mathieu Even,Clément Berenfeld,Linus Bleistein,Tudor Cebere,Julie Josse,Aurélien Bellet

【10】Evaluating False Alarm and Missing Attacks in CAN IDS
标题：评估CAN IDS中的虚警和缺失攻击
链接：https://arxiv.org/abs/2602.02781

作者：Nirab Hossain,Pablo Moriano
备注：8 pages, 2 figures, and 8 tables

【11】Exposing Vulnerabilities in Explanation for Time Series Classifiers via Dual-Target Attacks
标题：通过双目标攻击暴露时间序列分类器解释中的漏洞
链接：https://arxiv.org/abs/2602.02763

作者：Bohan Wang,Zewen Liu,Lu Lin,Hui Liu,Li Xiong,Ming Jin,Wei Jin

【12】TabPFN for Zero-shot Parametric Engineering Design Generation
标题：TabPFN用于Zero-Shot参数化工程设计生成
链接：https://arxiv.org/abs/2602.02735

作者：Ke Wang,Yifan Tang,Nguyen Gia Hien Vu,Faez Ahmed,G. Gary Wang
备注：14 pages, 8 figures

【13】Search-Augmented Masked Diffusion Models for Constrained Generation
标题：约束生成的搜索增强掩蔽扩散模型
链接：https://arxiv.org/abs/2602.02727

作者：Huu Binh Ta,Michael Cardei,Alvaro Velasquez,Ferdinando Fioretto
备注：Huu Binh Ta and Michael Cardei contributed equally to this work

【14】Expert-Data Alignment Governs Generation Quality in Decentralized Diffusion Models
标题：分散扩散模型中的专家数据一致性控制发电质量
链接：https://arxiv.org/abs/2602.02685

作者：Marcos Villagra,Bidhan Roy,Raihan Seraj,Zhiying Jiang
备注：15 pages, 4 figures

【15】Trajectory Consistency for One-Step Generation on Euler Mean Flows
标题：欧拉平均流一步生成的轨迹一致性
链接：https://arxiv.org/abs/2602.02571

作者：Zhiqi Li,Yuchen Sun,Duowen Chen,Jinjin He,Bo Zhu
备注：40 pages, 27 figures

【16】MathlibLemma: Folklore Lemma Generation and Benchmark for Formal Mathematics
标题：MathlibLemma：民间引理生成和形式数学基准
链接：https://arxiv.org/abs/2602.02561

作者：Xinyu Liu,Zixuan Xie,Amir Moeini,Claire Chen,Shuze Daniel Liu,Yu Meng,Aidong Zhang,Shangtong Zhang

【17】Enhancing Quantum Diffusion Models for Complex Image Generation
标题：增强的量子扩散模型在复杂图像生成中的应用
链接：https://arxiv.org/abs/2602.03405

作者：Jeongbin Jo,Santanam Wishal,Shah Md Khalil Ullah,Shan Kowalski,Dikshant Dulai
备注：18 pages, 6 figures

半/弱/无/有监督|不确定性|主动学习(16篇)

【1】Explanations Leak: Membership Inference with Differential Privacy and Active Learning Defense
标题：解释泄露：具有差异隐私和主动学习防御的会员推断
链接：https://arxiv.org/abs/2602.03611

作者：Fatima Ezzeddine,Osama Zammar,Silvia Giordano,Omran Ayoub

【2】The Label Horizon Paradox: Rethinking Supervision Targets in Financial Forecasting
标题：标签地平线悖论：重新思考金融预测中的监管目标
链接：https://arxiv.org/abs/2602.03395

作者：Chen-Hui Song,Shuoling Liu,Liyuan Chen

【3】From Vicious to Virtuous Cycles: Synergistic Representation Learning for Unsupervised Video Object-Centric Learning
标题：从恶性循环到良性循环：无监督视频以对象为中心学习的协同表示学习
链接：https://arxiv.org/abs/2602.03390

作者：Hyun Seok Seong,WonJun Moon,Jae-Pil Heo
备注：ICLR 2026; Code is available at https://github.com/hynnsk/SRL

【4】Feature, Alignment, and Supervision in Category Learning: A Comparative Approach with Children and Neural Networks
标题：类别学习中的特征、一致和监督：儿童和神经网络的比较方法
链接：https://arxiv.org/abs/2602.03124

作者：Fanxiao Wani Qiu,Oscar Leong

【5】TMS: Trajectory-Mixed Supervision for Reward-Free, On-Policy SFT
标题：TMS：针对免费奖励、政策性SFT的轨迹混合监管
链接：https://arxiv.org/abs/2602.03073

作者：Rana Muhammad Shahroz Khan,Zijie Liu,Zhen Tan,Charles Fleming,Tianlong Chen

【6】Variational Sparse Paired Autoencoders (vsPAIR) for Inverse Problems and Uncertainty Quantification
标题：用于逆问题和不确定性量化的变分稀疏配对自动编码器（vsPAIR）
链接：https://arxiv.org/abs/2602.02948

作者：Jack Michael Solomon,Rishi Leburu,Matthias Chung

【7】Cross-Temporal Attention Fusion (CTAF) for Multimodal Physiological Signals in Self-Supervised Learning
标题：自我监督学习中多峰生理信号的跨时间注意力融合（CTAF）
链接：https://arxiv.org/abs/2602.02784

作者：Arian Khorasani,Théophile Demazure

【8】BiTimeCrossNet: Time-Aware Self-Supervised Learning for Pediatric Sleep
标题：BiTimeCrossNet：儿科睡眠的时间感知自我监督学习
链接：https://arxiv.org/abs/2602.02769

作者：Saurav Raj Pandey,Harlin Lee

【9】On the Sample Efficiency of Inverse Dynamics Models for Semi-Supervised Imitation Learning
标题：半监督模仿学习逆动力学模型的样本效率
链接：https://arxiv.org/abs/2602.02762

作者：Sacha Morin,Moonsub Byeon,Alexia Jolicoeur-Martineau,Sébastien Lachapelle

【10】Sparsely Supervised Diffusion
标题：缺乏监督的扩散
链接：https://arxiv.org/abs/2602.02699

作者：Wenshuai Zhao,Zhiyuan Li,Yi Zhao,Mohammad Hassan Vali,Martin Trapp,Joni Pajarinen,Juho Kannala,Arno Solin
备注：20 pages, 11 figures

【11】A Semi-Supervised Pipeline for Generalized Behavior Discovery from Animal-Borne Motion Time Series
标题：从动物传播的运动时间序列中发现广义行为的半监督管道
链接：https://arxiv.org/abs/2602.02618

作者：Fatemeh Karimi Nejadasl,Judy Shamoun-Baranes,Eldar Rakhimberdiev

【12】BatCoder: Self-Supervised Bidirectional Code-Documentation Learning via Back-Translation
标题：BatCoder：通过反向翻译进行自我监督的双向代码文档学习
链接：https://arxiv.org/abs/2602.02554

作者：Jingwen Xu,Yiyang Lu,Zisu Huang,Changze Lv,Xiaohua Wang,Shizheng Li,Zhibo Xu,Zhengkang Guo,Zhengyuan Wang,Muzhao Tian,Xuanjing Huang,Xiaoqing Zheng

【13】Hypersonic Flow Control: Generalized Deep Reinforcement Learning for Hypersonic Intake Unstart Control under Uncertainty
标题：高超音速流量控制：不确定性下高超音速入口不启动控制的广义深度强化学习
链接：https://arxiv.org/abs/2602.02531

作者：Trishit Mondal,Ameya D. Jagtap
备注：34 Pages, 23 Figures

【14】Score-based diffusion models for diffuse optical tomography with uncertainty quantification
标题：基于分数的扩散光学层析成像不确定性量化模型
链接：https://arxiv.org/abs/2602.03449

作者：Fabian Schneider,Meghdoot Mozumder,Konstantin Tamarov,Leila Taghizadeh,Tanja Tarvainen,Tapio Helin,Duc-Lam Duong

【15】Multiparameter Uncertainty Mapping in Quantitative Molecular MRI using a Physics-Structured Variational Autoencoder (PS-VAE)
标题：使用物理结构变分自动编码器（PS-VAE）进行定量分子MRI的多参数不确定性映射
链接：https://arxiv.org/abs/2602.03317

作者：Alex Finkelstein,Ron Moneta,Or Zohar,Michal Rivlin,Moritz Zaiss,Dinora Friedmann Morvinski,Or Perlman
备注：Submitted to IEEE Transactions on Medical Imaging. This project was funded by the European Union (ERC, BabyMagnet, project no. 101115639). Views and opinions expressed are, however, those of the authors only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them

【16】CryoLVM: Self-supervised Learning from Cryo-EM Density Maps with Large Vision Models
标题：CryoLCC：使用大视觉模型从Cryo-EM密度图中进行自我监督学习
链接：https://arxiv.org/abs/2602.02620

作者：Weining Fu,Kai Shu,Kui Xu,Qiangfeng Cliff Zhang

迁移|Zero/Few/One-Shot|自适应(11篇)

【1】Soft Sensor for Bottom-Hole Pressure Estimation in Petroleum Wells Using Long Short-Term Memory and Transfer Learning
标题：使用长短期记忆和转移学习的油井底部压力软测量
链接：https://arxiv.org/abs/2602.03737

作者：M. A. Fernandes,E. Gildin,M. A. Sampaio

【2】Neural Attention Search Linear: Towards Adaptive Token-Level Hybrid Attention Models
标题：线性神经注意力搜索：迈向自适应代币级混合注意力模型
链接：https://arxiv.org/abs/2602.03681

作者：Difan Deng,Andreas Bentzen Winje,Lukas Fehring,Marius Lindauer
备注：17 pages, 8 figures

【3】Information-Theoretic Multi-Model Fusion for Target-Oriented Adaptive Sampling in Materials Design
标题：材料设计中面向目标自适应抽样的信息论多模型融合
链接：https://arxiv.org/abs/2602.03319

作者：Yixuan Zhang,Zhiyuan Li,Weijia He,Mian Dai,Chen Shen,Teng Long,Hongbin Zhang
备注：37 pages, 5 figures, 2 tables

【4】RDT2: Exploring the Scaling Limit of UMI Data Towards Zero-Shot Cross-Embodiment Generalization
标题：RDT 2：探索UMI数据的缩放限制以实现Zero-Shot跨实施例概括
链接：https://arxiv.org/abs/2602.03310

作者：Songming Liu,Bangguo Li,Kai Ma,Lingxuan Wu,Hengkai Tan,Xiao Ouyang,Hang Su,Jun Zhu

【5】Training and Simulation of Quadrupedal Robot in Adaptive Stair Climbing for Indoor Firefighting: An End-to-End Reinforcement Learning Approach
标题：室内消防自适应爬楼梯四足机器人的训练和仿真：端到端强化学习方法
链接：https://arxiv.org/abs/2602.03087

作者：Baixiao Huang,Baiyu Huang,Yu Hou
备注：8 pages, 9 figures, 43rd International Symposium on Automation and Robotics in Construction

【6】From Zero to Hero: Advancing Zero-Shot Foundation Models for Tabular Outlier Detection
标题：从零到英雄：推进Zero-Shot基础模型用于表格异常值检测
链接：https://arxiv.org/abs/2602.03018

作者：Xueying Ding,Haomin Wen,Simon Klütterman,Leman Akoglu
备注：37 pages

【7】Adaptive Batch Sizes Using Non-Euclidean Gradient Noise Scales for Stochastic Sign and Spectral Descent
标题：使用非欧几里得梯度噪音标度进行随机符号和谱下降的自适应批量大小
链接：https://arxiv.org/abs/2602.03001

作者：Hiroki Naganuma,Shagun Gupta,Youssef Briki,Ioannis Mitliagkas,Irina Rish,Parameswaran Raman,Hao-Jun Michael Shi
备注：8 pages, 2 figures, 4 tables

【8】FlexRank: Nested Low-Rank Knowledge Decomposition for Adaptive Model Deployment
标题：FlexRank：用于自适应模型部署的嵌套低等级知识分解
链接：https://arxiv.org/abs/2602.02680

作者：Riccardo Zaccone,Stefanos Laskaridis,Marco Ciccone,Samuel Horváth

【9】naPINN: Noise-Adaptive Physics-Informed Neural Networks for Recovering Physics from Corrupted Measurement
标题：naPINN：用于从受损坏的测量中恢复物理的噪音自适应物理信息神经网络
链接：https://arxiv.org/abs/2602.02547

作者：Hankyeol Kim,Pilsung Kang

【10】CADENT: Gated Hybrid Distillation for Sample-Efficient Transfer in Reinforcement Learning
标题：学员：门控混合蒸馏在强化学习中实现样本高效传输
链接：https://arxiv.org/abs/2602.02532

作者：Mahyar Alinejad,Yue Wang,George Atia

【11】Rethinking Test-Time Training: Tilting The Latent Distribution For Few-Shot Source-Free Adaptation
标题：重新思考测试时间训练：倾斜潜在分布以实现Few-Shot的无源适应
链接：https://arxiv.org/abs/2602.02633

作者：Tahir Qasim Syed,Behraj Khan

强化学习(13篇)

【1】IntentRL: Training Proactive User-intent Agents for Open-ended Deep Research via Reinforcement Learning
标题：IntentRL：通过强化学习训练主动用户意图代理进行开放式深度研究
链接：https://arxiv.org/abs/2602.03468

作者：Haohao Luo,Zexi Li,Yuexiang Xie,Wenhao Zhang,Yaliang Li,Ying Shen
备注：Preprint

【2】medR: Reward Engineering for Clinical Offline Reinforcement Learning via Tri-Drive Potential Functions
标题：medR：通过三驱动潜力函数进行临床离线强化学习的奖励工程
链接：https://arxiv.org/abs/2602.03305

作者：Qianyi Xu,Gousia Habib,Feng Wu,Yanrui Du,Zhihui Chen,Swapnil Mishra,Dilruk Perera,Mengling Feng

【3】From Scalar Rewards to Potential Trends: Shaping Potential Landscapes for Model-Based Reinforcement Learning
标题：从量化奖励到潜在趋势：塑造基于模型的强化学习的潜在景观
链接：https://arxiv.org/abs/2602.03201

作者：Yao-Hui Li,Zeyu Wang,Xin Li,Wei Pang,Yingfang Yuan,Zhengkun Chen,Boya Zhang,Riashat Islam,Alex Lamb,Yonggang Zhang
备注：26 pages, 20 figures.Preprint. Work in progress

【4】StepScorer: Accelerating Reinforcement Learning with Step-wise Scoring and Psychological Regret Modeling
标题：StepScorer：通过分步评分和心理遗憾建模加速强化学习
链接：https://arxiv.org/abs/2602.03171

作者：Zhe Xu
备注：10 pages, 5 figures, 1 table

【5】Neural Predictor-Corrector: Solving Homotopy Problems with Reinforcement Learning
标题：神经预测-修正器：用强化学习解决同伦问题
链接：https://arxiv.org/abs/2602.03086

作者：Jiayao Mai,Bangyan Liao,Zhenjun Zhao,Yingping Zeng,Haoang Li,Javier Civera,Tailin Wu,Yi Zhou,Peidong Liu

【6】Human-Centric Traffic Signal Control for Equity: A Multi-Agent Action Branching Deep Reinforcement Learning Approach
标题：以人为本的公平交通信号控制：一种多智能体动作分支深度强化学习方法
链接：https://arxiv.org/abs/2602.02959

作者：Xiaocai Zhang,Neema Nassir,Lok Sang Chan,Milad Haghani

【7】How Does the Lagrangian Guide Safe Reinforcement Learning through Diffusion Models?
标题：拉格朗日如何通过扩散模型指导安全强化学习？
链接：https://arxiv.org/abs/2602.02924

作者：Xiaoyuan Cheng,Wenxuan Yuan,Boyang Li,Yuanchao Xu,Yiming Yang,Hao Liang,Bei Peng,Robert Loftin,Zhuo Sun,Yukun Hu

【8】Manifold-Constrained Energy-Based Transition Models for Offline Reinforcement Learning
标题：基于流形约束能量的离线强化学习转移模型
链接：https://arxiv.org/abs/2602.02900

作者：Zeyu Fang,Zuyuan Zhang,Mahdi Imani,Tian Lan

【9】Causal Flow Q-Learning for Robust Offline Reinforcement Learning
标题：用于稳健离线强化学习的因果流Q学习
链接：https://arxiv.org/abs/2602.02847

作者：Mingxuan Li,Junzhe Zhang,Elias Bareinboim

【10】Hierarchical Entity-centric Reinforcement Learning with Factored Subgoal Diffusion
标题：基于因子子目标扩散的分层以效率为中心的强化学习
链接：https://arxiv.org/abs/2602.02722

作者：Dan Haramati,Carl Qi,Tal Daniel,Amy Zhang,Aviv Tamar,George Konidaris
备注：ICLR 2026

【11】Maximum Likelihood Reinforcement Learning
标题：最大可能性强化学习
链接：https://arxiv.org/abs/2602.02710

作者：Fahim Tajwar,Guanning Zeng,Yueer Zhou,Yuda Song,Daman Arora,Yiding Jiang,Jeff Schneider,Ruslan Salakhutdinov,Haiwen Feng,Andrea Zanette
备注：Project website and code: https://zanette-labs.github.io/MaxRL/

【12】Learning to Explore with Parameter-Space Noise: A Deep Dive into Parameter-Space Noise for Reinforcement Learning with Verifiable Rewards
标题：学习利用参数空间噪音进行探索：深入研究参数空间噪音，以实现强化学习并具有可验证奖励
链接：https://arxiv.org/abs/2602.02555

作者：Bizhe Bai,Xinyue Wang,Peng Ye,Tao Chen
备注：17 pages, 10 Figures

【13】Formulating Reinforcement Learning for Human-Robot Collaboration through Off-Policy Evaluation
标题：通过政策外评估制定人机协作强化学习
链接：https://arxiv.org/abs/2602.02530

作者：Saurav Singh,Rodney Sanchez,Alexander Ororbia,Jamison Heard

分层学习(1篇)

【1】Joint Learning of Hierarchical Neural Options and Abstract World Model
标题：分层神经选项和抽象世界模型的联合学习
链接：https://arxiv.org/abs/2602.02799

作者：Wasu Top Piriyakulkij,Wolfgang Lehrach,Kevin Ellis,Kevin Murphy

医学相关(9篇)

【1】EHRWorld: A Patient-Centric Medical World Model for Long-Horizon Clinical Trajectories
标题：EHRWorld：以患者为中心的长期临床轨迹医疗世界模型
链接：https://arxiv.org/abs/2602.03569

作者：Linjie Mu,Zhongzhen Huang,Yannian Gu,Shengqian Qin,Shaoting Zhang,Xiaofan Zhang

【2】HypCBC: Domain-Invariant Hyperbolic Cross-Branch Consistency for Generalizable Medical Image Analysis
标题：HypCBC：可推广医学图像分析的领域不变双曲跨分支一致性
链接：https://arxiv.org/abs/2602.03264

作者：Francesco Di Salvo,Sebastian Doerrich,Jonas Alle,Christian Ledig
备注：Accepted to Transactions on Machine Learning Research (TMLR)

【3】Fully Kolmogorov-Arnold Deep Model in Medical Image Segmentation
标题：医学图像分割中的完全Kolmogorov-Arnold深度模型
链接：https://arxiv.org/abs/2602.03156

作者：Xingyu Qiu,Xinghua Ma,Dong Liang,Gongning Luo,Wei Wang,Kuanquan Wang,Shuo Li
备注：11 pages, 5 figures, conference

【4】Synthetic Data Augmentation for Medical Audio Classification: A Preliminary Evaluation
标题：医学音频分类的合成数据增强：初步评估
链接：https://arxiv.org/abs/2602.02955

作者：David McShannon,Anthony Mella,Nicholas Dietrich
备注：5 pages, 1 figure

【5】Weighted Temporal Decay Loss for Learning Wearable PPG Data with Sparse Clinical Labels
标题：学习具有稀疏临床标签的可穿戴PPV数据的加权时间衰减损失
链接：https://arxiv.org/abs/2602.02917

作者：Yunsung Chung,Keum San Chun,Migyeong Gwak,Han Feng,Yingshuo Liu,Chanho Lim,Viswam Nathan,Nassir Marrouche,Sharanya Arcot Desai
备注：ICASSP 2026

【6】DoubleTake: Contrastive Reasoning for Faithful Decision-Making in Medical Imaging
标题：DoubleTake：医学成像忠实决策的对比推理
链接：https://arxiv.org/abs/2602.02894

作者：Daivik Patel,Shrenik Patel

【7】VerIde ECG Biometrics: Verification and Identification
标题：VerIde心电图生物识别：验证和识别
链接：https://arxiv.org/abs/2602.02776

作者：Scagnetto Arjuna

【8】Auditing Sybil: Explaining Deep Lung Cancer Risk Prediction Through Generative Interventional Attributions
标题：审核Sybil：通过生成性干预归因解释深度肺癌风险预测
链接：https://arxiv.org/abs/2602.02560

作者：Bartlomiej Sobieski,Jakub Grzywaczewski,Karol Dobiczek,Mateusz Wójcik,Tomasz Bartczak,Patryk Szatkowski,Przemysław Bombiński,Matthew Tivnan,Przemyslaw Biecek
备注：Preprint

【9】What Drives Length of Stay After Elective Spine Surgery? Insights from a Decade of Predictive Modeling
标题：是什么决定了选择性脊柱手术后的住院时间？十年预测建模的见解
链接：https://arxiv.org/abs/2602.02517

作者：Ha Na Cho,Seungmin Jeong,Yawen Guo,Alexander Lopez,Hansen Bow,Kai Zheng

蒸馏|知识提取(2篇)

【1】SAFE-KD: Risk-Controlled Early-Exit Distillation for Vision Backbones
标题：SAFE-KD：视觉主干的风险可控早期蒸馏
链接：https://arxiv.org/abs/2602.03043

作者：Salim Khazem
备注：Submitted to IJCNN

【2】Embodiment-Aware Generalist Specialist Distillation for Unified Humanoid Whole-Body Control
标题：具有适应意识的多面手专家蒸馏，用于统一的人形全身控制
链接：https://arxiv.org/abs/2602.02960

作者：Quanquan Peng,Yunfeng Lin,Yufei Xue,Jiangmiao Pang,Weinan Zhang

聚类(3篇)

【1】NPCNet: Navigator-Driven Pseudo Text for Deep Clustering of Early Sepsis Phenotyping
标题：NPCNet：导航器驱动的伪文本，用于早期脓毒症表型的深度聚集
链接：https://arxiv.org/abs/2602.03562

作者：Pi-Ju Tsai,Charkkri Limbud,Kuan-Fu Chen,Yi-Ju Tseng

【2】Vector Quantized Latent Concepts: A Scalable Alternative to Clustering-Based Concept Discovery
标题：矢量量化潜在概念：一种可扩展的基于概念发现的替代方案
链接：https://arxiv.org/abs/2602.02726

作者：Xuemin Yu,Ankur Garg,Samira Ebrahimi Kahou,Hassan Sajjad

【3】NeuralFLoC: Neural Flow-Based Joint Registration and Clustering of Functional Data
标题：NeuralFLoC：基于神经流的功能数据联合配准和聚类
链接：https://arxiv.org/abs/2602.03169

作者：Xinyang Xiong,Siyuan jiang,Pengcheng Zeng

超分辨率|去噪|去模糊|去雾(1篇)

【1】Tiled Prompts: Overcoming Prompt Underspecification in Image and Video Super-Resolution
标题：拼贴提示：克服图像和视频超分辨率中未规范的提示
链接：https://arxiv.org/abs/2602.03342

作者：Bryan Sangwoo Kim,Jonghyun Park,Jong Chul Ye
备注：13 pages, 7 figures

自动驾驶|车辆|车道检测等(3篇)

【1】Rethinking Benign Relearning: Syntax as the Hidden Driver of Unlearning Failures
标题：重新思考良性重新学习：学习是忘记失败的隐藏驱动力
链接：https://arxiv.org/abs/2602.03379

作者：Sangyeon Yoon,Hyesoo Hong,Wonje Jeung,Albert No
备注：Accepted at ICLR 2026

【2】SATORIS-N: Spectral Analysis based Traffic Observation Recovery via Informed Subspaces and Nuclear-norm minimization
标题：SATORIS-N：通过知情子空间和核规范最小化基于谱分析的流量观测恢复
链接：https://arxiv.org/abs/2602.03138

作者：Sampad Mohanty,Bhaskar Krishnamachari

【3】Incident-Guided Spatiotemporal Traffic Forecasting
标题：事件引导的时空交通预测
链接：https://arxiv.org/abs/2602.02528

作者：Lixiang Fan,Bohao Li,Tao Zou,Bowen Du,Junchen Ye

点云|SLAM|雷达|激光|深度RGBD相关(1篇)

【1】Sparsity is Combinatorial Depth: Quantifying MoE Expressivity via Tropical Geometry
标题：稀疏性即组合深度：通过热带几何量化MoE表现力
链接：https://arxiv.org/abs/2602.03204

作者：Ye Su,Huayi Tang,Zixuan Gong,Yong Liu

联邦学习|隐私保护|加密(3篇)

【1】Achieving Linear Speedup for Composite Federated Learning
标题：实现复合联邦学习的线性加速
链接：https://arxiv.org/abs/2602.03357

作者：Kun Huang,Shi Pu
备注：27 pages, 12 figures

【2】Trustworthy Blockchain-based Federated Learning for Electronic Health Records: Securing Participant Identity with Decentralized Identifiers and Verifiable Credentials
标题：值得信赖的基于区块链的电子健康记录联邦学习：通过分散的标识符和可验证的凭证保护参与者身份
链接：https://arxiv.org/abs/2602.02629

作者：Rodrigo Tertulino,Ricardo Almeida,Laercio Alencar

【3】VR-VFL: Joint Rate and Client Selection for Vehicular Federated Learning Under Imperfect CSI
标题：VR-VFL：不完美SI下车辆联邦学习的联合速率和客户端选择
链接：https://arxiv.org/abs/2602.03711

作者：Metehan Karatas,Subhrakanti Dey,Christian Rohner,Jose Mairton Barros da Silva
备注：This paper has been accepted for presentation at IEEE ICC 2026

推理|分析|理解|解释(28篇)

【1】Understanding and Exploiting Weight Update Sparsity for Communication-Efficient Distributed RL
标题：理解和利用权重更新稀疏性来实现通信高效的分布式RL
链接：https://arxiv.org/abs/2602.03839

作者：Erfan Miahi,Eugene Belilovsky
备注：32 pages, 14 figures

【2】Conformal Thinking: Risk Control for Reasoning on a Compute Budget
标题：保形思维：计算预算推理的风险控制
链接：https://arxiv.org/abs/2602.03814

作者：Xi Wang,Anushri Suresh,Alvin Zhang,Rishi More,William Jurayj,Benjamin Van Durme,Mehrdad Farajtabar,Daniel Khashabi,Eric Nalisnick

【3】Should I use Synthetic Data for That? An Analysis of the Suitability of Synthetic Data for Data Sharing and Augmentation
标题：我应该使用合成数据来实现这一目标吗？合成数据对数据共享和增强的适用性分析
链接：https://arxiv.org/abs/2602.03791

作者：Bogdan Kulynych,Theresa Stadler,Jean Louis Raisaro,Carmela Troncoso
备注：BK and TS contributed equally

【4】Inference-time Unlearning Using Conformal Prediction
标题：使用保形预测的推理时取消学习
链接：https://arxiv.org/abs/2602.03787

作者：Somnath Basu Roy Chowdhury,Rahul Kidambi,Avinava Dubey,David Wang,Gokhan Mergen,Amr Ahmed,Aranyak Mehta

【5】Reasoning Cache: Continual Improvement Over Long Horizons via Short-Horizon RL
标题：推理缓存：通过短期RL在长期视野中持续改进
链接：https://arxiv.org/abs/2602.03773

作者：Ian Wu,Yuxiao Qu,Amrith Setlur,Aviral Kumar
备注：preprint

【6】DALI: A Workload-Aware Offloading Framework for Efficient MoE Inference on Local PCs
标题：Dali：一个工作负载感知卸载框架，用于在本地PC上进行高效MoE推理
链接：https://arxiv.org/abs/2602.03495

作者：Zeyu Zhu,Gang Li,Peisong Wang,Zitao Mo,Minnan Pei,Zhuoran Song,Xiaoyao Liang,Jian Cheng

【7】Causal Inference on Networks under Misspecified Exposure Mappings: A Partial Identification Framework
标题：错误指定暴露映射下网络的因果推理：部分识别框架
链接：https://arxiv.org/abs/2602.03459

作者：Maresa Schröder,Miruna Oprescu,Stefan Feuerriegel,Nathan Kallus

【8】Symbol-Aware Reasoning with Masked Discrete Diffusion for Handwritten Mathematical Expression Recognition
标题：用于手写数学公式识别的掩蔽离散扩散符号感知推理
链接：https://arxiv.org/abs/2602.03370

作者：Takaya Kawakatsu,Ryo Ishiyama

【9】Token Sparse Attention: Efficient Long-Context Inference with Interleaved Token Selection
标题：令牌稀疏注意力：采用交织令牌选择的高效长上下文推理
链接：https://arxiv.org/abs/2602.03216

作者：Dongwon Jo,Beomseok Kang,Jiwon Song,Jae-Joon Kim

【10】ForesightKV: Optimizing KV Cache Eviction for Reasoning Models by Learning Long-Term Contribution
标题：ForesightKN：通过学习长期贡献优化推理模型的NV缓存驱逐
链接：https://arxiv.org/abs/2602.03203

作者：Zican Dong,Peiyu Liu,Junyi Li,Zhipeng Chen,Han Peng,Shuo Wang,Wayne Xin Zhao

【11】Prompt Augmentation Scales up GRPO Training on Mathematical Reasoning
标题：即时增强扩大数学推理方面的GRPO训练
链接：https://arxiv.org/abs/2602.03190

作者：Wenquan Lu,Hai Huang,Randall Balestriero

【12】MemCast: Memory-Driven Time Series Forecasting with Experience-Conditioned Reasoning
标题：MemCast：使用经验条件推理的记忆驱动时间序列预测
链接：https://arxiv.org/abs/2602.03164

作者：Xiaoyu Tao,Mingyue Cheng,Ze Guo,Shuo Yu,Yaguo Liu,Qi Liu,Shijin Wang

【13】Aligning Forest and Trees in Images and Long Captions for Visually Grounded Understanding
标题：在图像和长说明中对齐森林和树木，以实现基于视觉的理解
链接：https://arxiv.org/abs/2602.02977

作者：Byeongju Woo,Zilin Wang,Byeonghyun Pak,Sangwoo Mo,Stella X. Yu
备注：Preprint

【14】A Multi-scale Linear-time Encoder for Whole-Slide Image Analysis
标题：用于整片图像分析的多尺度线性时间编码器
链接：https://arxiv.org/abs/2602.02918

作者：Jagan Mohan Reddy Dwarampudi,Joshua Wong,Hien Van Nguyen,Tania Banerjee
备注：Accepted to ISBI 2026, 4 pages with 2 figures

【15】STEER: Inference-Time Risk Control via Constrained Quality-Diversity Search
标题：STEER：通过受约束的质量多样性搜索进行推理时风险控制
链接：https://arxiv.org/abs/2602.02862

作者：Eric Yang,Jong Ha Lee,Jonathan Amar,Elissa Ye,Yugang Jia
备注：20 pages

【16】When pre-training hurts LoRA fine-tuning: a dynamical analysis via single-index models
标题：当预训练伤害LoRA微调时：通过单指标模型的动态分析
链接：https://arxiv.org/abs/2602.02855

作者：Gibbs Nwemadji,Bruno Loureiro,Jean Barbier

【17】Causality--Δ: Jacobian-Based Dependency Analysis in Flow Matching Models
标题：因果关系--Δ：流量匹配模型中基于Jacobian的依赖性分析
链接：https://arxiv.org/abs/2602.02793

作者：Reza Rezvan,Gustav Gille,Moritz Schauer,Richard Torkar
备注：11 pages, 5 figures. Code: https://github.com/rezaarezvan/causdiff

【18】LEMON: Local Explanations via Modality-aware OptimizatioN
标题：LEMON：通过模式感知优化进行本地化
链接：https://arxiv.org/abs/2602.02786

作者：Yu Qin,Phillip Sloan,Raul Santos-Rodriguez,Majid Mirmehdi,Telmo de Menezes e Silva Filho

【19】Towards Understanding Steering Strength
标题：了解转向强度
链接：https://arxiv.org/abs/2602.02712

作者：Magamed Taimeskhanov,Samuel Vaiter,Damien Garreau
备注：33 pages (including appendix)

【20】QuantLRM: Quantization of Large Reasoning Models via Fine-Tuning Signals
标题：QuantLRM：通过微调信号量化大型推理模型
链接：https://arxiv.org/abs/2602.02581

作者：Nan Zhang,Eugene Kwek,Yusen Zhang,Muyu Pan,Suhang Wang,Prasenjit Mitra,Rui Zhang

【21】Reward Shaping for Inference-Time Alignment: A Stackelberg Game Perspective
标题：推理时间一致的奖励塑造：斯塔克伯格游戏的视角
链接：https://arxiv.org/abs/2602.02572

作者：Haichuan Wang,Tao Lin,Lingkai Kong,Ce Li,Hezi Jiang,Milind Tambe

【22】EEO-TFV: Escape-Explore Optimizer for Web-Scale Time-Series Forecasting and Vision Analysis
标题：EEO-TFV：Web规模时间序列预测和愿景分析的逃避探索优化器
链接：https://arxiv.org/abs/2602.02551

作者：Hua Wang,Jinghao Lu,Fan Zhang
备注：Main paper: 12 pages

【23】Beyond Alignment: Expanding Reasoning Capacity via Manifold-Reshaping Policy Optimization
标题：超越一致：通过多元化重塑政策优化扩大推理能力
链接：https://arxiv.org/abs/2602.02545

作者：Dayu Wang,Jiaye Yang,Weikang Li,Jiahui Liang,Yang Li

【24】From Sparse Decisions to Dense Reasoning: A Multi-attribute Trajectory Paradigm for Multimodal Moderation
标题：从稀疏决策到密集推理：多模式调节的多属性轨迹范式
链接：https://arxiv.org/abs/2602.02536

作者：Tianle Gu,Kexin Huang,Lingyu Li,Ruilin Luo,Shiyang Huang,Zongqi Wang,Yujiu Yang,Yan Teng,Yingchun Wang

【25】Enhancing Psychologists' Understanding through Explainable Deep Learning Framework for ADHD Diagnosis
标题：通过可解释的深度学习框架增强心理学家对ADHD诊断的理解
链接：https://arxiv.org/abs/2602.02535

作者：Abdul Rehman,Ilona Heldal,Jerry Chun-Wei Lin

【26】Improved Analysis of the Accelerated Noisy Power Method with Applications to Decentralized PCA
标题：加速降噪方法的改进分析及其在分散PCA中的应用
链接：https://arxiv.org/abs/2602.03682

作者：Pierre Aguié,Mathieu Even,Laurent Massoulié

【27】Simulation-Based Inference via Regression Projection and Batched Discrepancies
标题：通过回归投影和批量偏差进行基于模拟的推理
链接：https://arxiv.org/abs/2602.03613

作者：Arya Farahi,Jonah Rose,Paul Torrey
备注：comments are welcome,

【28】Unified Inference Framework for Single and Multi-Player Performative Prediction: Method and Asymptotic Optimality
标题：单人和多人表演预测的统一推理框架：方法和渐进最优性
链接：https://arxiv.org/abs/2602.03049

作者：Zhixian Zhang,Xiaotian Hou,Linjun Zhang

检测相关(9篇)

【1】ContraLog: Log File Anomaly Detection with Contrastive Learning and Masked Language Modeling
标题：Contrast Log：利用对比学习和掩蔽语言建模进行日志文件异常检测
链接：https://arxiv.org/abs/2602.03678

作者：Simon Dietz,Kai Klede,An Nguyen,Bjoern M Eskofier
备注：26 pages with 16 figures

【2】Efficient Sequential Neural Network with Spatial-Temporal Attention and Linear LSTM for Robust Lane Detection Using Multi-Frame Images
标题：具有时空注意力和线性LSTM的高效序列神经网络，用于使用多帧图像进行鲁棒的车道检测
链接：https://arxiv.org/abs/2602.03669

作者：Sandeep Patil,Yongqi Dong,Haneen Farah,Hans Hellendoorn
备注：14 pages, 9 figures, under review by IEEE T-ITS

【3】SAGE-5GC: Security-Aware Guidelines for Evaluating Anomaly Detection in the 5G Core Network
标题：SAGE-5GC：评估5G核心网络异常检测的安全意识指南
链接：https://arxiv.org/abs/2602.03596

作者：Cristian Manca,Christian Scano,Giorgio Piras,Fabio Brau,Maura Pintor,Battista Biggio
备注：ITASEC-2026

【4】Anomaly Detection via Mean Shift Density Enhancement
标题：通过均移密度增强进行异常检测
链接：https://arxiv.org/abs/2602.03293

作者：Pritam Kar,Rahul Bordoloi,Olaf Wolkenhauer,Saptarshi Bej

【5】Rare Event Early Detection: A Dataset of Sepsis Onset for Critically Ill Trauma Patients
标题：罕见事件早期检测：危重创伤患者败血症发病数据集
链接：https://arxiv.org/abs/2602.02930

作者：Yin Jin,Tucker R. Stewart,Deyi Zhou,Chhavi Gupta,Arjita Nema,Scott C. Brakenridge,Grant E. O'Keefe,Juhua Hu

【6】Refining Decision Boundaries In Anomaly Detection Using Similarity Search Within the Feature Space
标题：利用特征空间内相似性搜索细化异常检测中的决策边界
链接：https://arxiv.org/abs/2602.02925

作者：Sidahmed Benabderrahmane,Petko Valtchev,James Cheney,Talal Rahwan

【7】Late-Stage Generalization Collapse in Grokking: Detecting anti-grokking with Weightwatcher
标题：Grokking的后期概括崩溃：使用Weightwatchcher检测反Grokking
链接：https://arxiv.org/abs/2602.02859

作者：Hari K Prakash,Charles H Martin
备注：27 pages

【8】LPCVAE: A Conditional VAE with Long-Term Dependency and Probabilistic Time-Frequency Fusion for Time Series Anomaly Detection
标题：LPCVAE：一种具有长期依赖性和概率时频融合的条件VAE，用于时间序列异常检测
链接：https://arxiv.org/abs/2510.10915

作者：Hanchang Cheng,Weimin Mu,Fan Liu,Weilin Zhu,Can Ma

【9】Downscaling land surface temperature data using edge detection and block-diagonal Gaussian process regression
标题：使用边缘检测和块对角高斯过程回归缩减地表温度数据
链接：https://arxiv.org/abs/2602.02813

作者：Sanjit Dandapanthula,Margaret Johnson,Madeleine Pascolini-Campbell,Glynn Hulley,Mikael Kuusela

分类|识别(5篇)

【1】Enhancing Imbalanced Node Classification via Curriculum-Guided Feature Learning and Three-Stage Attention Network
标题：通过课程引导特征学习和三阶段注意力网络增强不平衡节点分类
链接：https://arxiv.org/abs/2602.03808

作者：Abdul Joseph Fofanah,Lian Wen,David Chen,Shaoyang Zhang

【2】BinaryPPO: Efficient Policy Optimization for Binary Classification
标题：BinaryPPO：二元分类的高效策略优化
链接：https://arxiv.org/abs/2602.02708

作者：Punya Syon Pandey,Zhijing Jin

【3】Auto-Augmentation Contrastive Learning for Wearable-based Human Activity Recognition
标题：基于可穿戴设备的人类活动识别的自动增强对比学习
链接：https://arxiv.org/abs/2602.02542

作者：Qingyu Wu,Jianfei Shen,Feiyi Fan,Yang Gu,Chenyang Xu,Yiqiang Chen

【4】How Much Information Can a Vision Token Hold? A Scaling Law for Recognition Limits in VLMs
标题：Vision Token可以容纳多少信息？VLM中识别限制的比例定律
链接：https://arxiv.org/abs/2602.02539

作者：Shuxin Zhuang,Zi Liang,Runsheng Yu,Hongzong Li,Rong Feng,Shiqin Tang,Youzhi Zhang

【5】Plug-In Classification of Drift Functions in Diffusion Processes Using Neural Networks
标题：使用神经网络对扩散过程漂移函数进行插件分类
链接：https://arxiv.org/abs/2602.02791

作者：Yuzhen Zhao,Jiarong Fan,Yating Liu

表征(6篇)

【1】Asymmetric Hierarchical Anchoring for Audio-Visual Joint Representation: Resolving Information Allocation Ambiguity for Robust Cross-Modal Generalization
标题：视听联合表示的非对称分层锚定：解决信息分配模糊性以实现鲁棒跨模式概括
链接：https://arxiv.org/abs/2602.03570

作者：Bixing Wu,Yuhong Zhao,Zongli Ye,Jiachen Lian,Xiangyu Yue,Gopala Anumanchipalli
备注：18 pages, 11 figures

【2】Robust Representation Learning in Masked Autoencoders
标题：掩蔽自动编码器中的鲁棒表示学习
链接：https://arxiv.org/abs/2602.03531

作者：Anika Shrivastava,Renu Rameshan,Samar Agnihotri
备注：11 pages, 8 figures, and 3 tables

【3】What Makes a Good Example? Modeling Exemplar Selection with Neural Network Representations
标题：什么是一个好例子？用神经网络表示对样本选择进行建模
链接：https://arxiv.org/abs/2602.03144

作者：Fanxiao Wani Qiu,Oscar Leong,Alexander LaTourrette

【4】Notes on the Reward Representation of Posterior Updates
标题：关于后期更新的奖励表示的注释
链接：https://arxiv.org/abs/2602.02912

作者：Pedro A. Ortega
备注：Technical report, 9 pages

【5】Fubini Study geometry of representation drift in high dimensional data
标题：富比尼研究多维数据中表示漂移的几何
链接：https://arxiv.org/abs/2602.02596

作者：Arturo Tozzi
备注：8 pages, 1 figure

【6】Learning ORDER-Aware Multimodal Representations for Composite Materials Design
标题：学习用于复合材料设计的订单感知多峰表示
链接：https://arxiv.org/abs/2602.02513

作者：Xinyao Li,Hangwei Qian,Jingjing Li,Ivor Tsang

3D|3D重建等相关(2篇)

【1】3D-Learning: Diffusion-Augmented Distributionally Robust Decision-Focused Learning
标题：3D学习：扩散增强的分布稳健的决策中心学习
链接：https://arxiv.org/abs/2602.02943

作者：Jiaqi Wen,Lei Fan,Jianyi Yang

【2】Position: 3D Gaussian Splatting Watermarking Should Be Scenario-Driven and Threat-Model Explicit
标题：立场：3D高斯飞溅水印应该是场景驱动且威胁模型显式的
链接：https://arxiv.org/abs/2602.02602

作者：Yangfan Deng,Anirudh Nakra,Min Wu

优化|敛散性(13篇)

【1】Riemannian Neural Optimal Transport
标题：雷曼神经最优传输
链接：https://arxiv.org/abs/2602.03566

作者：Alessandro Micheli,Yueqi Cao,Anthea Monod,Samir Bhatt
备注：58 pages

【2】Generative Decompression: Optimal Lossy Decoding Against Distribution Mismatch
标题：生成式解压缩：针对分布不匹配的最佳有损解码
链接：https://arxiv.org/abs/2602.03505

作者：Saeed R. Khosravirad,Ahmed Alkhateeb,Ingrid van de Voorde

【3】Reparameterization Flow Policy Optimization
标题：重新参数化流程政策优化
链接：https://arxiv.org/abs/2602.03501

作者：Hai Zhong,Zhuoran Li,Xun Wang,Longbo Huang

【4】An Approximate Ascent Approach To Prove Convergence of PPO
标题：证明PPO收敛性的近似上升方法
链接：https://arxiv.org/abs/2602.03386

作者：Leif Doering,Daniel Schmidt,Moritz Melcher,Sebastian Kassing,Benedikt Wille,Tilman Aach,Simon Weissmann

【5】Dynamic Topology Optimization for Non-IID Data in Decentralized Learning
标题：分散学习中非IID数据的动态布局优化
链接：https://arxiv.org/abs/2602.03383

作者：Bart Cox,Antreas Ioannou,Jérémie Decouchant
备注：10 pages, 11 figures. Accepted for publication in the 13th IEEE International Conference on Big Data (BigData 2025). To appear

【6】From Inexact Gradients to Byzantine Robustness: Acceleration and Optimization under Similarity
标题：从不精确的继承者到拜占庭式的稳健性：相似性下的加速和优化
链接：https://arxiv.org/abs/2602.03329

作者：Renaud Gaucher,Aymeric Dieuleveut,Hadrien Hendrikx

【7】BayeSQP: Bayesian Optimization through Sequential Quadratic Programming
标题：BayeSQP：通过序列二次规划的Bayesian优化
链接：https://arxiv.org/abs/2602.03232

作者：Paul Brunzema,Sebastian Trimpe

【8】PRISM: Structured Optimization via Anisotropic Spectral Shaping
标题：PRism：通过各向异性谱整形的结构化优化
链接：https://arxiv.org/abs/2602.03096

作者：Yujie Yang

【9】FlashSinkhorn: IO-Aware Entropic Optimal Transport
标题：Flash Sinkhorn：IO-aware的Entropic最佳运输
链接：https://arxiv.org/abs/2602.03067

作者：Felix X. -F. Ye,Xingjie Li,An Yu,Ming-Ching Chang,Linsong Chu,Davis Wertheimer

【10】Co2PO: Coordinated Constrained Policy Optimization for Multi-Agent RL
标题：Co2 PO：多代理RL的协调约束政策优化
链接：https://arxiv.org/abs/2602.02970

作者：Shrenik Patel,Christine Truong

【11】A Reduction from Delayed to Immediate Feedback for Online Convex Optimization with Improved Guarantees
标题：在线凸优化从延迟反馈减少到立即反馈并改善保证
链接：https://arxiv.org/abs/2602.02634

作者：Alexander Ryabchenko,Idan Attias,Daniel M. Roy

【12】ContextEvolve: Multi-Agent Context Compression for Systems Code Optimization
标题：ContextEvolve：用于系统代码优化的多代理上下文压缩
链接：https://arxiv.org/abs/2602.02597

作者：Hongyuan Su,Yu Zheng,Yong Li

【13】Design and Evaluation of Whole-Page Experience Optimization for E-commerce Search
标题：电子商务搜索整页体验优化的设计与评估
链接：https://arxiv.org/abs/2602.02514

作者：Pratik Lahiri,Bingqing Ge,Zhou Qin,Aditya Jumde,Shuning Huo,Lucas Scottini,Yi Liu,Mahmoud Mamlouk,Wenyang Liu

预测|估计(15篇)

【1】Prediction of Critical Heat Flux in Rod Bundles Using Tube-Based Hybrid Machine Learning Models in CTF
标题：在CTF中使用基于管的混合机器学习模型预测棒束中的临界热通量
链接：https://arxiv.org/abs/2602.03805

作者：Aidan Furlong,Robert Salko,Xingang Zhao,Xu Wu
备注：Submitted to the 2026 American Nuclear Society Annual Meeting

【2】Efficient Estimation of Kernel Surrogate Models for Task Attribution
标题：任务属性核代理模型的有效估计
链接：https://arxiv.org/abs/2602.03783

作者：Zhenshuo Zhang,Minxuan Duan,Hongyang R. Zhang
备注：27 pages. To appear in ICLR 2026

【3】Decision-oriented benchmarking to transform AI weather forecast access: Application to the Indian monsoon
标题：以决策为导向的基准，改变人工智能天气预报的访问方式：应用于印度季风
链接：https://arxiv.org/abs/2602.03767

作者：Rajat Masiwal,Colin Aitken,Adam Marchakitus,Mayank Gupta,Katherine Kowal,Hamid A. Pahlavan,Tyler Yang,Y. Qiang Sun,Michael Kremer,Amir Jina,William R. Boos,Pedram Hassanzadeh

【4】CoGenCast: A Coupled Autoregressive-Flow Generative Framework for Time Series Forecasting
标题：CoGenCast：用于时间序列预测的耦合自回归流生成框架
链接：https://arxiv.org/abs/2602.03564

作者：Yaguo Liu,Mingyue Cheng,Daoyu Wang,Xiaoyu Tao,Qi Liu

【5】A Minimal Task Reveals Emergent Path Integration and Object-Location Binding in a Predictive Sequence Model
标题：最小任务揭示预测序列模型中的紧急路径集成和对象位置绑定
链接：https://arxiv.org/abs/2602.03490

作者：Linda Ariel Ventura,Victoria Bosch,Tim C Kietzmann,Sushrut Thorat
备注：7 pages, 4 figures

【6】Accurate Failure Prediction in Agents Does Not Imply Effective Failure Prevention
标题：代理中的准确故障预测并不意味着有效的故障预防
链接：https://arxiv.org/abs/2602.03338

作者：Rakshith Vasudev,Melisa Russak,Dan Bikel,Waseem Alshikh

【7】Bayesian Conformal Prediction as a Decision Risk Problem
标题：作为决策风险问题的Bayesian Conformal预测
链接：https://arxiv.org/abs/2602.03331

作者：Fanyi Wu,Veronika Lohmanova,Samuel Kaski,Michele Caprio
备注：18 pages, 5 figures. Accepted at EIML 2025 at Eurips

【8】Enhanced Parcel Arrival Forecasting for Logistic Hubs: An Ensemble Deep Learning Approach
标题：物流中心增强的包裹到达预测：一种令人惊叹的深度学习方法
链接：https://arxiv.org/abs/2602.03135

作者：Xinyue Pan,Yujia Xu,Benoit Montreuil

【9】Koopman Autoencoders with Continuous-Time Latent Dynamics for Fluid Dynamics Forecasting
标题：具有连续时间潜在动力学的Koopman自动编码器用于流体动力学预测
链接：https://arxiv.org/abs/2602.02832

作者：Rares Grozavescu,Pengyu Zhang,Etienne Meunier,Mark Girolami

【10】Copula-Based Aggregation and Context-Aware Conformal Prediction for Reliable Renewable Energy Forecasting
标题：基于Copula聚合和上下文感知的保形预测在可再生能源可靠预测中的应用
链接：https://arxiv.org/abs/2602.02583

作者：Alireza Moradi,Mathieu Tanneau,Reza Zandehshahvar,Pascal Van Hentenryck

【11】IceBench-S2S: A Benchmark of Deep Learning for Challenging Subseasonal-to-Seasonal Daily Arctic Sea Ice Forecasting in Deep Latent Space
标题：IceBench-S2 S：深度学习基准，用于在深潜空间进行亚季节至季节每日北极海冰预报
链接：https://arxiv.org/abs/2602.02567

作者：Jingyi Xu,Shengnan Wang,Weidong Yang,Siwei Tu,Lei Bai,Ben Fei
备注：9 pages, 6 figures

【12】A Comparative Simulation Study of the Fairness and Accuracy of Predictive Policing Systems in Baltimore City
标题：巴尔的摩市预测警务系统公平性和准确性的比较模拟研究
链接：https://arxiv.org/abs/2602.02566

作者：Samin Semsar,Kiran Laxmikant Prabhu,Gabriella Waters,James Foulds
备注：36 pages, 27 figures

【13】A General ReLearner: Empowering Spatiotemporal Prediction by Re-learning Input-label Residual
标题：一般的重新学习者：通过重新学习输入标签残留来增强时空预测
链接：https://arxiv.org/abs/2602.02563

作者：Jiaming Ma,Binwu Wang,Pengkun Wang,Xu Wang,Zhengyang Zhou,Yang Wang

【14】Efficient Variance-reduced Estimation from Generative EHR Models: The SCOPE and REACH Estimators
标题：生成式EHR模型的有效方差降低估计：范围和REACH估计器
链接：https://arxiv.org/abs/2602.03730

作者：Luke Solo,Matthew B. A. McDermott,William F. Parker,Bashar Ramadan,Michael C. Burkhart,Brett K. Beaulieu-Jones
备注：10 pages, 2 figures

【15】Online Conformal Prediction via Universal Portfolio Algorithms
标题：通过通用投资组合算法进行在线保形预测
链接：https://arxiv.org/abs/2602.03168

作者：Tuo Liu,Edgar Dobriban,Francesco Orabona

其他神经网络|深度学习|模型|建模(57篇)

【1】PLATE: Plasticity-Tunable Efficient Adapters for Geometry-Aware Continual Learning
标题：PLATE：可塑性可调的高效适配器，用于具有几何意识的持续学习
链接：https://arxiv.org/abs/2602.03846

作者：Romain Cosentino

【2】Robust Intervention Learning from Emergency Stop Interventions
标题：从紧急停止干预中学习强有力的干预
链接：https://arxiv.org/abs/2602.03825

作者：Ethan Pronovost,Khimya Khetarpal,Siddhartha Srinivasa

【3】Anytime Pretraining: Horizon-Free Learning-Rate Schedules with Weight Averaging
标题：随时预训练：具有体重平均的无地平线学习率计划
链接：https://arxiv.org/abs/2602.03702

作者：Alexandru Meterez,Pranav Ajit Nair,Depen Morwani,Cengiz Pehlevan,Sham Kakade

【4】QuAIL: Quality-Aware Inertial Learning for Robust Training under Data Corruption
标题：QuAIL：质量感知的惯性学习，用于数据损坏下的鲁棒训练
链接：https://arxiv.org/abs/2602.03686

作者：Mattia Sabella,Alberto Archetti,Pietro Pinoli,Matteo Matteucci,Cinzia Cappiello

【5】Universal One-third Time Scaling in Learning Peaked Distributions
标题：学习峰值分布中的通用三分之一时间缩放
链接：https://arxiv.org/abs/2602.03685

作者：Yizhou Liu,Ziming Liu,Cengiz Pehlevan,Jeff Gore
备注：24 pages, 6 main text figures, 27 figures in total

【6】Sequential Group Composition: A Window into the Mechanics of Deep Learning
标题：顺序群组组成：深入学习机制的窗口
链接：https://arxiv.org/abs/2602.03655

作者：Giovanni Luca Marchetti,Daniel Kunin,Adele Myers,Francisco Acosta,Nina Miolane

【7】Quantization-Aware Regularizers for Deep Neural Networks Compression
标题：用于深度神经网络压缩的量化感知调节器
链接：https://arxiv.org/abs/2602.03614

作者：Dario Malchiodi,Mattia Ferraretto,Marco Frasca

【8】APEX: Probing Neural Networks via Activation Perturbation
标题：APEX：通过激活微扰探测神经网络
链接：https://arxiv.org/abs/2602.03586

作者：Tao Ren,Xiaoyu Luo,Qiongxiu Li

【9】$V_0$: A Generalist Value Model for Any Policy at State Zero
标题：$V_0$：零状态下任何政策的通才价值模型
链接：https://arxiv.org/abs/2602.03584

作者：Yi-Kai Zhang,Zhiyuan Yao,Hongyan Hao,Yueqing Sun,Qi Gu,Hui Su,Xunliang Cai,De-Chuan Zhan,Han-Jia Ye

【10】How to Train Your Resistive Network: Generalized Equilibrium Propagation and Analytical Learning
标题：如何训练你的阻力网络：广义均衡传播和分析学习
链接：https://arxiv.org/abs/2602.03546

作者：Jonathan Lin,Aman Desai,Frank Barrows,Francesco Caravelli
备注：8 pages double column; plus 16 supp mat.;

【11】Sparse Training of Neural Networks based on Multilevel Mirror Descent
标题：基于多层镜像下降的神经网络稀疏训练
链接：https://arxiv.org/abs/2602.03535

作者：Yannick Lunk,Sebastian J. Scott,Leon Bungert

【12】WARP Logic Neural Networks
标题：WARP逻辑神经网络
链接：https://arxiv.org/abs/2602.03527

作者：Lino Gerlach,Thore Gerlach,Liv Våge,Elliott Kauffman,Isobel Ojalvo
备注：Under review

【13】Live or Lie: Action-Aware Capsule Multiple Instance Learning for Risk Assessment in Live Streaming Platforms
标题：直播还是谎言：实时流媒体平台中风险评估的感知胶囊多实例学习
链接：https://arxiv.org/abs/2602.03520

作者：Yiran Qiao,Jing Chen,Xiang Ao,Qiwei Zhong,Yang Liu,Qing He

【14】A Function-Space Stability Boundary for Generalization in Interpolating Learning Systems
标题：插值学习系统推广的函数空间稳定边界
链接：https://arxiv.org/abs/2602.03514

作者：Ronald Katende
备注：10 pages, 8 figures,

【15】DeepDFA: Injecting Temporal Logic in Deep Learning for Sequential Subsymbolic Applications
标题：DeepPFA：在深度学习中注入时态逻辑以实现顺序子符号应用
链接：https://arxiv.org/abs/2602.03486

作者：Elena Umili,Francesco Argenziano,Roberto Capobianco

【16】ScDiVa: Masked Discrete Diffusion for Joint Modeling of Single-Cell Identity and Expression
标题：ScDiVA：用于单细胞身份和表达联合建模的掩蔽离散扩散
链接：https://arxiv.org/abs/2602.03477

作者：Mingxuan Wang,Cheng Chen,Gaoyang Jiang,Zijia Ren,Chuangxin Zhao,Lu Shi,Yanbiao Ma
备注：19 pages, 11 figures

【17】Scaling Continual Learning with Bi-Level Routing Mixture-of-Experts
标题：通过双层路由混合专家扩展持续学习
链接：https://arxiv.org/abs/2602.03473

作者：Meng Lou,Yunxiang Fu,Yizhou Yu

【18】Soft-Radial Projection for Constrained End-to-End Learning
标题：用于受约束端到端学习的软辐射投影
链接：https://arxiv.org/abs/2602.03461

作者：Philipp J. Schneider,Daniel Kuhn

【19】CRL-VLA: Continual Vision-Language-Action Learning
标题：CRL-VLA：持续视觉-语言-动作学习
链接：https://arxiv.org/abs/2602.03445

作者：Qixin Zeng,Shuo Zhang,Hongyin Zhang,Renjie Wang,Han Zhao,Libang Zhao,Runze Li,Donglin Wang,Chao Huang

【20】PACE: Pretrained Audio Continual Learning
标题：PACE：预训练的音频持续学习
链接：https://arxiv.org/abs/2602.03355

作者：Chang Li,Kanglei Zhou,Liyuan Wang
备注：Accepted at ICLR 2026

【21】Building Interpretable Models for Moral Decision-Making
标题：构建道德决策的可解释模型
链接：https://arxiv.org/abs/2602.03351

作者：Mayank Goel,Aritra Das,Paras Chopra
备注：8 pages, 4 figures, accepted to AAAI'26 Machine Ethics Workshop

【22】Periodic Regularized Q-Learning
标题：定期定期的Q学习
链接：https://arxiv.org/abs/2602.03301

作者：Hyukjun Yang,Han-Dong Lim,Donghwan Lee

【23】Lipschitz Multiscale Deep Equilibrium Models: A Theoretically Guaranteed and Accelerated Approach
标题：Lipschitz多尺度深度均衡模型：一种理论保证和加速方法
链接：https://arxiv.org/abs/2602.03297

作者：Naoki Sato,Hideaki Iiduka
备注：Accepted at AISTATS2026

【24】Lookahead Sample Reward Guidance for Test-Time Scaling of Diffusion Models
标题：扩散模型测试时间缩放的前瞻样本奖励指南
链接：https://arxiv.org/abs/2602.03211

作者：Yeongmin Kim,Donghyeok Shin,Byeonghu Na,Minsang Park,Richard Lee Kim,Il-Chul Moon
备注：Under Review

【25】Shortcut Features as Top Eigenfunctions of NTK: A Linear Neural Network Case and More
标题：NTK顶级特征函数的RST功能：线性神经网络案例及更多
链接：https://arxiv.org/abs/2602.03066

作者：Jinwoo Lim,Suhyun Kim,Soo-Mook Moon

【26】Fedcompass: Federated Clustered and Periodic Aggregation Framework for Hybrid Classical-Quantum Models
标题：Fedcompass：混合经典量子模型的联邦迭代和周期性聚合框架
链接：https://arxiv.org/abs/2602.03052

作者：Yueheng Wang,Xing He,Zinuo Cai,Rui Zhang,Ruhui Ma,Yuan Liu,Rajkumar Buyya
备注：Accepted by the 2026 IEEE International Conference on Acoustics, Speech, and Signal Processing(ICASSP 2026)

【27】Consistency Deep Equilibrium Models
标题：一致性深度均衡模型
链接：https://arxiv.org/abs/2602.03024

作者：Junchao Lin,Zenan Ling,Jingwen Xu,Robert C. Qiu

【28】Methods and Open Problems in Differentiable Social Choice: Learning Mechanisms, Decisions, and Alignment
标题：差异社会选择的方法和悬而未决的问题：学习机制、决策和一致
链接：https://arxiv.org/abs/2602.03003

作者：Zhiyu An,Wan Du

【29】Learning to Repair Lean Proofs from Compiler Feedback
标题：学习从更简单的反馈中修复精益证明
链接：https://arxiv.org/abs/2602.02990

作者：Evan Wang,Simon Chess,Daniel Lee,Siyuan Ge,Ajit Mallavarapu,Vasily Ilin
备注：15 pages, 6 figures

【30】Why Some Models Resist Unlearning: A Linear Stability Perspective
标题：为什么一些模型抵制放弃学习：线性稳定性的角度
链接：https://arxiv.org/abs/2602.02986

作者：Wei-Kai Chang,Rajiv Khanna

【31】Learning Fast Monomial Orders for Gröbner Basis Computations
标题：学习Gröbner基计算的快速单元阶
链接：https://arxiv.org/abs/2602.02972

作者：R. Caleb Bunch,Alperen A. Ergür,Melika Golestani,Jessie Tong,Malia Walewski,Yunus E. Zeytuncu

【32】Q-ShiftDP: A Differentially Private Parameter-Shift Rule for Quantum Machine Learning
标题：Q-ShiftDP：量子机器学习的差异私有参数转移规则
链接：https://arxiv.org/abs/2602.02962

作者：Hoang M. Ngo,Nhat Hoang-Xuan,Quan Nguyen,Nguyen Do,Incheol Shin,My T. Thai

【33】Distance Marching for Generative Modeling
标题：生成建模的距离行军
链接：https://arxiv.org/abs/2602.02928

作者：Zimo Wang,Ishit Mehta,Haolin Lu,Chung-En Sun,Ge Yan,Tsui-Wei Weng,Tzu-Mao Li

【34】A Reproducible Framework for Bias-Resistant Machine Learning on Small-Sample Neuroimaging Data
标题：小样本神经影像数据抗偏机器学习的可复制框架
链接：https://arxiv.org/abs/2602.02920

作者：Jagan Mohan Reddy Dwarampudi,Jennifer L Purks,Joshua Wong,Renjie Hu,Tania Banerjee
备注：Accepted to ISBI 2026, 5 pages with 1 figure

【35】A Random Matrix Theory Perspective on the Consistency of Diffusion Models
标题：扩散模型一致性的随机矩阵理论观点
链接：https://arxiv.org/abs/2602.02908

作者：Binxu Wang,Jacob Zavatone-Veth,Cengiz Pehlevan
备注：65 pages; 53 figures

【36】Self-Soupervision: Cooking Model Soups without Labels
标题：自我反省：没有标签的烹饪模型汤
链接：https://arxiv.org/abs/2602.02890

作者：Anthony Fuller,James R. Green,Evan Shelhamer
备注：code: https://github.com/antofuller/self_soupervision data: https://huggingface.co/datasets/antofuller/mini-VTAB

【37】Recurrent Equivariant Constraint Modulation: Learning Per-Layer Symmetry Relaxation from Data
标题：循环等变约束调制：从数据学习每层对称松弛
链接：https://arxiv.org/abs/2602.02853

作者：Stefanos Pertigkiozoglou,Mircea Petrache,Shubhendu Trivedi,Kostas Daniilidis

【38】Semantics-Aware Generative Latent Data Augmentation for Learning in Low-Resource Domains
标题：用于低资源领域学习的语义感知生成潜在数据增强
链接：https://arxiv.org/abs/2602.02841

作者：Jae-Sung Bae,Minje Kim

【39】Structure-Preserving Learning Improves Geometry Generalization in Neural PDEs
标题：结构保留学习改进神经PDEs中的几何推广
链接：https://arxiv.org/abs/2602.02788

作者：Benjamin D. Shaffer,Shawn Koohy,Brooks Kinch,M. Ani Hsieh,Nathaniel Trask

【40】Provable Effects of Data Replay in Continual Learning: A Feature Learning Perspective
标题：连续学习中数据回放的可证明效果：特征学习的角度
链接：https://arxiv.org/abs/2602.02767

作者：Meng Ding,Jinhui Xu,Kaiyi Ji
备注：AISTATS 2026

【41】NSC-SL: A Bandwidth-Aware Neural Subspace Compression for Communication-Efficient Split Learning
标题：NS-SL：一种带宽感知神经子空间压缩，用于通信高效的分离学习
链接：https://arxiv.org/abs/2602.02696

作者：Zhen Fang,Miao Yang,Zehang Lin,Zheng Lin,Zihan Fang,Zongyuan Zhang,Tianyang Duan,Dong Huang,Shunzhi Zhu
备注：5 pages, 3 figures

【42】Learning Better Certified Models from Empirically-Robust Teachers
标题：从经验强大的教师那里学习更好的认证模型
链接：https://arxiv.org/abs/2602.02626

作者：Alessandro De Palma

【43】Learning Consistent Causal Abstraction Networks
标题：学习一致因果抽象网络
链接：https://arxiv.org/abs/2602.02623

作者：Gabriele D'Acunto,Paolo Di Lorenzo,Sergio Barbarossa
备注：To be published in the proceedings of ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). arXiv admin note: substantial text overlap with arXiv:2509.25236

【44】Product Interaction: An Algebraic Formalism for Deep Learning Architectures
标题：产品交互：深度学习架构的代数形式主义
链接：https://arxiv.org/abs/2602.02573

作者：Haonan Dong,Chun-Wun Cheng,Angelica I. Aviles-Rivero

【45】PA-MIL: Phenotype-Aware Multiple Instance Learning Guided by Language Prompting and Genotype-to-Phenotype Relationships
标题：PA-MIL：由语言预设和基因型与表型关系指导的表型感知多实例学习
链接：https://arxiv.org/abs/2602.02558

作者：Zekang Yang,Hong Liu,Xiangdong Wang

【46】The Alignment Curse: Cross-Modality Jailbreak Transfer in Omni-Models
标题：一致诅咒：全方位模型中的跨模式越狱转移
链接：https://arxiv.org/abs/2602.02557

作者：Yupeng Chen,Junchi Yu,Aoxi Liu,Philip Torr,Adel Bibi

【47】Toward Ultra-Long-Horizon Sequential Model Editing
标题：迈向超长视野序列模型编辑
链接：https://arxiv.org/abs/2602.02543

作者：Mingda Liu,Zhenghan Zhu,Ze'an Miao,Katsuki Fujisawa

【48】TabularMath: Evaluating Computational Extrapolation in Tabular Learning via Program-Verified Synthesis
标题：TabularMath：通过程序验证的合成评估表格学习中的计算外推
链接：https://arxiv.org/abs/2602.02523

作者：Zerui Cheng,Jiashuo Liu,Jianzhu Yao,Pramod Viswanath,Ge Zhang,Wenhao Huang
备注：30 pages; TabularMath technical report

【49】Learning-augmented smooth integer programs with PAC-learnable oracles
标题：学习增广光滑整数规划与PAC-可学习预言机
链接：https://arxiv.org/abs/2602.02505

作者：Hao-Yuan He,Ming Li

【50】Sparse Adapter Fusion for Continual Learning in NLP
标题：用于NLP中连续学习的稀疏适配器融合
链接：https://arxiv.org/abs/2602.02502

作者：Min Zeng,Xi Chen,Haiqin Yang,Yike Guo
备注：This paper has been accepted to EACL 2026

【51】Test-Time Detoxification without Training or Learning Anything
标题：无需训练或学习任何内容即可进行考试时非自愿化
链接：https://arxiv.org/abs/2602.02498

作者：Baturay Saglam,Dionysis Kalogerias

【52】RLAnything: Forge Environment, Policy, and Reward Model in Completely Dynamic RL System
标题：RL Anything：完全动态RL系统中的锻造环境、政策和奖励模型
链接：https://arxiv.org/abs/2602.02488

作者：Yinjie Wang,Tianbao Xie,Ke Shen,Mengdi Wang,Ling Yang
备注：Code: https://github.com/Gen-Verse/Open-AgentRL

【53】Preference-based Conditional Treatment Effects and Policy Learning
标题：基于偏好的有条件待遇效果和政策学习
链接：https://arxiv.org/abs/2602.03823

作者：Dovid Parnas,Mathieu Even,Julie Josse,Uri Shalit
备注：Accepted to AISTATS 2026; 10 pages + appendix

【54】Acceleration of Atomistic NEGF: Algorithms, Parallelization, and Machine Learning
标题：原子NGF的加速：算法、并行化和机器学习
链接：https://arxiv.org/abs/2602.03438

作者：Mathieu Luisier,Nicolas Vetsch,Alexander Maeder,Vincent Maillou,Anders Winka,Leonard Deuschle,Chen Hao Xia,Manasa Kaniselvan,Marko Mladenovic,Jiang Cao,Alexandros Nikolaos Ziogas

【55】Latent Neural-ODE for Model-Informed Precision Dosing: Overcoming Structural Assumptions in Pharmacokinetics
标题：模型知情精确给药的潜在神经ODE：克服药代动力学中的结构假设
链接：https://arxiv.org/abs/2602.03215

作者：Benjamin Maurel,Agathe Guilloux,Sarah Zohar,Moreno Ursino,Jean-Baptiste Woillard

【56】Weighted Sum-of-Trees Model for Clustered Data
标题：离散数据的加权树和模型
链接：https://arxiv.org/abs/2602.02931

作者：Kevin McCoy,Zachary Wooten,Katarzyna Tomczak,Christine B. Peterson
备注：14 pages, 8 figures, 3 tables

【57】Training-Free Self-Correction for Multimodal Masked Diffusion Models
标题：多峰掩蔽扩散模型的免训练自修正
链接：https://arxiv.org/abs/2602.02927

作者：Yidong Ouyang,Panwen Hu,Zhengyan Wan,Zhe Wang,Liyan Xie,Dmitriy Bespalov,Ying Nian Wu,Guang Cheng,Hongyuan Zha,Qiang Sun

其他(64篇)

【1】Investigating Quantum Circuit Designs Using Neuro-Evolution
标题：使用神经进化研究量子电路设计
链接：https://arxiv.org/abs/2602.03840

作者：Devroop Kar,Daniel Krutz,Travis Desell
备注：Submitted to The Genetic and Evolutionary Computation Conference (GECCO) 2026. Under Review

【2】Antidistillation Fingerprinting
标题：反蒸馏指纹
链接：https://arxiv.org/abs/2602.03812

作者：Yixuan Even Xu,John Kirchenbauer,Yash Savani,Asher Trockman,Alexander Robey,Tom Goldstein,Fei Fang,J. Zico Kolter
备注：26 pages, 11 figures

【3】Manifold Random Features
标题：多种随机特征
链接：https://arxiv.org/abs/2602.03797

作者：Ananya Parashar,Derek Long,Dwaipayan Saha,Krzysztof Choromanski

【4】Reward Redistribution for CVaR MDPs using a Bellman Operator on L-infinity
标题：使用L-无限上Bellman运算符对CVaR MDP进行奖励重新分配
链接：https://arxiv.org/abs/2602.03778

作者：Aneri Muni,Vincent Taboga,Esther Derman,Pierre-Luc Bacon,Erick Delage

【5】UniGeM: Unifying Data Mixing and Selection via Geometric Exploration and Mining
标题：UniGeM：通过几何探索和挖掘统一数据混合和选择
链接：https://arxiv.org/abs/2602.03772

作者：Changhao Wang,Yunfei Yu,Xinhao Yao,Jiaolong Yang,Riccardo Cantoro,Chaobo Li,Qing Cui,Jun Zhou

【6】Fast-MWEM: Private Data Release in Sublinear Time
标题：Fast-MWEM：亚线性时间的私有数据发布
链接：https://arxiv.org/abs/2602.03732

作者：Themistoklis Haris,Steve Choi,Mutiraj Laksanawisit

【7】Efficient Training of Boltzmann Generators Using Off-Policy Log-Dispersion Regularization
标题：使用非策略日志分散正规化有效训练Boltzmann生成器
链接：https://arxiv.org/abs/2602.03729

作者：Henrik Schopmans,Christopher von Klitzing,Pascal Friederich

【8】Conflict-Resolving and Sharpness-Aware Minimization for Generalized Knowledge Editing with Multiple Updates
标题：具有多个更新的广义知识编辑的预算解析和敏锐性最小化
链接：https://arxiv.org/abs/2602.03696

作者：Duy Nguyen,Hanqi Xiao,Archiki Prasad,Elias Stengel-Eskin,Hyunji Lee,Mohit Bansal
备注：22 pages, 8 figures. Code link: https://github.com/duykhuongnguyen/CoRSA

【9】Equilibrium Propagation for Non-Conservative Systems
标题：非保守系统的平衡传播
链接：https://arxiv.org/abs/2602.03670

作者：Antonino Emanuele Scurria,Dimitri Vanden Abeele,Bortolo Matteo Mognetti,Serge Massar
备注：19 pages (9 pages main text), 7 figures

【10】Mitigating Conversational Inertia in Multi-Turn Agents
标题：缓解多回合代理中的对话惰性
链接：https://arxiv.org/abs/2602.03664

作者：Yang Wan,Zheng Cao,Zhenhao Zhang,Zhengwen Zeng,Shuheng Shen,Changhua Meng,Linchao Zhu

【11】Reinforcement Fine-Tuning for History-Aware Dense Retriever in RAG
标题：RAG中历史感知密集检索器的强化微调
链接：https://arxiv.org/abs/2602.03645

作者：Yicheng Zhang,Zhen Qin,Zhaomin Wu,Wenqi Zhang,Shuiguang Deng
备注：On going work. Codes are released at https://github.com/zyc140345/HARR

【12】TRE: Encouraging Exploration in the Trust Region
标题：TRE：鼓励信托区的勘探
链接：https://arxiv.org/abs/2602.03635

作者：Chao Huang,Yujing Lu,Quangang Li,Shenghe Wang,Yan Wang,Yueyang Zhang,Long Xia,Jiashu Zhao,Zhiyuan Sun,Daiting Shi,Tingwen Liu

【13】Ultra Fast PDE Solving via Physics Guided Few-step Diffusion
标题：通过物理引导的几步扩散实现超快速偏出方程求解
链接：https://arxiv.org/abs/2602.03627

作者：Cindy Xiangrui Kong,Yueqi Wang,Haoyang Zheng,Weijian Luo,Guang Lin

【14】EVE: Efficient Verification of Data Erasure through Customized Perturbation in Approximate Unlearning
标题：EVE：通过近似取消学习中的定制扰动对数据擦除进行有效验证
链接：https://arxiv.org/abs/2602.03567

作者：Weiqi Wang,Zhiyi Tian,Chenhan Zhang,Luoyu Chen,Shui Yu

【15】MatGPTQ: Accurate and Efficient Post-Training Matryoshka Quantization
标题：MatGPTQ：准确有效的训练后Matryoshka量化
链接：https://arxiv.org/abs/2602.03537

作者：Maximilian Kleinegger,Elvir Crnčević,Dan Alistarh
备注：Preprint

【16】Rank-Learner: Orthogonal Ranking of Treatment Effects
标题：排名学习者：治疗效果的垂直排名
链接：https://arxiv.org/abs/2602.03517

作者：Henri Arno,Dennis Frauen,Emil Javurek,Thomas Demeester,Stefan Feuerriegel

【17】Mitigating Staleness in Asynchronous Pipeline Parallelism via Basis Rotation
标题：通过基础旋转缓解非同步管道并行主义中的停滞
链接：https://arxiv.org/abs/2602.03515

作者：Hyunji Jung,Sungbin Shin,Namhoon Lee
备注：Preprint. Under review

【18】Least but not Last: Fine-tuning Intermediate Principal Components for Better Performance-Forgetting Trade-Offs
标题：最不重要但不是最后：微调中间主成分以获得更好的性能忘记权衡
链接：https://arxiv.org/abs/2602.03493

作者：Alessio Quercia,Arya Bangun,Ira Assent,Hanno Scharr

【19】Beyond Variance: Prompt-Efficient RLVR via Rare-Event Amplification and Bidirectional Pairing
标题：超越方差：通过罕见事件放大和双向配对实现预算高效的WLVR
链接：https://arxiv.org/abs/2602.03452

作者：Xin Sheng,Jiaxin Li,Yujuan Pang,Ran Peng,Yong Ma

【20】DiscoverLLM: From Executing Intents to Discovering Them
标题：发现LLM：从执行意图到发现意图
链接：https://arxiv.org/abs/2602.03429

作者：Tae Soo Kim,Yoonjoo Lee,Jaesang Yu,John Joon Young Chung,Juho Kim

【21】CoCoEmo: Composable and Controllable Human-Like Emotional TTS via Activation Steering
标题：CoCoEmo：通过激活转向的可组合和可控的类人情感TTS
链接：https://arxiv.org/abs/2602.03420

作者：Siyi Wang,Shihong Tan,Siyi Liu,Hong Jia,Gongping Huang,James Bailey,Ting Dang

【22】Chain-of-Goals Hierarchical Policy for Long-Horizon Offline Goal-Conditioned RL
标题：长期离线目标条件RL的目标链分层政策
链接：https://arxiv.org/abs/2602.03389

作者：Jinwoo Choi,Sang-Hyun Lee,Seung-Woo Seo
备注：22 pages

【23】Robustness as an Emergent Property of Task Performance
标题：稳健性作为任务绩效的紧急属性
链接：https://arxiv.org/abs/2602.03344

作者：Shir Ashury-Tahan,Ariel Gera,Elron Bandel,Michal Shmueli-Scheuer,Leshem Choshen

【24】Universal Approximation of Continuous Functionals on Compact Subsets via Linear Measurements and Scalar Nonlinearities
标题：通过线性测量和纯量非线性对紧子集上连续函式的普适逼近
链接：https://arxiv.org/abs/2602.03290

作者：Andrey Krylov,Maksim Penkin
备注：10 pages

【25】BlockRR: A Unified Framework of RR-type Algorithms for Label Differential Privacy
标题：BlockRR：标签差异隐私RR型算法的统一框架
链接：https://arxiv.org/abs/2602.03277

作者：Haixia Liu,Yi Ding
备注：19 pages, 2 figures

【26】TAME: A Trustworthy Test-Time Evolution of Agent Memory with Systematic Benchmarking
标题：TAME：具有系统基准的值得信赖的代理内存测试时进化
链接：https://arxiv.org/abs/2602.03224

作者：Yu Cheng,Jiuan Zhou,Yongkang Hu,Yihang Chen,Huichi Zhou,Mingang Chen,Zhizhong Zhang,Kun Shao,Yuan Xie,Zhaoxia Yin

【27】Probe-then-Commit Multi-Objective Bandits: Theoretical Benefits of Limited Multi-Arm Feedback
标题：先探测后投入多目标盗贼：有限多臂反馈的理论好处
链接：https://arxiv.org/abs/2602.03175

作者：Ming Shi

【28】TextME: Bridging Unseen Modalities Through Text Descriptions
标题：文本ME：通过文本描述弥合不可见的模式
链接：https://arxiv.org/abs/2602.03098

作者：Soyeon Hong,Jinchan Kim,Jaegook You,Seungtaek Choi,Suha Kwak,Hyunsouk Cho

【29】Geometry-Preserving Neural Architectures on Manifolds with Boundary
标题：有边界的多边形上保持几何形状的神经结构
链接：https://arxiv.org/abs/2602.03082

作者：Karthik Elamvazhuthi,Shiba Biswal,Kian Rosenblum,Arushi Katyal,Tianli Qu,Grady Ma,Rishi Sonthalia

【30】LatentMem: Customizing Latent Memory for Multi-Agent Systems
标题：LatentMem：为多代理系统定制潜在内存
链接：https://arxiv.org/abs/2602.03036

作者：Muxin Fu,Guibin Zhang,Xiangyuan Xue,Yafu Li,Zefeng He,Siyuan Huang,Xiaoye Qu,Yu Cheng,Yang Yang

【31】VOILA: Value-of-Information Guided Fidelity Selection for Cost-Aware Multimodal Question Answering
标题：VOILA：信息价值引导的富达选择，用于成本意识的多模式问题解答
链接：https://arxiv.org/abs/2602.03007

作者：Rahul Atul Bhope,K. R. Jayaram,Vinod Muthusamy,Ritesh Kumar,Vatche Isahagian,Nalini Venkatasubramanian

【32】DeltaEvolve: Accelerating Scientific Discovery through Momentum-Driven Evolution
标题：Delta Evolve：通过动量驱动的进化加速科学发现
链接：https://arxiv.org/abs/2602.02919

作者：Jiachen Jiang,Tianyu Ding,Zhihui Zhu

【33】Controlled disagreement improves generalization in decentralized training
标题：受控分歧提高了分散训练中的概括性
链接：https://arxiv.org/abs/2602.02899

作者：Zesen Wang,Mikael Johansson

【34】Mixture of Concept Bottleneck Experts
标题：概念瓶颈专家的混合
链接：https://arxiv.org/abs/2602.02886

作者：Francesco De Santis,Gabriele Ciravegna,Giovanni De Felice,Arianna Casanova,Francesco Giannini,Michelangelo Diligenti,Mateo Espinosa Zarlenga,Pietro Barbiero,Johannes Schneider,Danilo Giordano

【35】A Geometry-Aware Efficient Algorithm for Compositional Entropic Risk Minimization
标题：一种几何感知的组合熵风险最小化算法
链接：https://arxiv.org/abs/2602.02877

作者：Xiyuan Wei,Linli Zhou,Bokun Wang,Chih-Jen Lin,Tianbao Yang
备注：36 pages, 7 figures

【36】IMAGINE: Intelligent Multi-Agent Godot-based Indoor Networked Exploration
标题：想象：基于Godot的智能多智能体室内网络探索
链接：https://arxiv.org/abs/2602.02858

作者：Tiago Leite,Maria Conceição,António Grilo
备注：12 pages, submitted to a journal

【37】Beyond Content: Behavioral Policies Reveal Actors in Information Operations
标题：超越内容：行为政策揭示信息运营中的行为者
链接：https://arxiv.org/abs/2602.02838

作者：Philipp J. Schneider,Lanqin Yuan,Marian-Andrei Rizoiu

【38】TopoPrune: Robust Data Pruning via Unified Latent Space Topology
标题：TopoPrune：通过统一潜在空间布局进行稳健的数据修剪
链接：https://arxiv.org/abs/2602.02739

作者：Arjun Roy,Prajna G. Malettira,Manish Nagaraj,Kaushik Roy
备注：Preprint. Under Review

【39】Automated Dysphagia Screening Using Noninvasive Neck Acoustic Sensing
标题：使用无创颈部声学传感的自动吞咽困难筛查
链接：https://arxiv.org/abs/2602.02725

作者：Jade Chng,Rong Xing,Yunfei Luo,Kristen Linnemeyer-Risser,Tauhidur Rahman,Andrew Yousef,Philip A Weissbrod
备注：Accepted to 2026 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2026)

【40】Neural Probabilistic Amplitude Shaping for Nonlinear Fiber Channels
标题：非线性光纤通道的神经概率幅度整形
链接：https://arxiv.org/abs/2602.02716

作者：Mohammad Taha Askari,Lutz Lampe,Amirhossein Ghazisaeidi
备注：3 pages, 2 figures, Submitted to Optical Fiber Communication Conference (OFC) 2026

【41】MARA: Continuous SE(3)-Equivariant Attention for Molecular Force Fields
标题：MARA：分子力场的连续SE（3）-等变注意力
链接：https://arxiv.org/abs/2602.02671

作者：Francesco Leonardi,Boris Bonev,Kaspar Riesen

【42】hSNMF: Hybrid Spatially Regularized NMF for Image-Derived Spatial Transcriptomics
标题：hSNMF：用于图像衍生空间转录组学的混合空间正规化NMF
链接：https://arxiv.org/abs/2602.02638

作者：Md Ishtyaq Mahmud,Veena Kochat,Suresh Satpati,Jagan Mohan Reddy Dwarampudi,Humaira Anzum,Kunal Rai,Tania Banerjee
备注：The paper is accepted to the 2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI); 5 pages, 1 figure

【43】daVinci-Agency: Unlocking Long-Horizon Agency Data-Efficiently
标题：daVinci-Agency：高效解锁长视野机构数据
链接：https://arxiv.org/abs/2602.02619

作者：Mohan Jiang,Dayuan Fu,Junhao Shi,Ji Zeng,Weiye Si,Keyu Li,Xuefeng Li,Yang Xiao,Wenjie Li,Dequan Wang,Pengfei Liu

【44】Discovering Data Manifold Geometry via Non-Contracting Flows
标题：通过非承包流发现数据管汇几何
链接：https://arxiv.org/abs/2602.02611

作者：David Vigouroux,Lucas Drumetz,Ronan Fablet,François Rousseau

【45】RAP: KV-Cache Compression via RoPE-Aligned Pruning
标题：RAP：通过RoPE对齐修剪的KV-缓存压缩
链接：https://arxiv.org/abs/2602.02599

作者：Jihao Xin,Tian Lvu,Hatem Ltaief,David Keyes,Marco Canini

【46】Effective Frontiers: A Unification of Neural Scaling Laws
标题：有效边界：神经缩放定律的统一
链接：https://arxiv.org/abs/2602.02593

作者：Jiaxuan Zou,Zixuan Gong,Ye Su,Huayi Tang,Yong Liu

【47】Mitigating Task-Order Sensitivity and Forgetting via Hierarchical Second-Order Consolidation
标题：通过分层二阶合并缓解任务顺序敏感性和遗忘
链接：https://arxiv.org/abs/2602.02568

作者：Protik Nag,Krishnan Raghavan,Vignesh Narayanan
备注：21 pages, 8 figures

【48】High Rank Matrix Completion via Grassmannian Proxy Fusion
标题：通过格拉斯曼代理融合实现高等级矩阵
链接：https://arxiv.org/abs/2602.02565

作者：Huanran Li,Jeremy Johnson,Daniel Pimentel-Alarcón

【49】Label Curation Using Agentic AI
标题：使用抽象人工智能进行标签护理
链接：https://arxiv.org/abs/2602.02564

作者：Subhodeep Ghosh,Bayan Divaaniaazar,Md Ishat-E-Rabban,Spencer Clarke,Senjuti Basu Roy

【50】Experience-Driven Multi-Agent Systems Are Training-free Context-aware Earth Observers
标题：体验驱动的多智能体系统是无需训练的上下文感知地球观察员
链接：https://arxiv.org/abs/2602.02559

作者：Pengyu Dai,Weihao Xuan,Junjue Wang,Hongruixuan Chen,Jian Song,Yafei Ou,Naoto Yokoya
备注：21 pages, 6 figures

【51】ToolTok: Tool Tokenization for Efficient and Generalizable GUI Agents
标题：Tools Tok：高效且可通用的图形用户界面代理的工具代币化
链接：https://arxiv.org/abs/2602.02548

作者：Xiaoce Wang,Guibin Zhang,Junzhe Li,Jinzhe Tu,Chun Li,Ming Li
备注：8 pages main paper, 18 pages total, 8 figures, 5 tables, code at https://github.com/ZephinueCode/ToolTok

【52】Enhancing Post-Training Quantization via Future Activation Awareness
标题：通过未来激活意识增强训练后量化
链接：https://arxiv.org/abs/2602.02538

作者：Zheqi Lv,Zhenxuan Fan,Qi Tian,Wenqiao Zhang,Yueting Zhuang

【53】The "Robert Boulton" Singularity: Semantic Tunneling and Manifold Unfolding in Recursive AI
链接：https://arxiv.org/abs/2602.02526

作者：Pengyue Hou
备注：Companion paper to arXiv:2601.11594. Provides empirical validation of the MNCIS framework in Large Language Models (GPT-2) using a recursive training protocol (N=1500). Includes complete, reproducible Python implementation of Adaptive Spectral Negative Coupling (ASNC) and Effective Rank metrics in the Appendix

【54】Scaled Dot-Product Attention implements projection of inputs onto a common surface
标题：缩放点产品注意力实现将输入投影到公共表面上
链接：https://arxiv.org/abs/2602.02521

作者：Terence D Sanger

【55】Measuring Individual User Fairness with User Similarity and Effectiveness Disparity
标题：通过用户相似性和有效性差异衡量个人用户公平性
链接：https://arxiv.org/abs/2602.02516

作者：Theresia Veronika Rampisela,Maria Maistro,Tuukka Ruotsalo,Christina Lioma
备注：Preprint of a work that has been accepted to ECIR 2026 Full Papers track as a Findings paper

【56】UNSO: Unified Newton Schulz Orthogonalization
标题：UNSO：统一牛顿·舒尔茨国家化
链接：https://arxiv.org/abs/2602.02500

作者：Chen Hu,Qianxi Zhao,Yuming Li,Mingyu Zhou,Xiyin Li

【57】Kimi K2.5: Visual Agentic Intelligence
标题：Kimi K2.5：视觉统计智能
链接：https://arxiv.org/abs/2602.02276

作者：Kimi Team,Tongtong Bai,Yifan Bai,Yiping Bao,S. H. Cai,Yuan Cao,Y. Charles,H. S. Che,Cheng Chen,Guanduo Chen,Huarong Chen,Jia Chen,Jiahao Chen,Jianlong Chen,Jun Chen,Kefan Chen,Liang Chen,Ruijue Chen,Xinhao Chen,Yanru Chen,Yanxu Chen,Yicun Chen,Yimin Chen,Yingjiang Chen,Yuankun Chen,Yujie Chen,Yutian Chen,Zhirong Chen,Ziwei Chen,Dazhi Cheng,Minghan Chu,Jialei Cui,Jiaqi Deng,Muxi Diao,Hao Ding,Mengfan Dong,Mengnan Dong,Yuxin Dong,Yuhao Dong,Angang Du,Chenzhuang Du,Dikang Du,Lingxiao Du,Yulun Du,Yu Fan,Shengjun Fang,Qiulin Feng,Yichen Feng,Garimugai Fu,Kelin Fu,Hongcheng Gao,Tong Gao,Yuyao Ge,Shangyi Geng,Chengyang Gong,Xiaochen Gong,Zhuoma Gongque,Qizheng Gu,Xinran Gu,Yicheng Gu,Longyu Guan,Yuanying Guo,Xiaoru Hao,Weiran He,Wenyang He,Yunjia He,Chao Hong,Hao Hu,Jiaxi Hu,Yangyang Hu,Zhenxing Hu,Ke Huang,Ruiyuan Huang,Weixiao Huang,Zhiqi Huang,Tao Jiang,Zhejun Jiang,Xinyi Jin,Yu Jing,Guokun Lai,Aidi Li,C. Li,Cheng Li,Fang Li,Guanghe Li,Guanyu Li,Haitao Li,Haoyang Li,Jia Li,Jingwei Li,Junxiong Li,Lincan Li,Mo Li,Weihong Li,Wentao Li,Xinhang Li,Xinhao Li,Yang Li,Yanhao Li,Yiwei Li,Yuxiao Li,Zhaowei Li,Zheming Li,Weilong Liao,Jiawei Lin,Xiaohan Lin,Zhishan Lin,Zichao Lin,Cheng Liu,Chenyu Liu,Hongzhang Liu,Liang Liu,Shaowei Liu,Shudong Liu,Shuran Liu,Tianwei Liu,Tianyu Liu,Weizhou Liu,Xiangyan Liu,Yangyang Liu,Yanming Liu,Yibo Liu,Yuanxin Liu,Yue Liu,Zhengying Liu,Zhongnuo Liu,Enzhe Lu,Haoyu Lu,Zhiyuan Lu,Junyu Luo,Tongxu Luo,Yashuo Luo,Long Ma,Yingwei Ma,Shaoguang Mao,Yuan Mei,Xin Men,Fanqing Meng,Zhiyong Meng,Yibo Miao,Minqing Ni,Kun Ouyang,Siyuan Pan,Bo Pang,Yuchao Qian,Ruoyu Qin,Zeyu Qin,Jiezhong Qiu,Bowen Qu,Zeyu Shang,Youbo Shao,Tianxiao Shen,Zhennan Shen,Juanfeng Shi,Lidong Shi,Shengyuan Shi,Feifan Song,Pengwei Song,Tianhui Song,Xiaoxi Song,Hongjin Su,Jianlin Su,Zhaochen Su,Lin Sui,Jinsong Sun,Junyao Sun,Tongyu Sun,Flood Sung,Yunpeng Tai,Chuning Tang,Heyi Tang,Xiaojuan Tang,Zhengyang Tang,Jiawen Tao,Shiyuan Teng,Chaoran Tian,Pengfei Tian,Ao Wang,Bowen Wang,Chensi Wang,Chuang Wang,Congcong Wang,Dingkun Wang,Dinglu Wang,Dongliang Wang,Feng Wang,Hailong Wang,Haiming Wang,Hengzhi Wang,Huaqing Wang,Hui Wang,Jiahao Wang,Jinhong Wang,Jiuzheng Wang,Kaixin Wang,Linian Wang,Qibin Wang,Shengjie Wang,Shuyi Wang,Si Wang,Wei Wang,Xiaochen Wang,Xinyuan Wang,Yao Wang,Yejie Wang,Yipu Wang,Yiqin Wang,Yucheng Wang,Yuzhi Wang,Zhaoji Wang,Zhaowei Wang,Zhengtao Wang,Zhexu Wang,Zihan Wang,Zizhe Wang,Chu Wei,Ming Wei,Chuan Wen,Zichen Wen,Chengjie Wu,Haoning Wu,Junyan Wu,Rucong Wu,Wenhao Wu,Yuefeng Wu,Yuhao Wu,Yuxin Wu,Zijian Wu,Chenjun Xiao,Jin Xie,Xiaotong Xie,Yuchong Xie,Yifei Xin,Bowei Xing,Boyu Xu,Jianfan Xu,Jing Xu,Jinjing Xu,L. H. Xu,Lin Xu,Suting Xu,Weixin Xu,Xinbo Xu,Xinran Xu,Yangchuan Xu,Yichang Xu,Yuemeng Xu,Zelai Xu,Ziyao Xu,Junjie Yan,Yuzi Yan,Guangyao Yang,Hao Yang,Junwei Yang,Kai Yang,Ningyuan Yang,Ruihan Yang,Xiaofei Yang,Xinlong Yang,Ying Yang,Yi Yang,Yi Yang,Zhen Yang,Zhilin Yang,Zonghan Yang,Haotian Yao,Dan Ye,Wenjie Ye,Zhuorui Ye,Bohong Yin,Chengzhen Yu,Longhui Yu,Tao Yu,Tianxiang Yu,Enming Yuan,Mengjie Yuan,Xiaokun Yuan,Yang Yue,Weihao Zeng,Dunyuan Zha,Haobing Zhan,Dehao Zhang,Hao Zhang,Jin Zhang,Puqi Zhang,Qiao Zhang,Rui Zhang,Xiaobin Zhang,Y. Zhang,Yadong Zhang,Yangkun Zhang,Yichi Zhang,Yizhi Zhang,Yongting Zhang,Yu Zhang,Yushun Zhang,Yutao Zhang,Yutong Zhang,Zheng Zhang,Chenguang Zhao,Feifan Zhao,Jinxiang Zhao,Shuai Zhao,Xiangyu Zhao,Yikai Zhao,Zijia Zhao,Huabin Zheng,Ruihan Zheng,Shaojie Zheng,Tengyang Zheng,Junfeng Zhong,Longguang Zhong,Weiming Zhong,M. Zhou,Runjie Zhou,Xinyu Zhou,Zaida Zhou,Jinguo Zhu,Liya Zhu,Xinhao Zhu,Yuxuan Zhu,Zhen Zhu,Jingze Zhuang,Weiyu Zhuang,Ying Zou,Xinxing Zu
备注：Kimi K2.5 tech report

【58】Fast Sampling for Flows and Diffusions with Lazy and Point Mass Stochastic Interpolants
标题：利用懒惰和点质量随机插值对流动和扩散进行快速采样
链接：https://arxiv.org/abs/2602.03789

作者：Gabriel Damsholt,Jes Frellsen,Susanne Ditlevsen

【59】Conditional Flow Matching for Visually-Guided Acoustic Highlighting
标题：用于视觉引导声学突出显示的条件流匹配
链接：https://arxiv.org/abs/2602.03762

作者：Hugo Malard,Gael Le Lan,Daniel Wong,David Lou Alon,Yi-Chiao Wu,Sanjeel Parekh

【60】Improving the Linearized Laplace Approximation via Quadratic Approximations
标题：通过二次逼近改进线性化拉普拉斯逼近
链接：https://arxiv.org/abs/2602.03394

作者：Pedro Jiménez,Luis A. Ortega,Pablo Morales-Álvarez,Daniel Hernández-Lobato
备注：6 pages, 1 table. Accepted at European Symposium on Artificial Neural Networks (ESANN 2026) as poster presentation

【61】A Novel approach to portfolio construction
标题：投资组合构建的新颖方法
链接：https://arxiv.org/abs/2602.03325

作者：T. Di Matteo,L. Riso,M. G. Zoia

【62】Principled Federated Random Forests for Heterogeneous Data
标题：面向异类数据的原则性联邦随机森林
链接：https://arxiv.org/abs/2602.03258

作者：Rémi Khellaf,Erwan Scornet,Aurélien Bellet,Julie Josse

【63】Near-Universal Multiplicative Updates for Nonnegative Einsum Factorization
标题：非负Einsum因式分解的近通用乘性更新
链接：https://arxiv.org/abs/2602.02759

作者：John Hood,Aaron Schein
备注：26 pages, 5 figures

【64】Relaxed Triangle Inequality for Kullback-Leibler Divergence Between Multivariate Gaussian Distributions
标题：多元高斯分布Kullback-Leibler分歧的松弛三角不等式
链接：https://arxiv.org/abs/2602.02577

作者：Shiji Xiao,Yufeng Zhang,Chubo Liu,Yan Ding,Keqin Li,Kenli Li

机器翻译由腾讯交互翻译提供，仅供参考

点击“阅读原文”获取带摘要的学术速递