|
创作新主题 |
社区所有版块导航
DATA
| docker Elasticsearch |
WEB开发
| linux MongoDB Redis DATABASE NGINX 其他Web框架 web工具 zookeeper tornado NoSql Bootstrap js peewee Git bottle IE MQ Jquery |
机器学习
| 机器学习算法 |
产品
| 短视频 |
印度
| 印度 |
一周十大热门主题
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377篇
大模型相关(62篇)
【1】Mapping Post-Training Forgetting in Language Models at Scale
标题：大规模绘制语言模型中训练后遗忘的地图
链接：https://arxiv.org/abs/2510.17776
备注：43 pages,15 figures
摘要：Scaled post-training now drives many of the largest capability gains in language models (LMs), yet its effect on pretrained knowledge remains poorly understood. Not all forgetting is equal: Forgetting one fact (e.g., a U.S. president or an API call) does not "average out" by recalling another. Hence, we propose a sample-wise paradigm to measure what is forgotten and when backward transfer occurs. Our metric counts 1->0 transitions (correct before post-training, incorrect after) to quantify forgetting and 0->1 transitions to quantify backward transfer. Traditional task averages conflate these effects and obscure large changes. For multiple-choice benchmarks, we add chance-adjusted variants that subtract the expected contribution of random guessing from pre- and post-training accuracies. We apply this framework across post-training stages, model sizes, and data scales. Our large-scale analysis shows that: (1) Domain-continual pretraining induces moderate forgetting with low-to-moderate backward transfer; (2) RL/SFT post-training applied to base models and Instruction tuning yields moderate-to-large backward transfer on math and logic with overall low-to-moderate forgetting; (3) Applying RL/SFT to instruction-tuned models is sensitive on data scale: at small scales, both forgetting and backward transfer are small; at larger scales, effects are mixed and warrant further study with better controls; (4) Model merging does not reliably mitigate forgetting. Overall, our framework offers a practical yardstick for mapping how post-training alters pretrained knowledge at scale -- enabling progress towards generally capable AI systems.
【2】VERA-V: Variational Inference Framework for Jailbreaking Vision-Language Models
标题：VERA-V：越狱视觉语言模型的变分推理框架
链接：https://arxiv.org/abs/2510.17759
备注：18 pages, 7 Figures,
摘要：Vision-Language Models (VLMs) extend large language models with visual reasoning, but their multimodal design also introduces new, underexplored vulnerabilities. Existing multimodal red-teaming methods largely rely on brittle templates, focus on single-attack settings, and expose only a narrow subset of vulnerabilities. To address these limitations, we introduce VERA-V, a variational inference framework that recasts multimodal jailbreak discovery as learning a joint posterior distribution over paired text-image prompts. This probabilistic view enables the generation of stealthy, coupled adversarial inputs that bypass model guardrails. We train a lightweight attacker to approximate the posterior, allowing efficient sampling of diverse jailbreaks and providing distributional insights into vulnerabilities. VERA-V further integrates three complementary strategies: (i) typography-based text prompts that embed harmful cues, (ii) diffusion-based image synthesis that introduces adversarial signals, and (iii) structured distractors to fragment VLM attention. Experiments on HarmBench and HADES benchmarks show that VERA-V consistently outperforms state-of-the-art baselines on both open-source and frontier VLMs, achieving up to 53.75% higher attack success rate (ASR) over the best baseline on GPT-4o.
【3】Enabling Fine-Grained Operating Points for Black-Box LLMs
标题：为黑匣子LLM启用细粒度操作点
链接：https://arxiv.org/abs/2510.17727
备注：35 pages
摘要：Black-box Large Language Models (LLMs) provide practical and accessible alternatives to other machine learning methods, as they require minimal labeled data and machine learning expertise to develop solutions for various decision making problems. However, for applications that need operating with constraints on specific metrics (e.g., precision $\geq$ 95%), decision making with black-box LLMs remains unfavorable, due to their low numerical output cardinalities. This results in limited control over their operating points, preventing fine-grained adjustment of their decision making behavior. In this paper, we study using black-box LLMs as classifiers, focusing on efficiently improving their operational granularity without performance loss. Specifically, we first investigate the reasons behind their low-cardinality numerical outputs and show that they are biased towards generating rounded but informative verbalized probabilities. Then, we experiment with standard prompt engineering, uncertainty estimation and confidence elicitation techniques, and observe that they do not effectively improve operational granularity without sacrificing performance or increasing inference cost. Finally, we propose efficient approaches to significantly increase the number and diversity of available operating points. Our proposed approaches provide finer-grained operating points and achieve comparable to or better performance than the benchmark methods across 11 datasets and 3 LLMs.
【4】AcademicEval: Live Long-Context LLM Benchmark
标题：AcademicEval：实时长上下文LLM基准
链接：https://arxiv.org/abs/2510.17725
备注：Accepted by TMLR. Code is available at this https URL
摘要 ：Large Language Models (LLMs) have recently achieved remarkable performance in long-context understanding. However, current long-context LLM benchmarks are limited by rigid context length, labor-intensive annotation, and the pressing challenge of label leakage issues during LLM training. Therefore, we propose \textsc{AcademicEval}, a live benchmark for evaluating LLMs over long-context generation tasks. \textsc{AcademicEval} adopts papers on arXiv to introduce several academic writing tasks with long-context inputs, \textit{i.e.}, \textsc{Title}, \textsc{Abstract}, \textsc{Introduction}, and \textsc{Related Work}, which cover a wide range of abstraction levels and require no manual labeling. Moreover, \textsc{AcademicEval} integrates high-quality and expert-curated few-shot demonstrations from a collected co-author graph to enable flexible context length. Especially, \textsc{AcademicEval} features an efficient live evaluation, ensuring no label leakage. We conduct a holistic evaluation on \textsc{AcademicEval}, and the results illustrate that LLMs perform poorly on tasks with hierarchical abstraction levels and tend to struggle with long few-shot demonstrations, highlighting the challenge of our benchmark. Through experimental analysis, we also reveal some insights for enhancing LLMs' long-context modeling capabilities. Code is available at https://github.com/ulab-uiuc/AcademicEval
【5】LLM-as-a-Prophet: Understanding Predictive Intelligence with Prophet Arena
标题：先知法学硕士：通过先知竞技场了解预测智能
链接：https://arxiv.org/abs/2510.17638
备注：his https URL
摘要：Forecasting is not only a fundamental intellectual pursuit but also is of significant importance to societal systems such as finance and economics. With the rapid advances of large language models (LLMs) trained on Internet-scale data, it raises the promise of employing LLMs to forecast real-world future events, an emerging paradigm we call "LLM-as-a-Prophet". This paper systematically investigates such predictive intelligence of LLMs. To this end, we build Prophet Arena, a general evaluation benchmark that continuously collects live forecasting tasks and decomposes each task into distinct pipeline stages, in order to support our controlled and large-scale experimentation. Our comprehensive evaluation reveals that many LLMs already exhibit impressive forecasting capabilities, reflected in, e.g., their small calibration errors, consistent prediction confidence and promising market returns. However, we also uncover key bottlenecks towards achieving superior predictive intelligence via LLM-as-a-Prophet, such as LLMs' inaccurate event recalls, misunderstanding of data sources and slower information aggregation compared to markets when resolution nears.
【6】HGAdapter: Hypergraph-based Adapters in Language Models for Code Summarization and Clone Detection
标题：HGAdaptor：用于代码摘要和克隆检测的语言模型中基于Hypergraph的适配器
链接：https://arxiv.org/abs/2510.17591
备注：Accepted by the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025) as a findings long paper
摘要：Pre-trained language models (PLMs) are increasingly being applied to code-related tasks. Although PLMs have achieved good results, they do not take into account potential high-order data correlations within the code. We propose three types of high-order correlations in code tokens, i.e. abstract syntax tree family correlation, lexical correlation, and line correlation. We design a tokens and hyperedges generator to capture these high-order data correlations. We improve the architecture of hypergraph neural networks and combine it with adapter tuning to propose a novel hypergraph-based adapter (HGAdapter) to fine-tune PLMs. HGAdapter can encode high-order data correlations and is allowed to be inserted into various PLMs to enhance performance. Experiments were conducted on several public datasets, including six languages of code summarization and code clone detection tasks. Our methods improved the performance of PLMs in datasets to varying degrees. Experimental results validate the introduction of high-order data correlations that contribute to improved effectiveness.
【7】OncoReason: Structuring Clinical Reasoning in LLMs for Robust and Interpretable Survival Prediction
标题：OncoReason：在LLM中构建临床推理，以实现稳健且可解释的生存预测
链接：https://arxiv.org/abs/2510.17532
摘要：Predicting cancer treatment outcomes requires models that are both accurate and interpretable, particularly in the presence of heterogeneous clinical data. While large language models (LLMs) have shown strong performance in biomedical NLP, they often lack structured reasoning capabilities critical for high-stakes decision support. We present a unified, multi-task learning framework that aligns autoregressive LLMs with clinical reasoning for outcome prediction on the MSK-CHORD dataset. Our models are trained to jointly perform binary survival classification, continuous survival time regression, and natural language rationale generation. We evaluate three alignment strategies: (1) standard supervised fine-tuning (SFT), (2) SFT with Chain-of-Thought (CoT) prompting to elicit step-by-step reasoning, and (3) Group Relative Policy Optimization (GRPO), a reinforcement learning method that aligns model outputs to expert-derived reasoning trajectories. Experiments with LLaMa3-8B and Med42-8B backbones demonstrate that CoT prompting improves F1 by +6.0 and reduces MAE by 12%, while GRPO achieves state-of-the-art interpretability and predictive performance across BLEU, ROUGE, and BERTScore. We further show that existing biomedical LLMs often fail to produce valid reasoning traces due to architectural constraints. Our findings underscore the importance of reasoning-aware alignment in multi-task clinical modeling and set a new benchmark for interpretable, trustworthy LLMs in precision oncology.
【8】SimBench: Benchmarking the Ability of Large Language Models to Simulate Human Behaviors
标题：SimBench：对大型语言模型模拟人类行为的能力进行基准测试
链接：https://arxiv.org/abs/2510.17516
备注：Project Website: this http URL Data: this https URL
摘要 ：Large language model (LLM) simulations of human behavior have the potential to revolutionize the social and behavioral sciences, if and only if they faithfully reflect real human behaviors. Current evaluations are fragmented, based on bespoke tasks and metrics, creating a patchwork of incomparable results. To address this, we introduce SimBench, the first large-scale, standardized benchmark for a robust, reproducible science of LLM simulation. By unifying 20 diverse datasets covering tasks from moral decision-making to economic choice across a large global participant pool, SimBench provides the necessary foundation to ask fundamental questions about when, how, and why LLM simulations succeed or fail. We show that, while even the best LLMs today have limited simulation ability (score: 40.80/100), performance scales log-linearly with model size. Simulation performance is not improved by increased inference-time compute. We demonstrate an alignment-simulation trade-off: instruction-tuning improves performance on low-entropy (consensus) questions but degrades it on high-entropy (diverse) ones. Models particularly struggle when simulating specific demographic groups. Finally, we demonstrate that simulation ability correlates most strongly with deep, knowledge-intensive reasoning (MMLU-Pro, r=0.939). By making progress measurable, we aim to accelerate the development of more faithful LLM simulators.
【9】I-RAVEN-X: Benchmarking Generalization and Robustness of Analogical and Mathematical Reasoning in Large Language and Reasoning Models
标题：I-RAVEN-X：大型语言和推理模型中类比和数学推理的基准泛化和鲁棒性
链接：https://arxiv.org/abs/2510.17496
备注：Accepted at the 5th Workshop on Mathematical Reasoning and AI (MATH-AI), NeurIPS 2025
摘要：We introduce I-RAVEN-X, a symbolic benchmark designed to evaluate generalization and robustness in analogical and mathematical reasoning for Large Language Models (LLMs) and Large Reasoning Models (LRMs). I-RAVEN-X extends I-RAVEN by increasing operand complexity, attribute range, and introducing perceptual uncertainty. Compared to LLMs, empirical results show that LRMs achieve improved productivity and systematicity on longer reasoning relations and wider attribute ranges, respectively. However, LRMs are still significantly challenged by reasoning under uncertainty and cannot effectively explore multiple probabilistic outcomes.
【10】From Spatial to Actions: Grounding Vision-Language-Action Model in Spatial Foundation Priors
标题：从空间到行动：空间基础优先事项中的视觉-语言-行动模型的基础
链接：https://arxiv.org/abs/2510.17439
备注：Project page: this https URL
摘要：Existing vision-language-action (VLA) models act in 3D real-world but are typically built on 2D encoders, leaving a spatial reasoning gap that limits generalization and adaptability. Recent 3D integration techniques for VLAs either require specialized sensors and transfer poorly across modalities, or inject weak cues that lack geometry and degrade vision-language alignment. In this work, we introduce FALCON (From Spatial to Action), a novel paradigm that injects rich 3D spatial tokens into the action head. FALCON leverages spatial foundation models to deliver strong geometric priors from RGB alone, and includes an Embodied Spatial Model that can optionally fuse depth, or pose for higher fidelity when available, without retraining or architectural changes. To preserve language reasoning, spatial tokens are consumed by a Spatial-Enhanced Action Head rather than being concatenated into the vision-language backbone. These designs enable FALCON to address limitations in spatial representation, modality transferability, and alignment. In comprehensive evaluations across three simulation benchmarks and eleven real-world tasks, our proposed FALCON achieves state-of-the-art performance, consistently surpasses competitive baselines, and remains robust under clutter, spatial-prompt conditioning, and variations in object scale and height.
【11】Leveraging Group Relative Policy Optimization to Advance Large Language Models in Traditional Chinese Medicine
标题：利用集团相对政策优化推进中医大语言模型
链接：https://arxiv.org/abs/2510.17402
摘要：Traditional Chinese Medicine (TCM) presents a rich and structurally unique knowledge system that challenges conventional applications of large language models (LLMs). Although previous TCM-specific LLMs have shown progress through supervised fine-tuning, they often face limitations in alignment, data quality, and evaluation consistency. In this study, we introduce Ladder-base, the first TCM-focused LLM trained with Group Relative Policy Optimization (GRPO), a reinforcement learning method that improves reasoning and factual consistency by optimizing response selection based on intra-group comparisons. Ladder-base is built upon the Qwen2.5-7B-Instruct foundation model and trained exclusively on the textual subset of the TCM-Ladder benchmark, using 80 percent of the data for training and the remaining 20 percent split evenly between validation and test sets. Through standardized evaluation, Ladder-base demonstrates superior performance across multiple reasoning metrics when compared to both state-of-the-art general-purpose LLMs such as GPT-4, Gemini 2.5, Claude 3, and Qwen3 and domain-specific TCM models including BenTsao, HuatuoGPT2, and Zhongjing. These findings suggest that GRPO provides an effective and efficient strategy for aligning LLMs with expert-level reasoning in traditional medical domains and supports the development of trustworthy and clinically grounded TCM artificial intelligence systems.
【12】TabR1: Taming GRPO for tabular reasoning LLMs
标题：TabR1：驯服表格推理LLM的GRPO
链接：https://arxiv.org/abs/2510.17385
摘要 ：Tabular prediction has traditionally relied on gradient-boosted decision trees and specialized deep learning models, which excel within tasks but provide limited interpretability and weak transfer across tables. Reasoning large language models (LLMs) promise cross-task adaptability with trans- parent reasoning traces, yet their potential has not been fully realized for tabular data. This paper presents TabR1, the first reasoning LLM for tabular prediction with multi-step reasoning. At its core is Permutation Relative Policy Optimization (PRPO), a simple yet efficient reinforcement learning method that encodes column-permutation invariance as a structural prior. By construct- ing multiple label-preserving permutations per sample and estimating advantages both within and across permutations, PRPO transforms sparse rewards into dense learning signals and improves generalization. With limited supervision, PRPO activates the reasoning ability of LLMs for tabular prediction, enhancing few-shot and zero-shot performance as well as interpretability. Comprehensive experiments demonstrate that TabR1 achieves performance comparable to strong baselines under full-supervision fine-tuning. In the zero-shot setting, TabR1 approaches the performance of strong baselines under the 32-shot setting. Moreover, TabR1 (8B) substantially outperforms much larger LLMs across various tasks, achieving up to 53.17% improvement over DeepSeek-R1 (685B).
【13】Bridging Embodiment Gaps: Deploying Vision-Language-Action Models on Soft Robots
标题：弥合体现差距：在软机器人上部署视觉-语言-动作模型
链接：https://arxiv.org/abs/2510.17369
备注：Accepted by NeurIPS 2025 SpaVLE workshop. 4 pages, 2 figures(in main paper, excluding references and supplements)
摘要：Robotic systems are increasingly expected to operate in human-centered, unstructured environments where safety, adaptability, and generalization are essential. Vision-Language-Action (VLA) models have been proposed as a language guided generalized control framework for real robots. However, their deployment has been limited to conventional serial link manipulators. Coupled by their rigidity and unpredictability of learning based control, the ability to safely interact with the environment is missing yet critical. In this work, we present the deployment of a VLA model on a soft continuum manipulator to demonstrate autonomous safe human-robot interaction. We present a structured finetuning and deployment pipeline evaluating two state-of-the-art VLA models (OpenVLA-OFT and $\pi_0$) across representative manipulation tasks, and show while out-of-the-box policies fail due to embodiment mismatch, through targeted finetuning the soft robot performs equally to the rigid counterpart. Our findings highlight the necessity of finetuning for bridging embodiment gaps, and demonstrate that coupling VLA models with soft robots enables safe and flexible embodied AI in human-shared environments.
【14】Recurrent Attention-based Token Selection for Efficient Streaming Video-LLMs
标题：高效流媒体视频LLM的基于循环注意的令牌选择
链接：https://arxiv.org/abs/2510.17364
备注：NeurIPS 2025
摘要：Video Large Language Models (Video-LLMs) excel at understanding videos in-context, provided they have full access to the video when answering queries. However, these models face challenges in streaming scenarios where hour-long videos must be processed online, and questions need timely responses. In this work, we propose a training-free approach compatible with standard Video-LLMs, leveraging three key concepts: 1) LLM-informed selection of visual tokens to identify those that the LLM has attended to and contributed to its understanding of each short clip. Our attention-based selection allows us to discard up to ~95% of unimportant visual tokens with minimal performance loss; 2) Recurrent processing of past selected tokens to generate temporally coherent understanding of each processed clip; 3) Caption-based question answering for lightweight and accurate responses. Our method achieves state-of-the-art performance on streaming video benchmarks, striking a balance between efficiency and effectiveness.
【15】Localist LLMs with Recruitment Learning
标题：具有招聘学习的本地法学硕士
链接：https://arxiv.org/abs/2510.17358
摘要：We present a novel framework for training large language models with continuously adjustable internal representations that span the full spectrum from localist (interpretable, rule-based) to distributed (generalizable, efficient) encodings. The key innovations are (1) a locality dial, a tunable parameter that dynamically controls the degree of localization during both training and inference without requiring model retraining, (2) an information-theoretic recruitment mechanism that adaptively allocates semantic blocks as needed, eliminating the requirement for complete domain knowledge at initialization, and (3) a hierarchical recruitment framework that extends capacity allocation to entire specialized LLMs, enabling multi-granularity architectural adaptation. This is achieved through group sparsity penalties on attention mechanisms, information-theoretic anchor design, dynamic rule injection, and principled recruitment criteria based on penalized likelihood with explicit units. We provide rigorous mathematical results establishing explicit threshold conditions under which attention provably concentrates on semantically relevant blocks at stationary points, with exact bounds on attention entropy and pointer fidelity. The hierarchical recruitment mechanism provides convergence guarantees at both the block level (fine-grained, within-LLM) and the LLM level (coarse-grained, cross-domain), ensuring the system discovers semantic partitions that balance model complexity against data encoding efficiency. This framework enables practitioners to continuously interpolate between interpretable and high-performance modes while adapting architectural capacity at multiple granularities, supporting applications in regulated domains requiring both transparency and capability.
【16】MemoryBench: A Benchmark for Memory and Continual Learning in LLM Systems
标题：MemoryBench：LLM系统中记忆和持续学习的基准
链接：https://arxiv.org/abs/2510.17281
摘要 ：Scaling up data, parameters, and test-time computation has been the mainstream methods to improve LLM systems (LLMsys), but their upper bounds are almost reached due to the gradual depletion of high-quality data and marginal gains obtained from larger computational resource consumption. Inspired by the abilities of human and traditional AI systems in learning from practice, constructing memory and continual learning frameworks for LLMsys has become an important and popular research direction in recent literature. Yet, existing benchmarks for LLM memory often focus on evaluating the system on homogeneous reading comprehension tasks with long-form inputs rather than testing their abilities to learn from accumulated user feedback in service time. Therefore, we propose a user feedback simulation framework and a comprehensive benchmark covering multiple domains, languages, and types of tasks to evaluate the continual learning abilities of LLMsys. Experiments show that the effectiveness and efficiency of state-of-the-art baselines are far from satisfying, and we hope this benchmark could pave the way for future studies on LLM memory and optimization algorithms.
【17】Soft-Masked Diffusion Language Models
标题：软屏蔽扩散语言模型
链接：https://arxiv.org/abs/2510.17206
摘要：Diffusion models have demonstrated strong potential in language modeling, offering various advantages over traditional autoregressive approaches. Their ability to generate and revise entire responses in parallel enables faster generation and built-in self-correction mechanisms. Most modern diffusion-based language models employ masked diffusion, where decoding involves iteratively processing masked tokens based on a binary decision: either retaining the mask or replacing it with the predicted token. However, this binary choice discards valuable predictive information when the mask is retained. To address this limitation, we introduce soft-masking (SM), a novel method that dynamically blends the embedding of the mask token with the embeddings of the top-$k$ predicted tokens from the previous decoding step, for each retained mask. This provides the model with a more informative prior, preserving context from earlier computations and allowing partial information about masked tokens to propagate beyond a single step. We propose a training methodology that adapts a pretrained masked diffusion language model to incorporate SM. We demonstrate that continuing pretraining a 169M parameter model with SM leads to improved perplexity and MAUVE scores. Furthermore, we finetune two state-of-the-art diffusion models, Dream-7B and Dream-Coder-7B, with SM. SM consistently improves performance across multiple coding benchmarks, particularly in high-throughput settings.
【18】Do LLMs Recognize Your Latent Preferences? A Benchmark for Latent Information Discovery in Personalized Interaction
标题：LLM认识到您的潜在偏好吗？个性化交互中潜在信息发现的基准
链接：https://arxiv.org/abs/2510.17132
摘要：Large Language Models (LLMs) excel at producing broadly relevant text, but this generality becomes a limitation when user-specific preferences are required, such as recommending restaurants or planning travel. In these scenarios, users rarely articulate every preference explicitly; instead, much of what they care about remains latent, waiting to be inferred. This raises a fundamental question: Can LLMs uncover and reason about such latent information through conversation? We address this problem by introducing a unified benchmark for evaluating latent information discovery - the ability of LLMs to reveal and utilize hidden user attributes through multi-turn interaction. The benchmark spans three progressively realistic settings: the classic 20 Questions game, Personalized Question Answering, and Personalized Text Summarization. All tasks share a tri-agent framework (User, Assistant, Judge) enabling turn-level evaluation of elicitation and adaptation. Our results reveal that while LLMs can indeed surface latent information through dialogue, their success varies dramatically with context: from 32% to 98%, depending on task complexity, topic, and number of hidden attributes. This benchmark provides the first systematic framework for studying latent information discovery in personalized interaction, highlighting that effective preference inference remains an open frontier for building truly adaptive AI systems.
【19】Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Survey
标题：用于预定操纵的高效视觉-语言-动作模型：系统性调查
链接：https://arxiv.org/abs/2510.17111
摘要：Vision-Language-Action (VLA) models extend vision-language models to embodied control by mapping natural-language instructions and visual observations to robot actions. Despite their capabilities, VLA systems face significant challenges due to their massive computational and memory demands, which conflict with the constraints of edge platforms such as on-board mobile manipulators that require real-time performance. Addressing this tension has become a central focus of recent research. In light of the growing efforts toward more efficient and scalable VLA systems, this survey provides a systematic review of approaches for improving VLA efficiency, with an emphasis on reducing latency, memory footprint, and training and inference costs. We categorize existing solutions into four dimensions: model architecture, perception feature, action generation, and training/inference strategies, summarizing representative techniques within each category. Finally, we discuss future trends and open challenges, highlighting directions for advancing efficient embodied intelligence.
【20】The Ends Justify the Thoughts: RL-Induced Motivated Reasoning in LLMs
标题：目的证明思想的合理性：LLM中RL诱导的动机推理
链接：https://arxiv.org/abs/2510.17057
备注：26 pages
【21】Mapping from Meaning: Addressing the Miscalibration of Prompt-Sensitive Language Models
标题：意义映射：解决预算敏感语言模型的错误校准
链接：https://arxiv.org/abs/2510.17028
备注：None
【22】Forgetting to Forget: Attention Sink as A Gateway for Backdooring LLM Unlearning
标题
：忘记忘记：注意力下沉作为后备LLM遗忘的门户
链接：https://arxiv.org/abs/2510.17021
【23】Justitia: Fair and Efficient Scheduling for LLM Applications
标题：Justitia：LLM应用程序公平有效的调度
链接：https://arxiv.org/abs/2510.17015
【24】EEschematic: Multimodal-LLM Based AI Agent for Schematic Generation of Analog Circuit
标题：EEschemic：基于多模块LLM的人工智能代理，用于模拟电路原理图生成
链接：https://arxiv.org/abs/2510.17002
【25】Bits Leaked per Query: Information-Theoretic Bounds on Adversarial Attacks against LLMs
标题：每次查询泄露的位：针对LLM的对抗性攻击的信息理论界限
链接：https://arxiv.org/abs/2510.17000
备注：NeurIPS 2025 (spotlight)
【26】Parameter-Efficient Fine-Tuning for Low-Resource Languages: A Comparative Study of LLMs for Bengali Hate Speech Detection
标题：低资源语言的参数高效微调：孟加拉仇恨语音检测LLM的比较研究
链接：https://arxiv.org/abs/2510.16985
备注：Accepted to IEEE COMPAS 2025. 6 pages, 3 figures, 6 tables
【27】Peering Inside the Black Box: Uncovering LLM Errors in Optimization Modelling through Component-Level Evaluation
标题：窥视黑匣子内部：通过学生级评估发现优化建模中的LLM错误
链接：https://arxiv.org/abs/2510.16943
【28】SolverLLM: Leveraging Test-Time Scaling for Optimization Problem via LLM-Guided Search
标题：SolverLLM：通过LLM引导搜索利用测试时间缩放来解决优化问题
链接：https://arxiv.org/abs/2510.16916
备注：NeurIPS 2025
【29】Utility-Diversity Aware Online Batch Selection for LLM Supervised Fine-tuning
标题：LLM监督微调的实用多样性感知在线批量选择
链接：https://arxiv.org/abs/2510.16882
【30】Mixed-Precision Quantization for Language Models: Techniques and Prospects
标题：语言模型的混合精度量化：技术与展望
链接：https://arxiv.org/abs/2510.16805
备注：46 pages, 6 figures, 5 tables
【31】Black-box Optimization of LLM Outputs by Asking for Directions
标题：通过询问方向进行LLM输出的黑匣子优化
链接：https://arxiv.org/abs/2510.16794
【32】DistilLock: Safeguarding LLMs from Unauthorized Knowledge Distillation on the Edge
标题：DistilLock：保护LLM免受边缘未经授权的知识蒸馏
链接：https://arxiv.org/abs/2510.16716
【33】Unleashing Diverse Thinking Modes in LLMs through Multi-Agent Collaboration
标题：通过多代理协作在法学硕士中释放多元化思维模式
链接：https://arxiv.org/abs/2510.16645
【34】Prior Makes It Possible: From Sublinear Graph Algorithms to LLM Test-Time Methods
标题：Prior使其成为可能：从次线性图算法到LLM测试时方法
链接：https://arxiv.org/abs/2510.16609
【35】Atom-anchored LLMs speak Chemistry: A Retrosynthesis Demonstration
标题：原子锚定LLM讲化学：逆合成演示
链接：https://arxiv.org/abs/2510.16590
备注：Alan Kai Hassen and Andrius Bernatavicius contributed equally to this work
【36】Language over Content: Tracing Cultural Understanding in Multilingual Large Language Models
标题：语言优于内容：在多语种大型语言模型中追踪文化理解
链接：https://arxiv.org/abs/2510.16565
备注：Accepted to CIKM 2025 Workshop on Human Centric AI
【37】LANPO: Bootstrapping Language and Numerical Feedback for Reinforcement Learning in LLMs
标题：LANPO：LLM强化学习的引导语言和数字反馈
链接：https://arxiv.org/abs/2510.16552
【38】Realizing LLMs' Causal Potential Requires Science-Grounded, Novel Benchmarks
标题：实现LLM的因果潜力需要基于科学的新颖基准
链接：https://arxiv.org/abs/2510.16530
【39】MoReBench: Evaluating Procedural and Pluralistic Moral Reasoning in Language Models, More than Outcomes
标题：MoreBench：在语言模型中评估程序和多元道德推理，而不是结果
链接：https://arxiv.org/abs/2510.16380
备注：46 pages, 8 figures, 10 tables. Preprint
【40】QSVD: Efficient Low-rank Approximation for Unified Query-Key-Value Weight Compression in Low-Precision Vision-Language Models
标题：QSVD：低精度视觉语言模型中统一查询键值权重压缩的有效低秩近似
链接：https://arxiv.org/abs/2510.16292
备注：Accepted as Spotlight paper by NeurIPS 2025
【41】Do What You Say: Steering Vision-Language-Action Models via Runtime Reasoning-Action Alignment Verification
标题：照你说的做：通过任务推理-动作一致验证引导视觉-语言-动作模型
链接：https://arxiv.org/abs/2510.16281
【42】Publication Trend Analysis and Synthesis via Large Language Model: A Case Study of Engineering in PNAS
标题：基于大型语言模型的出版趋势分析与综合：PNAS工程案例研究
链接：https://arxiv.org/abs/2510.16152
备注：35 pages, 10 figures
【43】Facts in Stats: Impacts of Pretraining Diversity on Language Model Generalization
标题：统计数据中的事实：预训练多样性对语言模型概括的影响
链接：https://arxiv.org/abs/2510.16096
备注：28 pages, 15 figures
【44】STABLE: Gated Continual Learning for Large Language Models
标题：稳定：大型语言模型的门控持续学习
链接：https://arxiv.org/abs/2510.16089
【45】Breaking Memorization Barriers in LLM Code Fine-Tuning via Information Bottleneck for Improved Generalization
标题：打破LLM代码微调中的并行化障碍，通过信息瓶颈以改进通用性
链接：https://arxiv.org/abs/2510.16022
【46】Stratos: An End-to-End Distillation Pipeline for Customized LLMs under Distributed Cloud Environments
标题：Stratos：分布式云环境下定制LLM的端到端蒸馏管道
链接：https://arxiv.org/abs/2510.15992
【47】Can GRPO Help LLMs Transcend Their Pretraining Origin?
标题：GRPO能否帮助LLM超越其预训练起源？
链接：https://arxiv.org/abs/2510.15990
【48】Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
标题：语言模型中的逻辑基元与推理的组合几何
链接：https://arxiv.org/abs/2510.15987
【49】AMiD: Knowledge Distillation for LLMs with $α$-mixture Assistant Distribution
标题：AMiD：具有$a $-混合物辅助分配的LLM知识蒸馏
链接：https://arxiv.org/abs/2510.15982
【50】Cog-Rethinker: Hierarchical Metacognitive Reinforcement Learning for LLM Reasoning
标题：Cog-Rethinker：LLM推理的分层元认知强化学习
链接
：https://arxiv.org/abs/2510.15979
备注：22 Pages, 8 figures, 4 tables
【51】LinearizeLLM: An Agent-Based Framework for LLM-Driven Exact Linear Reformulation of Nonlinear Optimization Problems
标题：LinearizeLLM：一个基于Agent的LLM驱动的非线性优化问题精确线性重构框架
链接：https://arxiv.org/abs/2510.15969
【52】Long Exposure: Accelerating Parameter-Efficient Fine-Tuning for LLMs under Shadowy Sparsity
标题：长时间曝光：加速Shadowy Sparsity下LLM的参数高效微调
链接：https://arxiv.org/abs/2510.15964
备注：None
【53】CTR-LoRA: Curvature-Aware and Trust-Region Guided Low-Rank Adaptation for Large Language Models
标题：CTR-LoRA：大型语言模型的曲线感知和信任区域引导的低等级适应
链接：https://arxiv.org/abs/2510.15962
【54】How Good Are LLMs at Processing Tool Outputs?
标题：LLM在处理工具输出方面有多好？
链接：https://arxiv.org/abs/2510.15955
【55】BEACON: Bayesian Optimal Stopping for Efficient LLM Sampling
标题：BEACON：有效LLM采样的Bayesian最佳停止
链接：https://arxiv.org/abs/2510.15945
备注：Under review on ARR
【56】TeLLMe v2: An Efficient End-to-End Ternary LLM Prefill and Decode Accelerator with Table-Lookup Matmul on Edge FPGAs
标题：TeLLMe v2：一个高效的端到端三值LLM预填充和解码加速器，在Edge FPGA上使用表可扩展Matmul
链接：https://arxiv.org/abs/2510.15926
【57】LLM-VeriPPA: Power, Performance, and Area Optimization aware Verilog Code Generation with Large Language Models
标题：LLM-VeriPPA：具有功耗、性能和面积优化意识的Verilog代码生成，采用大型语言模型
链接：https://arxiv.org/abs/2510.15899
【58】Certified Self-Consistency: Statistical Guarantees and Test-Time Training for Reliable Reasoning in LLMs
标题：认证的自我一致性：LLM中可靠推理的统计保证和测试时训练
链接：https://arxiv.org/abs/2510.17472
【59】From Reviews to Actionable Insights: An LLM-Based Approach for Attribute and Feature Extraction
标题：从评论到可操作洞察：基于LLM的属性和特征提取方法
链接：https://arxiv.org/abs/2510.16551
【60】Few-Label Multimodal Modeling of SNP Variants and ECG Phenotypes Using Large Language Models for Cardiovascular Risk Stratification
标题：使用大型语言模型对SNP变体和心电图表型进行少标签多模式建模用于心血管风险分层
链接：https://arxiv.org/abs/2510.16536
【61】Interpretable RNA-Seq Clustering with an LLM-Based Agentic Evidence-Grounded Framework
标题：使用基于LLM的统计证据框架进行可解释的RN-Seq集群
链接：https://arxiv.org/abs/2510.16082
【62】Aligning Language Models with Investor and Market Behavior for Financial Recommendations
标题：将语言模型与投资者和市场行为保持一致以提供财务建议
链接：https://arxiv.org/abs/2510.15993
Graph相关(图学习|图神经网络|图优化等)(17篇)
【1】Executable Knowledge Graphs for Replicating AI Research
标题：可执行的知识图用于复制AI研究
链接：https://arxiv.org/abs/2510.17795
备注：Work in progress
【2】The Marked Edge Walk: A Novel MCMC Algorithm for Sampling of Graph Partitions
标题：标记边游走：一种用于图分区采样的新型MCMC算法
链接：https://arxiv.org/abs/2510.17714
【3】Deeper with Riemannian Geometry: Overcoming Oversmoothing and Oversquashing for Graph Foundation Models
标题：黎曼几何更深入：克服图基础模型的过度平滑和过度挤压
链接：https://arxiv.org/abs/2510.17457
备注：Accept by NeurIPS 25
【4】Graph Attention-Guided Search for Dense Multi-Agent Pathfinding
标题：密集多智能体寻路的图形注意力引导搜索
链接：https://arxiv.org/abs/2510.17382
【5】Model Metamers Reveal Invariances in Graph Neural Networks
标题：模型元数据揭示图神经网络中的不变性
链接：https://arxiv.org/abs/2510.17378
【6】Robustness in Text-Attributed Graph Learning: Insights, Trade-offs, and New Defenses
标题：文本属性图学习的鲁棒性：见解、权衡和新防御
链接：https://arxiv.org/abs/2510.17185
【7】Fighter: Unveiling the Graph Convolutional Nature of Transformers in Time Series Modeling
标题：斗士：在时间序列建模中揭开Transformer的图卷积本质
链接：https://arxiv.org/abs/2510.17106
备注：Preprint
【8】SNOMED CT-powered Knowledge Graphs for Structured Clinical Data and Diagnostic Reasoning
标题：用于结构化临床数据和诊断推理的SNOMED CT驱动知识图
链接：https://arxiv.org/abs/2510.16899
【9】UniGTE: Unified Graph-Text Encoding for Zero-Shot Generalization across Graph Tasks and Domains
标题：UniGTE：统一图形文本编码，用于跨图形任务和域的Zero-Shot通用
链接：https://arxiv.org/abs/2510.16885
备注：None
【10】Graph Learning is Suboptimal in Causal Bandits
标题：图学习在因果盗贼中次优
链接：https://arxiv.org/abs/2510.16811
备注：31 pages, 5 figures
【11】3D-GSRD: 3D Molecular Graph Auto-Encoder with Selective Re-mask Decoding
标题：3D-GSRD：具有选择性重新掩蔽解码的3D分子图自动编码器
链接：https://arxiv.org/abs/2510.16780
【12】Modeling Expert Interactions in Sparse Mixture of Experts via Graph Structures
标题：基于图结构的稀疏混合专家交互模型
链接：https://arxiv.org/abs/2510.16411
【13】MGTS-Net: Exploring Graph-Enhanced Multimodal Fusion for Augmented Time Series Forecasting
标题：MGTS-Net：探索用于增强时间序列预测的图形增强多峰融合
链接：https://arxiv.org/abs/2510.16350
【14】PassREfinder-FL: Privacy-Preserving Credential Stuffing Risk Prediction via Graph-Based Federated Learning for Representing Password Reuse between Websites
标题：PassREfinder-FL：基于图联邦学习的隐私保护凭证填充风险预测
链接：https://arxiv.org/abs/2510.16083
备注：Accepted by Elsevier Expert Systems with Applications
【15】ESCA: Contextualizing Embodied Agents via Scene-Graph Generation
标题：ESCA：通过场景图生成将预定代理上下文化
链接：https://arxiv.org/abs/2510.15963
备注：Accepted as a Spotlight Paper at NeurIPS 2025
【16】Interpretable Graph-Language Modeling for Detecting Youth Illicit Drug Use
标题：用于检测青少年非法药物使用的可解释图形语言建模
链接
：https://arxiv.org/abs/2510.15961
【17】AGNES: Adaptive Graph Neural Network and Dynamic Programming Hybrid Framework for Real-Time Nanopore Seed Chaining
标题：AGNES：用于实时纳米孔种子链的自适应图神经网络和动态规划混合框架
链接：https://arxiv.org/abs/2510.16013
备注：31 pages, 12 figures, 6 tables. Submitted to ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (ACM-BCB). Includes comprehensive evaluation with statistical validation, ablation studies, and open-source implementation
Transformer(8篇)
【1】ZACH-ViT: A Zero-Token Vision Transformer with ShuffleStrides Data Augmentation for Robust Lung Ultrasound Classification
标题：ZACH-ViT：具有ShuffleStrides数据增强的零令牌视觉Transformer，用于稳健的肺部超声分类
链接：https://arxiv.org/abs/2510.17650
备注：14 pages, 6 figures, 2 tables. Primary subject: cs.LG (Machine Learning) Cross-listed to: cs.CV (Computer Vision and Pattern Recognition), eess.IV (Image and Video Processing). Code available at: this https URL Installation: pip install zachvit Paper licensed under CC BY-NC-ND 4.0. Code released under Apache 2.0 License
【2】The Free Transformer
标题：自由Transformer
链接：https://arxiv.org/abs/2510.17558
【3】Layer Specialization Underlying Compositional Reasoning in Transformers
标题：Transformer中成分推理背后的层专业化
链接：https://arxiv.org/abs/2510.17469
【4】SOLE: Hardware-Software Co-design of Softmax and LayerNorm for Efficient Transformer Inference
标题：SOLE：Softmax和LayerNorm的软硬件协同设计实现高效的Transformer推理
链接：https://arxiv.org/abs/2510.17189
【5】Closing the Curvature Gap: Full Transformer Hessians and Their Implications for Scaling Laws
标题：缩小弯曲差距：全Transformer黑森人及其对比例定律的影响
链接：https://arxiv.org/abs/2510.16927
备注：38 pages, 12 figures. Submitted to ICLR 2026
【6】Renaissance of RNNs in Streaming Clinical Time Series: Compact Recurrence Remains Competitive with Transformers
标题：流媒体临床时间序列中RNN的复兴：紧凑复发仍然与Transformer具有竞争力
链接：https://arxiv.org/abs/2510.16677
【7】Sparse Transformer Architectures via Regularized Wasserstein Proximal Operator with $L_1$ Prior
标题：通过具有L_1 $ Prior的正规化Wasserstein Proximal运算符的稀疏Transformer架构
链接：https://arxiv.org/abs/2510.16356
【8】Early-stopping for Transformer model training
标题：Transformer模型训练提前停止
链接：https://arxiv.org/abs/2510.16074
GAN|对抗|攻击|生成相关(20篇)
【1】GAS: Improving Discretization of Diffusion ODEs via Generalized Adversarial Solver
标题：GAS：通过广义对抗求解器改进扩散常微分方程的离散化
链接：https://arxiv.org/abs/2510.17699
【2】Quantum Synthetic Data Generation for Industrial Bioprocess Monitoring
标题：用于工业生物过程监控的量子合成数据生成
链接：https://arxiv.org/abs/2510.17688
【3】Handling Extreme Class Imbalance: Using GANs in Data Augmentation for Suicide Prediction
标题：处理极端阶级失衡：在数据增强中使用GAN进行自杀预测
链接：https://arxiv.org/abs/2510.17661
【4】Reasoning Distillation and Structural Alignment for Improved Code Generation
标题：推理蒸馏和结构对齐以改进代码生成
链接：https://arxiv.org/abs/2510.17598
【5】Mitigating Clever Hans Strategies in Image Classifiers through Generating Counterexamples
标题：通过生成反例来缓解图像分类器中聪明的汉斯策略
链接：https://arxiv.org/abs/2510.17524
【6】AWARE: Audio Watermarking with Adversarial Resistance to Edits
标题：意识：具有对抗性编辑抵抗的音频水印
链接：https://arxiv.org/abs/2510.17512
【7】Latent Spaces Beyond Synthesis: From GANs to Diffusion Models
标题：超越合成的潜在空间：从GANs到扩散模型
链接：https://arxiv.org/abs/2510.17383
备注：Presented and published at Ethics and Aesthetics of Artificial Intelligence Conference (EA-AI'25)
【8】Towards Mixed-Modal Retrieval for Universal Retrieval-Augmented Generation
标题：走向通用检索增强一代的混合模式检索
链接：https://arxiv.org/abs/2510.17354
备注：This work is in progress
【9】Matricial Free Energy as a Gaussianizing Regularizer: Enhancing Autoencoders for Gaussian Code Generation
标题：作为高斯化调节器的矩阵自由能：增强高斯码生成的自动编码器
链接：https://arxiv.org/abs/2510.17120
【10】Adapting to Stochastic and Adversarial Losses in Episodic MDPs with Aggregate Bandit Feedback
标题：利用总体强盗反馈适应情节MDP中的随机和对抗损失
链接：https://arxiv.org/abs/2510.17103
备注：49 pages
【11】Differentially Private Linear Regression and Synthetic Data Generation with Statistical Guarantees
标题：具有统计保证的差异私有线性回归和合成数据生成
链接：https://arxiv.org/abs/2510.16974
【12】UNDREAM: Bridging Differentiable Rendering and Photorealistic Simulation for End-to-end Adversarial Attacks
标题：UNDREAM：为端到端对抗性攻击搭建差异渲染和真实感模拟的桥梁
链接：https://arxiv.org/abs/2510.16923
【13】U-Codec: Ultra Low Frame-rate Neural Speech Codec for Fast High-fidelity Speech Generation
标题：U-Codec：超低帧率神经语音编解码器，用于快速高保真语音生成
链接：https://arxiv.org/abs/2510.16718
【14】Universal and Transferable Attacks on Pathology Foundation Models
标题：对病理学基础模型的普遍和可转移攻击
链接：https://arxiv.org/abs/2510.16660
备注：38 Pages, 8 Figures
【15】A Versatile Framework for Designing Group-Sparse Adversarial Attacks
标题：设计群稀疏对抗攻击的通用框架
链接：https://arxiv.org/abs/2510.16637
【16】Colliding with Adversaries at ECML-PKDD 2025 Adversarial Attack Competition 1st Prize Solution
标题：在ECML-PKDD 2025年对抗性攻击竞赛中与对手发生冲突一等奖解决方案
链接：https://arxiv.org/abs/2510.16440
【17】DiffusionX: Efficient Edge-Cloud Collaborative Image Generation with Multi-Round Prompt Evolution
标题：DistusionX：具有多轮快速进化的高效边缘云协作图像生成
链接：https://arxiv.org/abs/2510.16326
【18】Bridging Symmetry and Robustness: On the Role of Equivariance in Enhancing Adversarial Robustness
标题：弥合对称性和鲁棒性：论等方差在增强对抗鲁棒性中的作用
链接：https://arxiv.org/abs/2510.16171
备注 ：Accepted for the proceedings of 39th Conference on Neural Information Processing Systems (NeurIPS 2025)
【19】The Hidden Cost of Modeling P(X): Vulnerability to Membership Inference Attacks in Generative Text Classifiers
标题：建模P（X）的隐藏成本：生成式文本分类器中成员推理攻击的脆弱性
链接：https://arxiv.org/abs/2510.16122
【20】On-Chain Decentralized Learning and Cost-Effective Inference for DeFi Attack Mitigation
标题：链上去中心化学习和具有成本效益的推理来缓解DeFi攻击
链接：https://arxiv.org/abs/2510.16024
备注：Published in the 7th Conference on Advances in Financial Technologies (AFT 2025)
半/弱/无/有监督|不确定性|主动学习(9篇)
【1】Efficient Algorithms for Mitigating Uncertainty and Risk in Reinforcement Learning
标题：缓解强化学习中不确定性和风险的有效算法
链接：https://arxiv.org/abs/2510.17690
备注：Dissertation
【2】Semi-supervised Latent Bayesian Optimization for Designing Antimicrobial Peptides
标题：抗菌肽设计的半监督潜在Bayesian优化
链接：https://arxiv.org/abs/2510.17569
备注：19 pages, 9 figures
【3】Uncertainty-aware data assimilation through variational inference
标题：通过变分推理实现不确定性感知数据同化
链接：https://arxiv.org/abs/2510.17268
【4】Quantile Regression, Variational Autoencoders, and Diffusion Models for Uncertainty Quantification: A Spatial Analysis of Sub-seasonal Wind Speed Prediction
标题：用于不确定性量化的分位数回归、变分自动编码器和扩散模型：亚季节风速预测的空间分析
链接：https://arxiv.org/abs/2510.16958
备注：This Work has been submitted to Monthly Weather Review. Copyright in this Work may be transferred without further notice
【5】Needles in the Landscape: Semi-Supervised Pseudolabeling for Archaeological Site Discovery under Label Scarcity
标题：风景中的针：在“稀缺”标签下对考古遗址发现进行半监督伪标签
链接：https://arxiv.org/abs/2510.16814
【6】SAMOSA: Sharpness Aware Minimization for Open Set Active learning
标题：SAMOSA：开放集主动学习的Shareduary Aware最小化
链接：https://arxiv.org/abs/2510.16757
【7】Asymptotically Stable Quaternion-valued Hopfield-structured Neural Network with Periodic Projection-based Supervised Learning Rules
标题：具有周期投影监督学习规则的渐近稳定四元数Hopfield结构神经网络
链接：https://arxiv.org/abs/2510.16607
【8】SCALAR: Self-Calibrating Adaptive Latent Attention Representation Learning
标题：SCAlar：自校准自适应隐性注意力代表学习
链接：https://arxiv.org/abs/2510.16474
【9】Self-Attention to Operator Learning-based 3D-IC Thermal Simulation
标题：自我关注基于操作员学习的3D-IC热模拟
链接：https://arxiv.org/abs/2510.15968
迁移|Zero/Few/One-Shot|自适应(23篇)
【1】On-the-Fly OVD Adaptation with FLAME: Few-shot Localization via Active Marginal-Samples Exploration
标题：使用FLAME进行实时OVD调整：通过活动边缘样本探索进行Few-Shot定位
链接：https://arxiv.org/abs/2510.17670
【2】DAMSDAN: Distribution-Aware Multi-Source Domain Adaptation Network for Cross-Domain EEG-based Emotion Recognition
标题：DAMSDAN：用于基于跨域脑电波的情感识别的分布感知多源域自适应网络
链接：https://arxiv.org/abs/2510.17475
备注：14 pages, 9 figures
【3】Adaptive Discretization for Consistency Models
标题：一致性模型的自适应离散化
链接：https://arxiv.org/abs/2510.17266
备注：Accepted by NeurIPS 2025
【4】ALPINE: A Lightweight and Adaptive Privacy-Decision Agent Framework for Dynamic Edge Crowdsensing
标题：ALPINE：用于动态边缘人群感知的轻量级、自适应隐私决策代理框架
链接：https://arxiv.org/abs/2510.17162
备注：12 pages, 8 figures, 4 tables. Submitted to The Web Conference (WWW 2026)
【5】Explainable Heterogeneous Anomaly Detection in Financial Networks via Adaptive Expert Routing
标题：基于自适应专家路由的金融网络异构异常检测
链接：https://arxiv.org/abs/2510.17088
【6】Consistent Zero-Shot Imitation with Contrastive Goal Inference
标题：一致的Zero-Shot模仿与对比目标推理
链接：https://arxiv.org/abs/2510.17059
【7】Extended LSTM: Adaptive Feature Gating for Toxic Comment Classification
标题：扩展LSTM：有毒评论分类的自适应特征门控
链接：https://arxiv.org/abs/2510.17018
【8】Adaptive Online Learning with LSTM Networks for Energy Price Prediction
标题：利用LSTM网络进行能源价格预测的自适应在线学习
链接：https://arxiv.org/abs/2510.16898
【9】Zero-Shot Performance Prediction for Probabilistic Scaling Laws
标题：概率缩放定律的零发射性能预测
链接：https://arxiv.org/abs/2510.16743
备注：Accepted to NeurIPS 2025
【10】Resolution-Aware Retrieval Augmented Zero-Shot Forecasting
标题：分辨率感知检索增强Zero-Shot预测
链接：https://arxiv.org/abs/2510.16695
【11】Zero-shot World Models via Search in Memory
标题：通过记忆搜索Zero-Shot世界模型
链接：https://arxiv.org/abs/2510.16123
备注：10 pages, 8 figures in main text + appendices
【12】BPL: Bias-adaptive Preference Distillation Learning for Recommender System
标题：BPL：推荐系统的偏自适应偏好蒸馏学习
链接：https://arxiv.org/abs/2510.16076
备注：\c{opyright} 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works
【13】AMS-QUANT: Adaptive Mantissa Sharing for Floating-point Quantization
标题：AMS-QUANT：用于浮点量化的自适应尾数共享
链接：https://arxiv.org/abs/2510.16045
备注：12 pages, 6 figures
【14】Transfer learning strategies for accelerating reinforcement-learning-based flow control
标题：加速基于业务学习的流控制的迁移学习策略
链接：https://arxiv.org/abs/2510.16016
【15】Decision-focused Sensing and Forecasting for Adaptive and Rapid Flood Response: An Implicit Learning Approach
标题：以决策为中心的感知和预测，以适应性和快速洪水响应：一种隐式学习方法
链接：https://arxiv.org/abs/2510.16015
【16】AMStraMGRAM: Adaptive Multi-cutoff Strategy Modification for ANaGRAM
标题：自适应多截断策略改进算法AMStraMatigue
链接：https://arxiv.org/abs/2510.15998
【17】Gains: Fine-grained Federated Domain Adaptation in Open Set
标题：收益：开放集中的细粒度联邦域适应
链接：https://arxiv.org/abs/2510.15967
备注：Accepted by NeurIPS2025
【18】Lyapunov-Stable Adaptive Control for Multimodal Concept Drift
标题：多峰概念漂移的Lyapunov稳定自适应控制
链接：https://arxiv.org/abs/2510.15944
【19】Plasma Shape Control via Zero-shot Generative Reinforcement Learning
标题：通过零激发生成强化学习实现等离子体形状控制
链接：https://arxiv.org/abs/2510.17531
【20】Adaptive Sample Sharing for Linear Regression
标题：线性回归的自适应样本共享
链接：https://arxiv.org/abs/2510.16986
【21】ARCO-BO: Adaptive Resource-aware COllaborative Bayesian Optimization for Heterogeneous Multi-Agent Design
标题：ARCO-BO：用于异类多代理设计的自适应资源感知协作式Bayesian优化
链接：https://arxiv.org/abs/2510.16652
【22】The Invisible Handshake: Tacit Collusion between Adaptive Market Agents
标题：隐形握手：适应性市场主体之间的隐性勾结
链接：https://arxiv.org/abs/2510.15995
【23】FinFlowRL: An Imitation-Reinforcement Learning Framework for Adaptive Stochastic Control in Finance
标题：FinFlowRL：金融自适应随机控制的模仿-强化学习框架
链接：https://arxiv.org/abs/2510.15883
备注：21 pages, 5 algorithms, 4 tables, 5 figures
强化学习(9篇)
【1】A Principle of Targeted Intervention for Multi-Agent Reinforcement Learning
标题：多Agent强化学习的目标干预原则
链接：https://arxiv.org/abs/2510.17697
备注：Accepted to NeurIPS 2025
【2】An Empirical Study of Lagrangian Methods in Safe Reinforcement Learning
标题：安全强化学习中拉格朗日方法的实证研究
链接：https://arxiv.org/abs/2510.17564
【3】Optimizing Energy Management of Smart Grid using Reinforcement Learning aided by Surrogate models built using Physics-informed Neural Networks
标题：使用强化学习优化智能电网的能源管理，并在使用物理知识神经网络构建的代理模型的辅助下
链接：https://arxiv.org/abs/2510.17380
备注：None
【4】Continuous Q-Score Matching: Diffusion Guided Reinforcement Learning for Continuous-Time Control
标题：连续Q得分匹配：用于连续时间控制的扩散引导强化学习
链接：https://arxiv.org/abs/2510.17122
【5】AoI-Aware Task Offloading and Transmission Optimization for Industrial IoT Networks: A Branching Deep Reinforcement Learning Approach
标题：工业物联网网络的AoI感知任务卸载和传输优化：一种分支深度强化学习方法
链接：https://arxiv.org/abs/2510.16414
备注：15 pages, 13 figures, submitted to IEEE journal for potential publication
【6】WEBSERV: A Browser-Server Environment for Efficient Training of Reinforcement Learning-based Web Agents at Scale
标题：WEBSERV：一个用于大规模有效训练基于强化学习的Web代理的浏览器-服务器环境
链接：https://arxiv.org/abs/2510.16252
【7】Human-Allied Relational Reinforcement Learning
标题：人际关系强化学习
链接：https://arxiv.org/abs/2510.16188
备注：Proceedings of the Twelfth Annual Conference on Advances in Cognitive Systems, ACS-2025 (143-159)
【8】The Formalism-Implementation Gap in Reinforcement Learning Research
标题：强化学习研究中的形式主义实施差距
链接：https://arxiv.org/abs/2510.16175
【9】Feature-driven reinforcement learning for photovoltaic in continuous intraday trading
标题：连续盘中交易中的企业驱动的太阳能强化学习
链接：https://arxiv.org/abs/2510.16021
元学习(1篇)
【1】Buzz, Choose, Forget: A Meta-Bandit Framework for Bee-Like Decision Making
标题：嗡嗡、选择、忘记：类似蜜蜂决策的元强盗框架
链接：https://arxiv.org/abs/2510.16462
符号|符号学习(2篇)
【1】Curiosity-driven RL for symbolic equation solving
标题：好奇心驱动的RL用于符号方程求解
链接：https://arxiv.org/abs/2510.17022
备注：Accepted at the NeurIPS 2025 MATH-AI Workshop
【2】Hey Pentti, We Did It Again!: Differentiable vector-symbolic types that prove polynomial termination
标题：嘿彭蒂，我们又做到了！：证明多项终止的可区分的载体符号类型
链接：https://arxiv.org/abs/2510.16533
备注：None
分层学习(1篇)
【1】DETree: DEtecting Human-AI Collaborative Texts via Tree-Structured Hierarchical Representation Learning
标题：DETree：通过树结构分层表示学习检测人机协作文本
链接：https://arxiv.org/abs/2510.17489
备注：To appear in NeurIPS 2025
医学相关(8篇)
【1】CEPerFed: Communication-Efficient Personalized Federated Learning for Multi-Pulse MRI Classification
标题：CEPerFed：用于多脉冲MRI分类的通信高效个性化联邦学习
链接：https://arxiv.org/abs/2510.17584
【2】MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
标题：MambaX-Net：用于纵向MRI分割的双输入Mamba-增强交叉注意网络
链接：https://arxiv.org/abs/2510.17529
【3】SAFE-D: A Spatiotemporal Detection Framework for Abnormal Driving Among Parkinson's Disease-like Drivers
标题：SAFE-D：帕金森病样驾驶员异常驾驶的时空检测框架
链接：https://arxiv.org/abs/2510.17517
【4】CrossStateECG: Multi-Scale Deep Convolutional Network with Attention for Rest-Exercise ECG Biometrics
标题：CrossState心电图：多尺度深度卷积网络，关注静息运动心电图生物统计学
链接：https://arxiv.org/abs/2510.17467
【5】Temporally Detailed Hypergraph Neural ODEs for Type 2 Diabetes Progression Modeling
标题：用于2型糖尿病进展建模的时间详细超图神经ODE
链接：https://arxiv.org/abs/2510.17211
【6】Time-Embedded Algorithm Unrolling for Computational MRI
标题：用于计算MRI的时间嵌入算法展开
链接：https://arxiv.org/abs/2510.16321
备注：Neural Information Processing Systems (NeurIPS), 2025
【7】Lung Cancer Classification from CT Images Using ResNet
标题：基于ResNet的CT图像肺癌分类
链接：https://arxiv.org/abs/2510.16310
备注：9 pages,4 figures, 3 tables
【8】Identifying multi-omics interactions for lung cancer drug targets discovery using Kernel Machine Regression
标题：使用核机器回归识别肺癌药物靶点发现的多组学相互作用
链接：https://arxiv.org/abs/2510.16093
蒸馏|知识提取(2篇)
【1】Diffusion Models as Dataset Distillation Priors
标题：扩散模型作为数据集蒸馏先验
链接：https://arxiv.org/abs/2510.17421
【2】Leave It to the Experts: Detecting Knowledge Distillation via MoE Expert Signatures
标题：将其交给专家：通过MoE专家签名检测知识提炼
链接：https://arxiv.org/abs/2510.16968
备注：Code is at this https URL
聚类(1篇)
【1】User Profiles of Sleep Disorder Sufferers: Towards Explainable Clustering and Differential Variable Analysis
标题：睡眠障碍患者的用户概况：可解释的聚类和差异变量分析
链接：https://arxiv.org/abs/2510.15986
备注：in French language, Plate-Forme Intelligence Artificielle, Jun 2025, Dijon (FRANCE), France
超分辨率|去噪|去模糊|去雾(1篇)
【1】Prominence-Aware Artifact Detection and Dataset for Image Super-Resolution
标题：图像超分辨率的突出感知预设检测和数据集
链接：https://arxiv.org/abs/2510.16752
自动驾驶|车辆|车道检测等(4篇)
【1】TrajMamba: An Efficient and Semantic-rich Vehicle Trajectory Pre-training Model
标题：TrajMamba：一种高效且语义丰富的车辆轨迹预训练模型
链接：https://arxiv.org/abs/2510.17545
备注：Accepted by NeurIPS2025
【2】DrivAerStar: An Industrial-Grade CFD Dataset for Vehicle Aerodynamic Optimization
标题：DrivAerStar：用于车辆空气动力学优化的工业级计算流体动力学数据集
链接：https://arxiv.org/abs/2510.16857
【3】LSTM-Based Forecasting and Analysis of EV Charging Demand in a Dense Urban Campus
标题：基于LSTM的城市密集校园电动汽车充电需求预测与分析
链接：https://arxiv.org/abs/2510.16719
【4】FUSE-Traffic: Fusion of Unstructured and Structured Data for Event-aware Traffic Forecasting
标题：FUSE-Traffic：融合非结构化和结构化数据的事件感知流量预测
链接：https://arxiv.org/abs/2510.16053
点云|SLAM|雷达|激光|深度RGBD相关(1篇)
【1】MNO: Multiscale Neural Operator for Computational Fluid Dynamics with 3D Point Cloud Data
标题：MNO：具有3D点云数据的计算流体动力学多尺度神经运算符
链接：https://arxiv.org/abs/2510.16071
联邦学习|隐私保护|加密(4篇)
【1】Frugal Federated Learning for Violence Detection: A Comparison of LoRA-Tuned VLMs and Personalized CNNs
标题：用于暴力检测的节俭联邦学习：LoRA调谐的VLM和个性化CNN的比较
链接：https://arxiv.org/abs/2510.17651
备注：7 pages, 1 figure, FLTA 2025
【2】CLIP: Client-Side Invariant Pruning for Mitigating Stragglers in Secure Federated Learning
标题：CLIP：客户端不变修剪，以缓解安全联邦学习中的落后者
链接：https://arxiv.org/abs/2510.16694
【3】FedPURIN: Programmed Update and Reduced INformation for Sparse Personalized Federated Learning
标题：FedPURIN：稀疏个性化联邦学习的编程更新和简化信息
链接：https://arxiv.org/abs/2510.16065
【4】Quantum Federated Learning: Architectural Elements and Future Directions
标题：量子联邦学习：架构元素和未来方向
链接：https://arxiv.org/abs/2510.17642
备注：28 PAGES, 11 figures, introductory review article (book chapter), to be published in a book with springer
推理|分析|理解|解释(29篇)
【1】Foundational Automatic Evaluators: Scaling Multi-Task Generative Evaluator Training for Reasoning-Centric Domains
标题：基础自动评估器：扩展以推理为中心的领域的多任务生成评估器训练
链接：https://arxiv.org/abs/2510.17793
备注：29 pages, 9 tables, 6 figures
【2】Inference-Time Compute Scaling For Flow Matching
标题：流匹配的推理时计算缩放
链接：https://arxiv.org/abs/2510.17786
【3】MIRAGE: Agentic Framework for Multimodal Misinformation Detection with Web-Grounded Reasoning
标题：MISYS：利用网络推理进行多模式错误信息检测的抽象框架
链接：https://arxiv.org/abs/2510.17590
备注：16 pages, 3 tables, 1 figure
【4】Reliable Inference in Edge-Cloud Model Cascades via Conformal Alignment
标题：通过保形对齐实现边云模型级联的可靠推理
链接：https://arxiv.org/abs/2510.17543
备注：Under Review
【5】The Graphon Limit Hypothesis: Understanding Neural Network Pruning via Infinite Width Analysis
标题：图形极限假设：通过无限宽度分析理解神经网络修剪
链接：https://arxiv.org/abs/2510.17515
备注：NeurIPS 2025 Spotlight
【6】Towards geological inference with process-based and deep generative modeling, part 2: inversion of fluvial deposits and latent-space disentanglement
标题：利用基于过程的深度生成建模进行地质推断，第2部分：河流沉积物的逆求和潜空间解纠缠
链接：https://arxiv.org/abs/2510.17478
备注：52 pages, 42 figures
【7】Explainable AI for microseismic event detection
标题：用于微震事件检测的可解释人工智能
链接：https://arxiv.org/abs/2510.17458
备注：Submitted to Artificial Intelligence in Geosciences
【8】Quantifying Climate Policy Action and Its Links to Development Outcomes: A Cross-National Data-Driven Analysis
标题：量化气候政策行动及其与发展成果的联系：跨国数据驱动分析
链接：https://arxiv.org/abs/2510.17425
备注：This paper/proposal has been accepted as a poster in the NeurIPS 2025
【9】Understanding and Improving Length Generalization in Hierarchical Sparse Attention Models
标题：理解和改进分层稀疏注意力模型中的长度概括
链接：https://arxiv.org/abs/2510.17196
备注：Preprint. Work in progress
【10】Video Reasoning without Training
标题：无需训练的视频推理
链接：https://arxiv.org/abs/2510.17045
【11】Diverse Influence Component Analysis: A Geometric Approach to Nonlinear Mixture Identifiability
标题：多样化影响分量分析：非线性混合可识别性的几何方法
链接：https://arxiv.org/abs/2510.17040
备注：30 pages, 3 figures
【12】Towards Interpretable and Trustworthy Time Series Reasoning: A BlueSky Vision
标题：迈向可解释且值得信赖的时间序列推理：蓝天愿景
链接：https://arxiv.org/abs/2510.16980
【13】Predicting life satisfaction using machine learning and explainable AI
标题：使用机器学习和可解释人工智能预测生活满意度
链接：https://arxiv.org/abs/2510.16547
备注：None
【14】Cataract-LMM: Large-Scale, Multi-Source, Multi-Task Benchmark for Deep Learning in Surgical Video Analysis
标题：Cataract-LMM：手术视频分析中深度学习的大规模、多源、多任务基准
链接：https://arxiv.org/abs/2510.16371
备注：20 pages, 11 figures, 11 tables. Data descriptor for the Cataract-LMM benchmark dataset. Source code and dataset are available
【15】Machine Learning for Climate Policy: Understanding Policy Progression in the European Green Deal
标题：气候政策的机器学习：理解欧洲绿色协议中的政策进展
链接：https://arxiv.org/abs/2510.16233
【16】A Minimal-Assumption Analysis of Q-Learning with Time-Varying Policies
标题：时变策略下Q学习的最小假设分析
链接：https://arxiv.org/abs/2510.16132
备注：43 pages, 4 figures
【17】FSRF: Factorization-guided Semantic Recovery for Incomplete Multimodal Sentiment Analysis
标题：FSRF：基于因子分解的不完全多模态情感分析语义恢复
链接：https://arxiv.org/abs/2510.16086
备注：6 pages,3 figures
【18】Data-Driven Analysis of Intersectional Bias in Image Classification: A Framework with Bias-Weighted Augmentation
标题：图像分类中交叉偏差的数据驱动分析：具有偏差加权增强的框架
链接：https://arxiv.org/abs/2510.16072
备注：18 pages
【19】Membership Inference over Diffusion-models-based Synthetic Tabular Data
标题：基于扩散模型的合成表格数据的隶属度推断
链接：https://arxiv.org/abs/2510.16037
【20】One Token Embedding Is Enough to Deadlock Your Large Reasoning Model
标题：一个令牌嵌入足以让你的大型推理模型陷入僵局
链接：https://arxiv.org/abs/2510.15965
备注：NeurIPS 2025
【21】Hydrogen production from blended waste biomass: pyrolysis, thermodynamic-kinetic analysis and AI-based modelling
标题：混合废弃生物质制氢：高温分解、热力学动力学分析和基于人工智能的建模
链接：https://arxiv.org/abs/2510.15960
备注：41 pages, 21 figures
【22】Time Series Analysis in Frequency Domain: A Survey of Open Challenges, Opportunities and Benchmarks
标题：频域时间序列分析：开放挑战、机遇和基准调查
链接：https://arxiv.org/abs/2504.07099
备注：35 pages
【23】Mode Collapse of Mean-Field Variational Inference
标题：平均场变分推理的模式崩溃
链接：https://arxiv.org/abs/2510.17063
【24】Prediction-Augmented Trees for Reliable Statistical Inference
标题：用于可靠统计推断的预测增强树
链接：https://arxiv.org/abs/2510.16937
备注：45 pages, 9 Figures
【25】Extending Prediction-Powered Inference through Conformal Prediction
标题：通过保形预测扩展预测动力推理
链接：https://arxiv.org/abs/2510.16166
【26】The Cultural Mapping and Pattern Analysis (CMAP) Visualization Toolkit: Open Source Text Analysis for Qualitative and Computational Social Science
标题：文化制图和模式分析（CMAP）可视化工具包：定性和计算社会科学的开源文本分析
链接：https://arxiv.org/abs/2510.16140
备注：V1
【27】Learning density ratios in causal inference using Bregman-Riesz regression
标题：使用Bregman-Riesz回归进行因果推理的学习密度比
链接：https://arxiv.org/abs/2510.16127
备注：Replication code is available from this https URL
【28】Dynamic Factor Analysis of Price Movements in the Philippine Stock Exchange
标题：菲律宾证券交易所价格变动的动态因素分析
链接：https://arxiv.org/abs/2510.15938
【29】Geometric Dynamics of Consumer Credit Cycles: A Multivector-based Linear-Attention Framework for Explanatory Economic Analysis
标题：消费者信贷周期的几何动力学：用于解释性经济分析的基于多因素的线性注意力框架
链接：https://arxiv.org/abs/2510.15892
备注：29 pages, 7 figures
检测相关(14篇)
【1】Formally Exploring Time-Series Anomaly Detection Evaluation Metrics
标题：正式探索时间序列异常检测评估工作表
链接：https://arxiv.org/abs/2510.17562
备注：73 pages, 13 figures
【2】Beyond Binary Out-of-Distribution Detection: Characterizing Distributional Shifts with Multi-Statistic Diffusion Trajectories
标题：超越二元分布外检测：用多统计扩散轨迹描述分布漂移
链接：https://arxiv.org/abs/2510.17381
备注：11 Pages, 6 Figures
【3】Fair and Interpretable Deepfake Detection in Videos
标题：视频中公平且可解释的Deepfake检测
链接：https://arxiv.org/abs/2510.17264
备注：10 pages (including References)
【4】High-Level Multi-Robot Trajectory Planning And Spurious Behavior Detection
标题：高层次多机器人轨迹规划与虚假行为检测
链接：https://arxiv.org/abs/2510.17261
备注：6 pages,3 figures, Iberian Robotics Conference 2025
【5】QRïS: A Preemptive Novel Method for Quishing Detection Through Structural Features of QR
标题：QRïS：一种通过QR结构特征进行Quaring检测的先发制人的新方法
链接：https://arxiv.org/abs/2510.17175
备注：13 pages, 11 figures, and 7 tables
【6】AI-Generated Text Detection in Low-Resource Languages: A Case Study on Urdu
标题：低资源语言中的人工智能生成文本检测：乌尔都语案例研究
链接：https://arxiv.org/abs/2510.16573
【7】Structured Temporal Causality for Interpretable Multivariate Time Series Anomaly Detection
标题：可解释多元时间序列异常检测的结构化时间因果关系
链接：https://arxiv.org/abs/2510.16511
备注：Accepted by NeurIPS 2025
【8】iWatchRoadv2: Pothole Detection, Geospatial Mapping, and Intelligent Road Governance
标题：iWatchRoadv 2：坑洞检测、地理空间映射和智能道路治理
链接：https://arxiv.org/abs/2510.16375
备注：Under review
【9】Benchmarking noisy label detection methods
标题：基准测试有噪标签检测方法
链接：https://arxiv.org/abs/2510.16211
【10】A Novel GPT-Based Framework for Anomaly Detection in System Logs
标题：一种基于GPT的系统异常检测框架
链接：https://arxiv.org/abs/2510.16044
【11】STAR: Boosting Time Series Foundation Models for Anomaly Detection through State-aware Adapter
标题：STAR：通过状态感知适配器增强异常检测的时间序列基础模型
链接：https://arxiv.org/abs/2510.16014
【12】Bolster Hallucination Detection via Prompt-Guided Data Augmentation
标题：通过预算引导数据增强增强幻觉检测
链接：https://arxiv.org/abs/2510.15977
【13】Learning from Mistakes: Enhancing Harmful Meme Detection via Misjudgment Risk Patterns
标题：从错误中学习：通过误判风险模式增强有害模因检测
链接：https://arxiv.org/abs/2510.15946
备注：12 Pages, Submitted to WWW'26
【14】Detecting and Preventing Harmful Behaviors in AI Companions: Development and Evaluation of the SHIELD Supervisory System
标题：检测和预防人工智能伴侣的有害行为：SHIELD监控系统的开发和评估
链接：https://arxiv.org/abs/2510.15891
分类|识别(3篇)
【1】A Prototypical Network with an Attention-based Encoder for Drivers Identification Application
标题：用于驾驶员识别应用的具有基于注意力的编码器的原型网络
链接：https://arxiv.org/abs/2510.17250
【2】WaveNet's Precision in EEG Classification
标题：WaveNet在脑电分类中的精确度
链接：https://arxiv.org/abs/2510.15947
备注：6 pages, 5 figures and 3 tables. Includes main text and bibliography
【3】Optimal Best Arm Identification under Differential Privacy
标题：差异隐私下的最佳手臂识别
链接：https://arxiv.org/abs/2510.17348
备注：92 pages, 2 figures, 2 tables. To be published in the Thirty-Ninth Annual Conference on Neural Information Processing Systems
表征(8篇)
【1】Atlas-based Manifold Representations for Interpretable Riemannian Machine Learning
标题：可解释Riemann机器学习的基于托马斯的Manifold表示
链接：https://arxiv.org/abs/2510.17772
【2】Disentanglement Beyond Static vs. Dynamic: A Benchmark and Evaluation Framework for Multi-Factor Sequential Representations
标题：超越静态与动态的理清：多因素序列表示的基准和评估框架
链接：https://arxiv.org/abs/2510.17313
【3】Fly-CL: A Fly-Inspired Framework for Enhancing Efficient Decorrelation and Reduced Training Time in Pre-trained Model-based Continual Representation Learning
标题：Fly-CL：一个受飞行启发的框架，用于在预训练的基于模型的连续表示学习中增强高效去相关并减少训练时间
链接：https://arxiv.org/abs/2510.16877
【4】Improving Model Representation and Reducing KV Cache via Skip Connections with First Value Heads
标题：通过跳过具有第一值头的连接来改进模型表示并减少KV缓存
链接：https://arxiv.org/abs/2510.16807
备注：The code is available at: \url{this https URL}
【5】Copy-Augmented Representation for Structure Invariant Template-Free Retrosynthesis
标题：结构不变无模板逆合成的拷贝增强表示
链接：https://arxiv.org/abs/2510.16588
【6】Blending Learning to Rank and Dense Representations for Efficient and Effective Cascades
标题：将学习与等级和密集表示相结合，以实现高效和有效的级联
链接：https://arxiv.org/abs/2510.16393
【7】Humanoid-inspired Causal Representation Learning for Domain Generalization
标题：面向领域概括的人文启发的因果表示学习
链接：https://arxiv.org/abs/2510.16382
【8】MEET-Sepsis: Multi-Endogenous-View Enhanced Time-Series Representation Learning for Early Sepsis Prediction Representation Learning for Early Sepsis Prediction
标题：MEET-脓毒症：用于早期脓毒症预测的多内生视图增强时间序列表示学习
链接：https://arxiv.org/abs/2510.15985
备注：Accepted to PRICAI 2025
3D|3D重建等相关(2篇)
【1】Learning to play: A Multimodal Agent for 3D Game-Play
标题：学习游戏：3D游戏的多模式代理
链接：https://arxiv.org/abs/2510.16774
备注：International Conference on Computer Vision Workshop on Multi-Modal Reasoning for Agentic Intelligence
【2】Accelerating Frontier MoE Training with 3D Integrated Optics
标题：利用3D集成光学加速前沿MoE训练
链接：https://arxiv.org/abs/2510.15893
备注：12 pages, 11 figures. To be published in Hot Interconnects 2025
编码器(2篇)
【1】Diagnosis of Fuel Cell Health Status with Deep Sparse Auto-Encoder Neural Network
标题：利用深度稀疏自动编码器神经网络诊断燃料电池健康状况
链接：https://arxiv.org/abs/2510.17214
【2】An Efficient Semantic Segmentation Decoder for In-Car or Distributed Applications
标题：用于车载或分布式应用的高效语义分割解码器
链接：https://arxiv.org/abs/2510.16747
优化|敛散性(12篇)
【1】LILO: Bayesian Optimization with Interactive Natural Language Feedback
标题：LILO：具有交互式自然语言反馈的Bayesian优化
链接：https://arxiv.org/abs/2510.17671
【2】Convergence Rates for Gradient Descent on the Edge of Stability in Overparametrised Least Squares
标题：过参数化最小二乘稳定边缘梯度下降的收敛速度
链接：https://arxiv.org/abs/2510.17506
备注：NeurIPS2025. Code available at this https URL
【3】Stochastic Difference-of-Convex Optimization with Momentum
标题：带动量的随机凸差优化
链接：https://arxiv.org/abs/2510.17503
【4】On the Universal Near Optimality of Hedge in Combinatorial Settings
标题：组合环境下对冲的普适近乎最优性
链接：https://arxiv.org/abs/2510.17099
备注：28 pages, 1 Figure
【5】Convergence of Regret Matching in Potential Games and Constrained Optimization
标题：潜在博弈和约束优化中遗憾匹配的收敛性
链接：https://arxiv.org/abs/2510.17067
【6】A Control-Theoretic Approach to Dynamic Payment Routing for Success Rate Optimization
标题：成功率优化的动态支付路径控制理论方法
链接：https://arxiv.org/abs/2510.16735
备注：7 Pages, 8 Figures
【7】Input Domain Aware MoE: Decoupling Routing Decisions from Task Optimization in Mixture of Experts
标题：输入域感知MoE：将路由决策与混合专家的任务优化脱钩
链接：https://arxiv.org/abs/2510.16448
备注：ACM MM25
【8】Optimization of the quantization of dense neural networks from an exact QUBO formulation
标题：根据精确的QUBO公式优化密集神经网络的量化
链接：https://arxiv.org/abs/2510.16075
【9】Residual Correction Models for AC Optimal Power Flow Using DC Optimal Power Flow Solutions
标题：使用直流最佳潮流解的交流最佳潮流的剩余修正模型
链接：https://arxiv.org/abs/2510.16064
【10】Airfoil optimization using Design-by-Morphing with minimized design-space dimensionality
标题：使用Design by-Morphing进行机翼优化，最小化设计空间维度
链接：https://arxiv.org/abs/2510.16020
【11】Near-Optimal Quantum Algorithms for Computing (Coarse) Correlated Equilibria of General-Sum Games
标题
：计算广义和博弈（粗）相关均衡的近优量子算法
链接：https://arxiv.org/abs/2510.16782
备注：Accepted at NeurIPS 2025, 27 pages
【12】Escaping Model Collapse via Synthetic Data Verification: Near-term Improvements and Long-term Convergence
标题：通过合成数据验证避免模型崩溃：近期改进和长期收敛
链接：https://arxiv.org/abs/2510.16657
备注：26 pages, 6 figures
预测|估计(13篇)
【1】Prediction of Sea Ice Velocity and Concentration in the Arctic Ocean using Physics-informed Neural Network
标题：使用物理信息神经网络预测北冰洋海冰速度和浓度
链接：https://arxiv.org/abs/2510.17756
备注：49 pages, 7 figures, submitted to Environmental Modelling & Software
【2】A Conditional Diffusion Model for Probabilistic Prediction of Battery Capacity Degradation
标题：电池容量退化概率预测的条件扩散模型
链接：https://arxiv.org/abs/2510.17414
【3】S4ECG: Exploring the impact of long-range interactions for arrhythmia prediction
标题：S4心电图：探索远程相互作用对心律失常预测的影响
链接：https://arxiv.org/abs/2510.17406
【4】HyperSearch: Prediction of New Hyperedges through Unconstrained yet Efficient Search
标题：HyperSearch：通过不受约束但高效的搜索预测新的超雷达
链接：https://arxiv.org/abs/2510.17153
备注：IEEE International Conference on Data Mining (ICDM) 2025
【5】A Primer on Kolmogorov-Arnold Networks (KANs) for Probabilistic Time Series Forecasting
标题：用于概率时间序列预测的Kolmogorov-Arnold网络（KAN）入门
链接：https://arxiv.org/abs/2510.16940
【6】A Lightweight DL Model for Smart Grid Power Forecasting with Feature and Resolution Mismatch
标题：具有特征和分辨率不匹配的智能电网电力预测轻量级DL模型
链接：https://arxiv.org/abs/2510.16911
备注：5 pages, 3 figures, The IEEE PES ISGT Middle East 2025 (ISGT-ME 2025) November 23-26th 2025, Dubai, UAE
【7】ProtoMol: Enhancing Molecular Property Prediction via Prototype-Guided Multimodal Learning
标题：ProtoMol：通过原型引导的多模式学习增强分子性质预测
链接：https://arxiv.org/abs/2510.16824
【8】Still Competitive: Revisiting Recurrent Models for Irregular Time Series Prediction
标题：仍然有竞争力：重新审视不规则时间序列预测的循环模型
链接：https://arxiv.org/abs/2510.16161
【9】Learning a Generalized Model for Substation Level Voltage Estimation in Distribution Networks
标题：学习配电网变电站级电压估计的广义模型
链接：https://arxiv.org/abs/2510.16063
【10】Layer-Aware Influence for Online Data Valuation Estimation
标题：分层感知对在线数据估值估计的影响
链接：https://arxiv.org/abs/2510.16007
【11】DAWP: A framework for global observation forecasting via Data Assimilation and Weather Prediction in satellite observation space
标题：DAWP：通过卫星观测空间中的数据同化和天气预测进行全球观测预测的框架
链接：https://arxiv.org/abs/2510.15978
备注：None
【12】A Storm-Centric 250 m NEXRAD Level-II Dataset for High-Resolution ML Nowcasting
标题：用于高分辨率ML即时预报的以风暴为中心的250 m NEXRAD Level II数据集
链接：https://arxiv.org/abs/2510.16031
备注：6 pages, 4 figures
【13】Bitcoin Price Forecasting Based on Hybrid Variational Mode Decomposition and Long Short Term Memory Network
标题：基于混合变分模分解和长短期记忆网络的比特币价格预测
链接：https://arxiv.org/abs/2510.15900
其他神经网络|深度学习|模型|建模(41篇)
【1】Functional Distribution Networks (FDN)
标题：功能分销网络（FDN）
链接：https://arxiv.org/abs/2510.17794
备注：Submitted to ICLR 2026. Code will be released upon acceptance
【2】SoftMimic: Learning Compliant Whole-body Control from Examples
标题：SoftMimic：从示例中学习合规的全身控制
链接：https://arxiv.org/abs/2510.17792
备注：Website: this https URL
【3】Just-In-Time Piecewise-Linear Semantics for ReLU-type Networks
标题：ReLU型网络的实时分段线性语义
链接：https://arxiv.org/abs/2510.17622
【4】Curiosity Meets Cooperation: A Game-Theoretic Approach to Long-Tail Multi-Label Learning
标题：好奇心遇见合作：长尾多标签学习的游戏理论方法
链接：https://arxiv.org/abs/2510.17520
备注：Under review
【5】Local properties of neural networks through the lens of layer-wise Hessians
标题：从分层黑森人的角度看神经网络的局部属性
链接：https://arxiv.org/abs/2510.17486
备注：Comments: 22 pages, 8 figures. Submitted to arXiv:cs.LG
【6】Unified Privacy Guarantees for Decentralized Learning via Matrix Factorization
标题：通过矩阵分解实现去中心化学习的统一隐私保证
链接：https://arxiv.org/abs/2510.17480
备注：21 pages, 5 figures
【7】Navigating the Alignment-Calibration Trade-off: A Pareto-Superior Frontier via Model Merging
标题：应对对准-校准权衡：通过模型合并实现帕累托-卓越前沿
链接：https://arxiv.org/abs/2510.17426
【8】MILES: Modality-Informed Learning Rate Scheduler for Balancing Multimodal Learning
标题：MILES：用于平衡多模式学习的基于模式的学习率指标
链接：https://arxiv.org/abs/2510.17394
备注：Accepted and presented at the 2025 International Joint Conference on Neural Networks (IJCNN'25). The paper was awarded an honorable mention (best 4 papers)
【9】M2H: Multi-Task Learning with Efficient Window-Based Cross-Task Attention for Monocular Spatial Perception
标题：M2 H：多任务学习，具有高效的基于窗口的跨任务注意力，以实现单目空间感知
链接：https://arxiv.org/abs/2510.17363
备注：Accepted to the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025). 8 pages, 7 figures
【10】Auto-Rubric: Learning to Extract Generalizable Criteria for Reward Modeling
标题：Auto-Rubric：学习提取奖励建模的可推广标准
链接：https://arxiv.org/abs/2510.17314
【11】Symmetries in PAC-Bayesian Learning
标题：Pac-Bayesian学习中的对称性
链接：https://arxiv.org/abs/2510.17303
【12】Learning After Model Deployment
标题：模型部署后学习
链接：https://arxiv.org/abs/2510.17160
备注：Published at ECAI-2025
【13】In-situ Autoguidance: Eliciting Self-Correction in Diffusion Models
标题：现场自动引导：在扩散模型中激发自我纠正
链接：https://arxiv.org/abs/2510.17136
备注：6 pages, 3 figures. ICML 2025 Workshop submission
【14】Hephaestus: Mixture Generative Modeling with Energy Guidance for Large-scale QoS Degradation
标题：Hepaestus：具有能量指导的混合生成建模用于大规模服务质量降级
链接：https://arxiv.org/abs/2510.17036
备注：62 pages, 19 figures, Neural Information Processing Systems (NeurIPS 2025)
【15】Graph4MM: Weaving Multimodal Learning with Structural Information
标题：Graph4 MM：利用结构信息编织多模式学习
链接：https://arxiv.org/abs/2510.16990
备注：ICML 2025
【16】One-step Diffusion Models with Bregman Density Ratio Matching
标题：Bregman密度比匹配的一步扩散模型
链接：https://arxiv.org/abs/2510.16983
备注：work in progress
【17】Domain Generalizable Continual Learning
标题：领域可推广的持续学习
链接：https://arxiv.org/abs/2510.16914
备注：25 pages
【18】DeepChem Equivariant: SE(3)-Equivariant Support in an Open-Source Molecular Machine Learning Library
标题：DeepChem Equivariant：SE（3）-开源分子机器学习库中的Equivariant支持
链接：https://arxiv.org/abs/2510.16897
备注：Presented at Machine Learning Symposium - BayLearn (2025)
【19】Simulation-free Structure Learning for Stochastic Dynamics
标题：随机动力学的无模拟结构学习
链接：https://arxiv.org/abs/2510.16656
【20】NeurIPT: Foundation Model for Neural Interfaces
标题：NeurAPT：神经接口的基础模型
链接：https://arxiv.org/abs/2510.16548
备注：Accepted by The Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025). Project Page: this https URL
【21】Colliding with Adversaries at ECML-PKDD 2025 Model Robustness Competition 1st Prize Solution
标题：ECML-PKDD 2025模型稳健性竞赛一等奖解决方案与对手发生冲突
链接：https://arxiv.org/abs/2510.16443
【22】Toward General Digraph Contrastive Learning: A Dual Spatial Perspective
标题：走向广义有向图对比学习：双重空间视角
链接：https://arxiv.org/abs/2510.16311
【23】One-Bit Quantization for Random Features Models
标题：随机特征模型的一比特量化
链接：https://arxiv.org/abs/2510.16250
【24】AtomBench: A Benchmark for Generative Atomic Structure Models using GPT, Diffusion, and Flow Architectures
标题：AtomBench：使用GPT、扩散和流架构的生成原子结构模型的基准
链接：https://arxiv.org/abs/2510.16165
【25】Zeroth-Order Sharpness-Aware Learning with Exponential Tilting
标题：具有指数倾斜的零阶敏锐度感知学习
链接：https://arxiv.org/abs/2510.16157
【26】Compressing Many-Shots in In-Context Learning
标题：在上下文学习中压缩多镜头
链接：https://arxiv.org/abs/2510.16092
【27】Differentiable, Bit-shifting, and Scalable Quantization without training neural network from scratch
标题：无需从头开始训练神经网络即可进行可区分、移位和可扩展量化
链接：https://arxiv.org/abs/2510.16088
【28】Continual Knowledge Consolidation LORA for Domain Incremental Learning
标题：用于领域增量学习的持续知识整合LORA
链接：https://arxiv.org/abs/2510.16077
【29】Beyond Accuracy: Are Time Series Foundation Models Well-Calibrated?
标题：超越准确性：时间序列基金会模型是否经过良好校准？
链接：https://arxiv.org/abs/2510.16060
【30】Vector Quantization in the Brain: Grid-like Codes in World Models
标题：大脑中的载体量化：世界模型中的网格式代码
链接：https://arxiv.org/abs/2510.16039
【31】Nondeterminism-Aware Optimistic Verification for Floating-Point Neural Networks
标题：浮点神经网络的不确定性意识乐观验证
链接：https://arxiv.org/abs/2510.16028
备注：17 pages, 7 figures
【32】Unifying Polymer Modeling and Design via a Conformation-Centric Generative Foundation Model
标题：通过以形态为中心的生成基础模型统一聚合物建模和设计
链接：https://arxiv.org/abs/2510.16023
【33】Using Kolmogorov-Smirnov Distance for Measuring Distribution Shift in Machine Learning
标题：利用Kolmogorov-Smirnov距离度量机器学习中的分布偏移
链接：https://arxiv.org/abs/2510.15996
【34】Cross-dataset Multivariate Time-series Model for Parkinson's Diagnosis via Keyboard Dynamics
标题：通过键盘动力学诊断帕金森氏症的跨数据集多元时间序列模型
链接：https://arxiv.org/abs/2510.15950
备注：Proceedings of the Workshop on Artificial Intelligence for Biomedical Data (AIBio 2025), 28th European Conference on Artificial Intelligence 2025, Springer CCIS
【35】Estimating Orbital Parameters of Direct Imaging Exoplanet Using Neural Network
标题：利用神经网络估计直接成像系外行星的轨道参数
链接：https://arxiv.org/abs/2510.17459
【36】DFNN: A Deep Fréchet Neural Network Framework for Learning Metric-Space-Valued Responses
标题：DFNN：用于学习度量空间值响应的Deep Fréchet神经网络框架
链接：https://arxiv.org/abs/2510.17072
【37】A Topological Approach to Parameterizing Deep Hedging Networks
标题：深度对冲网络参数化的拓扑方法
链接：https://arxiv.org/abs/2510.16938
【38】A three-step machine learning approach to predict market bubbles with financial news
标题：通过金融新闻预测市场泡沫的三步机器学习方法
链接：https://arxiv.org/abs/2510.16636
【39】Accelerated Learning on Large Scale Screens using Generative Library Models
标题：使用生成库模型在大规模屏幕上加速学习
链接：https://arxiv.org/abs/2510.16612
【40】Personalized Collaborative Learning with Affinity-Based Variance Reduction
标题：基于亲和力的方差减少的个性化协作学习
链接：https://arxiv.org/abs/2510.16232
【41】Quantum and Classical Machine Learning in Decentralized Finance: Comparative Evidence from Multi-Asset Backtesting of Automated Market Makers
标题：去中心化金融中的量子和经典机器学习：来自自动做市商多资产回测的比较证据
链接：https://arxiv.org/abs/2510.15903
其他(80篇)
【1】Unbiased Gradient Low-Rank Projection
标题：无偏梯度低等级投影
链接：https://arxiv.org/abs/2510.17802
【2】Glyph: Scaling Context Windows via Visual-Text Compression
标题：字形：通过视觉文本压缩扩展上下文窗口
链接：https://arxiv.org/abs/2510.17800
【3】Efficient Tensor Completion Algorithms for Highly Oscillatory Operators
标题：高振动性运算符的高效张量完成算法
链接：https://arxiv.org/abs/2510.17734
【4】Train for Truth, Keep the Skills: Binary Retrieval-Augmented Reward Mitigates Hallucinations
标题：训练真理，保持技能：二进制检索增强奖励减轻幻觉
链接：https://arxiv.org/abs/2510.17733
【5】Closing the Sim2Real Performance Gap in RL
标题：缩小RL中的Sim 2Real性能差距
链接：https://arxiv.org/abs/2510.17709
【6】RESample: A Robust Data Augmentation Framework via Exploratory Sampling for Robotic Manipulation
标题：RESample：通过探索性采样实现机器人操纵的稳健数据增强框架
链接：https://arxiv.org/abs/2510.17640
备注：9 pages,7 figures, submitted to ICRA2026
【7】How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime?
标题：标签噪音梯度下降如何改善低SNR条件下的通用性？
链接：https://arxiv.org/abs/2510.17526
备注：40 pages
【8】The Parameterized Complexity of Computing the VC-Dimension
标题：VC维计算的参数化复杂性
链接：https://arxiv.org/abs/2510.17451
备注：To appear in the proceedings of NeurIPS 2025
【9】RINS-T: Robust Implicit Neural Solvers for Time Series Linear Inverse Problems
标题：RINS-T：时间序列线性反问题的鲁棒隐式神经求解器
链接：https://arxiv.org/abs/2510.17396
备注：Accepted to IEEE Transactions on Instrumentation and Measurement
【10】Finite-Time Bounds for Average-Reward Fitted Q-Iteration
标题：平均回报匹配Q迭代的延迟时间界限
链接：https://arxiv.org/abs/2510.17391
【11】Exploration via Feature Perturbation in Contextual Bandits
标题：背景盗贼的特征扰动探索
链接：https://arxiv.org/abs/2510.17390
备注：Accepted at NeurIPS 2025 (spotlight)
【12】Breaking and Fixing Defenses Against Control-Flow Hijacking in Multi-Agent Systems
标题：打破和修复多智能体系统中针对控制流劫持的防御
链接：https://arxiv.org/abs/2510.17276
【13】D2C-HRHR: Discrete Actions with Double Distributional Critics for High-Risk-High-Return Tasks
标题：D2C-HRHR：针对高风险高回报任务的具有双重分布批评者的离散行动
链接：https://arxiv.org/abs/2510.17212
【14】A Standardized Benchmark for Machine-Learned Molecular Dynamics using Weighted Ensemble Sampling
标题：使用加权整体抽样的机器学习分子动力学标准化基准
链接：https://arxiv.org/abs/2510.17187
备注：37 Pages (Main Text), 10 Figures, Submitted to Journal of Physical Chemistry B
【15】Verification-Aware Planning for Multi-Agent Systems
标题：多智能体系统的验证感知规划
链接：https://arxiv.org/abs/2510.17109
备注：Submission for ARR Oct
【16】Data Reliability Scoring
标题：数据可靠性评分
链接：https://arxiv.org/abs/2510.17085
备注：39 pages, 5 figures
【17】Bitwidth-Specific Logarithmic Arithmetic for Future Hardware-Accelerated Training
标题：面向未来硬件加速训练的特定比特算术
链接：https://arxiv.org/abs/2510.17058
【18】MuonBP: Faster Muon via Block-Periodic Orthogonalization
标题：MuonBP：通过块周期子化实现更快的Muon
链接：https://arxiv.org/abs/2510.16981
【19】Schrödinger Bridge Mamba for One-Step Speech Enhancement
标题：Schrödinger Bridge Mamba用于一步语音增强
链接：https://arxiv.org/abs/2510.16834
备注：5 pages, 1 figure
【20】Finding Manifolds With Bilinear Autoencoders
标题：用双线性自编码器求流形
链接：https://arxiv.org/abs/2510.16820
【21】Trace Regularity PINNs: Enforcing $\mathrm{H}^{\frac{1}{2}}(\partial Ω)$ for Boundary Data
链接：https://arxiv.org/abs/2510.16817
【22】Efficient High-Accuracy PDEs Solver with the Linear Attention Neural Operator
标题：具有线性注意力神经运算符的高效高准确性PDEs求解器
链接：https://arxiv.org/abs/2510.16816
备注：31 pages, 8 figures
【23】Computational Budget Should Be Considered in Data Selection
标题：数据选择应考虑计算预算
链接：https://arxiv.org/abs/2510.16806
【24】More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
标题：少花钱多：高效编码代理的回合控制策略的实证研究
链接：https://arxiv.org/abs/2510.16786
【25】Connecting Domains and Contrasting Samples: A Ladder for Domain Generalization
标题：连接领域和对比样本：领域概括的阶梯
链接：https://arxiv.org/abs/2510.16704
备注：Accepted by KDD 2025
【26】On the Granularity of Causal Effect Identifiability
标题：论因果效应可识别性的粒度
链接：https://arxiv.org/abs/2510.16703
【27】High-Dimensional Privacy-Utility Dynamics of Noisy Stochastic Gradient Descent on Least Squares
标题：最小平方上有噪随机梯度下降的多维隐私-效用动力学
链接：https://arxiv.org/abs/2510.16687
【28】Active Target Discovery under Uninformative Prior: The Power of Permanent and Transient Memory
标题：无信息先验下的主动目标发现：永久和暂时记忆的力量
链接：https://arxiv.org/abs/2510.16676
备注：32 pages, 20 figures, Accepted to NeurIPS 2025
【29】Evaluating protein binding interfaces with PUMBA
标题：用DEMABA评价蛋白质结合界面
链接：https://arxiv.org/abs/2510.16674
【30】All You Need is One: Capsule Prompt Tuning with a Single Vector
标题：只需一个即可：使用单个载体进行胶囊提示调谐
链接：https://arxiv.org/abs/2510.16670
备注：NeurIPS 2025
【31】Robust Dynamic Staffing with Predictions
标题：带预测的鲁棒动态人员配置
链接：https://arxiv.org/abs/2510.16663
【32】Safire: Similarity Framework for Visualization Retrieval
标题：Safire：可视化检索相似性框架
链接：https://arxiv.org/abs/2510.16662
备注：To appear in IEEE VIS 2025
【33】On the Impossibility of Retrain Equivalence in Machine Unlearning
标题：论机器取消学习中重新训练等效的不可能性
链接：https://arxiv.org/abs/2510.16629
备注：Code available at this https URL
【34】Symmetry and Generalisation in Neural Approximations of Renormalisation Transformations
标题：重正化变换的神经逼近中的对称性和推广
链接：https://arxiv.org/abs/2510.16591
【35】Urban-R1: Reinforced MLLMs Mitigate Geospatial Biases for Urban General Intelligence
标题：Urban-R1：加强MLLM减轻城市通用智能的地理空间偏见
链接：https://arxiv.org/abs/2510.16555
【36】eDCF: Estimating Intrinsic Dimension using Local Connectivity
标题：eDCF：使用本地连接性估计内在维度
链接：https://arxiv.org/abs/2510.16513
备注：58 pages (35 (main) + 23 (appendix)), 54 figures (27 (main) + 27 (appendix))
【37】Automated Composition of Agents: A Knapsack Approach for Agentic Component Selection
标题：代理的自动组合：统计组件选择的背包方法
链接：https://arxiv.org/abs/2510.16499
备注：Accepted to NeurIPS 2025 Conference
【38】Edge-Based Speech Transcription and Synthesis for Kinyarwanda and Swahili Languages
标题：基尼亚卢旺达语和斯瓦希里语基于边缘的语音转录和合成
链接：https://arxiv.org/abs/2510.16497
【39】VIPAMIN: Visual Prompt Initialization via Embedding Selection and Subspace Expansion
标题：VIPAMIN：通过嵌入选择和子空间扩展的视觉提示收件箱
链接：https://arxiv.org/abs/2510.16446
备注：NeurIPS 2025
【40】The Burden of Interactive Alignment with Inconsistent Preferences
标题：偏好不一致的互动一致的负担
链接：https://arxiv.org/abs/2510.16368
备注：Published as a conference paper at NeurIPS 2025
【41】MLCPD: A Unified Multi-Language Code Parsing Dataset with Universal AST Schema
标题：MLCPD：具有通用AST模式的统一多语言代码解析数据集
链接：https://arxiv.org/abs/2510.16357
备注：12 pages, 7 figures, 4 tables, 2 algorithms, and 34 references. HuggingFace: this https URL GitHub: this https URL
【42】RL makes MLLMs see better than SFT
标题：RL让MLLM比SFT看得更好
链接：https://arxiv.org/abs/2510.16333
【43】Memorizing Long-tail Data Can Help Generalization Through Composition
标题：精简长尾数据可以通过合成帮助概括
链接：https://arxiv.org/abs/2510.16322
备注：30 pages
【44】Scaffold-Aware Generative Augmentation and Reranking for Enhanced Virtual Screening
标题：支架感知生成增强和重新排序以增强虚拟筛选
链接：https://arxiv.org/abs/2510.16306
【45】Disentangling Hyperedges through the Lens of Category Theory
标题：超边的范畴论解读
链接：https://arxiv.org/abs/2510.16289
备注：Accepted to NeurIPS 2025
【46】What Limits Agentic Systems Efficiency?
标题：是什么限制了大型系统的效率？
链接：https://arxiv.org/abs/2510.16276
备注：27 pages, 15 figures
【47】Protein Folding with Neural Ordinary Differential Equations
标题：利用神经常微方程折叠蛋白质
链接：https://arxiv.org/abs/2510.16253
【48】ScholarEval: Research Idea Evaluation Grounded in Literature
标题：学者评价：立足于文学的研究理念评价
链接：https://arxiv.org/abs/2510.16234
【49】Explore-then-Commit for Nonstationary Linear Bandits with Latent Dynamics
标题：具有潜在动力学的非平稳线性盗贼探索后投入
链接：https://arxiv.org/abs/2510.16208
【50】Expressive Reward Synthesis with the Runtime Monitoring Language
标题：基于情感监控语言的表达性奖励合成
链接：https://arxiv.org/abs/2510.16185
【51】Alignment is Localized: A Causal Probe into Preference Layers
标题：对齐是局部化的：偏好层的因果关系探索
链接：https://arxiv.org/abs/2510.16167
【52】Expert Merging in Sparse Mixture of Experts with Nash Bargaining
标题：纳什讨价还价的稀疏专家混合中的专家合并
链接：https://arxiv.org/abs/2510.16138
备注：10 pages in the main text. Under Review
【53】Aria Gen 2 Pilot Dataset
标题：Aria Gen 2飞行员数据集
链接：https://arxiv.org/abs/2510.16134
【54】Narrowing Action Choices with AI Improves Human Sequential Decisions
标题：人工智能缩小行动选择范围改善人类顺序决策
链接：https://arxiv.org/abs/2510.16097
备注：Accepted at the Human-AI Complementarity for Decision Making Workshop 2025 by the NSF AI Institute for Societal Decision Making
【55】Near-Equilibrium Propagation training in nonlinear wave systems
标题：非线性波动系统的近平衡传播训练
链接：https://arxiv.org/abs/2510.16084
备注：7 figures
【56】Algorithmic Fairness in AI Surrogates for End-of-Life Decision-Making
标题：人工智能替代生命终结决策的数学公平性
链接：https://arxiv.org/abs/2510.16056
【57】GUIrilla: A Scalable Framework for Automated Desktop UI Exploration
标题：GUIrilla：一个可扩展的自动桌面UI探索框架
链接：https://arxiv.org/abs/2510.16051
备注：22 pages
【58】In the Mood to Exclude: Revitalizing Trespass to Chattels in the Era of GenAI Scraping
标题：在情绪上的破坏：振兴侵入动产在时代的GenAI刮
链接：https://arxiv.org/abs/2510.16049
【59】Open Shouldn't Mean Exempt: Open-Source Exceptionalism and Generative AI
标题：开放不应该意味着豁免：开源例外论和生成式AI
链接：https://arxiv.org/abs/2510.16048
【60】RoBCtrl: Attacking GNN-Based Social Bot Detectors via Reinforced Manipulation of Bots Control Interaction
标题：RoBck：通过加强机器人控制交互的操纵来攻击基于GNN的社交机器人检测器
链接：https://arxiv.org/abs/2510.16035
备注：27 pages, 10 figures
【61】Disaster Management in the Era of Agentic AI Systems: A Vision for Collective Human-Machine Intelligence for Augmented Resilience
标题：人工智能系统时代的灾害管理：增强复原力的集体人机智能愿景
链接：https://arxiv.org/abs/2510.16034
【62】A tutorial on discovering and quantifying the effect of latent causal sources of multimodal EHR data
标题：关于发现和量化多模式EHR数据潜在因果来源影响的教程
链接：https://arxiv.org/abs/2510.16026
【63】Predict Training Data Quality via Its Geometry in Metric Space
标题：利用度量空间中训练数据的几何性质预测训练数据的质量
链接：https://arxiv.org/abs/2510.15970
备注：Accepted to the NeurIPS 2025 Workshop on New Perspectives in Graph Machine Learning
【64】Fire-EnSF: Wildfire Spread Data Assimilation using Ensemble Score Filter
标题：Fire-EnSF：使用Ensemble Score过滤器进行野火传播数据同化
链接：https://arxiv.org/abs/2510.15954
【65】Attention to Non-Adopters
标题：注意非吸烟者
链接：https://arxiv.org/abs/2510.15951
【66】Lean Finder: Semantic Search for Mathlib That Understands User Intents
标题：精益收件箱：理解用户意图的Mathlib语义搜索
链接：https://arxiv.org/abs/2510.15940
【67】FlexLink: Boosting your NVLink Bandwidth by 27% without accuracy concern
标题：FlexLink：将NVLink带宽提高27%，而无需担心准确性
链接：https://arxiv.org/abs/2510.15882
【68】Multimodal Chip Physical Design Engineer Assistant
标题：多模式芯片物理设计工程师助理
链接：https://arxiv.org/abs/2510.15872
【69】A Semantic Generalization of Shannon's Information Theory and Applications
标题：香农信息理论的语义概括及其应用
链接：https://arxiv.org/abs/2510.15871
备注：45 pages, 18 Figures, a review paper
【70】Advancing Routing-Awareness in Analog ICs Floorplanning
标题：提高模拟IC布局规划中的布线意识
链接：https://arxiv.org/abs/2510.15387
【71】Non-asymptotic error bounds for probability flow ODEs under weak log-concavity
标题：弱log条件下概率流ODE的非渐进误差界
链接：https://arxiv.org/abs/2510.17608
【72】Kernel-Based Nonparametric Tests For Shape Constraints
标题：形状约束的基于核的非参数测试
链接：https://arxiv.org/abs/2510.16745
备注：31 pages, 1 figure
【73】Local regression on path spaces with signature metrics
标题：具有特征度量的路径空间上的局部回归
链接：https://arxiv.org/abs/2510.16728
【74】Infinite Neural Operators: Gaussian processes on functions
标题：无限神经算子：函数上的高斯过程
链接：https://arxiv.org/abs/2510.16675
备注：Accepted at the Conference on Neural Information Processing Systems (NeurIPS) 2025
【75】Multi-Marginal Schrödinger Bridge Matching
标题：多边缘薛定汉桥匹配
链接：https://arxiv.org/abs/2510.16587
【76】A Relative Error-Based Evaluation Framework of Heterogeneous Treatment Effect Estimators
标题：基于相对误差的异类治疗效果估计器评估框架
链接：https://arxiv.org/abs/2510.16419
【77】Synergizing chemical and AI communities for advancing laboratories of the future
标题：协同化学和人工智能社区推进未来实验室
链接：https://arxiv.org/abs/2510.16293
【78】Cash Flow Underwriting with Bank Transaction Data: Advancing MSME Financial Inclusion in Malaysia
标题：利用银行交易数据进行现金流承保：推进马来西亚中小微企业金融包容性
链接：https://arxiv.org/abs/2510.16066
备注：Accepted at the FinREM Workshop, ICAIF 2025
【79】Data for Inclusion: The Redistributive Power of Data Economics
标题：数据包容性：数据经济学的再分配力量
链接：https://arxiv.org/abs/2510.16009
【80】Convolutional Attention in Betting Exchange Markets
标题：博彩交易市场中的卷积关注
链接：https://arxiv.org/abs/2510.16008
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
Python社区是高质量的Python/Django开发社区
本文地址：http://www.python88.com/topic/188153
本文地址：http://www.python88.com/topic/188153