![机器学习学术速递[10.14]](https://cdn.10100.com/user/90e220ca11dd41cbaa03eba41e3d7f79.png?x-oss-process=style/180x)
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355篇
大模型相关(49篇)
【1】QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs
标题：QeRL：超越效率--针对LLM的量化增强强化学习
链接：https://arxiv.org/abs/2510.11696
备注：Code is available at this https URL
摘要：We propose QeRL, a Quantization-enhanced Reinforcement Learning framework for large language models (LLMs). While RL is essential for LLMs' reasoning capabilities, it is resource-intensive, requiring substantial GPU memory and long rollout durations. QeRL addresses these issues by combining NVFP4 quantization with Low-Rank Adaptation (LoRA), accelerating rollout phase of RL while reducing memory overhead. Beyond efficiency, our findings show that quantization noise increases policy entropy, enhancing exploration, and enabling the discovery of better strategies during RL. To further optimize exploration, QeRL introduces an Adaptive Quantization Noise (AQN) mechanism, which dynamically adjusts noise during training. Experiments demonstrate that QeRL delivers over 1.5 times speedup in the rollout phase. Moreover, this is the first framework to enable RL training of a 32B LLM on a single H100 80GB GPU, while delivering overall speedups for RL training. It also achieves faster reward growth and higher final accuracy than 16-bit LoRA and QLoRA, while matching the performance of full-parameter fine-tuning on mathematical benchmarks such as GSM8K (90.8%) and MATH 500 (77.4%) in the 7B model. These results establish QeRL as an efficient and effective framework for RL training in LLMs.
【2】Representation-Based Exploration for Language Models: From Test-Time to Post-Training
标题：基于表示的语言模型探索：从测试时到训练后
链接：https://arxiv.org/abs/2510.11686
备注：Website and code: this https URL
摘要：Reinforcement learning (RL) promises to expand the capabilities of language models, but it is unclear if current RL techniques promote the discovery of novel behaviors, or simply sharpen those already present in the base model. In this paper, we investigate the value of deliberate exploration -- explicitly incentivizing the model to discover novel and diverse behaviors -- and aim to understand how the knowledge in pre-trained models can guide this search. Our main finding is that exploration with a simple, principled, representation-based bonus derived from the pre-trained language model's hidden states significantly improves diversity and pass@k rates -- both for post-training, and in a novel inference-time scaling setting we introduce. For inference-time, exploration with representation-based diversity improves efficiency, consistently improving pass@k rates across a variety of models and reasoning tasks. For example, for Qwen-2.5-14b-Instruct we obtain over 50% improvement in verifier efficiency on almost all tasks. For post-training, we show that integrating this exploration strategy into an RL pipeline improves reasoning performance over that of the initial model and over standard RL post-training. For example, on AIME 2024, our post-trained Qwen-2.5-7b-Instruct's pass@80 matches the pass@256 of GRPO on the same model, demonstrating a 3x improvement in test-time sample efficiency. Overall, our findings suggest that deliberate exploration -- with the right notion of diversity -- is a practical path toward discovery of new behaviors beyond sharpening.
【3】Boundary-Guided Policy Optimization for Memory-efficient RL of Diffusion Large Language Models
标题：扩散大语言模型内存高效RL的边界引导策略优化
链接：https://arxiv.org/abs/2510.11683
摘要：A key challenge in applying reinforcement learning (RL) to diffusion large language models (dLLMs) lies in the intractability of their likelihood functions, which are essential for the RL objective, necessitating corresponding approximation in each training step. While existing methods approximate the log-likelihoods by their evidence lower bounds (ELBOs) via customized Monte Carlo (MC) sampling, the forward computational graphs of all MC samples need to be retained for the gradient computation of non-linear terms in the RL objective, resulting in significant memory overhead. This constraint restricts feasible sample sizes, leading to imprecise likelihood approximations and ultimately distorting the RL objective. To overcome this limitation, we propose \emph{Boundary-Guided Policy Optimization} (BGPO), a memory-efficient RL algorithm that maximizes a specially constructed lower bound of the ELBO-based objective. This lower bound is carefully designed to satisfy two key properties: (1) Linearity: it is formulated in a linear sum where each term depends only on a single MC sample, thereby enabling gradient accumulation across samples and ensuring constant memory usage; (2) Equivalence: Both the value and gradient of this lower bound are equal to those of the ELBO-based objective in on-policy training, making it also an effective approximation for the original RL objective. These properties allow BGPO to adopt a large MC sample size, resulting in more accurate likelihood approximations and improved RL objective estimation, which in turn leads to enhanced performance. Experiments show that BGPO significantly outperforms previous RL algorithms for dLLMs in math problem solving, code generation, and planning tasks.
【4】ReLook: Vision-Grounded RL with a Multimodal LLM Critic for Agentic Web Coding
标题：ReLook：基于视觉的RL，具有针对大型Web编码的多模式LLM评论
链接：https://arxiv.org/abs/2510.11498
摘要：While Large Language Models (LLMs) excel at algorithmic code generation, they struggle with front-end development, where correctness is judged on rendered pixels and interaction. We present ReLook, an agentic, vision-grounded reinforcement learning framework that empowers an agent to close a robust generate--diagnose--refine loop by invoking a multimodal LLM (MLLM) as a tool. During training, the agent uses the MLLM-in-the-loop both as a visual critic--scoring code with screenshots--and as a source of actionable, vision-grounded feedback; a strict zero-reward rule for invalid renders anchors renderability and prevents reward hacking. To prevent behavioral collapse, we introduce Forced Optimization, a strict acceptance rule that admits only improving revisions, yielding monotonically better trajectories. At inference, we decouple the critic and run a lightweight, critic-free self-edit cycle, keeping latency comparable to base decoding while retaining most of the gains. Across three widely used benchmarks, ReLook consistently outperforms strong baselines in vision-grounded front-end code generation, highlighting the benefits of agentic perception, visual rewards, and training-inference decoupling.
【5】Leveraging LLMs for Semi-Automatic Corpus Filtration in Systematic Literature Reviews
标题：在系统性文献评论中利用LLM进行半自动数据库过滤
链接：https://arxiv.org/abs/2510.11409
摘要：The creation of systematic literature reviews (SLR) is critical for analyzing the landscape of a research field and guiding future research directions. However, retrieving and filtering the literature corpus for an SLR is highly time-consuming and requires extensive manual effort, as keyword-based searches in digital libraries often return numerous irrelevant publications. In this work, we propose a pipeline leveraging multiple large language models (LLMs), classifying papers based on descriptive prompts and deciding jointly using a consensus scheme. The entire process is human-supervised and interactively controlled via our open-source visual analytics web interface, LLMSurver, which enables real-time inspection and modification of model outputs. We evaluate our approach using ground-truth data from a recent SLR comprising over 8,000 candidate papers, benchmarking both open and commercial state-of-the-art LLMs from mid-2024 and fall 2025. Results demonstrate that our pipeline significantly reduces manual effort while achieving lower error rates than single human annotators. Furthermore, modern open-source models prove sufficient for this task, making the method accessible and cost-effective. Overall, our work demonstrates how responsible human-AI collaboration can accelerate and enhance systematic literature reviews within academic workflows.
【6】Medical Interpretability and Knowledge Maps of Large Language Models
标题：大型语言模型的医学可解释性和知识地图
链接：https://arxiv.org/abs/2510.11390
备注：29 pages, 34 figures, 5 tables
摘要：We present a systematic study of medical-domain interpretability in Large Language Models (LLMs). We study how the LLMs both represent and process medical knowledge through four different interpretability techniques: (1) UMAP projections of intermediate activations, (2) gradient-based saliency with respect to the model weights, (3) layer lesioning/removal and (4) activation patching. We present knowledge maps of five LLMs which show, at a coarse-resolution, where knowledge about patient's ages, medical symptoms, diseases and drugs is stored in the models. In particular for Llama3.3-70B, we find that most medical knowledge is processed in the first half of the model's layers. In addition, we find several interesting phenomena: (i) age is often encoded in a non-linear and sometimes discontinuous manner at intermediate layers in the models, (ii) the disease progression representation is non-monotonic and circular at certain layers of the model, (iii) in Llama3.3-70B, drugs cluster better by medical specialty rather than mechanism of action, especially for Llama3.3-70B and (iv) Gemma3-27B and MedGemma-27B have activations that collapse at intermediate layers but recover by the final layers. These results can guide future research on fine-tuning, un-learning or de-biasing LLMs for medical tasks by suggesting at which layers in the model these techniques should be applied.
【7】When Does Supervised Training Pay Off? The Hidden Economics of Object Detection in the Era of Vision-Language Models
标题：监督训练什么时候有回报？视觉语言模型时代对象检测的隐性经济学
链接：https://arxiv.org/abs/2510.11302
备注：23 pages, 4 figures, 4 tables
摘要：Object detection systems have traditionally relied on supervised learning with manually annotated bounding boxes, achieving high accuracy at the cost of substantial annotation investment. The emergence of Vision-Language Models (VLMs) offers an alternative paradigm enabling zero-shot detection through natural language queries, eliminating annotation requirements but operating with reduced accuracy. This paper presents the first comprehensive cost-effectiveness analysis comparing supervised detection (YOLO) with zero-shot VLM inference (Gemini Flash 2.5). Through systematic evaluation on 1,000 stratified COCO images and 200 diverse product images spanning consumer electronics and rare categories, combined with detailed Total Cost of Ownership modeling, we establish quantitative break-even thresholds governing architecture selection. Our findings reveal that supervised YOLO achieves 91.2% accuracy versus 68.5% for zero-shot Gemini on standard categories, representing a 22.7 percentage point advantage that costs $10,800 in annotation for 100-category systems. However, this advantage justifies investment only beyond 55 million inferences, equivalent to 151,000 images daily for one year. Zero-shot Gemini demonstrates 52.3% accuracy on diverse product categories (ranging from highly web-prevalent consumer electronics at 75-85% to rare specialized equipment at 25-40%) where supervised YOLO achieves 0% due to architectural constraints preventing detection of untrained classes. Cost per Correct Detection analysis reveals substantially lower per-detection costs for Gemini ($0.00050 vs $0.143) at 100,000 inferences despite accuracy deficits. We develop decision frameworks demonstrating that optimal architecture selection depends critically on deployment volume, category stability, budget constraints, and accuracy requirements rather than purely technical performance metrics.
【8】Vision-LLMs for Spatiotemporal Traffic Forecasting
标题：用于时空交通预测的视觉LLM
链接：https://arxiv.org/abs/2510.11282
摘要：Accurate spatiotemporal traffic forecasting is a critical prerequisite for proactive resource management in dense urban mobile networks. While Large Language Models (LLMs) have shown promise in time series analysis, they inherently struggle to model the complex spatial dependencies of grid-based traffic data. Effectively extending LLMs to this domain is challenging, as representing the vast amount of information from dense geographical grids can be inefficient and overwhelm the model's context. To address these challenges, we propose ST-Vision-LLM, a novel framework that reframes spatiotemporal forecasting as a vision-language fusion problem. Our approach leverages a Vision-LLM visual encoder to process historical global traffic matrices as image sequences, providing the model with a comprehensive global view to inform cell-level predictions. To overcome the inefficiency of LLMs in handling numerical data, we introduce an efficient encoding scheme that represents floating-point values as single tokens via a specialized vocabulary, coupled with a two-stage numerical alignment fine-tuning process. The model is first trained with Supervised Fine-Tuning (SFT) and then further optimized for predictive accuracy using Group Relative Policy Optimization (GRPO), a memory-efficient reinforcement learning method. Evaluations on real-world mobile traffic datasets demonstrate that ST-Vision-LLM outperforms existing methods by 15.6% in long-term prediction accuracy and exceeds the second-best baseline by over 30.04% in cross-domain few-shot scenarios. Our extensive experiments validate the model's strong generalization capabilities across various data-scarce environments.
【9】ENIGMA: The Geometry of Reasoning and Alignment in Large-Language Models
标题：谜：大型语言模型中推理和对齐的几何学
链接：https://arxiv.org/abs/2510.11278
备注：52 pages, 10 figures
摘要：We present Entropic Mutual-Information Geometry Large-Language Model Alignment (ENIGMA), a novel approach to Large-Language Model (LLM) training that jointly improves reasoning, alignment and robustness by treating an organisation's policies/principles as directions to move on a model's information manifold. Our single-loop trainer combines Group-Relative Policy Optimisation (GRPO), an on-policy, critic-free RL method with Chain-of-Thought (CoT)-format only rewards; a Self-Supervised Alignment with Mutual Information (SAMI)-style symmetric InfoNCE auxiliary; and an entropic Sinkhorn optimal-transport regulariser on hidden-state distributions to bound geometry drift. We also introduce infoNCE metrics that specialise to a standard MI lower bound under matched negatives to measure how strongly a model's CoT encodes these policies. These metrics include a Sufficiency Index (SI) that enables the selection and creation of principles that maximise downstream performance prior to training. In our experiments using small (1B) LLMs, high-SI principles predict steadier training dynamics and improved benchmark performance over GRPO ablations. Our information-geometry analysis of trained models validates desirable structural change in the manifold. These results support our hypothesis that reasoning, alignment, and robustness are projections of a single informationgeometric objective, and that models trained using ENIGMA demonstrate principled reasoning without the use of a reward model, offering a path to trusted capability
【10】Large Language Models Are Effective Code Watermarkers
标题：大型语言模型是有效的代码水印
链接：https://arxiv.org/abs/2510.11251
摘要：The widespread use of large language models (LLMs) and open-source code has raised ethical and security concerns regarding the distribution and attribution of source code, including unauthorized redistribution, license violations, and misuse of code for malicious purposes. Watermarking has emerged as a promising solution for source attribution, but existing techniques rely heavily on hand-crafted transformation rules, abstract syntax tree (AST) manipulation, or task-specific training, limiting their scalability and generality across languages. Moreover, their robustness against attacks remains limited. To address these limitations, we propose CodeMark-LLM, an LLM-driven watermarking framework that embeds watermark into source code without compromising its semantics or readability. CodeMark-LLM consists of two core components: (i) Semantically Consistent Embedding module that applies functionality-preserving transformations to encode watermark bits, and (ii) Differential Comparison Extraction module that identifies the applied transformations by comparing the original and watermarked code. Leveraging the cross-lingual generalization ability of LLM, CodeMark-LLM avoids language-specific engineering and training pipelines. Extensive experiments across diverse programming languages and attack scenarios demonstrate its robustness, effectiveness, and scalability.
【11】Neural Weight Compression for Language Models
标题：语言模型的神经权重压缩
链接：https://arxiv.org/abs/2510.11234
摘要：The efficient storage and transmission of language model weights is becoming increasingly important, as their scale and adoption continue to grow. However, as our understanding of this new data modality is limited, designing a good compression algorithm for language model weights heavily relies on manual, trial-and-error approaches. In this paper, we propose a learned compression framework that trains neural codecs directly from pretrained language model weights. Unlike conventional data (e.g., images), language model weights pose unique challenges: the sizes and shapes of weight tensors vary significantly, and the reconstruction quality must be judged by downstream model predictions rather than na\"ive MSE loss. To address this, we introduce Neural Weight Compression (NWC), a novel autoencoder-based neural codec tailored to model weight compression. The proposed method inherits the advantages of autoencoder-based codecs while incorporating three technical components: (1) column-wise tensor chunking and normalization; (2) an importance-aware training loss; (3) an inference-time error compensation mechanism guided by model outputs. Experiments on open-weight language models show that NWC achieves competitive or state-of-the-art accuracy-compression tradeoffs, with particularly strong results at 4-6 bit precisions where accuracy remains nearly on par with FP16 models.
【12】Discursive Circuits: How Do Language Models Understand Discourse Relations?
标题：话语回路：语言模型如何理解话语关系？
链接：https://arxiv.org/abs/2510.11210
备注：Accepted to EMNLP 2025 (Main Conference); 9 pages, 8 figures, 5 tables (20 pages, 12 figures, 14 tables including references and appendices)
摘要：Which components in transformer language models are responsible for discourse understanding? We hypothesize that sparse computational graphs, termed as discursive circuits, control how models process discourse relations. Unlike simpler tasks, discourse relations involve longer spans and complex reasoning. To make circuit discovery feasible, we introduce a task called Completion under Discourse Relation (CuDR), where a model completes a discourse given a specified relation. To support this task, we construct a corpus of minimal contrastive pairs tailored for activation patching in circuit discovery. Experiments show that sparse circuits ($\approx 0.2\%$ of a full GPT-2 model) recover discourse understanding in the English PDTB-based CuDR task. These circuits generalize well to unseen discourse frameworks such as RST and SDRT. Further analysis shows lower layers capture linguistic features such as lexical semantics and coreference, while upper layers encode discourse-level abstractions. Feature utility is consistent across frameworks (e.g., coreference supports Expansion-like relations).
【13】Efficient In-Memory Acceleration of Sparse Block Diagonal LLMs
标题：稀疏块对角LLM的高效内存内加速
链接：https://arxiv.org/abs/2510.11192
备注：8 pages, to appear in IEEE Cross-disciplinary Conference on Memory-Centric Computing (CCMCC)
摘要：Structured sparsity enables deploying large language models (LLMs) on resource-constrained systems. Approaches like dense-to-sparse fine-tuning are particularly compelling, achieving remarkable structured sparsity by reducing the model size by over 6.7x, while still maintaining acceptable accuracy. Despite this reduction, LLM inference, especially the decode stage being inherently memory-bound, is extremely expensive on conventional Von-Neumann architectures. Compute-in-memory (CIM) architectures mitigate this by performing computations directly in memory, and when paired with sparse LLMs, enable storing and computing the entire model in memory, eliminating the data movement on the off-chip bus and improving efficiency. Nonetheless, naively mapping sparse matrices onto CIM arrays leads to poor array utilization and diminished computational efficiency. In this paper, we present an automated framework with novel mapping and scheduling strategies to accelerate sparse LLM inference on CIM accelerators. By exploiting block-diagonal sparsity, our approach improves CIM array utilization by over 50%, achieving more than 4x reduction in both memory footprint and the number of required floating-point operations.
【14】Protein as a Second Language for LLMs
标题：蛋白质作为法学硕士的第二语言
链接：https://arxiv.org/abs/2510.11188
备注：Main paper: 9 pages, 6 figures. With references and appendix: 18 pages, 9 figures total. Submitted to ICLR 2026 (under review)
摘要：Deciphering the function of unseen protein sequences is a fundamental challenge with broad scientific impact, yet most existing methods depend on task-specific adapters or large-scale supervised fine-tuning. We introduce the "Protein-as-Second-Language" framework, which reformulates amino-acid sequences as sentences in a novel symbolic language that large language models can interpret through contextual exemplars. Our approach adaptively constructs sequence-question-answer triples that reveal functional cues in a zero-shot setting, without any further training. To support this process, we curate a bilingual corpus of 79,926 protein-QA instances spanning attribute prediction, descriptive understanding, and extended reasoning. Empirically, our method delivers consistent gains across diverse open-source LLMs and GPT-4, achieving up to 17.2% ROUGE-L improvement (average +7%) and even surpassing fine-tuned protein-specific language models. These results highlight that generic LLMs, when guided with protein-as-language cues, can outperform domain-specialized models, offering a scalable pathway for protein understanding in foundation models.
【15】Refining Hybrid Genetic Search for CVRP via Reinforcement Learning-Finetuned LLM
标题：通过强化学习改进CVRP的混合遗传搜索-Finetuned LLM
链接：https://arxiv.org/abs/2510.11121
摘要：While large language models (LLMs) are increasingly used as automated heuristic designers for vehicle routing problems (VRPs), current state-of-the-art methods predominantly rely on prompting massive, general-purpose models like GPT-4. This work challenges that paradigm by demonstrating that a smaller, specialized LLM, when meticulously fine-tuned, can generate components that surpass expert-crafted heuristics within advanced solvers. We propose RFTHGS, a novel Reinforcement learning (RL) framework for Fine-Tuning a small LLM to generate high-performance crossover operators for the Hybrid Genetic Search (HGS) solver, applied to the Capacitated VRP (CVRP). Our method employs a multi-tiered, curriculum-based reward function that progressively guides the LLM to master generating first compilable, then executable, and finally, superior-performing operators that exceed human expert designs. This is coupled with an operator caching mechanism that discourages plagiarism and promotes diversity during training. Comprehensive experiments show that our fine-tuned LLM produces crossover operators which significantly outperform the expert-designed ones in HGS. The performance advantage remains consistent, generalizing from small-scale instances to large-scale problems with up to 1000 nodes. Furthermore, RFTHGS exceeds the performance of leading neuro-combinatorial baselines, prompt-based methods, and commercial LLMs such as GPT-4o and GPT-4o-mini.
【16】Stronger Together: On-Policy Reinforcement Learning for Collaborative LLMs
标题：更强大：协作式LLM的政策强化学习
链接：https://arxiv.org/abs/2510.11062
摘要：Multi-agent systems (MAS) and reinforcement learning (RL) are widely used to enhance the agentic capabilities of large language models (LLMs). MAS improves task performance through role-based orchestration, while RL uses environmental rewards to learn stronger policies, such as GRPO-style optimization. However, applying on-policy RL to MAS remains underexplored and presents unique challenges. Algorithmically, standard GRPO grouping assumptions break down because prompts vary by role and by turn. System-wise, the training stack must support MAS-workflow rollouts and on-policy updates for both single-policy and multi-policy models. We propose AT-GRPO, which includes (i) an agent- and turn-wise grouped RL algorithm tailored to MAS and (ii) a training system that supports both single- and multi-policy regimes. Across game, planning, coding, and math tasks, AT-GRPO delivers substantial gains. On long-horizon planning, it increases accuracy from a 14.0 to 47.0 percent single-agent RL baseline to 96.0 to 99.5 percent. It also improves reasoning performance, with average gains of 3.87 to 7.62 percent on coding tasks and 9.0 to 17.93 percent on math. Code and environments are available at: https://github.com/pettingllms-ai/PettingLLMs.
【17】ABLEIST: Intersectional Disability Bias in LLM-Generated Hiring Scenarios
标题：ABLEIST：LLM生成的招聘场景中的交叉残疾偏见
链接：https://arxiv.org/abs/2510.10998
备注：28 pages, 11 figures, 16 tables. In submission
摘要：Large language models (LLMs) are increasingly under scrutiny for perpetuating identity-based discrimination in high-stakes domains such as hiring, particularly against people with disabilities (PwD). However, existing research remains largely Western-centric, overlooking how intersecting forms of marginalization--such as gender and caste--shape experiences of PwD in the Global South. We conduct a comprehensive audit of six LLMs across 2,820 hiring scenarios spanning diverse disability, gender, nationality, and caste profiles. To capture subtle intersectional harms and biases, we introduce ABLEIST (Ableism, Inspiration, Superhumanization, and Tokenism), a set of five ableism-specific and three intersectional harm metrics grounded in disability studies literature. Our results reveal significant increases in ABLEIST harms towards disabled candidates--harms that many state-of-the-art models failed to detect. These harms were further amplified by sharp increases in intersectional harms (e.g., Tokenism) for gender and caste-marginalized disabled candidates, highlighting critical blind spots in current safety tools and the need for intersectional safety evaluations of frontier models in high-stakes domains like hiring.
【18】Rediscovering Entropy Regularization: Adaptive Coefficient Unlocks Its Potential for LLM Reinforcement Learning
标题：重新发现熵正规化：自适应系数释放其LLM强化学习的潜力
链接：https://arxiv.org/abs/2510.10959
备注：16 pages, 4 figures
摘要：Reasoning ability has become a defining capability of Large Language Models (LLMs), with Reinforcement Learning with Verifiable Rewards (RLVR) emerging as a key paradigm to enhance it. However, RLVR training often suffers from policy entropy collapse, where the policy becomes overly deterministic, hindering exploration and limiting reasoning performance. While entropy regularization is a common remedy, its effectiveness is highly sensitive to the fixed coefficient, making it unstable across tasks and models. In this work, we revisit entropy regularization in RLVR and argue that its potential has been largely underestimated. Our analysis shows that (i) tasks of varying difficulty demand distinct exploration intensities, and (ii) balanced exploration may require the policy entropy to be maintained within a moderate range below its initial level. Therefore, we propose Adaptive Entropy Regularization (AER)--a framework that dynamically balances exploration and exploitation via three components: difficulty-aware coefficient allocation, initial-anchored target entropy, and dynamic global coefficient adjustment. Experiments on multiple mathematical reasoning benchmarks show that AER consistently outperforms baselines, improving both reasoning accuracy and exploration capability.
【19】FG-CLIP 2: A Bilingual Fine-grained Vision-Language Alignment Model
标题：FG-CLIP 2：双语细粒度视觉语言对齐模型
链接：https://arxiv.org/abs/2510.10921
摘要：Fine-grained vision-language understanding requires precise alignment between visual content and linguistic descriptions, a capability that remains limited in current models, particularly in non-English settings. While models like CLIP perform well on global alignment, they often struggle to capture fine-grained details in object attributes, spatial relations, and linguistic expressions, with limited support for bilingual comprehension. To address these challenges, we introduce FG-CLIP 2, a bilingual vision-language model designed to advance fine-grained alignment for both English and Chinese. Our approach leverages rich fine-grained supervision, including region-text matching and long-caption modeling, alongside multiple discriminative objectives. We further introduce the Textual Intra-modal Contrastive (TIC) loss to better distinguish semantically similar captions. Trained on a carefully curated mixture of large-scale English and Chinese data, FG-CLIP 2 achieves powerful bilingual performance. To enable rigorous evaluation, we present a new benchmark for Chinese multimodal understanding, featuring long-caption retrieval and bounding box classification. Extensive experiments on 29 datasets across 8 tasks show that FG-CLIP 2 outperforms existing methods, achieving state-of-the-art results in both languages. We release the model, code, and benchmark to facilitate future research on bilingual fine-grained alignment.
【20】Glance for Context: Learning When to Leverage LLMs for Node-Aware GNN-LLM Fusion
标题：上下文一瞥：学习何时利用LLM进行节点感知GNN-LLM融合
链接：https://arxiv.org/abs/2510.10849
【21】Is Implicit Knowledge Enough for LLMs? A RAG Approach for Tree-based Structures
标题：隐性知识对于法学硕士来说足够吗？基于树的结构的RAG方法
链接：https://arxiv.org/abs/2510.10806
备注：Waiting for Conference Response
【22】A Stochastic Differential Equation Framework for Multi-Objective LLM Interactions: Dynamical Systems Analysis with Code Generation Applications
标题：多目标LLM相互作用的随机方程框架：动态系统分析与代码生成应用
链接：https://arxiv.org/abs/2510.10739
备注：Peer-reviewed and accepted to the 39th Conference on Neural Information Processing Systems (NeurIPS 2025) DynaFront 2025 Workshop (this https URL)
【23】Merlin's Whisper: Enabling Efficient Reasoning in LLMs via Black-box Adversarial Prompting
标题：Merlin ' s Whisper：通过黑匣子对抗预算在LLM中实现高效推理
链接：https://arxiv.org/abs/2510.10528
【24】The Hidden DNA of LLM-Generated JavaScript: Structural Patterns Enable High-Accuracy Authorship Attribution
标题：LLM生成的JavaScript的隐藏DNA：结构模式实现高准确性作者归因
链接：https://arxiv.org/abs/2510.10493
【25】AnyBCQ: Hardware Efficient Flexible Binary-Coded Quantization for Multi-Precision LLMs
标题：AnyBCQ：用于多精度LLM的硬件高效灵活二进制编码量化
链接：https://arxiv.org/abs/2510.10467
【26】Rethinking LLM Evaluation: Can We Evaluate LLMs with 200x Less Data?
标题：重新思考LLM评估：我们可以用减少200倍的数据来评估LLM吗？
链接：https://arxiv.org/abs/2510.10457
备注：18 pages, 5 figures
【27】RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models
标题：推荐长凳：扎根语言模型中选择性拒绝的生成性评估
链接：https://arxiv.org/abs/2510.10390
【28】ArtPerception: ASCII Art-based Jailbreak on LLMs with Recognition Pre-test
标题：ArtPercept：LLC上的基于ASC艺术的越狱，并进行识别预测试
链接：https://arxiv.org/abs/2510.10281
备注：30 pages, 22 figures. This preprint has been accepted for publication in Elsevier JOURNAL OF NETWORK AND COMPUTER APPLICATIONS (JNCA)
【29】Simulating Viva Voce Examinations to Evaluate Clinical Reasoning in Large Language Models
标题：模拟Viva Voice检查以评估大型语言模型中的临床推理
链接：https://arxiv.org/abs/2510.10278
【30】Lost in the Middle: An Emergent Property from Information Retrieval Demands in LLMs
标题：LLM信息检索需求的一个新特性--“迷失在中间”
链接：https://arxiv.org/abs/2510.10276
【31】Reasoning-Enhanced Large Language Models for Molecular Property Prediction
标题：用于分子性质预测的推理增强大型语言模型
链接：https://arxiv.org/abs/2510.10248
【32】You only need 4 extra tokens: Synergistic Test-time Adaptation for LLMs
标题：您只需要4个额外的令牌：LLM的协同测试时适应
链接：https://arxiv.org/abs/2510.10223
备注：Under Review
【33】RLFR: Extending Reinforcement Learning for LLMs with Flow Environment
标题：WLFR：通过流环境扩展LLM的强化学习
链接：https://arxiv.org/abs/2510.10201
备注：Project Website: this https URL
【34】PermLLM: Learnable Channel Permutation for N:M Sparse Large Language Models
标题：PermLLM：N：M稀疏大型语言模型的可学习通道排列
链接：https://arxiv.org/abs/2510.10136
备注：Accepted by NeurIPS 2025
【35】Pharmacist: Safety Alignment Data Curation for Large Language Models against Harmful Fine-tuning
标题：药剂师：针对有害微调的大型语言模型的安全对齐数据修复
链接：https://arxiv.org/abs/2510.10085
【36】Efficient Onboard Vision-Language Inference in UAV-Enabled Low-Altitude Economy Networks via LLM-Enhanced Optimization
标题：通过LLM增强优化在支持无人机的低空经济网络中进行高效的机载视觉语言推理
链接：https://arxiv.org/abs/2510.10028
【37】Reinforcement Fine-Tuning of Flow-Matching Policies for Vision-Language-Action Models
标题：视觉-语言-动作模型流匹配策略的强化微调
链接：https://arxiv.org/abs/2510.09976
【38】Learning Bug Context for PyTorch-to-JAX Translation with LLMs
标题：使用LLM学习PyTorch-to-JAX翻译的漏洞上下文
链接：https://arxiv.org/abs/2510.09898
【39】ProxRouter: Proximity-Weighted LLM Query Routing for Improved Robustness to Outliers
标题：ProxRouter：邻近加权LLM查询路由，提高对离群值的鲁棒性
链接：https://arxiv.org/abs/2510.09852
【40】CALM: A Causal Analysis Language Model for Tabular Data in Complex Systems with Local Scores, Conditional Independence Tests, and Relation Attributes
标题：CALM：具有局部分数、条件独立性测试和关系属性的复杂系统中表格数据的因果分析语言模型
链接：https://arxiv.org/abs/2510.09846
【41】Large Language Models for Imbalanced Classification: Diversity makes the difference
标题：不平衡分类的大型语言模型：多样性带来差异
链接：https://arxiv.org/abs/2510.09783
【42】InterCorpRel-LLM: Enhancing Financial Relational Understanding with Graph-Language Models
标题：InterCorpRel-LLM：用图形语言模型增强财务关系理解
链接：https://arxiv.org/abs/2510.09735
【43】Evaluating LLM-Based Process Explanations under Progressive Behavioral-Input Reduction
标题：在渐进行为输入减少下评估基于LLM的流程解释
链接：https://arxiv.org/abs/2510.09732
备注：12 pages, 2 figures, 3 tables; to appear in Enterprise Design, Operations, and Computing. EDOC 2025 Workshops, Lecture Notes in Business Information Processing (LNBIP), Springer, 2025. Part of 29th International Conference on Enterprise Design, Operations, and Computing (EDOC)
【44】ICL-Router: In-Context Learned Model Representations for LLM Routing
标题：ICL路由器：LLM路由的上下文学习模型表示
链接：https://arxiv.org/abs/2510.09719
【45】All Code, No Thought: Current Language Models Struggle to Reason in Ciphered Language
标题：全代码，不思考：当前语言模型在Cipbitt语言中难以推理
链接：https://arxiv.org/abs/2510.09714
【46】Using LLMs to Directly Guess Conditional Expectations Can Improve Efficiency in Causal Estimation
标题：使用LLM直接猜测条件期望可以提高因果估计的效率
链接：https://arxiv.org/abs/2510.09684
【47】LMCache: An Efficient KV Cache Layer for Enterprise-Scale LLM Inference
标题：LMCache：用于大规模LLM推理的高效KV缓存层
链接：https://arxiv.org/abs/2510.09665
【48】Domain-Specific Constitutional AI: Enhancing Safety in LLM-Powered Mental Health Chatbots
标题：特定领域的宪法AI：提高LLM-Powered心理健康聊天机器人的安全性
链接：https://arxiv.org/abs/2509.16444
【49】Integrating Large Language Models and Reinforcement Learning for Sentiment-Driven Quantitative Trading
标题：集成大型语言模型和强化学习以实现情绪驱动的量化交易
链接：https://arxiv.org/abs/2510.10526
Graph相关(图学习|图神经网络|图优化等)(22篇)
【1】Lecture Notes on Verifying Graph Neural Networks
标题：字形图神经网络课堂笔记
链接：https://arxiv.org/abs/2510.11617
【2】Query-Specific GNN: A Comprehensive Graph Representation Learning Method for Retrieval Augmented Generation
标题：特定于查询的GNN：一种用于检索增强生成的综合图表示学习方法
链接：https://arxiv.org/abs/2510.11541
【3】Multi-View Graph Feature Propagation for Privacy Preservation and Feature Sparsity
标题：用于隐私保护和特征稀疏性的多视图图特征传播
链接：https://arxiv.org/abs/2510.11347
【4】Event-Aware Prompt Learning for Dynamic Graphs
标题：动态图的事件感知提示学习
链接：https://arxiv.org/abs/2510.11339
备注：Under review
【5】Learning the Structure of Connection Graphs
标题：学习连接图的结构
链接：https://arxiv.org/abs/2510.11245
【6】Enforcing convex constraints in Graph Neural Networks
标题：在图神经网络中强制凸约束
链接：https://arxiv.org/abs/2510.11227
【7】Graph Neural Network-Based Multicast Routing for On-Demand Streaming Services in 6G Networks
标题：6G网络中基于图神经网络的按需流媒体服务多播路由
链接：https://arxiv.org/abs/2510.11109
【8】Conformal Inference for Time Series over Graphs
标题：图上时间序列的保形推理
链接：https://arxiv.org/abs/2510.11049
【9】HeroFilter: Adaptive Spectral Graph Filter for Varying Heterophilic Relations
标题：Hero滤镜：用于变化异嗜关系的自适应谱图滤镜
链接：https://arxiv.org/abs/2510.10864
【10】GraphTARIF: Linear Graph Transformer with Augmented Rank and Improved Focus
标题：GraphTARIF：具有增强排名和改进焦点的线性图Transformer
链接：https://arxiv.org/abs/2510.10631
【11】Multi-Task Learning with Feature-Similarity Laplacian Graphs for Predicting Alzheimer's Disease Progression
标题：利用相似拉普拉斯图的多任务学习预测阿尔茨海默病进展
链接：https://arxiv.org/abs/2510.10433
【12】Controllable Graph Generation with Diffusion Models via Inference-Time Tree Search Guidance
标题：通过推理时间树搜索引导的扩散模型可控图生成
链接：https://arxiv.org/abs/2510.10402
【13】Multi-View Graph Learning with Graph-Tuple
标题：使用图组的多视图图学习
链接：https://arxiv.org/abs/2510.10341
备注：Submitted to TAG workshop
【14】ProGress: Structured Music Generation via Graph Diffusion and Hierarchical Music Analysis
标题：ProGress：通过图形扩散和分层音乐分析的结构化音乐生成
链接：https://arxiv.org/abs/2510.10249
【15】Hierarchical Bayesian Flow Networks for Molecular Graph Generation
标题：用于分子图生成的分层Bayesian流网络
链接：https://arxiv.org/abs/2510.10211
【16】The Hybrid Multimodal Graph Index (HMGI): A Comprehensive Framework for Integrated Relational and Vector Search
标题：混合多模式图索引（HMGI）：集成关系和载体搜索的综合框架
链接：https://arxiv.org/abs/2510.10123
【17】Lighter-X: An Efficient and Plug-and-play Strategy for Graph-based Recommendation through Decoupled Propagation
标题：Lighter-X：通过脱钩传播进行基于图形的推荐的高效即插即用策略
链接：https://arxiv.org/abs/2510.10105
【18】Learning Joint Embeddings of Function and Process Call Graphs for Malware Detection
标题：学习用于恶意软件检测的函数和进程调用图的联合嵌入
链接：https://arxiv.org/abs/2510.09984
备注：None
【19】Augmenting generative models with biomedical knowledge graphs improves targeted drug discovery
标题：利用生物医学知识图谱增强生成模型改善有针对性的药物发现
链接：https://arxiv.org/abs/2510.09914
备注：This paper has been accepted for publication in the IEEE Transactions on Artificial Intelligence, October 2025
【20】TAWRMAC: A Novel Dynamic Graph Representation Learning Method
标题：TAWRMAC：一种新颖的动态图表示学习方法
链接：https://arxiv.org/abs/2510.09884
【21】HeSRN: Representation Learning On Heterogeneous Graphs via Slot-Aware Retentive Network
标题：HeSRN：通过Slot-Aware保留网络在异类图上进行表示学习
链接：https://arxiv.org/abs/2510.09767
【22】Hound: Relation-First Knowledge Graphs for Complex-System Reasoning in Security Audits
标题：Hound：安全审计中复杂系统推理的对象优先知识图
链接：https://arxiv.org/abs/2510.09633
Transformer(9篇)
【1】Diffusion Transformers with Representation Autoencoders
标题：具有表示自动编码器的扩散变形机
链接：https://arxiv.org/abs/2510.11690
备注：Technical Report; Project Page: this https URL
【2】Softmax $\geq$ Linear: Transformers may learn to classify in-context by kernel gradient descent
链接：https://arxiv.org/abs/2510.10425
【3】Combo-Gait: Unified Transformer Framework for Multi-Modal Gait Recognition and Attribute Analysis
标题：Combo-Gait：多模态步态识别与属性分析的统一Transformer框架
链接：https://arxiv.org/abs/2510.10417
【4】Transformer Model Detects Antidepressant Use From a Single Night of Sleep, Unlocking an Adherence Biomarker
标题：Transformer模型在一夜睡眠中检测到抗抑郁药的使用，解锁粘附性生物标志物
链接：https://arxiv.org/abs/2510.10364
【5】What Makes Looped Transformers Perform Better Than Non-Recursive Ones (Provably)
标题：是什么让循环Transformer比非回归Transformer性能更好（可以证明）
链接：https://arxiv.org/abs/2510.10089
【6】Stability of Transformers under Layer Normalization
标题：分层规范下Transformer的稳定性
链接：https://arxiv.org/abs/2510.09904
【7】Why Do Transformers Fail to Forecast Time Series In-Context?
标题：为什么Transformer无法在上下文中预测时间序列？
链接：https://arxiv.org/abs/2510.09776
备注：Code: this https URL
【8】Heterogeneous Point Set Transformers for Segmentation of Multiple View Particle Detectors
标题：用于多视角粒子检测器分割的异类点集变换器
链接：https://arxiv.org/abs/2510.09659
备注：Submitted to Machine Learning and the Physical Sciences Workshop (ML4PS) at NeurIPS 2025
【9】Hierarchical Qubit-Merging Transformer for Quantum Error Correction
标题：用于量子纠错的分层量子位合并Transformer
链接：https://arxiv.org/abs/2510.11593
备注：6 pages, 5 figures
GAN|对抗|攻击|生成相关(18篇)
【1】Adversarial Attacks Leverage Interference Between Features in Superposition
标题：对抗性攻击利用叠加特征之间的干扰
链接：https://arxiv.org/abs/2510.11709
【2】A Framework for Low-Effort Training Data Generation for Urban Semantic Segmentation
标题：用于城市语义分割的低成本训练数据生成框架
链接：https://arxiv.org/abs/2510.11567
【3】EAGER: Entropy-Aware GEneRation for Adaptive Inference-Time Scaling
标题：EAGER：自适应推断时间缩放的具有感知性的GerneRation
链接：https://arxiv.org/abs/2510.11170
【4】Neutral Agent-based Adversarial Policy Learning against Deep Reinforcement Learning in Multi-party Open Systems
标题：多方开放系统中基于中立代理的对抗性政策学习与深度强化学习
链接：https://arxiv.org/abs/2510.10937
【5】Digital Twin-enabled Multi-generation Control Co-Design with Deep Reinforcement Learning
标题：具有深度强化学习的数字双胞胎多代控制协同设计
链接：https://arxiv.org/abs/2510.10694
备注：to be published in Journal of Mechanical Design
【6】ImpMIA: Leveraging Implicit Bias for Membership Inference Attack under Realistic Scenarios
标题：ImpMIA：在现实场景下利用隐式偏差进行会员推断攻击
链接：https://arxiv.org/abs/2510.10625
【7】Encoder Decoder Generative Adversarial Network Model for Stock Market Prediction
标题：用于股市预测的编码器解码器生成对抗网络模型
链接：https://arxiv.org/abs/2510.10617
【8】FusionGen: Feature Fusion-Based Few-Shot EEG Data Generation
标题：FusionGen：基于特征融合的少激发脑电数据生成
链接：https://arxiv.org/abs/2510.10604
【9】Multi-scale Frequency-Aware Adversarial Network for Parkinson's Disease Assessment Using Wearable Sensors
标题：基于可穿戴传感器的多尺度频率感知对抗网络用于帕金森病评估
链接：https://arxiv.org/abs/2510.10558
【10】A Hybrid Machine Learning Approach for Synthetic Data Generation with Post Hoc Calibration for Clinical Tabular Datasets
标题：用于合成数据生成的混合机器学习方法，并对临床表格数据集进行事后校准
链接：https://arxiv.org/abs/2510.10513
【11】Latent Retrieval Augmented Generation of Cross-Domain Protein Binders
标题：潜在检索增强跨结构域蛋白质结合剂的生成
链接：https://arxiv.org/abs/2510.10480
【12】Adversarial Attacks on Downstream Weather Forecasting Models: Application to Tropical Cyclone Trajectory Prediction
标题：下游天气预报模型的对抗攻击：在热带气旋轨迹预测中的应用
链接：https://arxiv.org/abs/2510.10140
【13】Tight Robustness Certificates and Wasserstein Distributional Attacks for Deep Neural Networks
标题：深度神经网络的严格鲁棒性证书和Wasserstein分布式攻击
链接：https://arxiv.org/abs/2510.10000
【14】Exploration of Incremental Synthetic Non-Morphed Images for Single Morphing Attack Detection
标题：增量式合成非变形图像用于单次变形攻击检测的探索
链接：https://arxiv.org/abs/2510.09836
备注：Workshop paper accepted NeurIPS 2025
【15】Combined Representation and Generation with Diffusive State Predictive Information Bottleneck
标题：具有扩散状态预测信息瓶颈的组合表示和生成
链接：https://arxiv.org/abs/2510.09784
【16】Group-Adaptive Adversarial Learning for Robust Fake News Detection Against Malicious Comments
标题：群体自适应对抗学习用于针对恶意评论的鲁棒假新闻检测
链接：https://arxiv.org/abs/2510.09712
备注：10 pages, 12 figures
【17】Adversarial Robustness in One-Stage Learning-to-Defer
标题：一阶段学习推迟中的对抗稳健性
链接：https://arxiv.org/abs/2510.10988
【18】Interactive Atmospheric Composition Emulation for Next-Generation Earth System Models
标题：下一代地球系统模型的交互式大气成分模拟
链接：https://arxiv.org/abs/2510.10654
半/弱/无/有监督|不确定性|主动学习(14篇)
【1】FUSE: Fast Semi-Supervised Node Embedding Learning via Structural and Label-Aware Optimization
标题：FUSE：通过结构和标签感知优化进行快速半监督节点嵌入学习
链接：https://arxiv.org/abs/2510.11250
【2】PhysioME: A Robust Multimodal Self-Supervised Framework for Physiological Signals with Missing Modalities
标题：PhysoME：针对缺失模式的生理信号的稳健多模式自我监督框架
链接：https://arxiv.org/abs/2510.11110
备注：9 pages, 2 figures
【3】Deep semi-supervised approach based on consistency regularization and similarity learning for weeds classification
标题：基于一致性正规化和相似性学习的深度半监督杂草分类方法
链接：https://arxiv.org/abs/2510.10573
备注：Submitted to EURASIP Journal on Image and Video Processing
【4】Understanding Self-supervised Contrastive Learning through Supervised Objectives
标题：通过监督目标理解自我监督对比学习
链接：https://arxiv.org/abs/2510.10572
备注：Accepted at TMLR 2025
【5】Reverse Supervision at Scale: Exponential Search Meets the Economics of Annotation
标题：大规模反向监管：指数搜索遇上注释经济学
链接：https://arxiv.org/abs/2510.10446
备注：10 pages
【6】Unveiling Gamer Archetypes through Multi modal feature Correlations and Unsupervised Learning
标题：通过多模式特征相关性和无监督学习揭开游戏玩家原型
链接：https://arxiv.org/abs/2510.10263
备注：Submitted to Peer Review Journal
【7】Uncertainty-Aware Post-Detection Framework for Enhanced Fire and Smoke Detection in Compact Deep Learning Models
标题：不确定性感知后检测框架，用于在紧凑深度学习模型中增强火灾和烟雾检测
链接：https://arxiv.org/abs/2510.10108
备注：Accepted and to be presented at the International Conference on Smart Multimedia (ICSM 2025) - this https URL
【8】Cooperative Pseudo Labeling for Unsupervised Federated Classification
标题：无监督联邦分类的协作伪标记
链接：https://arxiv.org/abs/2510.10100
备注：Accepted by ICCV 2025
【9】An Unsupervised Time Series Anomaly Detection Approach for Efficient Online Process Monitoring of Additive Manufacturing
标题：一种用于增材制造高效在线过程监控的无监督时间序列异常检测方法
链接：https://arxiv.org/abs/2510.09977
备注：2025 IEEE 21st International Conference on Automation Science and Engineering
【10】Myopic Bayesian Decision Theory for Batch Active Learning with Partial Batch Label Sampling
标题：部分批量标签抽样批量主动学习的短视Bayesian决策理论
链接：https://arxiv.org/abs/2510.09877
【11】Harnessing Self-Supervised Deep Learning and Geostationary Remote Sensing for Advancing Wildfire and Associated Air Quality Monitoring: Improved Smoke and Fire Front Masking using GOES and TEMPO Radiance Data
标题：利用自我监督深度学习和地球同步遥感推进野火和相关空气质量监测：使用GOES和TEMPO辐射数据改进烟雾和火灾前沿掩蔽
链接：https://arxiv.org/abs/2510.09845
备注：his https URL
【12】Reliable Active Learning from Unreliable Labels via Neural Collapse Geometry
标题：通过神经崩溃几何从不可靠标签进行可靠主动学习
链接：https://arxiv.org/abs/2510.09740
备注：Accepted to NeurIPS 2025 Workshop on Reliable ML from Unreliable Data
【13】Spatial Uncertainty Quantification in Wildfire Forecasting for Climate-Resilient Emergency Planning
标题：应对气候变化的应急规划野火预测中的空间不确定性量化
链接：https://arxiv.org/abs/2510.09666
备注：None
【14】Neural variational inference for cutting feedback during uncertainty propagation
标题：不确定性传播期间切割反馈的神经变分推理
链接：https://arxiv.org/abs/2510.10268
迁移|Zero/Few/One-Shot|自适应(17篇)
【1】Constraint-Aware Reinforcement Learning via Adaptive Action Scaling
标题：通过自适应动作缩放的约束感知强化学习
链接：https://arxiv.org/abs/2510.11491
【2】Test-Time Adaptation by Causal Trimming
标题：通过因果修剪来适应测试时间
链接：https://arxiv.org/abs/2510.11133
备注：Accepted to the Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025); Code is available at this https URL
【3】Lightweight Facial Landmark Detection in Thermal Images via Multi-Level Cross-Modal Knowledge Transfer
标题：通过多层跨模式知识转移在热图像中进行轻量级面部地标检测
链接：https://arxiv.org/abs/2510.11128
【4】Efficient Edge Test-Time Adaptation via Latent Feature Coordinate Correction
标题：通过潜在特征坐标修正进行高效的边缘测试时间自适应
链接：https://arxiv.org/abs/2510.11068
备注：Under review
【5】APLOT: Robust Reward Modeling via Adaptive Preference Learning with Optimal Transport
标题：APLOT：通过最佳传输的自适应偏好学习进行稳健奖励建模
链接：https://arxiv.org/abs/2510.10963
备注：EMNLP2025
【6】Preconditioned Norms: A Unified Framework for Steepest Descent, Quasi-Newton and Adaptive Methods
标题：预条件规范：最陡下降、拟牛顿和适应性方法的统一框架
链接：https://arxiv.org/abs/2510.10777
备注：22 pages, 2 figures, 8 tables
【7】Reinforced Domain Selection for Continuous Domain Adaptation
标题：连续域适应的强化域选择
链接：https://arxiv.org/abs/2510.10530
备注：None
【8】FLAMMABLE: A Multi-Model Federated Learning Framework with Multi-Model Engagement and Adaptive Batch Sizes
标题：FLAMMAPLE：一个具有多模型参与和自适应批量大小的多模型联邦学习框架
链接：https://arxiv.org/abs/2510.10380
【9】Opacity-Gradient Driven Density Control for Compact and Efficient Few-Shot 3D Gaussian Splatting
标题：不透明度梯度驱动的密度控制，实现紧凑高效的Few-Shot3D高斯飞溅
链接：https://arxiv.org/abs/2510.10257
【10】Preference-driven Knowledge Distillation for Few-shot Node Classification
标题：偏好驱动的Few-Shot节点分类知识提炼
链接：https://arxiv.org/abs/2510.10116
备注：Accepted at NeurIPS 2025
【11】ADEPT: Continual Pretraining via Adaptive Expansion and Dynamic Decoupled Tuning
标题：ADEPT：通过自适应扩展和动态脱钩调整进行连续预训练
链接：https://arxiv.org/abs/2510.10071
【12】Translution: Unifying Self-attention and Convolution for Adaptive and Relative Modeling
标题：翻译：统一自我注意力和卷积以实现自适应和相对建模
链接：https://arxiv.org/abs/2510.10060
备注：technical report
【13】Skill-Targeted Adaptive Training
标题：针对技能的适应性训练
链接：https://arxiv.org/abs/2510.10023
【14】ARROW: An Adaptive Rollout and Routing Method for Global Weather Forecasting
标题：ARROW：全球天气预报的自适应推出和路由方法
链接：https://arxiv.org/abs/2510.09734
备注：16 pages, 6 figures, conference
【15】AdaptAuth: Multi-Layered Behavioral and Credential Analysis for a Secure and Adaptive Authentication Framework for Password Security
标题：AdaptAuth：用于密码安全的安全和自适应身份验证框架的多层行为和凭证分析
链接：https://arxiv.org/abs/2510.09645
【16】Transfer Learning with Distance Covariance for Random Forest: Error Bounds and an EHR Application
标题：随机森林具有距离协方差的迁移学习：误差界和EHR应用
链接：https://arxiv.org/abs/2510.10870
【17】Quantifying Dataset Similarity to Guide Transfer Learning
标题：量化数据集相似性以指导迁移学习
链接：https://arxiv.org/abs/2510.10866
强化学习(13篇)
【1】Context-Aware Model-Based Reinforcement Learning for Autonomous Racing
标题：基于上下文感知模型的自主赛车强化学习
链接：https://arxiv.org/abs/2510.11501
备注：Accepted to IEEE ICAR 2025
【2】Offline Reinforcement Learning with Generative Trajectory Policies
标题：具有生成轨迹策略的离线强化学习
链接：https://arxiv.org/abs/2510.11499
备注：Preprint. Under review at ICLR 2026
【3】How Reinforcement Learning After Next-Token Prediction Facilitates Learning
标题：下一个令牌预测后的强化学习如何促进学习
链接：https://arxiv.org/abs/2510.11495
【4】Coordinated Strategies in Realistic Air Combat by Hierarchical Multi-Agent Reinforcement Learning
标题：分层多智能体强化学习实现现实空战协调策略
链接：https://arxiv.org/abs/2510.11474
备注：2025 IEEE International Conference on Agentic AI (ICA)
【5】Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers
标题：通过调整训练和推理路由器稳定MoE强化学习
链接：https://arxiv.org/abs/2510.11370
【6】Can Tool-Integrated Reinforcement Learning Generalize Across Diverse Domains?
标题：工具集成强化学习能否在不同领域推广？
链接：https://arxiv.org/abs/2510.11184
【7】Emergence of hybrid computational dynamics through reinforcement learning
标题：通过强化学习出现混合计算动力学
链接：https://arxiv.org/abs/2510.11162
备注：22 pages, 11 figures
【8】Chart-RVR: Reinforcement Learning with Verifiable Rewards for Explainable Chart Reasoning
标题：Chart-RVR：具有可解释图表推理可验证奖励的强化学习
链接：https://arxiv.org/abs/2510.10973
备注：23 pages
【9】PAC-Bayesian Reinforcement Learning Trains Generalizable Policies
标题：Pac-Bayesian强化学习训练可推广策略
链接：https://arxiv.org/abs/2510.10544
【10】Data-driven simulator of multi-animal behavior with unknown dynamics via offline and online reinforcement learning
标题：通过离线和在线强化学习，具有未知动态的多动物行为的数据驱动模拟器
链接：https://arxiv.org/abs/2510.10451
备注：21 pages, 7 figures
【11】One4Many-StablePacker: An Efficient Deep Reinforcement Learning Framework for the 3D Bin Packing Problem
标题：One 4 Many-StablePacker：针对3D垃圾箱装箱问题的高效深度强化学习框架
链接：https://arxiv.org/abs/2510.10057
【12】Experience-Efficient Model-Free Deep Reinforcement Learning Using Pre-Training
标题：使用预训练的体验高效的无模型深度强化学习
链接：https://arxiv.org/abs/2510.10029
【13】Structured Cooperative Multi-Agent Reinforcement Learning: a Bayesian Network Perspective
标题：结构化合作多智能体强化学习：Bayesian网络的视角
链接：https://arxiv.org/abs/2510.09937
元学习(1篇)
【1】In-Context Learning Is Provably Bayesian Inference: A Generalization Theory for Meta-Learning
标题：上下文内学习是可证明的Bayesian推理：元学习的概括理论
链接：https://arxiv.org/abs/2510.10981
医学相关(8篇)
【1】Reconstructing 12-Lead ECG from 3-Lead ECG using Variational Autoencoder to Improve Cardiac Disease Detection of Wearable ECG Devices
标题：使用变分自动编码器从3导心电图重建12导心电图以改进可穿戴心电图设备的心脏疾病检测
链接：https://arxiv.org/abs/2510.11442
备注：24 pages, 5 figures, submitted to Nature Communications
【2】MIEO: encoding clinical data to enhance cardiovascular event prediction
标题：MIEO：编码临床数据以增强心血管事件预测
链接：https://arxiv.org/abs/2510.11257
备注：Presented in the Poster Session of Computational Intelligence methods for Bioinformatics and Biostatistics (CIBB) 2025
【3】Vision4PPG: Emergent PPG Analysis Capability of Vision Foundation Models for Vital Signs like Blood Pressure
标题：Vision 4PPV：Vision Foundation模型对血压等生命体征的紧急PPV分析能力
链接：https://arxiv.org/abs/2510.10366
备注：BHI abstract extended
【4】MRI Brain Tumor Detection with Computer Vision
标题：计算机视觉MRI脑肿瘤检测
链接：https://arxiv.org/abs/2510.10250
备注：12 pages, 8 figures, final project report for CS4100 (Machine Learning), Northeastern University, April 2024
【5】Chain-of-Influence: Tracing Interdependencies Across Time and Features in Clinical Predictive Modelings
标题：影响链：追踪临床预测建模中跨时间和特征的相互依赖性
链接：https://arxiv.org/abs/2510.09895
【6】A Hybrid Computational Intelligence Framework with Metaheuristic Optimization for Drug-Drug Interaction Prediction
标题：用于药物相互作用预测的元启发式优化混合计算智能框架
链接：https://arxiv.org/abs/2510.09668
【7】Direct Routing Gradient (DRGrad): A Personalized Information Surgery for Multi-Task Learning (MTL) Recommendations
标题：直接路由梯度（DRGrad）：多任务学习（MTL）建议的个性化信息手术
链接：https://arxiv.org/abs/2510.09643
备注：None
【8】Detecting Conspiracy Theory Against COVID-19 Vaccines
标题：检测针对COVID-19疫苗的阴谋论
链接：https://arxiv.org/abs/2211.13003
备注：6 pages, 5 figures
蒸馏|知识提取(2篇)
【1】Find Your Optimal Teacher: Personalized Data Synthesis via Router-Guided Multi-Teacher Distillation
标题：寻找您的最佳教师：通过路由器引导的多教师蒸馏进行个性化数据合成
链接：https://arxiv.org/abs/2510.10925
备注：19 pages, 10 figures
【2】Semantic-Cohesive Knowledge Distillation for Deep Cross-modal Hashing
标题：深度跨模哈希的语义凝聚知识提炼
链接：https://arxiv.org/abs/2510.09664
推荐(2篇)
【1】Differentiable Fast Top-K Selection for Large-Scale Recommendation
标题：大规模推荐的差异化快速Top-K选择
链接：https://arxiv.org/abs/2510.11472
备注：12 pages, 5 figures
【2】Does Weighting Improve Matrix Factorization for Recommender Systems?
标题：加权能否改善推荐系统的矩阵分解？
链接：https://arxiv.org/abs/2510.10440
备注：In the proceedings of the Web Conference (WWW) 2025 (11 pages)
聚类(3篇)
【1】Learning-Augmented Streaming Algorithms for Correlation Clustering
标题：相关性集群的学习增强流媒体算法
链接：https://arxiv.org/abs/2510.10705
备注：NeurIPS 2025
【2】Clustering Result Re-guided Incomplete Multi-view Spectral Clustering
标题：重新引导的不完整多视图光谱聚集结果
链接：https://arxiv.org/abs/2510.09959
【3】Distributed clustering in partially overlapping feature spaces
标题：部分重叠特征空间中的分布式集群
链接：https://arxiv.org/abs/2510.09799
超分辨率|去噪|去模糊|去雾(1篇)
【1】Robust Photoplethysmography Signal Denoising via Mamba Networks
标题：通过Mamba网络实现稳健的光电体积脉搏波信号去噪
链接：https://arxiv.org/abs/2510.11058
备注：5 pages, 2 figures
自动驾驶|车辆|车道检测等(3篇)
【1】Align2Act: Instruction-Tuned Models for Human-Aligned Autonomous Driving
标题：Alignn 2Act：人性化自动驾驶的指导调整模型
链接：https://arxiv.org/abs/2510.10503
【2】Enhancing the Cross-Size Generalization for Solving Vehicle Routing Problems via Continual Learning
标题：通过持续学习增强跨规模推广以解决车辆路径问题
链接：https://arxiv.org/abs/2510.10262
【3】Enhanced Urban Traffic Management Using CCTV Surveillance Videos and Multi-Source Data Current State Prediction and Frequent Episode Mining
标题：利用闭路电视监控视频和多源数据增强城市交通管理现状预测和频繁事件挖掘
链接：https://arxiv.org/abs/2510.09644
备注：24 pages, 9 figures
点云|SLAM|雷达|激光|深度RGBD相关(3篇)
【1】Optimally Deep Networks -- Adapting Model Depth to Datasets for Superior Efficiency
标题：优化深度网络--根据数据集调整模型深度以实现卓越效率
链接：https://arxiv.org/abs/2510.10764
备注：6 pages, 3 figures, 1 table
【2】DAGLFNet:Deep Attention-Guided Global-Local Feature Fusion for Pseudo-Image Point Cloud Segmentation
标题：DAGLFNet：深度注意力引导的全局-局部特征融合用于伪图像点云分割
链接：https://arxiv.org/abs/2510.10471
【3】Machine Learning-Integrated Hybrid Fluid-Kinetic Framework for Quantum Electrodynamic Laser Plasma Simulations
标题：量子电动激光等离子体模拟的机器学习集成混合流体动力学框架
链接：https://arxiv.org/abs/2510.11174
联邦学习|隐私保护|加密(2篇)
【1】FedHybrid: Breaking the Memory Wall of Federated Learning via Hybrid Tensor Management
标题：FedHybrid：通过混合张量管理打破联邦学习的记忆墙
链接：https://arxiv.org/abs/2510.11400
备注：Sensys 2024
【2】Evaluation of Differential Privacy Mechanisms on Federated Learning
标题：联邦学习中差异隐私机制的评估
链接：https://arxiv.org/abs/2510.09691
备注：Supervised by Prof. Dr.-Ing. habil. Alois C. Knoll; Advisor: Nagacharan Teja Tangirala, this http URL
推理|分析|理解|解释(13篇)
【1】Are Large Reasoning Models Interruptible?
标题：大型推理模型会中断吗？
链接：https://arxiv.org/abs/2510.11713
备注：Project Page: this https URL
【2】MS-Mix: Unveiling the Power of Mixup for Multimodal Sentiment Analysis
标题：MS-Mix：揭示Mixup在多模式情绪分析中的力量
链接：https://arxiv.org/abs/2510.11579
备注：Under Review
【3】Iterative Amortized Inference: Unifying In-Context Learning and Learned Optimizers
标题：迭代摊销推理：统一上下文学习和习得优化器
链接：https://arxiv.org/abs/2510.11471
【4】Understanding the Generalization of Stochastic Gradient Adam in Learning Neural Networks
标题：理解随机梯度Adam在学习神经网络中的推广
链接：https://arxiv.org/abs/2510.11354
备注：71 pages, 12 figures, NeurIPS 2025
【5】Beyond single-model XAI: aggregating multi-model explanations for enhanced trustworthiness
标题：超越单一模型XAI：聚合多模型解释以增强可信度
链接：https://arxiv.org/abs/2510.11164
备注：Accepted at the European Workshop on Trustworthy Artificial Intelligence (TRUST-AI), co-located within ECAI 2025
【6】Not All Bits Are Equal: Scale-Dependent Memory Optimization Strategies for Reasoning Models
标题：并非所有位都是相等的：推理模型的规模相关内存优化策略
链接：https://arxiv.org/abs/2510.10964
备注：20 pages, 12 figures
【7】Understanding Sampler Stochasticity in Training Diffusion Models for RLHF
标题：RLHF扩散模型训练中样本随机性的理解
链接：https://arxiv.org/abs/2510.10767
【8】Explainable Human-in-the-Loop Segmentation via Critic Feedback Signals
标题：通过批评反馈信号进行可解释的人在环分割
链接：https://arxiv.org/abs/2510.09945
备注：Submitted to a computer vision conference (under review)
【9】Understanding Robust Machine Learning for Nonparametric Regression with Heavy-Tailed Noise
标题：了解具有重尾噪音的非参数回归的鲁棒机器学习
链接：https://arxiv.org/abs/2510.09888
【10】DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning
标题：DELTA：用于高效长上下文推理的动态层感知令牌注意力
链接：https://arxiv.org/abs/2510.09883
【11】Decomposer Networks: Deep Component Analysis and Synthesis
标题：分解器网络：深度组件分析和合成
链接：https://arxiv.org/abs/2510.09825
备注：13 Pages, 4 figures
【12】The Geometry of Reasoning: Flowing Logics in Representation Space
标题：推理的几何学：表象空间中的流动逻辑
链接：https://arxiv.org/abs/2510.09782
备注：Code: this https URL
【13】It's 2025 -- Narrative Learning is the new baseline to beat for explainable machine learning
标题：现在是2025年--叙事学习是可解释机器学习的新基准
链接：https://arxiv.org/abs/2510.09723
备注：18 pages, 5 figures
检测相关(12篇)
【1】NV3D: Leveraging Spatial Shape Through Normal Vector-based 3D Object Detection
标题：NV 3D：通过基于法向载体的3D对象检测利用空间形状
链接：https://arxiv.org/abs/2510.11632
【2】$Δ\mathrm{Energy}$: Optimizing Energy Change During Vision-Language Alignment Improves both OOD Detection and OOD Generalization
链接：https://arxiv.org/abs/2510.11296
备注：Accepted by NeruIPS2025
【3】Evaluating Line-level Localization Ability of Learning-based Code Vulnerability Detection Models
标题：评估基于学习的代码漏洞检测模型的行级本地化能力
链接：https://arxiv.org/abs/2510.11202
备注：Preprint
【4】A Comprehensive Forecasting-Based Framework for Time Series Anomaly Detection: Benchmarking on the Numenta Anomaly Benchmark (NAB)
标题：时间序列异常检测的全面基于预测的框架：Numenta异常基准（NAB）的基准
链接：https://arxiv.org/abs/2510.11141
【5】Causal Disentanglement Learning for Accurate Anomaly Detection in Multivariate Time Series
标题：因果解纠缠学习用于多元时间序列中的准确异常检测
链接：https://arxiv.org/abs/2510.11084
备注：20 pages, 4 Figures,
【6】LPCVAE: A Conditional VAE with Long-Term Dependency and Probabilistic Time-Frequency Fusion for Time Series Anomaly Detection
标题：LPCVAE：一种用于时间序列异常检测的长时相关概率时频融合条件VAE
链接：https://arxiv.org/abs/2510.10915
【7】YOLOv11-Litchi: Efficient Litchi Fruit Detection based on UAV-Captured Agricultural Imagery in Complex Orchard Environments
标题：YOLOv 11-荔枝：复杂果园环境中基于无人机捕获农业图像的高效荔枝果实检测
链接：https://arxiv.org/abs/2510.10141
【8】Gradient-based Model Shortcut Detection for Time Series Classification
标题：时间序列分类的基于对象模型的状态检测
链接：https://arxiv.org/abs/2510.10075
备注：Code available at: this https URL
【9】Advancing Intoxication Detection: A Smartwatch-Based Approach
标题：推进中毒检测：基于智能手表的方法
链接：https://arxiv.org/abs/2510.09916
【10】HIPPD: Brain-Inspired Hierarchical Information Processing for Personality Detection
标题：HIPPD：用于人格检测的脑启发分层信息处理
链接：https://arxiv.org/abs/2510.09893
【11】Kelp: A Streaming Safeguard for Large Models via Latent Dynamics-Guided Risk Detection
标题：Kelp：通过潜在动态引导风险检测为大型模型提供流媒体保护
链接：https://arxiv.org/abs/2510.09694
【12】Risk-Calibrated Bayesian Streaming Intrusion Detection with SRE-Aligned Decisions
标题：具有SRE一致决策的风险校准的Bayesian流入侵检测
链接：https://arxiv.org/abs/2510.09619
备注：11 pages, 7 figures. Primary category: cs.CR; cross-list: cs.LG, stat.ML. Implementation code and datasets are available from the corresponding author upon reasonable request. Code and reproducibility materials will be made available upon publication
分类|识别(7篇)
【1】Automatic Music Sample Identification with Multi-Track Contrastive Learning
标题：多轨对比学习自动音乐样本识别
链接：https://arxiv.org/abs/2510.11507
【2】LightPneumoNet: Lightweight Pneumonia Classifier
标题：LightPneumoNet：轻型肺炎分类器
链接：https://arxiv.org/abs/2510.11232
备注：13 pages (including references), 5 figures
【3】Mesh-Gait: A Unified Framework for Gait Recognition Through Multi-Modal Representation Learning from 2D Silhouettes
标题：网格步态：通过从2D剪影中进行多模式表示学习进行步态识别的统一框架
链接：https://arxiv.org/abs/2510.10406
【4】Applying non-negative matrix factorization with covariates to label matrix for classification
标题：应用带协变量的非负矩阵分解来标记矩阵进行分类
链接：https://arxiv.org/abs/2510.10375
备注：2 figures, R package: nmfkc published in GitHub, this https URL
【5】Improving Speech Emotion Recognition with Mutual Information Regularized Generative Model
标题：用互信息正规生成模型改进语音情感识别
链接：https://arxiv.org/abs/2510.10078
【6】Bidirectional Time-Frequency Pyramid Network for Enhanced Robust EEG Classification
标题：用于增强稳健脑电分类的双向时频金字塔网络
链接：https://arxiv.org/abs/2510.10004
备注：Accepted to IEEE BIBM 2025
【7】High-Power Training Data Identification with Provable Statistical Guarantees
标题：具有可证明统计保证的高性能训练数据识别
链接：https://arxiv.org/abs/2510.09717
表征(5篇)
【1】Instruction-aware User Embedding via Synergistic Language and Representation Modeling
标题：通过协同语言和表示建模实现教学感知用户嵌入
链接：https://arxiv.org/abs/2510.11016
【2】On the Optimal Representation Efficiency of Barlow Twins: An Information-Geometric Interpretation
标题：Barlow孪生子最佳表征效率的信息几何解释
链接：https://arxiv.org/abs/2510.10980
备注：7 pages
【3】INR-Bench: A Unified Benchmark for Implicit Neural Representations in Multi-Domain Regression and Reconstruction
标题：INR-Bench：多域回归和重建中隐式神经表示的统一基准
链接：https://arxiv.org/abs/2510.10188
【4】Beyond AlphaEarth: Toward Human-Centered Spatial Representation via POI-Guided Contrastive Learning
标题：超越AlphaEarth：通过兴趣点引导的对比学习实现以人为本的空间表示
链接：https://arxiv.org/abs/2510.09894
【5】Generative Modeling of Aerosol State Representations
标题：气溶胶状态表示的生成式建模
链接：https://arxiv.org/abs/2510.10361
备注：31 pages, 20 figures
编码器(2篇)
【1】Forward-Forward Autoencoder Architectures for Energy-Efficient Wireless Communications
标题：用于节能无线通信的前向自动编码器架构
链接：https://arxiv.org/abs/2510.11418
【2】ProteinAE: Protein Diffusion Autoencoders for Structure Encoding
标题：ProteinAE：用于结构编码的蛋白质扩散自动编码器
链接：https://arxiv.org/abs/2510.10634
优化|敛散性(8篇)
【1】Diffusion-DFL: Decision-focused Diffusion Models for Stochastic Optimization
标题：扩散-DFL：随机优化的以决策为中心的扩散模型
链接：https://arxiv.org/abs/2510.11590
【2】FedLoRA-Optimizer: Federated LoRA Fine-Tuning with Global and Local Optimization in Heterogeneous Data Scenarios
标题：FedLoRA-Optimizer：在异类数据场景中通过全局和本地优化进行联合LoRA微调
链接：https://arxiv.org/abs/2510.11274
【3】ELMO: Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces
标题：ELMO：通过大输出空间中的低精度和峰值内存优化提高效率
链接：https://arxiv.org/abs/2510.11168
备注：Accepted to ICML 2025
【4】LOOPerSet: A Large-Scale Dataset for Data-Driven Polyhedral Compiler Optimization
标题：LOOPerSet：一个用于数据驱动多边形调度器优化的大规模数据集
链接：https://arxiv.org/abs/2510.10209
【5】Accelerated stochastic first-order method for convex optimization under heavy-tailed noise
标题：重尾噪音下凸优化的加速随机一阶方法
链接：https://arxiv.org/abs/2510.11676
【6】Deep Signature and Neural RDE Methods for Path-Dependent Portfolio Optimization
标题：路径相关投资组合优化的深度签名和神经RDE方法
链接：https://arxiv.org/abs/2510.10728
备注：Accepted for presentation at the ACM International Conference on AI in Finance (ICAIF 2025), QuantAI Workshop, Singapore. 9 pages. Code available at: this https URL
【7】Mean-square and linear convergence of a stochastic proximal point algorithm in metric spaces of nonpositive curvature
标题：非正弯曲度量空间中随机近点算法的均方和线性收敛性
链接：https://arxiv.org/abs/2510.10697
备注：24 pages
【8】Second-order Optimization under Heavy-Tailed Noise: Hessian Clipping and Sample Complexity Limits
标题：重尾噪音下的二阶优化：黑森裁剪和样本复杂性限制
链接：https://arxiv.org/abs/2510.10690
备注：Accepted for publication at NeurIPS 2025
预测|估计(17篇)
【1】Cross-Scale Reservoir Computing for large spatio-temporal forecasting and modeling
标题：用于大时空预测和建模的跨尺度水库计算
链接：https://arxiv.org/abs/2510.11209
【2】Comparative Evaluation of Neural Network Architectures for Generalizable Human Spatial Preference Prediction in Unseen Built Environments
标题：用于不可见建筑环境中可推广人类空间偏好预测的神经网络架构比较评估
链接：https://arxiv.org/abs/2510.10954
备注：The 15th International Workshop on Structural Health Monitoring (IWSHM)
【3】Rethinking deep learning: linear regression remains a key benchmark in predicting terrestrial water storage
标题：重新思考深度学习：线性回归仍然是预测陆地蓄水量的关键基准
链接：https://arxiv.org/abs/2510.10799
【4】Structure Over Signal: A Globalized Approach to Multi-relational GNNs for Stock Prediction
标题：信号上的结构：一种用于股票预测的多关系GNN的全球化方法
链接：https://arxiv.org/abs/2510.10775
【5】Seeing My Future: Predicting Situated Interaction Behavior in Virtual Reality
标题：看到我的未来：预测虚拟现实中的情境交互行为
链接：https://arxiv.org/abs/2510.10742
备注：Project Page: this https URL
【6】Controllable Generative Trajectory Prediction via Weak Preference Alignment
标题：基于弱偏好对齐的可控生成轨迹预测
链接：https://arxiv.org/abs/2510.10731
【7】Attention-Enhanced LSTM Modeling for Improved Temperature and Rainfall Forecasting in Bangladesh
标题：提高注意力的LSTM建模用于改进孟加拉国的气温和降雨预报
链接：https://arxiv.org/abs/2510.10702
【8】Stock Prediction via a Dual Relation Fusion Network incorporating Static and Dynamic Relations
标题：通过结合静态和动态关系的双关系融合网络进行股票预测
链接：https://arxiv.org/abs/2510.10695
备注：11 pages
【9】Automatic Piecewise Linear Regression for Predicting Student Learning Satisfaction
标题：自动分段线性回归预测学生学习满意度
链接：https://arxiv.org/abs/2510.10639
【10】SDG-L: A Semiparametric Deep Gaussian Process based Framework for Battery Capacity Prediction
标题：SDG-L：基于半参数深高斯过程的电池容量预测框架
链接：https://arxiv.org/abs/2510.10621
备注：None
【11】LightSAE: Parameter-Efficient and Heterogeneity-Aware Embedding for IoT Multivariate Time Series Forecasting
标题：LightAE：用于物联网多元时间序列预测的参数高效和异源性感知嵌入
链接：https://arxiv.org/abs/2510.10465
备注：Submitted to IEEE IoT-J
【12】Exploration-free Algorithms for Multi-group Mean Estimation
标题：多组均值估计的免探索算法
链接：https://arxiv.org/abs/2510.10374
【13】A Unified Frequency Domain Decomposition Framework for Interpretable and Robust Time Series Forecasting
标题：用于可解释且稳健的时间序列预测的统一频域分解框架
链接：https://arxiv.org/abs/2510.10145
【14】SVTime: Small Time Series Forecasting Models Informed by "Physics" of Large Vision Model Forecasters
链接：https://arxiv.org/abs/2510.09780
【15】SeFEF: A Seizure Forecasting Evaluation Framework
标题：SeFEF：癫痫发作预测评估框架
链接：https://arxiv.org/abs/2510.11275
备注：main document: 14 pages, 9 figures, 2 tables; appendix: 7 pages, 2 figures, 3 tables, 2 algorithms
【16】High-Dimensional Learning Dynamics of Quantized Models with Straight-Through Estimator
标题：具有直接估计的量化模型的多维学习动态
链接：https://arxiv.org/abs/2510.10693
备注：27 pages, 14 figures
【17】On some practical challenges of conformal prediction
标题：关于共形预报的一些实际挑战
链接：https://arxiv.org/abs/2510.10324
其他神经网络|深度学习|模型|建模(56篇)
【1】MATH-Beyond: A Benchmark for RL to Expand Beyond the Base Model
标题：Math-Beyond：RL超越基本模型扩展的基准
链接：https://arxiv.org/abs/2510.11653
【2】Deconstructing Attention: Investigating Design Principles for Effective Language Modeling
标题：解构注意力：研究有效语言建模的设计原则
链接：https://arxiv.org/abs/2510.11602
【3】Ontolearn-A Framework for Large-scale OWL Class Expression Learning in Python
标题：Ontolearn-Python中大规模OWR类表达学习框架
链接：https://arxiv.org/abs/2510.11561
备注：None
【4】Knowledge-Guided Machine Learning Models to Upscale Evapotranspiration in the U.S. Midwest
标题：知识引导的机器学习模型扩大美国中西部的蒸散量
链接：https://arxiv.org/abs/2510.11505
【5】Learning to Make MISTAKEs: Modeling Incorrect Student Thinking And Key Errors
标题：学会犯错误：建模不正确的学生思维和关键错误
链接：https://arxiv.org/abs/2510.11502
【6】Rescaling-Aware Training for Efficient Deployment of Deep Learning Models on Full-Integer Hardware
标题：重新缩放感知训练，在全负载硬件上高效部署深度学习模型
链接：https://arxiv.org/abs/2510.11484
备注：Submitted to IEEE Embedded Systems Letters
【7】DiffStyleTS: Diffusion Model for Style Transfer in Time Series
标题：DistStyleTS：时间序列中风格转移的扩散模型
链接：https://arxiv.org/abs/2510.11335
【8】Diffusion-Link: Diffusion Probabilistic Model for Bridging the Audio-Text Modality Gap
标题：扩散-链接：弥合音频-文本情态差距的扩散概率模型
链接：https://arxiv.org/abs/2510.11330
备注：5 pages. Submitted to IEEE ICASSP 2026
【9】Network-Optimised Spiking Neural Network (NOS) Scheduling for 6G O-RAN: Spectral Margin and Delay-Tail Control
标题：6 G O-RAN的网络优化尖峰神经网络（NOS）调度：频谱裕度和延迟尾控制
链接：https://arxiv.org/abs/2510.11291
备注：6 pages, 5 figures, 1 table
【10】DUAL: Learning Diverse Kernels for Aggregated Two-sample and Independence Testing
标题：DUAL：学习用于聚合双样本和独立性测试的多样化核心
链接：https://arxiv.org/abs/2510.11140
【11】Temporal Alignment Guidance: On-Manifold Sampling in Diffusion Models
标题：时间对齐指导：扩散模型中的总管上采样
链接：https://arxiv.org/abs/2510.11057
备注：54 pages, 17 figures, 18 tables
【12】Catch-Only-One: Non-Transferable Examples for Model-Specific Authorization
标题：仅限一：型号特定授权的不可转让示例
链接：https://arxiv.org/abs/2510.10982
【13】MC#: Mixture Compressor for Mixture-of-Experts Large Models
标题：MC#：适用于专家混合大型型号的混合压缩机
链接：https://arxiv.org/abs/2510.10962
备注：15 pages, 13 figures
【14】Interpretable Machine Learning for Cognitive Aging: Handling Missing Data and Uncovering Social Determinant
标题：认知衰老的可解释机器学习：处理缺失数据并揭示社会决定因素
链接：https://arxiv.org/abs/2510.10952
【15】Redundancy as a Structural Information Principle for Learning and Generalization
标题：冗余作为学习和泛化的结构信息原则
链接：https://arxiv.org/abs/2510.10938
【16】DreamMakeup: Face Makeup Customization using Latent Diffusion Models
标题：DreamMakeup：使用潜在扩散模型的面部化妆定制
链接：https://arxiv.org/abs/2510.10918
【17】A Joint Learning Approach to Hardware Caching and Prefetching
标题：硬件缓存和预取的联合学习方法
链接：https://arxiv.org/abs/2510.10862
备注：Accepted at ML for Systems Workshop at the 39th Conference on Neural Information Processing Systems (NeurIPS 2025)
【18】Discrete State Diffusion Models: A Sample Complexity Perspective
标题：离散状态扩散模型：样本复杂性视角
链接：https://arxiv.org/abs/2510.10854
【19】Designing ReLU Generative Networks to Enumerate Trees with a Given Tree Edit Distance
标题：设计ReLU生成网络以列举具有给定树编辑距离的树
链接：https://arxiv.org/abs/2510.10706
【20】Multitask Learning with Learned Task Relationships
标题：具有习得任务关系的多任务学习
链接：https://arxiv.org/abs/2510.10570
【21】Population-Coded Spiking Neural Networks for High-Dimensional Robotic Control
标题：用于多维机器人控制的人口编码尖峰神经网络
链接：https://arxiv.org/abs/2510.10516
【22】Gradient Enhanced Self-Training Physics-Informed Neural Network (gST-PINN) for Solving Nonlinear Partial Differential Equations
标题：求解非线性偏微分方程的梯度增强自训练物理信息神经网络（gST-PINN）
链接：https://arxiv.org/abs/2510.10483
【23】Hierarchical LoRA MoE for Efficient CTR Model Scaling
标题：分层LoRA MoE用于高效的TLR模型缩放
链接：https://arxiv.org/abs/2510.10432
备注：13 pages, 9 figures
【24】Learning to Throw-Flip
标题：学习投掷翻转
链接：https://arxiv.org/abs/2510.10357
备注：Accepted to IROS 2025. Video Summary: this https URL
【25】Learning Operators through Coefficient Mappings in Fixed Basis Spaces
标题：通过固定基空间中的系数映射学习运算符
链接：https://arxiv.org/abs/2510.10350
【26】Sample-Efficient Online Learning in LM Agents via Hindsight Trajectory Rewriting
标题：通过事后诸葛亮轨迹重写在LM代理中进行样本高效在线学习
链接：https://arxiv.org/abs/2510.10304
【27】Grounded AI for Code Review: Resource-Efficient Large-Model Serving in Enterprise Pipelines
标题：用于代码审查的接地人工智能：在企业管道中服务资源高效的大型模型
链接：https://arxiv.org/abs/2510.10290
备注：Submitted to MLSys 2026
【28】Progressive Scale Convolutional Network for Spatio-Temporal Downscaling of Soil Moisture: A Case Study Over the Tibetan Plateau
标题：土壤湿度时空缩减的递进尺度卷积网络：青藏高原的案例研究
链接：https://arxiv.org/abs/2510.10244
【29】CauchyNet: Compact and Data-Efficient Learning using Holomorphic Activation Functions
标题：CauchyNet：使用全纯激活函数的紧凑且数据高效的学习
链接：https://arxiv.org/abs/2510.10195
【30】Robust Learning of Diffusion Models with Extremely Noisy Conditions
标题：具有极噪条件的扩散模型的鲁棒学习
链接：https://arxiv.org/abs/2510.10149
【31】PANTHER: Generative Pretraining Beyond Language for Sequential User Behavior Modeling
标题：PANTHER：超越语言的生成性预训练，用于序列用户行为建模
链接：https://arxiv.org/abs/2510.10102
【32】FOSSIL: Regret-Minimizing Curriculum Learning for Metadata-Free and Low-Data Mpox Diagnosis
标题：FOSSIL：针对无元数据和低数据Mpox诊断的最小化遗憾课程学习
链接：https://arxiv.org/abs/2510.10041
备注：35 pages, 11 figures, submitted to Computers in Biology and Medicine (Elsevier, under review)
【33】Phase-Aware Deep Learning with Complex-Valued CNNs for Audio Signal Applications
标题：利用复值CNN进行阶段感知深度学习用于音频信号应用
链接：https://arxiv.org/abs/2510.09926
【34】AutoGD: Automatic Learning Rate Selection for Gradient Descent
标题：AutoVD：梯度下降的自动学习率选择
链接：https://arxiv.org/abs/2510.09923
【35】Temporal Lifting as Latent-Space Regularization for Continuous-Time Flow Models in AI Systems
标题：人工智能系统中连续时间流模型的潜空间规则化时间提升
链接：https://arxiv.org/abs/2510.09805
备注：6 pages, 1 figure, 1 table, 1 algorithm
【36】Principled Operator Learning in Ocean Dynamics: The Role of Temporal Structure
标题：海洋动力学中的原则性操作员学习：时间结构的作用
链接：https://arxiv.org/abs/2510.09792
备注：Accepted at NeurIPS ML4PS 2025
【37】A Generic Machine Learning Framework for Radio Frequency Fingerprinting
标题：一种通用的射频指纹识别机器学习框架
链接：https://arxiv.org/abs/2510.09775
【38】Leveraging Shared Prototypes for a Multimodal Pulse Motion Foundation Model
标题：利用共享原型实现多模式脉搏运动基础模型
链接：https://arxiv.org/abs/2510.09764
【39】Machine learning methods fail to provide cohesive atheoretical construction of personality traits from semantic embeddings
标题：机器学习方法未能从语义嵌入中提供个性特征的凝聚力非理论构建
链接：https://arxiv.org/abs/2510.09739
备注：1 figure, 12 pages
【40】Operator Learning for Power Systems Simulation
标题：电力系统模拟的操作员学习
链接：https://arxiv.org/abs/2510.09704
【41】Vanishing Contributions: A Unified Approach to Smoothly Transition Neural Models into Compressed Form
标题：消失的贡献：将神经模型平稳转换为压缩形式的统一方法
链接：https://arxiv.org/abs/2510.09696
备注：Code available at this https URL
【42】On the Occurence of Critical Learning Periods in Neural Networks
标题：关于神经网络中关键学习期的发生
链接：https://arxiv.org/abs/2510.09687
备注：8 pages, 8 figures
【43】Deep Neural Networks Inspired by Differential Equations
标题：受微方程启发的深度神经网络
链接：https://arxiv.org/abs/2510.09685
备注：35 Pages, 3 figures
【44】A physics-aware deep learning model for shear band formation around collapsing pores in shocked reactive materials
标题：物理感知深度学习模型，用于在受冲击反应材料中塌陷的孔隙周围形成剪切带
链接：https://arxiv.org/abs/2510.09670
备注：None
【45】Assessment of different loss functions for fitting equivalent circuit models to electrochemical impedance spectroscopy data
标题：评估不同损失函数以将等效电路模型与电化学阻抗谱数据匹配
链接：https://arxiv.org/abs/2510.09662
【46】Learning What Matters: Steering Diffusion via Spectrally Anisotropic Forward Noise
标题：了解重要的事情：通过光谱各向异性正向噪音引导扩散
链接：https://arxiv.org/abs/2510.09660
【47】Gradient-Sign Masking for Task Vector Transport Across Pre-Trained Models
标题：预训练模型之间任务向量传输的学生签名掩蔽
链接：https://arxiv.org/abs/2510.09658
【48】Generative Models for Helmholtz Equation Solutions: A Dataset of Acoustic Materials
标题：Helmholtz方程解的生成模型：声学材料数据集
链接：https://arxiv.org/abs/2510.09657
备注：Accepted at EUSIPCO 2025
【49】Rounding-Guided Backdoor Injection in Deep Learning Model Quantization
标题：深度学习模型量化中的舍入引导后门注入
链接：https://arxiv.org/abs/2510.09647
备注：This paper is to appear in NDSS 2026
【50】Analyzing Data Quality and Decay in Mega-Constellations: A Physics-Informed Machine Learning Approach
标题：分析巨型星座中的数据质量和衰退：一种基于物理知识的机器学习方法
链接：https://arxiv.org/abs/2510.11242
备注：76th International Astronautical Congress
【51】Enhanced Sampling for Efficient Learning of Coarse-Grained Machine Learning Potentials
标题：增强采样以有效学习粗粒度机器学习潜力
链接：https://arxiv.org/abs/2510.11148
【52】Missing Data Multiple Imputation for Tabular Q-Learning in Online RL
标题：在线RL中表格Q-Learning的缺失数据多重插补
链接：https://arxiv.org/abs/2510.10709
备注：Working paper
【53】Uncovering Singularities in Feynman Integrals via Machine Learning
标题：通过机器学习揭示费曼积分中的奇异性
链接：https://arxiv.org/abs/2510.10099
【54】Calibrating Generative Models
标题：校准生成模型
链接：https://arxiv.org/abs/2510.10020
备注：Our codebase accompanying the paper is available at: this https URL
【55】Learning with Incomplete Context: Linear Contextual Bandits with Pretrained Imputation
标题：在不完整背景下学习：具有预先训练的线性背景盗贼
链接：https://arxiv.org/abs/2510.09908
【56】Performance of Machine Learning Methods for Gravity Inversion: Successes and Challenges
标题：重力倒置机器学习方法的性能：成功与挑战
链接：https://arxiv.org/abs/2510.09632
其他(68篇)
【1】Reinforced sequential Monte Carlo for amortised sampling
标题：用于摊销抽样的增强序贯蒙特卡罗方法
链接：https://arxiv.org/abs/2510.11711
备注：code: this https URL
【2】Tight Regret Upper and Lower Bounds for Optimistic Hedge in Two-Player Zero-Sum Games
标题：两人零和游戏中乐观对冲的上下限和下限
链接：https://arxiv.org/abs/2510.11691
备注：29 pages, 2 figures
【3】Chronologically Consistent Generative AI
标题：时间一致的生成人工智能
链接：https://arxiv.org/abs/2510.11677
【4】An Eulerian Perspective on Straight-Line Sampling
标题：直线抽样的欧拉观点
链接：https://arxiv.org/abs/2510.11657
【5】Continual Release of Densest Subgraphs: Privacy Amplification & Sublinear Space via Subsampling
标题：连续发布密集子图：通过二次采样实现隐私放大和次线性空间
链接：https://arxiv.org/abs/2510.11640
备注：to be published in SOSA'26
【6】Attention Factors for Statistical Arbitrage
标题：统计套利的注意因素
链接：https://arxiv.org/abs/2510.11616
备注：Accepted to the 6th ACM International Conference on AI in Finance
【7】SemCSE-Multi: Multifaceted and Decodable Embeddings for Aspect-Specific and Interpretable Scientific Domain Mapping
标题：SemCSE-Multi：用于航空航天特定和可解释科学领域映射的多面和可解码嵌入
链接：https://arxiv.org/abs/2510.11599
【8】Part II: ROLL Flash -- Accelerating RLVR and Agentic Training with Asynchrony
标题：第二部分：ROLL Flash --利用非同步加速WLVR和动态训练
链接：https://arxiv.org/abs/2510.11345
【9】LouisKV: Efficient KV Cache Retrieval for Long Input-Output Sequences
标题：LouisKV：针对长输入输出序列的高效KV缓存检索
链接：https://arxiv.org/abs/2510.11292
【10】Gym-TORAX: Open-source software for integrating RL with plasma control simulators
标题：Gym-TORAX：用于集成RL与等离子体控制模拟器的开源软件
链接：https://arxiv.org/abs/2510.11283
【11】DemoHLM: From One Demonstration to Generalizable Humanoid Loco-Manipulation
标题：DemoHLM：从一个演示到可推广的类人机器人操作
链接：https://arxiv.org/abs/2510.11258
【12】PhysHSI: Towards a Real-World Generalizable and Natural Humanoid-Scene Interaction System
标题：PhysHSI：迈向现实世界可概括和自然的类人场景交互系统
链接：https://arxiv.org/abs/2510.11072
备注：Project website: this https URL
【13】The Easy Path to Robustness: Coreset Selection using Sample Hardness
标题：鲁棒性的简单途径：使用样本硬度选择核心集
链接：https://arxiv.org/abs/2510.11018
【14】GrASP: A Generalizable Address-based Semantic Prefetcher for Scalable Transactional and Analytical Workloads
标题：GrISP：一个可推广的基于地址的语义预取器，用于可扩展的事务处理和分析工作负载
链接：https://arxiv.org/abs/2510.11011
备注：This is a preprint version
【15】Blade: A Derivative-free Bayesian Inversion Method using Diffusion Priors
标题：Blade：一种使用扩散先验的无导性Bayesian倒置方法
链接：https://arxiv.org/abs/2510.10968
【16】End-to-end Sequence Labeling via Bi-directional LSTM-CNNs-CRF: A Reproducibility Study
标题：通过双向LSTM-CNNS-RF进行端到端序列标记：一项重复性研究
链接：https://arxiv.org/abs/2510.10936
【17】Quantifying Information Disclosure During Gradient Descent Using Gradient Uniqueness
标题：使用梯度唯一性量化梯度下降期间的信息披露
链接：https://arxiv.org/abs/2510.10902
【18】Topological Alignment of Shared Vision-Language Embedding Space
标题：共享视觉语言嵌入空间的布局对齐
链接：https://arxiv.org/abs/2510.10889
备注：24 pages, 5 figures, 19 tables
【19】Fast and the Furious: Hot Starts in Pursuit-Evasion Games
标题：速度与激情：追逐逃避游戏的热门开始
链接：https://arxiv.org/abs/2510.10830
备注：Presented at AAMAS Workshop on Autonomous Robots and Multirobot Systems (ARMS)
【20】Aegis: A Correlation-Based Data Masking Advisor for Data Sharing Ecosystems
标题：Aegis：数据共享生态系统基于相关性的数据掩蔽顾问
链接：https://arxiv.org/abs/2510.10810
备注：Accepted at SIGMOD 2026
【21】Crisis-Aware Regime-Conditioned Diffusion with CVaR Allocation
标题：具有危机意识的制度条件扩散和CVaR分配
链接：https://arxiv.org/abs/2510.10807
备注：Code available at: this https URL
【22】PruneGCRN: Minimizing and explaining spatio-temporal problems through node pruning
标题：PruneGCRN：通过节点修剪最小化和解释时空问题
链接：https://arxiv.org/abs/2510.10803
【23】MSCloudCAM: Cross-Attention with Multi-Scale Context for Multispectral Cloud Segmentation
标题：MSCloudCAM：多尺度上下文的交叉注意力用于多光谱云分割
链接：https://arxiv.org/abs/2510.10802
备注：7 pages, 2 Figures
【24】BioOSS: A Bio-Inspired Oscillatory State System with Spatio-Temporal Dynamics
标题：BioOSS：具有时空动力学的生物启发振荡状态系统
链接：https://arxiv.org/abs/2510.10790
【25】ParsVoice: A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis
标题：ParsVoice：一个用于文语合成的大规模多人波斯语语音语料库
链接：https://arxiv.org/abs/2510.10774
【26】Provable Anytime Ensemble Sampling Algorithms in Nonlinear Contextual Bandits
标题：非线性上下文盗贼中可证明的随时集合采样算法
链接：https://arxiv.org/abs/2510.10730
备注：40 pages, 1 figure
【27】Trustworthy Retrosynthesis: Eliminating Hallucinations with a Diverse Ensemble of Reaction Scorers
标题：值得信赖的逆转录：通过反应评分者的多样化群体消除幻觉
链接：https://arxiv.org/abs/2510.10645
【28】DCP: Addressing Input Dynamism In Long-Context Training via Dynamic Context Parallelism
标题：DPP：通过动态上下文并行主义解决长上下文训练中的输入动态性
链接：https://arxiv.org/abs/2510.10620
备注：16 pages, 22 figures
【29】Budget Allocation for Unknown Value Functions in a Lipschitz Space
标题：Lipschitz空间中未知值函数的预算分配
链接：https://arxiv.org/abs/2510.10605
【30】Compositional Symmetry as Compression: Lie Pseudogroup Structure in Algorithmic Agents
标题：作为压缩的组成对称性：心律失常剂中的李伪群结构
链接：https://arxiv.org/abs/2510.10586
备注：Submitted to NeurReps 2025 (this https URL)
【31】Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?
标题：重新思考RL评估：基准能否真正揭示RL方法的失败？
链接：https://arxiv.org/abs/2510.10541
【32】MCE: Towards a General Framework for Handling Missing Modalities under Imbalanced Missing Rates
标题：MCE：建立一个在缺失率不平衡下处理缺失模式的一般框架
链接：https://arxiv.org/abs/2510.10534
备注：This is the accepted version of an article that has been published in \textbf{Pattern Recognition}. The final published version will be available soon
【33】f-INE: A Hypothesis Testing Framework for Estimating Influence under Training Randomness
标题：f-INE：估计训练随机性影响的假设测试框架
链接：https://arxiv.org/abs/2510.10510
【34】Anchor-based Maximum Discrepancy for Relative Similarity Testing
标题：基于锚点的相对相似性测试的最大偏差
链接：https://arxiv.org/abs/2510.10477
【35】Measuring What Matters: Connecting AI Ethics Evaluations to System Attributes, Hazards, and Harms
标题：衡量重要的事情：将人工智能道德评估与系统属性、危险和危害联系起来
链接：https://arxiv.org/abs/2510.10339
【36】SGM: A Statistical Godel Machine for Risk-Controlled Recursive Self-Modification
标题：SGM：一台用于风险控制的渐进式自我修改的统计Godel机
链接：https://arxiv.org/abs/2510.10232
【37】BrainForm: a Serious Game for BCI Training and Data Collection
标题：BrainForm：BCI训练和数据收集的严肃游戏
链接：https://arxiv.org/abs/2510.10169
备注：15 pages, 6 figures. Author-accepted version. Accepted for presentation at the Brain Informatics 2025 conference, to appear in Springer Lecture Notes in Artificial Intelligence (LNAI) Brain Informatics Books Series. The final authenticated version will be available via SpringerLink
【38】Rethinking Entropy Interventions in RLVR: An Entropy Change Perspective
标题：重新思考RLVR中的熵干预：一种熵变化的视角
链接：https://arxiv.org/abs/2510.10150
【39】CacheClip: Accelerating RAG with Effective KV Cache Reuse
标题：Cache：通过有效的KV缓存重用加速RAG
链接：https://arxiv.org/abs/2510.10129
【40】Rademacher Meets Colors: More Expressivity, but at What Cost ?
标题：拉德马赫遇见色彩：更具表现力，但代价是什么？
链接：https://arxiv.org/abs/2510.10101
【41】Diversity Augmentation of Dynamic User Preference Data for Boosting Personalized Text Summarizers
标题：动态用户偏好数据的多样性增强以增强个性化文本摘要器
链接：https://arxiv.org/abs/2510.10082
【42】Operationalizing AI: Empirical Evidence on MLOps Practices, User Satisfaction, and Organizational Context
标题：人工智能操作化：MLOps实践、用户满意度和组织环境的经验证据
链接：https://arxiv.org/abs/2510.09968
【43】Homomorphic Mappings for Value-Preserving State Aggregation in Markov Decision Processes
标题：Markov决策过程中保值状态集结的同态映射
链接：https://arxiv.org/abs/2510.09965
【44】Beyond Fertility: Analyzing STRR as a Metric for Multilingual Tokenization Evaluation
标题：超越生育率：分析STRR作为多语言标记化评估的指标
链接：https://arxiv.org/abs/2510.09947
备注：NeurIPS 2025 Workshop
【45】Conformal Sparsification for Bandwidth-Efficient Edge-Cloud Speculative Decoding
标题：带宽高效边缘云推测解码的共形稀疏化
链接：https://arxiv.org/abs/2510.09942
备注：39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: AI and ML for Next-Generation Wireless Communications and Networking (AI4NextG)
【46】MemPromptTSS: Persistent Prompt Memory for Iterative Multi-Granularity Time Series State Segmentation
标题：MembertTSS：用于迭代多粒度时间序列状态分割的持久提示记忆
链接：https://arxiv.org/abs/2510.09930
备注：This paper is currently under review. The code will be made available upon acceptance
【47】Probabilistic bias adjustment of seasonal predictions of Arctic Sea Ice Concentration
标题：北冰洋冰浓度季节预测的概率偏差调整
链接：https://arxiv.org/abs/2510.09891
【48】WARC-Bench: Web Archive Based Benchmark for GUI Subtask Executions
标题：WARC-Bench：基于Web存档的图形用户界面子任务执行基准
链接：https://arxiv.org/abs/2510.09872
【49】An Exploration of Non-Euclidean Gradient Descent: Muon and its Many Variants
标题：非欧梯度下降的探索：μ子及其多种变体
链接：https://arxiv.org/abs/2510.09827
【50】A Unified Framework for Lifted Training and Inversion Approaches
标题：提升训练和倒置方法的统一框架
链接：https://arxiv.org/abs/2510.09796
【51】Causality $\neq$ Decodability, and Vice Versa: Lessons from Interpreting Counting ViTs
标题：因果关系$ eq$ Decodability和Vice Versa：解释计算ViT的教训
链接：https://arxiv.org/abs/2510.09794
【52】Building a Foundational Guardrail for General Agentic Systems via Synthetic Data
标题：通过合成数据构建通用统计系统的基础保障
链接：https://arxiv.org/abs/2510.09781
【53】Scaling Laws and Symmetry, Evidence from Neural Force Fields
标题：标度律和对称性，来自神经力场的证据
链接：https://arxiv.org/abs/2510.09768
备注：22 pages, 10 figures
【54】PatentVision: A multimodal method for drafting patent applications
标题：PatentVision：起草专利申请的多模式方法
链接：https://arxiv.org/abs/2510.09762
【55】Patentformer: A demonstration of AI-assisted automated patent drafting
标题：Patentformer：人工智能辅助的自动化专利起草演示
链接：https://arxiv.org/abs/2510.09752
【56】Constructive Distortion: Improving MLLMs with Attention-Guided Image Warping
标题：结构性失真：通过注意力引导图像扭曲来改善MLLM
链接：https://arxiv.org/abs/2510.09741
【57】Federated k-Means via Generalized Total Variation Minimization
标题：基于广义全变差最小化的联邦k均值算法
链接：https://arxiv.org/abs/2510.09718
【58】A Multi-Component Reward Function with Policy Gradient for Automated Feature Selection with Dynamic Regularization and Bias Mitigation
标题：具有策略梯度的多分量奖励函数，用于自动特征选择，具有动态正规化和偏差缓解
链接：https://arxiv.org/abs/2510.09705
【59】Neural PDE Solvers with Physics Constraints: A Comparative Study of PINNs, DRM, and WANs
标题：具有物理约束的神经PED求解器：PINN、DRM和WAN的比较研究
链接：https://arxiv.org/abs/2510.09693
备注：50 pages, 13 figures
【60】Coupled Data and Measurement Space Dynamics for Enhanced Diffusion Posterior Sampling
标题：增强扩散后验抽样的耦合数据和测量空间动力学
链接：https://arxiv.org/abs/2510.09676
【61】Population synthesis with geographic coordinates
标题：具有地理坐标的人口合成
链接：https://arxiv.org/abs/2510.09669
【62】Efficient Group Lasso Regularized Rank Regression with Data-Driven Parameter Determination
标题：具有数据驱动参数确定的高效群Lasso正规化排序回归
链接：https://arxiv.org/abs/2510.11546
备注：36 pages, 4 figures, 8 tables
【63】PAC-Bayesian Bounds on Constrained f-Entropic Risk Measures
标题：约束f-熵风险指标的PAC-Bayesian界
链接：https://arxiv.org/abs/2510.11169
【64】torchsom: The Reference PyTorch Library for Self-Organizing Maps
标题：torchsom：自组织地图的PyTorch参考库
链接：https://arxiv.org/abs/2510.11147
备注：4 mains pages with 2 tables, 4 pages of references, 15 pages of appendices with 13 figures and 3 tables
【65】How Patterns Dictate Learnability in Sequential Data
标题：模式如何决定序列数据中的可学习性
链接：https://arxiv.org/abs/2510.10744
备注：NeurIPS 2025, 36 pages, 4 figures
【66】Kernel Treatment Effects with Adaptively Collected Data
标题：适应性收集数据的核心治疗效果
链接：https://arxiv.org/abs/2510.10245
【67】Neuro-inspired automated lens design
标题：神经启发的自动镜头设计
链接：https://arxiv.org/abs/2510.09979
【68】Egocentric Visual Navigation through Hippocampal Sequences
标题：通过海马序列的自我中心视觉导航
链接：https://arxiv.org/abs/2510.09951
备注：20 pages, 21 figures. This is a conference submission
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