|
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385篇
大模型相关(40篇)
【1】How Far Can Unsupervised RLVR Scale LLM Training?
标题：无监督的WLVR可以在多大程度上衡量LLM训练？
链接：https://arxiv.org/abs/2603.08660
备注：Accepted to the ICLR 2026
摘要：Unsupervised reinforcement learning with verifiable rewards (URLVR) offers a pathway to scale LLM training beyond the supervision bottleneck by deriving rewards without ground truth labels. Recent works leverage model intrinsic signals, showing promising early gains, yet their potential and limitations remain unclear. In this work, we revisit URLVR and provide a comprehensive analysis spanning taxonomy, theory and extensive experiments. We first classify URLVR methods into intrinsic versus external based on reward sources, then establish a unified theoretical framework revealing that all intrinsic methods converge toward sharpening the model's initial distribution This sharpening mechanism succeeds when initial confidence aligns with correctness but fails catastrophically when misaligned. Through systematic experiments, we show intrinsic rewards consistently follow a rise-then-fall pattern across methods, with collapse timing determined by model prior rather than engineering choices. Despite these scaling limits, we find intrinsic rewards remain valuable in test-time training on small datasets, and propose Model Collapse Step to measure model prior, serving as a practical indicator for RL trainability. Finally, we explore external reward methods that ground verification in computational asymmetries, showing preliminary evidence they may escape the confidence-correctness ceiling. Our findings chart boundaries for intrinsic URLVR while motivating paths toward scalable alternatives.
【2】PostTrainBench: Can LLM Agents Automate LLM Post-Training?
标题：PostTrainBench：LLM代理可以自动化LLM后训练吗？
链接：https://arxiv.org/abs/2603.08640
摘要：AI agents have become surprisingly proficient at software engineering over the past year, largely due to improvements in reasoning capabilities. This raises a deeper question: can these systems extend their capabilities to automate AI research itself? In this paper, we explore post-training, the critical phase that turns base LLMs into useful assistants. We introduce PostTrainBench to benchmark how well LLM agents can perform post-training autonomously under bounded compute constraints (10 hours on one H100 GPU). We ask frontier agents (e.g., Claude Code with Opus 4.6) to optimize the performance of a base LLM on a particular benchmark (e.g., Qwen3-4B on AIME). Importantly, we do not provide any predefined strategies to the agents and instead give them full autonomy to find necessary information on the web, run experiments, and curate data. We find that frontier agents make substantial progress but generally lag behind instruction-tuned LLMs from leading providers: 23.2% for the best agent vs. 51.1% for official instruction-tuned models. However, agents can exceed instruction-tuned models in targeted scenarios: GPT-5.1 Codex Max achieves 89% on BFCL with Gemma-3-4B vs. 67% for the official model. We also observe several failure modes worth flagging. Agents sometimes engage in reward hacking: training on the test set, downloading existing instruction-tuned checkpoints instead of training their own, and using API keys they find to generate synthetic data without authorization. These behaviors are concerning and highlight the importance of careful sandboxing as these systems become more capable. Overall, we hope PostTrainBench will be useful for tracking progress in AI R&D automation and for studying the risks that come with it. Website and code are available at https://posttrainbench.com/.
【3】Revealing Behavioral Plasticity in Large Language Models: A Token-Conditional Perspective
标题：揭示大型语言模型中的行为可塑性：标记条件视角
链接：https://arxiv.org/abs/2603.08398
备注：Work done during an internship at the Qwen Team, Alibaba Group
摘要：In this work, we reveal that Large Language Models (LLMs) possess intrinsic behavioral plasticity-akin to chameleons adapting their coloration to environmental cues-that can be exposed through token-conditional generation and stabilized via reinforcement learning. Specifically, by conditioning generation on carefully selected token prefixes sampled from responses exhibiting desired behaviors, LLMs seamlessly adapt their behavioral modes at inference time (e.g., switching from step-by-step reasoning to direct answering) without retraining. Based on this insight, we propose Token-Conditioned Reinforcement Learning (ToCoRL), a principled framework that leverages RL to internalize this chameleon-like plasticity, transforming transient inference-time adaptations into stable and learnable behavioral patterns. ToCoRL guides exploration with token-conditional generation and keep enhancing exploitation, enabling emergence of appropriate behaviors. Extensive experiments show that ToCoRL enables precise behavioral control without capability degradation. Notably, we show that large reasoning models, while performing strongly on complex mathematics, can be effectively adapted to excel at factual question answering, which was a capability previously hindered by their step-by-step reasoning patterns.
【4】Towards a more efficient bias detection in financial language models
标题：在金融语言模型中实现更有效的偏见检测
链接：https://arxiv.org/abs/2603.08267
摘要：Bias in financial language models constitutes a major obstacle to their adoption in real-world applications. Detecting such bias is challenging, as it requires identifying inputs whose predictions change when varying properties unrelated to the decision, such as demographic attributes. Existing approaches typically rely on exhaustive mutation and pairwise prediction analysis over large corpora, which is effective but computationally expensive-particularly for large language models and can become impractical in continuous retraining and releasing processes. Aiming at reducing this cost, we conduct a large-scale study of bias in five financial language models, examining similarities in their bias tendencies across protected attributes and exploring cross-model-guided bias detection to identify bias-revealing inputs earlier. Our study uses approximately 17k real financial news sentences, mutated to construct over 125k original-mutant pairs. Results show that all models exhibit bias under both atomic (0.58\%-6.05\%) and intersectional (0.75\%-5.97\%) settings. Moreover, we observe consistent patterns in bias-revealing inputs across models, enabling substantial reuse and cost reduction in bias detection. For example, up to 73\% of FinMA's biased behaviours can be uncovered using only 20\% of the input pairs when guided by properties derived from DistilRoBERTa outputs.
【5】The Struggle Between Continuation and Refusal: A Mechanistic Analysis of the Continuation-Triggered Jailbreak in LLMs
标题：延续与拒绝之间的斗争：法学硕士继续引发越狱的机制分析
链接：https://arxiv.org/abs/2603.08234
摘要：With the rapid advancement of large language models (LLMs), the safety of LLMs has become a critical concern. Despite significant efforts in safety alignment, current LLMs remain vulnerable to jailbreaking attacks. However, the root causes of such vulnerabilities are still poorly understood, necessitating a rigorous investigation into jailbreak mechanisms across both academic and industrial communities. In this work, we focus on a continuation-triggered jailbreak phenomenon, whereby simply relocating a continuation-triggered instruction suffix can substantially increase jailbreak success rates. To uncover the intrinsic mechanisms of this phenomenon, we conduct a comprehensive mechanistic interpretability analysis at the level of attention heads. Through causal interventions and activation scaling, we show that this jailbreak behavior primarily arises from an inherent competition between the model's intrinsic continuation drive and the safety defenses acquired through alignment training. Furthermore, we perform a detailed behavioral analysis of the identified safety-critical attention heads, revealing notable differences in the functions and behaviors of safety heads across different model architectures. These findings provide a novel mechanistic perspective for understanding and interpreting jailbreak behaviors in LLMs, offering both theoretical insights and practical implications for improving model safety.
【6】SERQ: Saliency-Aware Low-Rank Error Reconstruction for LLM Quantization
标题：SEN：LLM量化的显着性感知低等级错误重建
链接：https://arxiv.org/abs/2603.08185
备注：21 pages, 4 figures
摘要：Post-training quantization (PTQ) has emerged as a prevailing technique for deploying large language models (LLMs) efficiently in terms of both memory and computation, across edge devices and server platforms. Existing PTQ methods primarily aim to reduce precision in weights and activations by mitigating quantization errors caused by channel-wise outlier activations (e.g., pre-quantization scaling, online transformations, or low-rank error reconstruction). Among these approaches, error reconstruction with low-rank adaptation (LoRA) has proven particularly effective, as it introduces a lightweight auxiliary computation path without requiring heavy optimization or additional online layers. However, prior studies reveal severe accuracy degradation under W4A4 settings, and conventional low-rank adaptations rely on two sequential factors, necessitating intermediate quantization during inference and thereby limiting low-precision efficiency. In this work, we propose SERQ, a saliency-aware error reconstruction method for low-bit LLM inference that employs a single low-rank compensation matrix. SERQ preserves efficient 4-bit matrix multiplication in linear layers by jointly mitigating quantization errors arising from both activation and weight saliency through three stages: (1) static activation flattening, (2) saliency-aware error reconstruction, and (3) offline weight permutation. The method incurs additional computation only for low-rank error reconstruction via a single decomposition, while all other operations are performed offline, thereby keeping latency overhead minimal. Empirically, SERQ outperforms prior error reconstruction methods under both W4A8 and W4A4 settings, and achieves higher accuracy than state-of-the-art rotation-based W4A4 approaches, while substantially reducing calibration complexity.
【7】AutoAdapt: An Automated Domain Adaptation Framework for LLMs
标题：AutoAdapt：LLM的自动化领域适应框架
链接：https://arxiv.org/abs/2603.08181
摘要：Large language models (LLMs) excel in open domains but struggle in specialized settings with limited data and evolving knowledge. Existing domain adaptation practices rely heavily on manual trial-and-error processes, incur significant hyperparameter complexity, and are highly sensitive to data and user preferences, all under the high cost of LLM training. Moreover, the interactions and transferability of hyperparameter choices across models/domains remain poorly understood, making adaptation gains uncertain even with substantial effort. To solve these challenges, we present AutoAdapt, a novel end-to-end automated framework for efficient and reliable LLM domain adaptation. AutoAdapt leverages curated knowledge bases from literature and open-source resources to reduce expert intervention. To narrow the search space, we design a novel multi-agent debating system in which proposal and critic agents iteratively interact to align user intent and incorporate data signals and best practices into the planning process. To optimize hyperparameters under tight budgets, we propose AutoRefine, a novel LLM-based surrogate that replaces costly black-box search. Across 10 tasks, AutoAdapt achieves a 25% average relative accuracy improvement over state-of-the-art Automated Machine Learning baselines with minimal overhead.
【8】Covenant-72B: Pre-Training a 72B LLM with Trustless Peers Over-the-Internet
标题：Covenant-72 B：通过互联网与不信任的同行对72 B LLM进行预训练
链接：https://arxiv.org/abs/2603.08163
备注：26 pages, 6 figures, 4 tables
摘要：Recently, there has been increased interest in globally distributed training, which has the promise to both reduce training costs and democratize participation in building large-scale foundation models. However, existing models trained in a globally distributed manner are relatively small in scale and have only been trained with whitelisted participants. Therefore, they do not yet realize the full promise of democratized participation. In this report, we describe Covenant-72B, an LLM produced by the largest collaborative globally distributed pre-training run (in terms of both compute and model scale), which simultaneously allowed open, permissionless participation supported by a live blockchain protocol. We utilized a state-of-the-art communication-efficient optimizer, SparseLoCo, supporting dynamic participation with peers joining and leaving freely. Our model, pre-trained on approximately 1.1T tokens, performs competitively with fully centralized models pre-trained on similar or higher compute budgets, demonstrating that fully democratized, non-whitelisted participation is not only feasible, but can be achieved at unprecedented scale for a globally distributed pre-training run.
【9】Invisible Safety Threat: Malicious Finetuning for LLM via Steganography
标题：隐形安全威胁：通过隐写术对LLM进行恶意微调
链接：https://arxiv.org/abs/2603.08104
摘要：Understanding and addressing potential safety alignment risks in large language models (LLMs) is critical for ensuring their safe and trustworthy deployment. In this paper, we highlight an insidious safety threat: a compromised LLM can maintain a facade of proper safety alignment while covertly generating harmful content. To achieve this, we finetune the model to understand and apply a steganographic technique. At inference time, we input a prompt that contains a steganographically embedded malicious target question along with a plaintext cover question. The model, in turn, produces a target response similarly embedded within a benign-looking cover response. In this process, human observers only see the model being prompted with a cover question and generating a corresponding cover response, while the malicious content is hidden from view. We demonstrate this invisible safety threat on GPT-4.1 despite the OpenAI finetuning API's safeguards. The finetuned model produces steganographic malicious outputs in response to hidden malicious prompts, while the user interface displays only a fully benign cover interaction. We also replicate the attack on three open-source models, Llama-3.3-70B-Instruct, Phi-4, and Mistral-Small-24B-Base-2501, confirming the generality of our method. We quantitatively evaluate our method on the AdvBench dataset, using Llama-Guard-3-8B for content safety classification. Across all four models, all stegotexts containing malicious content are incorrectly classified as safe.
【10】Deterministic Differentiable Structured Pruning for Large Language Models
标题：大型语言模型的确定性可区分结构化修剪
链接：https://arxiv.org/abs/2603.08065
摘要：Structured pruning reduces LLM inference cost by removing low-importance architectural components. This can be viewed as learning a multiplicative gate for each component under an l0 sparsity constraint. Due to the discreteness of the l0 norm, prior work typically adopts stochastic hard-concrete relaxations to enable differentiable optimization; however, this stochasticity can introduce a train--test mismatch when sampled masks are discretized for deployment and restricts masks to a bounded, near-binary range. To address this, we propose Deterministic Differentiable Pruning (DDP), a mask-only optimization method that eliminates stochasticity by directly optimizing a deterministic soft surrogate of the discrete l0 objective. Compared with prior approaches, DDP offers greater expressiveness, reduced train--test mismatch, and faster convergence. We apply our method to several dense and MoE models, including Qwen3-32B and Qwen3-30B-A3B, achieving a performance loss as small as 1% on downstream tasks while outperforming previous methods at 20% sparsity. We further demonstrate end-to-end inference speedups in realistic deployment settings with vLLM.
【11】Capacity-Aware Mixture Law Enables Efficient LLM Data Optimization
标题：容量感知混合律实现高效LLM数据优化
链接：https://arxiv.org/abs/2603.08022
摘要：A data mixture refers to how different data sources are combined to train large language models, and selecting an effective mixture is crucial for optimal downstream performance. Existing methods either conduct costly searches directly on the target model or rely on mixture scaling laws that fail to extrapolate well to large model sizes. We address these limitations by introducing a compute-efficient pipeline for data mixture scaling. First, we propose CAMEL, a capacity-aware mixture law that models validation loss with the nonlinear interplay between model size and mixture. We also introduce a loss-to-benchmark prediction law that estimates benchmark accuracy from validation loss, enabling end-to-end performance prediction for the target model. Next, we study how to allocate a fixed compute budget across model scales to fit the law and reduce prediction error. Finally, we apply our method to Mixture-of-Experts models with up to 7B-A150M parameters to fit the law, and verify the optimal mixture derived from the law by extrapolating to a 55B-A1.2B target model. Compared to prior methods, we reduces mixture optimization costs by 50\% and improves downstream benchmark performance by up to 3\%.
【12】SmartThinker: Progressive Chain-of-Thought Length Calibration for Efficient Large Language Model Reasoning
标题：SmartThinker：渐进式思维链长度校准，用于高效的大型语言模型推理
链接：https://arxiv.org/abs/2603.08000
摘要 ：Large reasoning models (LRMs) like OpenAI o1 and DeepSeek-R1 achieve high accuracy on complex tasks by adopting long chain-of-thought (CoT) reasoning paths. However, the inherent verbosity of these processes frequently results in redundancy and overthinking. To address this issue, existing works leverage Group Relative Policy Optimization (GRPO) to reduce LRM output length, but their static length reward design cannot dynamically adapt according to the relative problem difficulty and response length distribution, causing over-compression and compromised accuracy. Therefore, we propose SmartThinker, a novel GRPO-based efficient reasoning method with progressive CoT length calibration. SmartThinker makes a two-fold contribution: First, it dynamically estimates the optimal length with peak accuracy during training and guides overlong responses toward it to reduce response length while sustaining accuracy. Second, it dynamically modulates the length reward coefficient to avoid the unwarranted penalization of correct reasoning paths. Extensive experiment results show that SmartThinker achieves up to 52.5% average length compression with improved accuracy, and achieves up to 16.6% accuracy improvement on challenging benchmarks like AIME25. The source code can be found at https://github.com/SJTU-RTEAS/SmartThinker.
【13】ELLMob: Event-Driven Human Mobility Generation with Self-Aligned LLM Framework
标题：ELLMob：采用自对准LLM框架的事件驱动的人类流动性生成
链接：https://arxiv.org/abs/2603.07946
备注：Accepted by ICLR 2026
摘要：Human mobility generation aims to synthesize plausible trajectory data, which is widely used in urban system research. While Large Language Model-based methods excel at generating routine trajectories, they struggle to capture deviated mobility during large-scale societal events. This limitation stems from two critical gaps: (1) the absence of event-annotated mobility datasets for design and evaluation, and (2) the inability of current frameworks to reconcile competitions between users' habitual patterns and event-imposed constraints when making trajectory decisions. This work addresses these gaps with a twofold contribution. First, we construct the first event-annotated mobility dataset covering three major events: Typhoon Hagibis, COVID-19, and the Tokyo 2021 Olympics. Second, we propose ELLMob, a self-aligned LLM framework that first extracts competing rationales between habitual patterns and event constraints, based on Fuzzy-Trace Theory, and then iteratively aligns them to generate trajectories that are both habitually grounded and event-responsive. Extensive experiments show that ELLMob wins state-of-the-art baselines across all events, demonstrating its effectiveness. Our codes and datasets are available at https://github.com/deepkashiwa20/ELLMob.
【14】DyQ-VLA: Temporal-Dynamic-Aware Quantization for Embodied Vision-Language-Action Models
标题：DyQ-VLA：用于预定视觉-语言-动作模型的时间动态感知量化
链接：https://arxiv.org/abs/2603.07904
摘要：Vision-Language-Action (VLA) models are dominant in embodied intelligence but are constrained by inference overheads. While model quantization alleviates these bottlenecks for edge deployment, static quantization approaches remain suboptimal for VLAs due to two critical challenges: (1) Temporal-dynamic sensitivity, where fixed precision wastes resources by ignoring stage-varying error tolerances; and (2) Real-time allocation, where identifying real-time sensitivity to guide bit allocation remains unsolved. To address these challenges, we propose DyQ-VLA, a dynamic quantization framework for VLAs. Specifically, a sensitivity-aware switching strategy leverages real-time kinematic proxies to trigger the bit-width switch, while a kinematic-guided module dynamically allocates the optimal bit-width. Experiments show that DyQ-VLA requires only 30.9% of the original memory footprint while maintaining 99.5% of its original performance, achieving 1.49x simulation and up to 1.43x real-world speedups.
【15】LeJOT-AutoML: LLM-Driven Feature Engineering for Job Execution Time Prediction in Databricks Cost Optimization
标题：LeJOT-AutoML：LLM-Driven Feature Engineering for Job Execution Time Prediction in Databricks Cost Optimization
链接：https://arxiv.org/abs/2603.07897
摘要：Databricks job orchestration systems (e.g., LeJOT) reduce cloud costs by selecting low-priced compute configurations while meeting latency and dependency constraints. Accurate execution-time prediction under heterogeneous instance types and non-stationary runtime conditions is therefore critical. Existing pipelines rely on static, manually engineered features that under-capture runtime effects (e.g., partition pruning, data skew, and shuffle amplification), and predictive signals are scattered across logs, metadata, and job scripts-lengthening update cycles and increasing engineering overhead. We present LeJOT-AutoML, an agent-driven AutoML framework that embeds large language model agents throughout the ML lifecycle. LeJOT-AutoML combines retrieval-augmented generation over a domain knowledge base with a Model Context Protocol toolchain (log parsers, metadata queries, and a read-only SQL sandbox) to analyze job artifacts, synthesize and validate feature-extraction code via safety gates, and train/select predictors. This design materializes runtime-derived features that are difficult to obtain through static analysis alone. On enterprise Databricks workloads, LeJOT-AutoML generates over 200 features and reduces the feature-engineering and evaluation loop from weeks to 20-30 minutes, while maintaining competitive prediction accuracy. Integrated into the LeJOT pipeline, it enables automated continuous model updates and achieves 19.01% cost savings in our deployment setting through improved orchestration.
【16】Reject, Resample, Repeat: Understanding Parallel Reasoning in Language Model Inference
标题：重新采样、重复：理解语言模型推理中的并行推理
链接：https://arxiv.org/abs/2603.07887
摘要 ：Inference-time methods that aggregate and prune multiple samples have emerged as a powerful paradigm for steering large language models, yet we lack any principled understanding of their accuracy-cost tradeoffs. In this paper, we introduce a route to rigorously study such approaches using the lens of *particle filtering* algorithms such as Sequential Monte Carlo (SMC). Given a base language model and a *process reward model* estimating expected terminal rewards, we ask: *how accurately can we sample from a target distribution given some number of process reward evaluations?* Theoretically, we identify (1) simple criteria enabling non-asymptotic guarantees for SMC; (2) algorithmic improvements to SMC; and (3) a fundamental limit faced by all particle filtering methods. Empirically, we demonstrate that our theoretical criteria effectively govern the *sampling error* of SMC, though not necessarily its final *accuracy*, suggesting that theoretical perspectives beyond sampling may be necessary.
【17】Hospitality-VQA: Decision-Oriented Informativeness Evaluation for Vision-Language Models
标题：Hospitality-VQA：视觉语言模型的面向决策的信息性评估
链接：https://arxiv.org/abs/2603.07868
备注：Accepted at EACL 2026 SRW. 16 pages
摘要：Recent advances in Vision-Language Models (VLMs) have demonstrated impressive multimodal understanding in general domains. However, their applicability to decision-oriented domains such as hospitality remains largely unexplored. In this work, we investigate how well VLMs can perform visual question answering (VQA) about hotel and facility images that are central to consumer decision-making. While many existing VQA benchmarks focus on factual correctness, they rarely capture what information users actually find useful. To address this, we first introduce Informativeness as a formal framework to quantify how much hospitality-relevant information an image-question pair provides. Guided by this framework, we construct a new hospitality-specific VQA dataset that covers various facility types, where questions are specifically designed to reflect key user information needs. Using this benchmark, we conduct experiments with several state-of-the-art VLMs, revealing that VLMs are not intrinsically decision-aware-key visual signals remain underutilized, and reliable informativeness reasoning emerges only after modest domain-specific finetuning.
【18】Using GPUs And LLMs Can Be Satisfying for Nonlinear Real Arithmetic Problems
标题：使用图形处理器和LLM可以满足非线性实算术问题
链接：https://arxiv.org/abs/2603.07764
备注：Workshop submission, minor errors fixed
摘要：Solving quantifier-free non-linear real arithmetic (NRA) problems is a computationally hard task. To tackle this problem, prior work proposed a promising approach based on gradient descent. In this work, we extend their ideas and combine LLMs and GPU acceleration to obtain an efficient technique. We have implemented our findings in the novel SMT solver GANRA (GPU Accelerated solving of Nonlinear Real Arithmetic problems). We evaluate GANRA on two different NRA benchmarks and demonstrate significant improvements over the previous state of the art. In particular, on the Sturm-MBO benchmark, we can prove satisfiability for more than five times as many instances in less than 1/20th of the previous state-of-the-art runtime.
【19】Reverse Distillation: Consistently Scaling Protein Language Model Representations
标题：反向蒸馏：一致缩放蛋白质语言模型表示
链接：https://arxiv.org/abs/2603.07710
备注：Proceedings of ICLR 2026
摘要：Unlike the predictable scaling laws in natural language processing and computer vision, protein language models (PLMs) scale poorly: for many tasks, models within the same family plateau or even decrease in performance, with mid-sized models often outperforming the largest in the family. We introduce Reverse Distillation, a principled framework that decomposes large PLM representations into orthogonal subspaces guided by smaller models of the same family. The resulting embeddings have a nested, Matryoshka-style structure: the first k dimensions of a larger model's embedding are exactly the representation from the smaller model. This ensures that larger reverse-distilled models consistently outperform smaller ones. A motivating intuition is that smaller models, constrained by capacity, preferentially encode broadly-shared protein features. Reverse distillation isolates these shared features and orthogonally extracts additional contributions from larger models, preventing interference between the two. On ProteinGym benchmarks, reverse-distilled ESM-2 variants outperform their respective baselines at the same embedding dimensionality, with the reverse-distilled 15 billion parameter model achieving the strongest performance. Our framework is generalizable to any model family where scaling challenges persist. Code and trained models are available at https://github.com/rohitsinghlab/plm_reverse_distillation.
【20】TS-MLLM: A Multi-Modal Large Language Model-based Framework for Industrial Time-Series Big Data Analysis
标题：TS-MLLM：用于工业时间序列大数据分析的基于多模式大语言模型的框架
链接：https://arxiv.org/abs/2603.07572
【21】wDPO: Winsorized Direct Preference Optimization for Robust LLM Alignment
标题：wDPO：用于稳健LLM对齐的Winsorized直接偏好优化
链接：https://arxiv.org/abs/2603.07211
【22】Making LLMs Optimize Multi-Scenario CUDA Kernels Like Experts
标题：让LLM像专家一样优化多场景CUDA核心
链接：https://arxiv.org/abs/2603.07169
【23】Entropy-Aware On-Policy Distillation of Language Models
标题：语言模型的信息感知性策略提炼
链接：https://arxiv.org/abs/2603.07079
备注：16 pages, 11 figures, preprint
【24】Resource-Adaptive Federated Text Generation with Differential Privacy
标题：具有差异隐私的资源自适应联邦文本生成
链接：https://arxiv.org/abs/2603.07027
备注：Accepted by DATA-FM workshop @ ICLR 2026
【25】Can Safety Emerge from Weak Supervision? A Systematic Analysis of Small Language Models
标题：监管薄弱能否保障安全？小语言模型的系统分析
链接：https://arxiv.org/abs/2603.07017
备注：19 pages, 10 tables, 7 figures, under Review
【26】Adaptive Discovery of Interpretable Audio Attributes with Multimodal LLMs for Low-Resource Classification
标题：利用多模式LLM自适应发现可解释音频属性以实现低资源分类
链接：https://arxiv.org/abs/2603.06991
备注：5 pages, 1 figure
【27】NerVE: Nonlinear Eigenspectrum Dynamics in LLM Feed-Forward Networks
标题：NerVE：LLM前向网络中的非线性特征谱动力学
链接：https://arxiv.org/abs/2603.06922
备注：Published at ICLR 2026
【28】Contextual Counterfactual Credit Assignment for Multi-Agent Reinforcement Learning in LLM Collaboration
标题：LLM协作中多智能体强化学习的上下文反事实学分分配
链接：https://arxiv.org/abs/2603.06859
【29】Enhancing Instruction Following of LLMs via Activation Steering with Dynamic Rejection
标题：通过具有动态拒绝的激活引导增强LLM的指令遵循
链接：https://arxiv.org/abs/2603.06745
备注：Accepted at ICLR 2026
【30】Stabilizing Reinforcement Learning for Diffusion Language Models
标题：稳定扩散语言模型的强化学习
链接：https://arxiv.org/abs/2603.06743
【31】Orion: Characterizing and Programming Apple's Neural Engine for LLM Training and Inference
标题：Orion：描述和编程Apple用于LLM训练和推理的神经引擎
链接：https://arxiv.org/abs/2603.06728
【32】HEARTS: Benchmarking LLM Reasoning on Health Time Series
标题：HeARTS：LLM推理对健康时间序列进行基准测试
链接：https://arxiv.org/abs/2603.06638
【33】SmartBench: Evaluating LLMs in Smart Homes with Anomalous Device States and Behavioral Contexts
标题：SmartBench：评估具有异常设备状态和行为背景的智能家居中的LLM
链接：https://arxiv.org/abs/2603.06636
【34】Graph Property Inference in Small Language Models: Effects of Representation and Inference Strategy
标题：小语言模型中的图属性推理：表示和推理策略的影响
链接：https://arxiv.org/abs/2603.06635
【35】Evo: Autoregressive-Diffusion Large Language Models with Evolving Balance
标题：Evo：具有不断演变的平衡的自回归扩散大型语言模型
链接：https://arxiv.org/abs/2603.06617
【36】RACER: Risk-Aware Calibrated Efficient Routing for Large Language Models
标题：RABER：针对大型语言模型的风险感知校准高效路由
链接：https://arxiv.org/abs/2603.06616
【37】Consensus is Not Verification: Why Crowd Wisdom Strategies Fail for LLM Truthfulness
标题：共识不是验证：为什么群体智慧策略在LLM真实性方面失败
链接：https://arxiv.org/abs/2603.06612
【38】CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training
标题：CapTrack：LLM后训练中遗忘的多方面评估
链接：https://arxiv.org/abs/2603.06610
【39】Know When You're Wrong: Aligning Confidence with Correctness for LLM Error Detection
标题：知道自己错了：将信心与正确性结合起来进行LLM错误检测
链接：https://arxiv.org/abs/2603.06604
【40】How Attention Sinks Emerge in Large Language Models: An Interpretability Perspective
标题：注意力如何在大型语言模型中出现：可解释性的角度
链接：https://arxiv.org/abs/2603.06591
Graph相关(图学习|图神经网络|图优化等)(18篇)
【1】Towards Effective and Efficient Graph Alignment without Supervision
标题：在没有监督的情况下实现有效且高效的图形对齐
链接：https://arxiv.org/abs/2603.08526
备注：World Wide Web Journal
【2】Graph-Instructed Neural Networks for parametric problems with varying boundary conditions
标题：图形指导神经网络用于变化边界条件的参数问题
链接：https://arxiv.org/abs/2603.08304
【3】SCL-GNN: Towards Generalizable Graph Neural Networks via Spurious Correlation Learning
标题：SCL-GNN：通过伪相关学习迈向可推广图神经网络
链接：https://arxiv.org/abs/2603.08270
【4】Mitigating Homophily Disparity in Graph Anomaly Detection: A Scalable and Adaptive Approach
标题：缓解图异常检测中的同质性差异：一种可扩展和自适应的方法
链接：https://arxiv.org/abs/2603.08137
备注：Accepted by WWW 2026
【5】GCGNet: Graph-Consistent Generative Network for Time Series Forecasting with Exogenous Variables
标题：GCGNet：具有外生变量的时间序列预测的图一致生成网络
链接：https://arxiv.org/abs/2603.08032
【6】Hide and Find: A Distributed Adversarial Attack on Federated Graph Learning
标题：隐藏和查找：对联邦图学习的分布式对抗攻击
链接：https://arxiv.org/abs/2603.07743
备注：Accepted at ICLR 2026 Workshop: Principled Design for Trustworthy AI
【7】A Dual-Graph Spatiotemporal GNN Surrogate for Nonlinear Response Prediction of Reinforced Concrete Beams under Four-Point Bending
标题：四点弯曲下钢筋混凝土梁非线性响应预测的时空双图GNN代理
链接：https://arxiv.org/abs/2603.07201
【8】Topology-Aware Reinforcement Learning over Graphs for Resilient Power Distribution Networks
标题：弹性配电网络的图上的布局感知强化学习
链接：https://arxiv.org/abs/2603.06964
【9】Not All Neighbors Matter: Understanding the Impact of Graph Sparsification on GNN Pipelines
标题：并非所有邻居都重要：了解图形稀疏化对GNN管道的影响
链接：https://arxiv.org/abs/2603.06952
【10】SpatialMAGIC: A Hybrid Framework Integrating Graph Diffusion and Spatial Attention for Spatial Transcriptomics Imputation
标题：SpatialMAGIC：一个集成图形扩散和空间注意力的混合框架，用于空间转录组学归因
链接：https://arxiv.org/abs/2603.06780
备注 ：30 pages, 6 figures
【11】HGT-Scheduler: Deep Reinforcement Learning for the Job Shop Scheduling Problem via Heterogeneous Graph Transformers
标题：HGT-SYS：通过异类图变换器解决车间调度问题的深度强化学习
链接：https://arxiv.org/abs/2603.06777
备注：23 pages, 6 figures
【12】Metalearning traffic assignment for network disruptions with graph convolutional neural networks
标题：利用图卷积神经网络应对网络中断的元收入流量分配
链接：https://arxiv.org/abs/2603.06763
【13】Approximate Nearest Neighbor Search for Modern AI: A Projection-Augmented Graph Approach
标题：现代人工智能的大约最近邻居搜索：投影增强图方法
链接：https://arxiv.org/abs/2603.06660
备注：Source code is available at https://github.com/KejingLu-810/PAG/
【14】How the Graph Construction Technique Shapes Performance in IoT Botnet Detection
标题：图构建技术如何提高物联网僵尸网络检测的性能
链接：https://arxiv.org/abs/2603.06654
【15】Leakage Safe Graph Features for Interpretable Fraud Detection in Temporal Transaction Networks
标题：时态交易网络中可解释欺诈检测的泄漏安全图特征
链接：https://arxiv.org/abs/2603.06632
备注：7 pages, 7 figures. Submitted to arXiv as a preprint
【16】Pavement Missing Condition Data Imputation through Collective Learning-Based Graph Neural Networks
标题：基于集体学习的图神经网络的路面缺失状况数据插补
链接：https://arxiv.org/abs/2603.06625
【17】GraphSkill: Documentation-Guided Hierarchical Retrieval-Augmented Coding for Complex Graph Reasoning
标题：GraphSkill：用于复杂图推理的文档引导分层检索增强编码
链接：https://arxiv.org/abs/2603.06620
备注：Under review
【18】Characterization and upgrade of a quantum graph neural network for charged particle tracking
标题：用于带电粒子跟踪的量子图神经网络的特征和升级
链接：https://arxiv.org/abs/2603.08667
备注：16 total pages, 15 figures
Transformer(17篇)
【1】Rethinking Attention Output Projection: Structured Hadamard Transforms for Efficient Transformers
标题：重新思考注意力输出预测：高效Transformer的结构化阿达玛变形
链接：https://arxiv.org/abs/2603.08343
备注：12 pages, 9 figures, 4 tables
【2】Bayesian Transformer for Probabilistic Load Forecasting in Smart Grids
标题：智能电网概率负荷预测的Bayesian Transformer
链接：https://arxiv.org/abs/2603.07899
【3】Fusion Complexity Inversion: Why Simpler Cross View Modules Outperform SSMs and Cross View Attention Transformers for Pasture Biomass Regression
标题：融合复杂性倒置：为什么更简单的交叉视图模块在牧场生物量回归方面优于RSM和交叉视图注意力转换器
链接：https://arxiv.org/abs/2603.07819
【4】Vision Transformers that Never Stop Learning
标题：永不停止学习的视觉Transformer
链接：https://arxiv.org/abs/2603.07787
【5】Interpretable-by-Design Transformers via Architectural Stream Independence
标题：通过建筑流独立性实现可设计解释的Transformer
链接：https://arxiv.org/abs/2603.07482
【6】Contact-Guided 3D Genome Structure Generation of E. coli via Diffusion Transformers
标题：接触引导的大肠杆菌3D基因组结构生成。大肠杆菌通过扩散转化器
链接：https://arxiv.org/abs/2603.07472
备注：Accepted at the Gen2 Workshop at ICLR 2026
【7】The Dual-Stream Transformer: Channelized Architecture for Interpretable Language Modeling
标题：双流Transformer：可解释语言建模的并行化架构
链接：https://arxiv.org/abs/2603.07461
【8】Discrete Tokenization Unlocks Transformers for Calibrated Tabular Forecasting
标题：离散代币化解锁变形器以实现校准表格预测
链接：https://arxiv.org/abs/2603.07448
【9】OrthoFormer: Instrumental Variable Estimation in Transformer Hidden States via Neural Control Functions
标题：OrthoFormer：基于神经控制函数的Transformer隐态辅助变量估计
链接：https://arxiv.org/abs/2603.07431
【10】Spectral Conditioning of Attention Improves Transformer Performance
标题：注意力的光谱调节提高了Transformer的性能
链接：https://arxiv.org/abs/2603.07162
备注：NeurIPS 2025
【11】RESCHED: Rethinking Flexible Job Shop Scheduling from a Transformer-based Architecture with Simplified States
标题：RESCHED：从具有简化状态的基于转换器的架构重新思考灵活的作业车间调度
链接：https://arxiv.org/abs/2603.07020
【12】A SISA-based Machine Unlearning Framework for Power Transformer Inter-Turn Short-Circuit Fault Localization
标题：基于SISA的电力Transformer线间短路故障定位机器去学习框架
链接：https://arxiv.org/abs/2603.06962
【13】Rank-Factorized Implicit Neural Bias: Scaling Super-Resolution Transformer with FlashAttention
标题：排名因子化的隐式神经偏差：使用Flash Attention缩放超分辨率Transformer
链接：https://arxiv.org/abs/2603.06738
【14】Safe Transformer: An Explicit Safety Bit For Interpretable And Controllable Alignment
标题：安全Transformer：用于可解释和可控制对齐的显式安全位
链接：https://arxiv.org/abs/2603.06727
【15】T-REX: Transformer-Based Category Sequence Generation for Grocery Basket Recommendation
标题：T-REX：针对杂货篮推荐的基于转换器的类别序列生成
链接：https://arxiv.org/abs/2603.06631
【16】Structure-Aware Set Transformers: Temporal and Variable-Type Attention Biases for Asynchronous Clinical Time Series
标题：结构感知集Transformer：非同步临床时间序列的时间和可变类型注意力偏差
链接：https://arxiv.org/abs/2603.06605
备注：Under review at the ICLR 2026 Workshop on Time Series in the Age of Large Models (TSALM)
【17】Electrocardiogram Classification with Transformers Using Koopman and Wavelet Features
标题：使用Koopman和子波特征的Transformer进行心电图分类
链接：https://arxiv.org/abs/2603.08339
GAN|对抗|攻击|生成相关(19篇)
【1】Context-free Self-Conditioned GAN for Trajectory Forecasting
标题：用于轨迹预测的无上下文自调节GAN
链接：https://arxiv.org/abs/2603.08658
备注：Accepted at the 2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)
【2】Foley-Flow: Coordinated Video-to-Audio Generation with Masked Audio-Visual Alignment and Dynamic Conditional Flows
标题：Foley-Flow：具有掩蔽视听对齐和动态条件流的协调视频到音频生成
链接：https://arxiv.org/abs/2603.08126
【3】Adversarial Domain Adaptation Enables Knowledge Transfer Across Heterogeneous RNA-Seq Datasets
标题：对抗性领域适应实现跨异类RN-Seq数据集中的知识转移
链接：https://arxiv.org/abs/2603.08062
备注：7 pages, 5 figures. Submitted to ECCB 2026
【4】CDRRM: Contrast-Driven Rubric Generation for Reliable and Interpretable Reward Modeling
标题：CDRRM：对比度驱动的条目生成，用于可靠且可解释的奖励建模
链接：https://arxiv.org/abs/2603.08035
【5】Analysis-Driven Procedural Generation of an Engine Sound Dataset with Embedded Control Annotations
标题：分析驱动的带有嵌入式控制注释的发动机声音数据集的过程生成
链接：https://arxiv.org/abs/2603.07584
备注：Preprint. 19 hours of engine audio, 5,935 files, sample-accurate annotations. Dataset publicly available at https://doi.org/10.5281/zenodo.16883336 and https://huggingface.co/datasets/rdoerfler/procedural-engine-sounds
【6】Revisiting the LiRA Membership Inference Attack Under Realistic Assumptions
标题：现实假设下的LiRA成员推断攻击
链接：https://arxiv.org/abs/2603.07567
备注：Accepted to PoPETs 2026(3)
【7】Adversarial Latent-State Training for Robust Policies in Partially Observable Domains
标题：部分可观察领域中稳健政策的对抗潜伏状态训练
链接：https://arxiv.org/abs/2603.07313
备注：20 pages, 3 figures
【8】Variational Flow Maps: Make Some Noise for One-Step Conditional Generation
标题：变分流图：为一步条件生成制造一些噪音
链接：https://arxiv.org/abs/2603.07276
【9】Retrieval-Augmented Generation for Predicting Cellular Responses to Gene Perturbation
标题：基于检索-扩增生成的细胞基因扰动响应预测
链接：https://arxiv.org/abs/2603.07233
备注：Accepted at ICLR 2026 Workshop: Generative AI in Genomics. 25 pages, 9 figures
【10】Physics-Informed Diffusion Model for Generating Synthetic Extreme Rare Weather Events Data
标题：用于生成合成极端罕见天气事件数据的物理信息扩散模型
链接：https://arxiv.org/abs/2603.06782
备注：24 pages, 10 figures, 4 tables. Submitted to MDPI journal
【11】Improved Constrained Generation by Bridging Pretrained Generative Models
标题：通过桥梁预训练生成模型改进约束生成
链接：https://arxiv.org/abs/2603.06742
【12】From Statistical Fidelity to Clinical Consistency: Scalable Generation and Auditing of Synthetic Patient Trajectories
标题：从统计保真度到临床一致性：合成患者轨迹的可扩展生成和审计
链接：https://arxiv.org/abs/2603.06720
备注：23 pages, 8 figures, 6 tables; Code:https://github.com/jameszhou-gl/Coogee
【13】HURRI-GAN: A Novel Approach for Hurricane Bias-Correction Beyond Gauge Stations using Generative Adversarial Networks
标题：HURRI-GAN：一种使用生成对抗网络在气象站之外进行飓风偏差修正的新型方法
链接：https://arxiv.org/abs/2603.06649
备注：18 pages, 6 figures
【14】Advances in GRPO for Generation Models: A Survey
标题：发电模型GRPO的进展：调查
链接：https://arxiv.org/abs/2603.06623
【15】Reward Under Attack: Analyzing the Robustness and Hackability of Process Reward Models
标题：攻击下的奖励：分析流程奖励模型的鲁棒性和可攻击性
链接：https://arxiv.org/abs/2603.06621
【16】Annealed Co-Generation: Disentangling Variables via Progressive Pairwise Modeling
标题：安妮联合发电：通过渐进成对建模解开变量
链接：https://arxiv.org/abs/2603.06615
备注：21 pages, 4 figures, 8 tables
【17】Hierarchical Embedding Fusion for Retrieval-Augmented Code Generation
标题：用于检索增强代码生成的分层嵌入融合
链接：https://arxiv.org/abs/2603.06593
【18】Hierarchical Latent Structures in Data Generation Process Unify Mechanistic Phenomena across Scale
标题：数据生成过程中的分层潜在结构统一跨规模的机械现象
链接：https://arxiv.org/abs/2603.06592
【19】Generative Adversarial Regression (GAR): Learning Conditional Risk Scenarios
标题：生成对抗回归（GAR）：学习条件风险场景
链接：https://arxiv.org/abs/2603.08553
半/弱/无/有监督|不确定性|主动学习(11篇)
【1】Data-Driven Priors for Uncertainty-Aware Deterioration Risk Prediction with Multimodal Data
标题：利用多峰数据进行不确定性意识恶化风险预测的数据驱动先验
链接：https://arxiv.org/abs/2603.08459
备注：24 pages, 5 figures, 8 tables
【2】TRIAGE: Type-Routed Interventions via Aleatoric-Epistemic Gated Estimation in Robotic Manipulation and Adaptive Perception -- Don't Treat All Uncertainty the Same
标题：分类：通过机器人操纵和适应性感知中的先验-认知门控估计进行类型路径干预--不要用同样的方式对待所有不确定性
链接：https://arxiv.org/abs/2603.08128
【3】Revisiting Unknowns: Towards Effective and Efficient Open-Set Active Learning
标题：重温未知：迈向有效和高效的开放式主动学习
链接
：https://arxiv.org/abs/2603.07898
备注：Accepted to CVPR 2026
【4】Uncertainty-Gated Generative Modeling
标题：不确定门控生成建模
链接：https://arxiv.org/abs/2603.07753
备注：Accepeted by ICLR 2026 Workshop Advances in Financial AI
【5】AutoResearch-RL: Perpetual Self-Evaluating Reinforcement Learning Agents for Autonomous Neural Architecture Discovery
标题：AutoResearch-RL：用于自主神经结构发现的永久自评估强化学习代理
链接：https://arxiv.org/abs/2603.07300
【6】Soft Equivariance Regularization for Invariant Self-Supervised Learning
标题：不变自我监督学习的软等方差正规化
链接：https://arxiv.org/abs/2603.06693
备注：14th International Conference on Learning Representations (ICLR 2026)
【7】High-Resolution Image Reconstruction with Unsupervised Learning and Noisy Data Applied to Ion-Beam Dynamics for Particle Accelerators
标题：无监督学习和有噪数据的高分辨率图像重建应用于粒子加速器离子束动力学
链接：https://arxiv.org/abs/2603.06689
【8】RECAP: Local Hebbian Prototype Learning as a Self-Organizing Readout for Reservoir Dynamics
标题：RECAP：本地赫布原型学习作为水库动力学的自组织读数
链接：https://arxiv.org/abs/2603.06639
备注：20 pages, 6 figures
【9】A new Uncertainty Principle in Machine Learning
标题：机器学习中的一种新的不确定性原理
链接：https://arxiv.org/abs/2603.06634
备注：24 pages
【10】Calibrated Credit Intelligence: Shift-Robust and Fair Risk Scoring with Bayesian Uncertainty and Gradient Boosting
标题：校准信用情报：具有Bayesian不确定性和梯度提升的Shift稳健和公平风险评分
链接：https://arxiv.org/abs/2603.06733
备注：7 pages, table 1, figure 5
【11】Uncertainty-Aware Solar Flare Regression
标题：具有不确定性的太阳耀斑回归
链接：https://arxiv.org/abs/2603.06712
迁移|Zero/Few/One-Shot|自适应(18篇)
【1】Adaptive Entropy-Driven Sensor Selection in a Camera-LiDAR Particle Filter for Single-Vessel Tracking
标题：单血管跟踪相机LiDART粒子过滤器中的自适应熵驱动传感器选择
链接：https://arxiv.org/abs/2603.08457
备注：8 pages, 5 figures, submitted to FUSION 2026 conference proceedings
【2】Grow, Assess, Compress: Adaptive Backbone Scaling for Memory-Efficient Class Incremental Learning
标题：增长、评估、压缩：自适应主干扩展，实现内存高效的类增量学习
链接：https://arxiv.org/abs/2603.08426
【3】Meta-RL with Shared Representations Enables Fast Adaptation in Energy Systems
标题：具有共享表示的Meta-RL实现能源系统的快速适应
链接：https://arxiv.org/abs/2603.08418
备注：accepted at PAKDD 2026, Hong Kong
【4】FedPrism: Adaptive Personalized Federated Learning under Non-IID Data
标题：FedPrism：非IID数据下的自适应个性化联邦学习
链接：https://arxiv.org/abs/2603.08252
【5】Model-based Offline RL via Robust Value-Aware Model Learning with Implicitly Differentiable Adaptive Weighting
标题：基于模型的离线RL，通过鲁棒的价值感知模型学习和内在可区分自适应加权
链接：https://arxiv.org/abs/2603.08118
备注：Accepted at ICLR 2026
【6】Guess & Guide: Gradient-Free Zero-Shot Diffusion Guidance
标题：猜测与指南：无干扰Zero-Shot扩散指南
链接：https://arxiv.org/abs/2603.07860
【7】SMAT: Staged Multi-Agent Training for Co-Adaptive Exoskeleton Control
标题：SMAT：用于协同适应外骨骼控制的分阶段多智能体训练
链接：https://arxiv.org/abs/2603.07618
【8】Compression as Adaptation: Implicit Visual Representation with Diffusion Foundation Models
标题：压缩作为适应：采用扩散基础模型的隐式视觉表示
链接：https://arxiv.org/abs/2603.07615
【9】A Unified Framework for Knowledge Transfer in Bidirectional Model Scaling
标题：双向模型缩放中知识转移的统一框架
链接：https://arxiv.org/abs/2603.07506
【10】SLNet: A Super-Lightweight Geometry-Adaptive Network for 3D Point Cloud Recognition
标题：SLNet：一种用于3D点云识别的超轻量级几何自适应网络
链接：https://arxiv.org/abs/2603.07454
备注：Accepted to the 2026 IEEE International Conference on Robotics and Automation (ICRA 2026)
【11】Adaptive Double-Booking Strategy for Outpatient Scheduling Using Multi-Objective Reinforcement Learning
标题：使用多目标强化学习的门诊调度自适应双重预约策略
链接：https://arxiv.org/abs/2603.07270
备注：26 pages, 10 figures
【12】Diversity-Aware Adaptive Collocation for Physics-Informed Neural Networks via Sparse QUBO Optimization and Hybrid Coresets
标题：通过稀疏QUBO优化和混合核心集实现物理信息神经网络的多样性感知自适应配置
链接：https://arxiv.org/abs/2603.06761
备注：9 pages, accepted to be published as a ICLR workshop paper
【13】XAI and Few-shot-based Hybrid Classification Model for Plant Leaf Disease Prognosis
标题：植物叶病预后的XAI和基于Few-Shot的混合分类模型
链接：https://arxiv.org/abs/2603.06676
备注：27 pages, 8 figures
【14】Distilling and Adapting: A Topology-Aware Framework for Zero-Shot Interaction Prediction in Multiplex Biological Networks
标题：提炼和适应：多重生物网络中Zero-Shot相互作用预测的一个具有全局意识的框架
链接：https://arxiv.org/abs/2603.06618
备注：Accepted by ICLR 2026
【15】A Robust Incomplete Multimodal Low-Rank Adaptation Approach for Emotion Recognition
标题：一种鲁棒的不完全多模式低等级自适应情绪识别方法
链接：https://arxiv.org/abs/2507.11202
【16】Robust Transfer Learning with Side Information
标题：具有辅助信息的稳健迁移学习
链接：https://arxiv.org/abs/2603.07921
【17】MetaSort: An Accelerated Approach for Non-uniform Compression and Few-shot Classification of Neural Spike Waveforms
标题：MetaSort：一种用于神经尖峰波形非均匀压缩和Few-Shot分类的加速方法
链接：https://arxiv.org/abs/2603.07602
备注：4 pages (under review paper)
【18】Towards Lightweight Adaptation of Speech Enhancement Models in Real-World Environments
标题：实现现实环境中语音增强模型的轻量级适应
链接：https://arxiv.org/abs/2603.07471
备注：Accepted to ICASSP 2026
强化学习(19篇)
【1】Towards Batch-to-Streaming Deep Reinforcement Learning for Continuous Control
标题：迈向批量到流的深度强化学习以实现连续控制
链接：https://arxiv.org/abs/2603.08588
【2】Impact of Connectivity on Laplacian Representations in Reinforcement Learning
标题：强化学习中连通性对拉普拉斯表示的影响
链接：https://arxiv.org/abs/2603.08558
【3】Breaking the Bias Barrier in Concave Multi-Objective Reinforcement Learning
标题：突破凹多目标强化学习中的偏差障碍
链接：https://arxiv.org/abs/2603.08518
【4】Integrating Lagrangian Neural Networks into the Dyna Framework for Reinforcement Learning
标题：将拉格朗日神经网络集成到Dyna框架中进行强化学习
链接：https://arxiv.org/abs/2603.08468
备注：5 pages, 3 figures
【5】A Recipe for Stable Offline Multi-agent Reinforcement Learning
标题：一个稳定的离线多智能体强化学习的方法
链接：https://arxiv.org/abs/2603.08399
备注：Preprint
【6】Toward Global Intent Inference for Human Motion by Inverse Reinforcement Learning
标题：通过反向强化学习实现人类运动的全球意图推理
链接：https://arxiv.org/abs/2603.07797
备注：8 pages, 6 figures
【7】Scaling Data Difficulty: Improving Coding Models via Reinforcement Learning on Fresh and Challenging Problems
标题：扩展数据难度：通过针对新鲜和棘手问题的强化学习改进编码模型
链接：https://arxiv.org/abs/2603.07779
【8】Breaking Training Bottlenecks: Effective and Stable Reinforcement Learning for Coding Models
标题：打破训练瓶颈：编码模型的有效稳定的强化学习
链接：https://arxiv.org/abs/2603.07777
【9】Helix: Evolutionary Reinforcement Learning for Open-Ended Scientific Problem Solving
标题：螺旋：用于开放式科学问题解决的进化强化学习
链接：https://arxiv.org/abs/2603.07642
备注：Accepted at ICLR 2026
【10】Reinforcement learning-based dynamic cleaning scheduling framework for solar energy system
标题：基于强化学习的太阳能系统动态清洁调度框架
链接：https://arxiv.org/abs/2603.07518
备注：16 pages, 6 figures, This is an accepted manuscript of the article published in Journal of Korean Institute of Intelligent Systems, 35(1), 84-97, 2025
【11】Generalization in Online Reinforcement Learning for Mobile Agents
标题：移动代理在线强化学习的推广
链接
：https://arxiv.org/abs/2603.07432
【12】Learning to Reflect: Hierarchical Multi-Agent Reinforcement Learning for CSI-Free mmWave Beam-Focusing
标题：学习反思：用于无CSI毫米波波束聚焦的分层多智能体强化学习
链接：https://arxiv.org/abs/2603.07370
【13】NePPO: Near-Potential Policy Optimization for General-Sum Multi-Agent Reinforcement Learning
标题：NePPO：通用和多智能体强化学习的近潜力策略优化
链接：https://arxiv.org/abs/2603.06977
【14】Chart-RL: Generalized Chart Comprehension via Reinforcement Learning with Verifiable Rewards
标题：Chart-RL：通过强化学习和可验证奖励的广义图表理解
链接：https://arxiv.org/abs/2603.06958
【15】Joint MDPs and Reinforcement Learning in Coupled-Dynamics Environments
标题：耦合动力学环境中的联合MDP和强化学习
链接：https://arxiv.org/abs/2603.06946
备注：25 pages, 7 figures
【16】Multi-Agent Reinforcement Learning with Submodular Reward
标题：具有子模块奖励的多智能体强化学习
链接：https://arxiv.org/abs/2603.06810
【17】Not all tokens are needed(NAT): token efficient reinforcement learning
标题：并非所有令牌都是需要的（RAT）：令牌高效强化学习
链接：https://arxiv.org/abs/2603.06619
【18】Scaling Strategy, Not Compute: A Stand-Alone, Open-Source StarCraft II Benchmark for Accessible Reinforcement Learning Research
标题：扩展策略，而不是计算：可达强化学习研究的独立、开源《星际争霸II》基准
链接：https://arxiv.org/abs/2603.06608
【19】Posterior Sampling Reinforcement Learning with Gaussian Processes for Continuous Control: Sublinear Regret Bounds for Unbounded State Spaces
标题：具有高斯过程的连续控制的后验抽样强化学习：无界状态空间的次线性遗憾界
链接：https://arxiv.org/abs/2603.08287
备注：37 pages, 8 figures
元学习(1篇)
【1】OptiRoulette Optimizer: A New Stochastic Meta-Optimizer for up to 5.3x Faster Convergence
标题：OptimRoulette优化器：一种新型随机元优化器，收敛速度可达5.3倍
链接：https://arxiv.org/abs/2603.06613
备注：23 pages, 10 figures, 7 tables
符号|符号学习(2篇)
【1】Turning Time Series into Algebraic Equations: Symbolic Machine Learning for Interpretable Modeling of Chaotic Time Series
标题：将时间序列转化为代数方程：用于混乱时间序列可解释建模的符号机器学习
链接：https://arxiv.org/abs/2603.07261
【2】Failure Detection in Chemical Processes using Symbolic Machine Learning: A Case Study on Ethylene Oxidation
标题：使用符号机器学习进行化学过程故障检测：乙烯氧化案例研究
链接：https://arxiv.org/abs/2603.06767
备注：Accepted at AAAI-MAKE 2026
分层学习(1篇)
【1】Learning Hierarchical Knowledge in Text-Rich Networks with Taxonomy-Informed Representation Learning
标题：通过分类学知情的表示学习在文本丰富的网络中学习分层知识
链接：https://arxiv.org/abs/2603.08159
备注：Accepted by KDD 2026. Extended version coming soon
医学相关(11篇)
【1】Echo2ECG: Enhancing ECG Representations with Cardiac Morphology from Multi-View Echos
标题：Echo2心电图：利用多视图回声的心脏形态增强心电图表示
链接：https://arxiv.org/abs/2603.08505
【2】A prospective clinical feasibility study of a conversational diagnostic AI in an ambulatory primary care clinic
标题：门诊初级保健诊所对话诊断人工智能的前瞻性临床可行性研究
链接：https://arxiv.org/abs/2603.08448
【3】Beyond Attention Heatmaps: How to Get Better Explanations for Multiple Instance Learning Models in Histopathology
标题：超越注意力热图：如何为组织学中的多实例学习模型获得更好的解释
链接：https://arxiv.org/abs/2603.08328
【4】TA-RNN-Medical-Hybrid: A Time-Aware and Interpretable Framework for Mortality Risk Prediction
标题：TA-RNN-医疗混合：死亡风险预测的时间感知和可解释框架
链接：https://arxiv.org/abs/2603.08278
【5】Hybrid Quantum Neural Network for Multivariate Clinical Time Series Forecasting
标题：混合量子神经网络用于多元临床时间序列预测
链接：https://arxiv.org/abs/2603.08072
【6】ECG Classification on PTB-XL: A Data-Centric Approach with Simplified CNN-VAE
标题：PTB-XL上的心电图分类：一种以数据为中心的方法，具有简化的CNN-VAE
链接：https://arxiv.org/abs/2603.07558
【7】Learning Clinical Representations Under Systematic Distribution Shift
标题：系统分布变化下的临床代表学习
链接：https://arxiv.org/abs/2603.07348
【8】LF2L: Loss Fusion Horizontal Federated Learning Across Heterogeneous Feature Spaces Using External Datasets Effectively: A Case Study in Second Primary Cancer Prediction
标题：LF 2L：有效使用外部数据集跨异类特征空间的损失融合水平联邦学习：第二次初级癌症预测的案例研究
链接：https://arxiv.org/abs/2603.07249
【9】LightMedSeg: Lightweight 3D Medical Image Segmentation with Learned Spatial Anchors
标题：LightMedSeg：使用学习空间先验知识的轻量级3D医学图像分割
链接：https://arxiv.org/abs/2603.07228
备注：8 pages, X figures. Submitted to CVPRW ECV 2026
【10】Enhancing SHAP Explainability for Diagnostic and Prognostic ML Models in Alzheimer Disease
标题：增强阿尔茨海默病诊断和预后ML模型的SHAP解释性
链接：https://arxiv.org/abs/2603.06758
【11】Subclass Classification of Gliomas Using MRI Fusion Technique
标题：利用MRI融合技术进行脑胶质瘤的亚类分类
链接：https://arxiv.org/abs/2502.18775
备注：15 pages, 7 figures, 1 algorithm, 4 tables, journal paper
蒸馏|知识提取(2篇)
【1】Unlocking Data Value in Finance: A Study on Distillation and Difficulty-Aware Training
标题：释放金融中的数据价值：蒸馏和困难意识训练研究
链接：https://arxiv.org/abs/2603.07223
【2】A Dynamic Self-Evolving Extraction System
标题：动态自演化萃取系统
链接：https://arxiv.org/abs/2603.06915
推荐(2篇)
【1】Exploration Space Theory: Formal Foundations for Prerequisite-Aware Location-Based Recommendation
标题：探索空间理论：先决条件感知基于位置的推荐的形式基础
链接：https://arxiv.org/abs/2603.06624
备注：Pre-print of a theoretical framework for prerequisite-aware recommendation using Knowledge Space Theory and Birkhoff representation
【2】Isotonic Layer: A Universal Framework for Generic Recommendation Debiasing
标题：等张层：通用推荐去偏置的通用框架
链接：https://arxiv.org/abs/2603.06589
备注：8 pages, 5 figures, submitted to KDD 2026
聚类(4篇)
【1】Single-pass Possibilistic Clustering with Damped Window Footprints
标题：具有衰减窗口足迹的单遍可能性聚集
链接：https://arxiv.org/abs/2603.06889
【2】Learning Unbiased Cluster Descriptors for Interpretable Imbalanced Concept Drift Detection
标题：学习无偏集群描述符用于可解释不平衡概念漂移检测
链接：https://arxiv.org/abs/2603.06757
备注：14 pages, 7 figures
【3】LegoNet: Memory Footprint Reduction Through Block Weight Clustering
标题：LegoNet：通过块权重聚类减少内存占用
链接：https://arxiv.org/abs/2603.06606
备注：7 pages, 24 figures, published to IEEE DASC 2022 (20th year)
【4】Khatri-Rao Clustering for Data Summarization
标题：用于数据总结的Khatri-Rao集群
链接：https://arxiv.org/abs/2603.06602
自动驾驶|车辆|车道检测等(3篇)
【1】NaviDriveVLM: Decoupling High-Level Reasoning and Motion Planning for Autonomous Driving
标题：NaviDriveVLM：自动驾驶的高级推理和运动规划脱钩
链接：https://arxiv.org/abs/2603.07901
【2】Toward Unified Multimodal Representation Learning for Autonomous Driving
标题：迈向自动驾驶的统一多模式表示学习
链接：https://arxiv.org/abs/2603.07874
【3】Constraints Matrix Diffusion based Generative Neural Solver for Vehicle Routing Problems
标题：基于约束扩散矩阵的生成神经元求解器的车辆路径问题
链接：https://arxiv.org/abs/2603.07568
点云|SLAM|雷达|激光|深度RGBD相关(3篇)
【1】Minor First, Major Last: A Depth-Induced Implicit Bias of Sharpness-Aware Minimization
标题：次要优先，主要最后：深度引发的敏锐意识最小化的隐性偏见
链接：https://arxiv.org/abs/2603.08290
备注：Accepted to ICLR 2026, 82 pages, 35 figures
【2】ALOOD: Exploiting Language Representations for LiDAR-based Out-of-Distribution Object Detection
标题：ALOOD：利用语言表示进行基于LiDART的非分布对象检测
链接：https://arxiv.org/abs/2603.08180
备注：Accepted for publication at the 2025 IEEE Intelligent Transportation Systems Conference (ITSC)
【3】Generative prediction of laser-induced rocket ignition with dynamic latent space representations
标题：利用动态潜空间表示的激光诱导火箭点火生成预测
链接：https://arxiv.org/abs/2603.07525
联邦学习|隐私保护|加密(5篇)
【1】Split Federated Learning Architectures for High-Accuracy and Low-Delay Model Training
标题：用于高准确度和低延迟模型训练的分离联邦学习架构
链接：https://arxiv.org/abs/2603.08687
【2】Revisiting Gradient Staleness: Evaluating Distance Metrics for Asynchronous Federated Learning Aggregation
标题：重新审视梯度停滞：评估同步联邦学习聚合的距离范围
链接：https://arxiv.org/abs/2603.08211
【3】Stabilized Fine-Tuning with LoRA in Federated Learning: Mitigating the Side Effect of Client Size and Rank via the Scaling Factor
标题：在联邦学习中使用LoRA进行稳定微调：通过比例因子减轻客户规模和排名的副作用
链接：https://arxiv.org/abs/2603.08058
【4】Trust Aware Federated Learning for Secure Bone Healing Stage Interpretation in e-Health
标题：信任感知的联合学习用于e-健康中的安全骨愈合阶段解释
链接：https://arxiv.org/abs/2603.06646
【5】Compressed Proximal Federated Learning for Non-Convex Composite Optimization on Heterogeneous Data
标题：用于异类数据非凸复合优化的压缩近端联邦学习
链接：https://arxiv.org/abs/2603.07654
备注：14 pages, 4 figures
推理|分析|理解|解释(25篇)
【1】NN-OpInf: an operator inference approach using structure-preserving composable neural networks
标题：NN-OpInf：一种使用结构保持可组合神经网络的操作员推理方法
链接：https://arxiv.org/abs/2603.08488
【2】Reasoning as Compression: Unifying Budget Forcing via the Conditional Information Bottleneck
标题：推理即压缩：通过条件信息瓶颈统一预算强制
链接：https://arxiv.org/abs/2603.08462
【3】LycheeCluster: Efficient Long-Context Inference with Structure-Aware Chunking and Hierarchical KV Indexing
标题：LycheStack：具有结构感知分块和分层KN索引的高效长上下文推理
链接：https://arxiv.org/abs/2603.08453
备注：17 pages, 12 figures
【4】SYNAPSE: Framework for Neuron Analysis and Perturbation in Sequence Encoding
标题：SYAPSE：序列编码中神经元分析和扰动的框架
链接：https://arxiv.org/abs/2603.08424
【5】Towards plausibility in time series counterfactual explanations
标题：走向时间序列反事实解释的合理性
链接：https://arxiv.org/abs/2603.08349
【6】Is continuous CoT better suited for multi-lingual reasoning?
标题：连续CoT更适合多语言推理吗？
链接：https://arxiv.org/abs/2603.08177
备注：Accepted at the ICLR latent reasoning workshop
【7】C$^2$FG: Control Classifier-Free Guidance via Score Discrepancy Analysis
标题：C $' 2$FG：通过分数差异分析控制无分类器指导
链接：https://arxiv.org/abs/2603.08155
【8】Explainable Condition Monitoring via Probabilistic Anomaly Detection Applied to Helicopter Transmissions
标题：应用于直升机变速箱的概率异常检测的可解释状态监控
链接：https://arxiv.org/abs/2603.08130
【9】DC-W2S: Dual-Consensus Weak-to-Strong Training for Reliable Process Reward Modeling in Biological Reasoning
标题：DC-W2 S：生物推理中可靠流程奖励建模的双共识弱到强训练
链接：https://arxiv.org/abs/2603.08095
【10】VLM-SubtleBench: How Far Are VLMs from Human-Level Subtle Comparative Reasoning?
标题：VLM-SubtleBench：VLM-SubtleBench距离人类层面的微妙比较推理有多远？
链接：https://arxiv.org/abs/2603.07888
备注：ICLR 2026
【11】Beyond Surrogates: A Quantitative Analysis for Inter-Metric Relationships
标题：超越代理人：跨指标关系的定量分析
链接：https://arxiv.org/abs/2603.07671
备注：18 pages, 1 figure
【12】Enhanced Random Subspace Local Projections for High-Dimensional Time Series Analysis
标题：用于多维时间序列分析的增强随机子空间局部投影
链接：https://arxiv.org/abs/2603.07500
备注：12 pages, 18 figures
【13】Trusting What You Cannot See: Auditable Fine-Tuning and Inference for Proprietary AI
标题：相信你看不到的东西：专有人工智能的可审核微调和推理
链接：https://arxiv.org/abs/2603.07466
【14】Context Channel Capacity: An Information-Theoretic Framework for Understanding Catastrophic Forgetting
标题：上下文通道容量：理解灾难性遗忘的信息理论框架
链接：https://arxiv.org/abs/2603.07415
备注：39 pages
【15】Learning Concept Bottleneck Models from Mechanistic Explanations
标题：学习概念来自机械解释的瓶颈模型
链接：https://arxiv.org/abs/2603.07343
备注：ICLR 2026
【16】Agentic Planning with Reasoning for Image Styling via Offline RL
标题：通过离线RL进行图像造型推理的统计规划
链接：https://arxiv.org/abs/2603.07148
备注：85 pages
【17】Physics-informed AI Accelerated Retention Analysis of Ferroelectric Vertical NAND: From Day-Scale TCAD to Second-Scale Surrogate Model
标题：基于物理信息的人工智能加速铁电垂直NAMA保留分析：从日规模TCAD到二级代理模型
链接：https://arxiv.org/abs/2603.06881
备注：4 pages, 6 figues, submitted to ICMC (International Compact Modeling Conference)
【18】Best-of-Tails: Bridging Optimism and Pessimism in Inference-Time Alignment
标题：最好的尾巴：在推理时间一致中弥合乐观与悲观
链接：https://arxiv.org/abs/2603.06797
【19】HyperTokens: Controlling Token Dynamics for Continual Video-Language Understanding
标题：HyperTokens：控制代币动态以实现连续的视频语言理解
链接：https://arxiv.org/abs/2603.06662
【20】EnsAug: Augmentation-Driven Ensembles for Human Motion Sequence Analysis
标题：EnsAug：用于人体运动序列分析的增强驱动套件
链接：https://arxiv.org/abs/2603.06661
【21】Correlation Analysis of Generative Models
标题：生成模型的相关性分析
链接：https://arxiv.org/abs/2603.06614
【22】A Novel Approach for Testing Water Safety Using Deep Learning Inference of Microscopic Images of Unincubated Water Samples
标题：一种使用深度学习推断未孵育水样微观图像来测试水安全性的新方法
链接：https://arxiv.org/abs/2603.06611
【23】Valid Feature-Level Inference for Tabular Foundation Models via the Conditional Randomization Test
标题：通过条件随机化测试对表格基础模型进行有效的制造水平推断
链接：https://arxiv.org/abs/2603.06609
【24】Probabilistic Inference and Learning with Stein's Method
标题：斯坦方法的概率推理和学习
链接：https://arxiv.org/abs/2603.07467
【25】Explainable and Hardware-Efficient Jamming Detection for 5G Networks Using the Convolutional Tsetlin Machine
标题：使用卷积Tsetlin机的5G网络可解释和硬件高效的干扰检测
链接：https://arxiv.org/abs/2603.07336
备注：6 pages, 4 figures. IEEE ICC 2026 Workshops (under submission)
检测相关(10篇)
【1】X-AVDT: Audio-Visual Cross-Attention for Robust Deepfake Detection
标题：X-AVDT：用于稳健的Deepfake检测的视听交叉注意力
链接：https://arxiv.org/abs/2603.08483
【2】The Boiling Frog Threshold: Criticality and Blindness in World Model-Based Anomaly Detection Under Gradual Drift
标题：沸腾的青蛙阈值：渐进漂移下基于世界模型的异常检测的临界性和盲目性
链接：https://arxiv.org/abs/2603.08455
备注：10 pages, 5 figures, preprint
【3】Evaluating Synthetic Data for Baggage Trolley Detection in Airport Logistics
标题：机场物流行李车检测综合数据评估
链接：https://arxiv.org/abs/2603.07645
【4】Integration of deep generative Anomaly Detection algorithm in high-speed industrial line
标题：深度生成异常检测算法在高速工业线中的集成
链接：https://arxiv.org/abs/2603.07577
备注：Preprint under review at a Springer Nature journal. 36 pages, 3 tables, 29 figures. Updated and expanded version of the SSRN preprint (abstract_id=4858664), with substantial revisions and Springer Nature formatting
【5】A Systematic Comparison of Training Objectives for Out-of-Distribution Detection in Image Classification
标题：图像分类中分布外检测训练目标的系统比较
链接：https://arxiv.org/abs/2603.07571
【6】GRD-Net: Generative-Reconstructive-Discriminative Anomaly Detection with Region of Interest Attention Module
标题：GRD-Net：具有感兴趣注意区域模块的生成重建区分异常检测
链接：https://arxiv.org/abs/2603.07566
备注：Peer-reviewed journal version published. 18 pages, 12 figures, 7 tables
【7】Online Continual Learning for Anomaly Detection in IoT under Data Distribution Shifts
标题：数据分布变化下的物联网异常检测在线持续学习
链接：https://arxiv.org/abs/2603.07507
备注：Manuscript submitted to EUSIPCO 2026. The copyright might be transferred without further notice
【8】Shaping Parameter Contribution Patterns for Out-of-Distribution Detection
标题：为分布外检测塑造参数贡献模式
链接：https://arxiv.org/abs/2603.07195
【9】Interpretable Maximum Margin Deep Anomaly Detection
标题：可解释最大裕度深度异常检测
链接：https://arxiv.org/abs/2603.07073
【10】An Interpretable Generative Framework for Anomaly Detection in High-Dimensional Financial Time Series
标题：用于多维金融时间序列异常检测的可解释生成框架
链接：https://arxiv.org/abs/2603.07864
分类|识别(1篇)
【1】Deterministic Fuzzy Triage for Legal Compliance Classification and Evidence Retrieval
标题：基于确定性模糊分类的合法性分类与证据检索
链接：https://arxiv.org/abs/2603.07390
备注：8 pages, 5 figures. Published in the Proceedings of the AAAI Bridge between Artificial Intelligence and Law 2026 (Full papers), pages 51-58
表征(7篇)
【1】Cost-Driven Representation Learning for Linear Quadratic Gaussian Control: Part II
标题：线性二次高斯控制的成本驱动表示学习：第二部分
链接：https://arxiv.org/abs/2603.07437
备注：38 pages; preliminary version appeared in IEEE CDC 2023; this is the extended journal version, with an end-to-end guarantee added
【2】Norm-Hierarchy Transitions in Representation Learning: When and Why Neural Networks Abandon Shortcuts
标题：表示学习中的规范-层次转变：神经网络何时以及为何放弃捷径
链接：https://arxiv.org/abs/2603.07323
备注：20 pages, 5 figs
【3】Dreamer-CDP: Improving Reconstruction-free World Models Via Continuous Deterministic Representation Prediction
标题：Dreamer-DPP：通过连续确定性表示预测改进无重建世界模型
链接：https://arxiv.org/abs/2603.07083
【4】Gauge Freedom and Metric Dependence in Neural Representation Spaces
标题：神经表示空间中的规范自由度和度量依赖性
链接：https://arxiv.org/abs/2603.06774
备注：14 pages, 4 figures
【5】Implementation of Quantum Implicit Neural Representation in Deterministic and Probabilistic Autoencoders for Image Reconstruction/Generation Tasks
标题：用于图像重建/生成任务的确定性和概率自动编码器中的量子隐式神经表示的实现
链接：https://arxiv.org/abs/2603.06755
【6】Grouter: Decoupling Routing from Representation for Accelerated MoE Training
标题：胶合剂：将路由与代表脱钩以加速MoE训练
链接：https://arxiv.org/abs/2603.06626
【7】How Private Are DNA Embeddings? Inverting Foundation Model Representations of Genomic Sequences
标题：DNA嵌入有多私密？基因组序列的倒置基础模型表示
链接：https://arxiv.org/abs/2603.06950
3D|3D重建等相关(1篇)
【1】Joint 3D Gravity and Magnetic Inversion via Rectified Flow and Ginzburg-Landau Guidance
标题：通过整流流和金茨堡-兰道引导联合3D重力和磁倒置
链接：https://arxiv.org/abs/2603.06829
编码器(2篇)
【1】Generalizing Linear Autoencoder Recommenders with Decoupled Expected Quadratic Loss
标题：具有脱钩期望二次损失的线性自动编码器推荐
链接：https://arxiv.org/abs/2603.07402
备注：Accepted at ICLR 2026 (https://openreview.net/forum?id=ANH044Wdje)
【2】Latent Autoencoder Ensemble Kalman Filter for Data assimilation
标题：用于数据同化的潜在自动编码器引入卡尔曼过滤器
链接：https://arxiv.org/abs/2603.06752
优化|敛散性(13篇)
【1】Pareto-Optimal Anytime Algorithms via Bayesian Racing
标题：通过Bayesian Racing的帕累托最优随时算法
链接：https://arxiv.org/abs/2603.08493
备注：32 pages, 12 figures, 2 tables, and 4 pages of appendix with additional details. Submitted to ACM Transactions on Evolutionary Learning and Optimization
【2】Beyond the Markovian Assumption: Robust Optimization via Fractional Weyl Integrals in Imbalanced Data
标题：超越马尔科夫假设：在不平衡数据中通过分数Weyl积分进行鲁棒优化
链接：https://arxiv.org/abs/2603.08377
备注：5 pages, 3 figures
【3】PolyFormer: learning efficient reformulations for scalable optimization under complex physical constraints
标题：PolyFormer：学习有效的重新配方，以在复杂物理约束下实现可扩展优化
链接：https://arxiv.org/abs/2603.08283
备注：Code availability: All the data and code are made openly available at https://github.com/wenyl16/PolyFormer
【4】Fibration Policy Optimization
标题：纤维化政策优化
链接：https://arxiv.org/abs/2603.08239
【5】Transferable Optimization Network for Cross-Domain Image Reconstruction
标题：跨域图像重建的可移植优化网络
链接：https://arxiv.org/abs/2603.07831
备注：30 pages, 7 figures
【6】Global Convergence of Average Reward Constrained MDPs with Neural Critic and General Policy Parameterization
标题：具有神经批判和一般策略参数化的平均报酬约束MDP的全局收敛性
链接：https://arxiv.org/abs/2603.07698
备注：Submitted to UAI 2026
【7】Data Agent: Learning to Select Data via End-to-End Dynamic Optimization
标题：数据代理：学习通过端到端动态优化选择数据
链接：https://arxiv.org/abs/2603.07433
【8】Feed m Birds with One Scone: Accelerating Multi-task Gradient Balancing via Bi-level Optimization
标题：用一个烤饼喂m只鸟：通过双层优化加速多任务梯度平衡
链接：https://arxiv.org/abs/2603.07389
【9】Statistical Contraction for Chance-Constrained Trajectory Optimization of Non-Gaussian Stochastic Systems
标题：非高斯随机系统机会约束轨迹优化的统计压缩
链接：https://arxiv.org/abs/2603.07092
【10】Conditional Unbalanced Optimal Transport Maps: An Outlier-Robust Framework for Conditional Generative Modeling
标题：条件不平衡最优运输映射：一个异常鲁棒的条件生成模型框架
链接：https://arxiv.org/abs/2603.06972
备注：15 pages, 5 figures
【11】Chart Deep Research in LVLMs via Parallel Relative Policy Optimization
标题：图表通过并行相对政策优化深入研究LVLM
链接：https://arxiv.org/abs/2603.06677
备注：Accepted at ICLR 2026
【12】Local Constrained Bayesian Optimization
标题：局部约束Bayesian优化
链接：https://arxiv.org/abs/2603.07965
【13】Post-Training with Policy Gradients: Optimality and the Base Model Barrier
标题
：政策受益者的后训练：最优性和基础模型障碍
链接：https://arxiv.org/abs/2603.06957
备注：36 pages, 2 figures
预测|估计(18篇)
【1】Impermanent: A Live Benchmark for Temporal Generalization in Time Series Forecasting
标题：非永久性：时间序列预测中时间概括的实时基准
链接：https://arxiv.org/abs/2603.08707
【2】Oracle-Guided Soft Shielding for Safe Move Prediction in Chess
标题：Oracle引导的软屏蔽用于国际象棋安全棋步预测
链接：https://arxiv.org/abs/2603.08506
备注：Accepted for publication at the 24th International Conference on Machine Learning and Applications (ICMLA), 2025. Preprint version in Arxiv
【3】Efficient Credal Prediction through Decalibration
标题：通过去校准高效的信任预测
链接：https://arxiv.org/abs/2603.08495
【4】FlowTouch: View-Invariant Visuo-Tactile Prediction
标题：FlowTouch：视图不变的视觉触觉预测
链接：https://arxiv.org/abs/2603.08255
【5】Optimising antibiotic switching via forecasting of patient physiology
标题：通过预测患者生理状况优化抗生素转换
链接：https://arxiv.org/abs/2603.08242
备注：32 pages, 8 figures
【6】Are We Winning the Wrong Game? Revisiting Evaluation Practices for Long-Term Time Series Forecasting
标题：我们赢错了游戏吗？重新审视长期时间序列预测的评估实践
链接：https://arxiv.org/abs/2603.08156
备注：First draft
【7】Domain-Specific Quality Estimation for Machine Translation in Low-Resource Scenarios
标题：低资源场景下机器翻译的特定领域质量估计
链接：https://arxiv.org/abs/2603.07372
备注：21 pages, 7 tables, 7 figures
【8】N-Tree Diffusion for Long-Horizon Wildfire Risk Forecasting
标题：长期野火风险预测的N树扩散
链接：https://arxiv.org/abs/2603.07361
备注：15 pages, 6 figures
【9】Retrieval-Augmented Multi-scale Framework for County-Level Crop Yield Prediction Across Large Regions
标题：大区域县级农作物产量预测检索增强多尺度框架
链接：https://arxiv.org/abs/2603.07305
【10】Regression Models Meet Foundation Models: A Hybrid-AI Approach to Practical Electricity Price Forecasting
标题：回归模型满足基础模型：实用电价预测的混合人工智能方法
链接：https://arxiv.org/abs/2603.06726
备注：15 pages. Preprint
【11】Bi Directional Feedback Fusion for Activity Aware Forecasting of Indoor CO2 and PM2.5
标题：双向反馈融合用于室内CO2和PM2.5活动感知预测
链接：https://arxiv.org/abs/2603.06724
备注：Journal Submission
【12】From ARIMA to Attention: Power Load Forecasting Using Temporal Deep Learning
标题：从ARIMA到注意力：使用时态深度学习进行电力负荷预测
链接：https://arxiv.org/abs/2603.06622
备注：5 pages; Published in IEEE
【13】Momentum SVGD-EM for Accelerated Maximum Marginal Likelihood Estimation
标题：加速最大边际似然估计的动量SVGD-EM
链接：https://arxiv.org/abs/2603.08676
备注：Accepted to AISTATS 2026
【14】Outlier-robust Autocovariance Least Square Estimation via Iteratively Reweighted Least Square
标题：通过迭代重加权最小平方的离群稳健自协方差最小平方估计
链接：https://arxiv.org/abs/2603.08158
备注：10 pages, 8 figures
【15】Beyond Data Splitting: Full-Data Conformal Prediction by Differential Privacy
标题：超越数据分割：通过差异隐私进行全数据保形预测
链接：https://arxiv.org/abs/2603.07522
【16】Deep Generative Spatiotemporal Engression for Probabilistic Forecasting of Epidemics
标题：流行病概率预测的深代时空参与
链接：https://arxiv.org/abs/2603.07108
【17】Prediction of Steady-State Flow through Porous Media Using Machine Learning Models
标题：使用机器学习模型预测多孔介质的稳态流动
链接：https://arxiv.org/abs/2603.06762
【18】GNN For Muon Particle Momentum estimation
标题：用于μ子粒子动量估计的GNN
链接：https://arxiv.org/abs/2603.06675
其他神经网络|深度学习|模型|建模(54篇)
【1】Benchmarking Language Modeling for Lossless Compression of Full-Fidelity Audio
标题：高保真音频无损压缩的基准语言建模
链接：https://arxiv.org/abs/2603.08683
备注：Submitted for review at Interspeech 2026, 7 pages, 5 figures
【2】Group Entropies and Mirror Duality: A Class of Flexible Mirror Descent Updates for Machine Learning
标题：群熵和镜像二元性：机器学习的一类灵活的镜像下降更新
链接：https://arxiv.org/abs/2603.08651
备注：36 pages, 5 figures
【3】Don't Look Back in Anger: MAGIC Net for Streaming Continual Learning with Temporal Dependence
标题：不要愤怒地回头看：用于具有时间依赖性的流媒体持续学习的MAGIC Net
链接：https://arxiv.org/abs/2603.08600
【4】DualFlexKAN: Dual-stage Kolmogorov-Arnold Networks with Independent Function Control
标题：DualFlexKAN：具有独立函数控制的双级Kolmogorov-Arnold网络
链接：https://arxiv.org/abs/2603.08583
备注：22 pages, 12 figures
【5】MUSA-PINN: Multi-scale Weak-form Physics-Informed Neural Networks for Fluid Flow in Complex Geometries
标题：MUSA-PINN：用于复杂几何结构中流体流动的多尺度弱形式物理信息神经网络
链接：https://arxiv.org/abs/2603.08465
【6】Leaderboard Incentives: Model Rankings under Strategic Post-Training
标题：排行榜激励措施：战略后训练下的模型排名
链接：https://arxiv.org/abs/2603.08371
【7】Airborne Magnetic Anomaly Navigation with Neural-Network-Augmented Online Calibration
标题：带神经网络增强在线校准的机载磁异常导航
链接：https://arxiv.org/abs/2603.08265
【8】Distributional Regression with Tabular Foundation Models: Evaluating Probabilistic Predictions via Proper Scoring Rules
标题：使用表格基础模型的分布回归：通过适当的评分规则评估概率预测
链接：https://arxiv.org/abs/2603.08206
【9】Training event-based neural networks with exact gradients via Differentiable ODE Solving in JAX
标题：通过JAX中的可微ODE求解以精确梯度训练基于事件的神经网络
链接：https://arxiv.org/abs/2603.08146
备注：9 pages, 3 figures
【10】Tau-BNO: Brain Neural Operator for Tau Transport Model
标题：Tau-BNO：Tau输运模型的脑神经算子
链接：https://arxiv.org/abs/2603.08108
【11】Tiny Autoregressive Recursive Models
标题：微小自回归递归模型
链接：https://arxiv.org/abs/2603.08082
【12】PSTNet: Physically-Structured Turbulence Network
标题：PSTNet：物理结构湍流网络
链接：https://arxiv.org/abs/2603.07957
备注：7 pages, 6 figures, 2 tables
【13】Rel-MOSS: Towards Imbalanced Relational Deep Learning on Relational Databases
标题：Rel-MOSS：在关系数据库上实现不平衡的关系深度学习
链接：https://arxiv.org/abs/2603.07916
【14】Slumbering to Precision: Enhancing Artificial Neural Network Calibration Through Sleep-like Processes
标题：潜入精确：通过类似睡眠的过程增强人工神经网络校准
链接：https://arxiv.org/abs/2603.07867
【15】Gradient Iterated Temporal-Difference Learning
标题：梯度迭代时间差异学习
链接：https://arxiv.org/abs/2603.07833
【16】Step-Size Decay and Structural Stagnation in Greedy Sparse Learning
标题：贪婪稀疏学习中的阶梯衰退和结构停滞
链接：https://arxiv.org/abs/2603.07703
【17】Scalable Training of Mixture-of-Experts Models with Megatron Core
标题：基于Megatron Core的混合专家模型的可扩展训练
链接：https://arxiv.org/abs/2603.07685
备注：Technical Report. 88 pages. 42 figures
【18】Mitigating the Memory Bottleneck with Machine Learning-Driven and Data-Aware Microarchitectural Techniques
标题：利用机器学习驱动和数据感知微架构技术缓解内存瓶颈
链接：https://arxiv.org/abs/2603.07683
【19】Partial Differential Equations in the Age of Machine Learning: A Critical Synthesis of Classical, Machine Learning, and Hybrid Methods
标题：机器学习时代的偏微方程：经典、机器学习和混合方法的批判综合
链接：https://arxiv.org/abs/2603.07655
【20】Exoskeleton Control through Learning to Reduce Biological Joint Moments in Simulations
标题：通过学习减少模拟中的生物关节矩来控制外骨骼
链接：https://arxiv.org/abs/2603.07629
【21】Accelerating Diffusion Models for Generative AI Applications with Silicon Photonics
标题：硅光电子生成人工智能应用的扩散加速模型
链接：https://arxiv.org/abs/2603.07626
【22】TT-Sparse: Learning Sparse Rule Models with Differentiable Truth Tables
标题：TT-Sparse：使用可微真值表学习稀疏规则模型
链接：https://arxiv.org/abs/2603.07606
【23】Models as Lego Builders: Assembling Malice from Benign Blocks via Semantic Blueprints
标题：乐高建造者模型：通过语义蓝图从良性积木中组装恶意
链接：https://arxiv.org/abs/2603.07590
【24】COOL-MC: Verifying and Explaining RL Policies for Multi-bridge Network Maintenance
标题：COOL-MC：验证并解释用于多桥网络维护的RL策略
链接：https://arxiv.org/abs/2603.07546
【25】DreamSAC: Learning Hamiltonian World Models via Symmetry Exploration
标题：DreamSAC：通过对称性探索学习Hamilton世界模型
链接：https://arxiv.org/abs/2603.07545
备注：19 pages, 5 figures
【26】Neural Dynamics-Informed Pre-trained Framework for Personalized Brain Functional Network Construction
标题：用于个性化脑功能网络构建的神经动力学预训练框架
链接：https://arxiv.org/abs/2603.07524
【27】One-for-All Model Initialization with Frequency-Domain Knowledge
标题：具有频域知识的一体化模型收件箱
链接：https://arxiv.org/abs/2603.07523
【28】A Unified View of Drifting and Score-Based Models
标题：漂移和基于分数的模型的统一视图
链接：https://arxiv.org/abs/2603.07514
【29】Scaling Laws in the Tiny Regime: How Small Models Change Their Mistakes
标题：小政权中的缩放定律：小模型如何改变错误
链接：https://arxiv.org/abs/2603.07365
备注：17 pages, 6 figures, 2 tables. Submitted to MDPI Machine Learning and Knowledge Extraction (MAKE)
【30】Latent Generative Models with Tunable Complexity for Compressed Sensing and other Inverse Problems
标题：用于压缩感知和其他反问题的具有可调复杂性的潜在生成模型
链接：https://arxiv.org/abs/2603.07357
【31】A Distributed Gaussian Process Model for Multi-Robot Mapping
标题：多机器人地图的分布式高斯过程模型
链接：https://arxiv.org/abs/2603.07351
备注：ICRA 2026, 8 pages
【32】StructSAM: Structure- and Spectrum-Preserving Token Merging for Segment Anything Models
标题：StructSam：Segment Anything模型的结构和光谱保留代币合并
链接：https://arxiv.org/abs/2603.07307
备注：Firsrt version
【33】Combining Adam and its Inverse Counterpart to Enhance Generalization of Deep Learning Optimizers
标题：结合Adam及其反向对应物以增强深度学习优化器的通用性
链接：https://arxiv.org/abs/2603.07122
【34】VLN-Cache: Enabling Token Caching for VLN Models with Visual/Semantic Dynamics Awareness
标题：VLN-缓存：为具有视觉/语义动态感知的VLN模型启用令牌缓存
链接：https://arxiv.org/abs/2603.07080
【35】The Talking Robot: Distortion-Robust Acoustic Models for Robot-Robot Communication
标题：会说话的机器人：机器人与机器人通信的失真鲁棒声学模型
链接：https://arxiv.org/abs/2603.07072
【36】Learning Quadruped Walking from Seconds of Demonstration
标题：从几秒钟的演示中学习四足步行
链接：https://arxiv.org/abs/2603.06961
【37】Physics-Consistent Neural Networks for Learning Deformation and Director Fields in Microstructured Media with Loss-Based Validation Criteria
标题：具有基于损失的验证标准的物理一致性神经网络用于学习微结构媒体中的变形和指向器场
链接：https://arxiv.org/abs/2603.06939
【38】Swimba: Switch Mamba Model Scales State Space Models
标题：Swimba：切换曼巴模型扩展状态空间模型
链接：https://arxiv.org/abs/2603.06938
【39】Learning From Design Procedure To Generate CAD Programs for Data Augmentation
标题：从设计过程中学习生成用于数据扩充的CAD程序
链接：https://arxiv.org/abs/2603.06894
备注：Accepted by NeurIPS 2025 Workshop: Deep Learning for Code in the Agentic Era
【40】NEST: Network- and Memory-Aware Device Placement For Distributed Deep Learning
标题：NEST：分布式深度学习的网络和内存感知设备放置
链接：https://arxiv.org/abs/2603.06798
备注：Accepted to MLSys 2026
【41】Heterogeneous Decentralized Diffusion Models
标题：异类分散扩散模型
链接：https://arxiv.org/abs/2603.06741
备注：Accepted to CVPR2026
【42】ProtAlign: Contrastive learning paradigm for Sequence and structure alignment
标题：ProtAlign：序列和结构对齐的对比学习范式
链接：https://arxiv.org/abs/2603.06722
备注：5 pages, 4 figures
【43】ERP-RiskBench: Leakage-Safe Ensemble Learning for Financial Risk
标题：ERP-RiskBench：针对金融风险的泄漏安全学生学习
链接：https://arxiv.org/abs/2603.06671
备注：12 pages, 11 figures, 8 tables
【44】Geodesic Gradient Descent: A Generic and Learning-rate-free Optimizer on Objective Function-induced Manifolds
标题：测地梯度下降：目标函数诱导的多边形上的通用且无学习率优化器
链接：https://arxiv.org/abs/2603.06651
【45】Roots Beneath the Cut: Uncovering the Risk of Concept Revival in Pruning-Based Unlearning for Diffusion Models
标题：削减之下的根源：揭示扩散模型基于修剪的取消学习中概念复兴的风险
链接：https://arxiv.org/abs/2603.06640
备注：Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026
【46】Switchable Activation Networks
标题：可切换激活网络
链接：https://arxiv.org/abs/2603.06601
备注：14 pages, 9 figures
【47】vLLM Hook v0: A Plug-in for Programming Model Internals on vLLM
标题：vLLM Hook v0：用于在vLLM上编程模型内部的插件
链接：https://arxiv.org/abs/2603.06588
【48】Structural Causal Bottleneck Models
标题：结构性因果瓶颈模型
链接：https://arxiv.org/abs/2603.08682
【49】Scaling Machine Learning Interatomic Potentials with Mixtures of Experts
标题：利用专家混合来扩大机器学习原子间潜力
链接：https://arxiv.org/abs/2603.07977
【50】Learning embeddings of non-linear PDEs: the Burgers' equation
标题：非线性偏出方程的学习嵌入：伯格斯方程
链接：https://arxiv.org/abs/2603.07812
备注：Accepted to ICLR2026
【51】Lindbladian Learning with Neural Differential Equations
标题：利用神经方程进行Lindbladian学习
链接：https://arxiv.org/abs/2603.07778
备注：22 pages, 15 figures
【52】Conditional Rank-Rank Regression via Deep Conditional Transformation Models
标题：通过深度条件转换模型的条件排名-排名回归
链接：https://arxiv.org/abs/2603.07230
【53】Fairness May Backfire: When Leveling-Down Occurs in Fair Machine Learning
标题：公平可能会适得其反：当公平机器学习中的分层工作时
链接：https://arxiv.org/abs/2603.06901
备注：Short version of the paper (Nov 20, 2025)
【54】Quantum Deep Learning: A Comprehensive Review
标题：量子深度学习：全面评论
链接：https://arxiv.org/abs/2603.06644
其他(78篇)
【1】Agentic Critical Training
标题：强化批判性训练
链接：https://arxiv.org/abs/2603.08706
备注：Project page: https://attention-is-all-i-need.github.io/ACT/
【2】A New Lower Bound for the Random Offerer Mechanism in Bilateral Trade using AI-Guided Evolutionary Search
标题：使用人工智能引导的进化搜索的双边贸易中随机报价机制的新下限
链接：https://arxiv.org/abs/2603.08679
【3】Divide and Predict: An Architecture for Input Space Partitioning and Enhanced Accuracy
标题：划分和预测：输入空间划分和增强准确性的架构
链接：https://arxiv.org/abs/2603.08649
备注：Under review; 24 pages; 8 figures
【4】Grow, Don't Overwrite: Fine-tuning Without Forgetting
标题：成长，不要覆盖：不忘微调
链接：https://arxiv.org/abs/2603.08647
【5】Retrieval-Augmented Gaussian Avatars: Improving Expression Generalization
标题：检索增强的高斯化身：改进表达式泛化
链接：https://arxiv.org/abs/2603.08645
【6】Integral Formulas for Vector Spherical Tensor Products
标题：向球张量积的积分公式
链接：https://arxiv.org/abs/2603.08630
备注：16 pages, 2 figures
【7】Drift-to-Action Controllers: Budgeted Interventions with Online Risk Certificates
标题：行动漂移控制者：通过在线风险证书进行强制干预
链接：https://arxiv.org/abs/2603.08578
备注：Published as a conference paper at CAO Workshop at ICLR 2026
【8】Trust via Reputation of Conviction
标题：通过定罪声誉来信任
链接：https://arxiv.org/abs/2603.08575
备注：19 pages, 4 figures
【9】Interactive World Simulator for Robot Policy Training and Evaluation
标题：用于机器人政策训练和评估的互动世界模拟器
链接：https://arxiv.org/abs/2603.08546
备注：Project Page: https://yixuanwang.me/interactive_world_sim
【10】The Neural Compass: Probabilistic Relative Feature Fields for Robotic Search
标题：神经指南针：机器人搜索的概率相对特征场
链接：https://arxiv.org/abs/2603.08544
备注：9 pages, 7 figures, 2 tables, submitted to IROS 2026
【11】STRIDE: Structured Lagrangian and Stochastic Residual Dynamics via Flow Matching
标题：STRIDE：通过流匹配的结构化拉格朗日和随机剩余动力学
链接：https://arxiv.org/abs/2603.08478
备注：9 pages, 7 figures
【12】IronEngine: Towards General AI Assistant
标题：IronEngine：走向通用人工智能助理
链接：https://arxiv.org/abs/2603.08425
备注：Technical Report
【13】Geometrically Constrained Outlier Synthesis
标题：几何约束离群点合成
链接：https://arxiv.org/abs/2603.08413
备注：18 pages, 6 figures
【14】Concept-Guided Fine-Tuning: Steering ViTs away from Spurious Correlations to Improve Robustness
标题：概念引导微调：引导ViT远离虚假相关性以提高稳健性
链接：https://arxiv.org/abs/2603.08309
备注：CVPR 2026 ; Project page: https://yonisgit.github.io/concept-ft/
【15】Wiener Chaos Expansion based Neural Operator for Singular Stochastic Partial Differential Equations
标题：奇异随机偏方程基于Wiener混乱展开的神经运算
链接：https://arxiv.org/abs/2603.08219
【16】Sequential Service Region Design with Capacity-Constrained Investment and Spillover Effect
标题：考虑能力限制投资和溢出效应的顺序服务区设计
链接：https://arxiv.org/abs/2603.08188
【17】DARC: Disagreement-Aware Alignment via Risk-Constrained Decoding
标题：DARC：通过风险约束解码实现分歧意识对齐
链接：https://arxiv.org/abs/2603.08145
【18】SaiVLA-0: Cerebrum--Pons--Cerebellum Tripartite Architecture for Compute-Aware Vision-Language-Action
标题：SaiVLA-0：大脑--Pons--大脑计算机感知视觉-语言-动作三方架构
链接：https://arxiv.org/abs/2603.08124
备注：14 pages, 3 figures
【19】EAGLE-Pangu: Accelerator-Safe Tree Speculative Decoding on Ascend NPUs
标题：EAGLE-Pangu：上行NPU上的加速器安全树推测解码
链接：https://arxiv.org/abs/2603.08088
备注：14 pages. 7 figures
【20】FedMomentum: Preserving LoRA Training Momentum in Federated Fine-Tuning
标题：FedMomentum：在联邦微调中保持LoRA训练势头
链接：https://arxiv.org/abs/2603.08014
【21】Amortizing Maximum Inner Product Search with Learned Support Functions
标题：利用习得的支持功能摊销最大的内部产品搜索
链接：https://arxiv.org/abs/2603.08001
【22】MJ1: Multimodal Judgment via Grounded Verification
标题：MJ 1：通过接地验证的多模式判断
链接：https://arxiv.org/abs/2603.07990
【23】\$OneMillion-Bench: How Far are Language Agents from Human Experts?
链接：https://arxiv.org/abs/2603.07980
备注：39 pages, 9 figures, 8 tables
【24】Semantic Risk Scoring of Aggregated Metrics: An AI-Driven Approach for Healthcare Data Governance
标题：聚合收件箱的语义风险评分：一种人工智能驱动的医疗保健数据治理方法
链接：https://arxiv.org/abs/2603.07924
备注：6 pages, 3 figures, 1 Table, Accepted for publication in the 21st Int. Conference on Data Science (ICDATA 25)
【25】SMGI: A Structural Theory of General Artificial Intelligence
标题：SMGI：通用人工智能的结构理论
链接：https://arxiv.org/abs/2603.07896
备注：Preprint. 77 pages, 1 figure, 3 tables
【26】Designing probabilistic AI monsoon forecasts to inform agricultural decision-making
标题：设计概率人工智能季风预测为农业决策提供信息
链接：https://arxiv.org/abs/2603.07893
【27】Viewpoint-Agnostic Grasp Pipeline using VLM and Partial Observations
标题：使用VLM和部分观察的观点不可知的抓取管道
链接：https://arxiv.org/abs/2603.07866
【28】Neural Precoding in Complex Projective Spaces
标题：复射影空间中的神经预编码
链接：https://arxiv.org/abs/2603.07811
【29】ProgAgent:A Continual RL Agent with Progress-Aware Rewards
标题：ProgAgent：具有进度感知奖励的连续RL代理
链接：https://arxiv.org/abs/2603.07784
【30】A Lightweight MPC Bidding Framework for Brand Auction Ads
标题：品牌拍卖广告的轻量级MPC竞标框架
链接：https://arxiv.org/abs/2603.07721
【31】Deep Incentive Design with Differentiable Equilibrium Blocks
标题：具有差异均衡块的深度激励设计
链接：https://arxiv.org/abs/2603.07705
备注：24 pages, 7 figures
【32】MAS-H2: A Hierarchical Multi-Agent System for Holistic Cloud-Native Autoscaling
标题：MAS-H2：一个用于整体云原生自动缩放的分层多代理系统
链接：https://arxiv.org/abs/2603.07607
【33】Shorter Thoughts, Same Answers: Difficulty-Scaled Segment-Wise RL for CoT Compression
标题：更短的想法，相同的答案：用于CoT压缩的难度扩展分段RL
链接：https://arxiv.org/abs/2603.07598
备注：12 pages, 3 figures. Preprint. Code available at the GitHub project repository
【34】Obliviator Reveals the Cost of Nonlinear Guardedness in Concept Erasure
标题：遗忘者揭示了概念擦除中非线性保护的成本
链接：https://arxiv.org/abs/2603.07529
备注：Accepted to NeurIPS 2025 [Poster]. Code available at: https://github.com/ramin-akbari/Obliviator
【35】Pushing Bistatic Wireless Sensing toward High Accuracy at the Sub-Wavelength Scale
标题：推动双站无线传感在亚波长尺度上实现高准确度
链接：https://arxiv.org/abs/2603.07492
【36】Dial: A Knowledge-Grounded Dialect-Specific NL2SQL System
标题：Dial：一个基于知识的特定于拨号的NL2SQL系统
链接：https://arxiv.org/abs/2603.07449
【37】Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning
标题：代币少，杠杆大：在微调期间通过限制安全代币来保持安全一致
链接：https://arxiv.org/abs/2603.07445
【38】DualSpec: Accelerating Deep Research Agents via Dual-Process Action Speculation
标题：DualSec：通过双流程行动推测加速深度研究代理
链接：https://arxiv.org/abs/2603.07416
【39】Sparsity and Out-of-Distribution Generalization
标题：稀疏性和分布外概括
链接：https://arxiv.org/abs/2603.07388
【40】ConfHit: Conformal Generative Design with Oracle Free Guarantees
标题：ConfHit：具有Oracle免费保证的保形生成设计
链接：https://arxiv.org/abs/2603.07371
备注：Accepted at ICLR 2026
【41】AgrI Challenge: A Data-Centric AI Competition for Cross-Team Validation in Agricultural Vision
标题：AgRI挑战：一场以数据为中心的人工智能农业愿景跨团队验证竞赛
链接
：https://arxiv.org/abs/2603.07356
备注：17 pages, 8 figures, 6 tables. Introduces the AgrI Challenge dataset containing 50,673 field images of six tree species collected by twelve independent teams
【42】ShakyPrepend: A Multi-Group Learner with Improved Sample Complexity
标题：ShakyPrepend：具有改进样本复杂性的多组学习者
链接：https://arxiv.org/abs/2603.07319
备注：29 pages, 10 figures, submitted to ICML2026
【43】Shutdown Safety Valves for Advanced AI
标题：适用于高级人工智能的安全阀
链接：https://arxiv.org/abs/2603.07315
【44】Spectral Discovery of Continuous Symmetries via Generalized Fourier Transforms
标题：通过广义傅里叶变换发现连续对称性
链接：https://arxiv.org/abs/2603.07299
【45】Rethinking Deep Research from the Perspective of Web Content Distribution Matching
标题：从网络内容分发匹配角度重新思考深度研究
链接：https://arxiv.org/abs/2603.07241
【46】Margin in Abstract Spaces
标题：抽象空间中的页边
链接：https://arxiv.org/abs/2603.07221
【47】Towards Objective Gastrointestinal Auscultation: Automated Segmentation and Annotation of Bowel Sound Patterns
标题：走向客观的胃肠道听诊：肠道声音模式的自动分割和注释
链接：https://arxiv.org/abs/2603.07215
【48】Countdown-Code: A Testbed for Studying The Emergence and Generalization of Reward Hacking in RLVR
标题：Countdown-Code：一个研究RLVR中奖励黑客的产生和推广的实验平台
链接：https://arxiv.org/abs/2603.07084
【49】Combinatorial Allocation Bandits with Nonlinear Arm Utility
标题：具有非线性手臂效用的组合分配盗贼
链接：https://arxiv.org/abs/2603.07005
备注：32 pages
【50】Diffusion Controller: Framework, Algorithms and Parameterization
标题：扩散控制器：框架、算法和参数化
链接：https://arxiv.org/abs/2603.06981
【51】CN-CBF: Composite Neural Control Barrier Function for Safe Robot Navigation in Dynamic Environments
标题：CN-CBF：动态环境中机器人安全导航的复合神经控制屏障函数
链接：https://arxiv.org/abs/2603.06921
【52】XGenBoost: Synthesizing Small and Large Tabular Datasets with XGBoost
标题：XGenboost：使用XGboost合成小型和大型表格数据集
链接：https://arxiv.org/abs/2603.06904
【53】Stochastic Attention via Langevin Dynamics on the Modern Hopfield Energy
标题：通过朗之万动力学对现代霍普菲尔德能量的随机注意力
链接：https://arxiv.org/abs/2603.06875
备注：Main body (including references excluding the appendix): 11 pages, 2 figures and 1 table. Total paper: 26 pages, 13 figures and 7 pages
【54】Symmetry-Constrained Language-Guided Program Synthesis for Discovering Governing Equations from Noisy and Partial Observations
标题：用于从有噪和部分观测中发现控制方程的对称约束网格引导程序综合
链接：https://arxiv.org/abs/2603.06869
备注：12 pages, 4 figures, 5 tables
【55】IGLU: The Integrated Gaussian Linear Unit Activation Function
标题：IGLU：积分高斯线性单位激活函数
链接：https://arxiv.org/abs/2603.06861
【56】Optimistic Policy Regularization
标题：乐观的政策规范化
链接：https://arxiv.org/abs/2603.06793
【57】xaitimesynth: A Python Package for Evaluating Attribution Methods for Time Series with Synthetic Ground Truth
标题：xaitimesynth：一个用于使用合成基本真相评估时间序列归因方法的Python包
链接：https://arxiv.org/abs/2603.06781
备注：9 pages, 1 figure, 2 tables, 1 listing
【58】Property-driven Protein Inverse Folding With Multi-Objective Preference Alignment
标题：具有多目标偏好比对的性质驱动的蛋白质反向折叠
链接：https://arxiv.org/abs/2603.06748
【59】PolyBlocks: A Compiler Infrastructure for AI Chips and Programming Frameworks
标题：PolyBlocks：人工智能芯片和编程框架的更简单基础设施
链接：https://arxiv.org/abs/2603.06731
备注：24 pages with 5 pages of appendix
【60】Don't Freeze, Don't Crash: Extending the Safe Operating Range of Neural Navigation in Dense Crowds
标题：不要冷冻，不要崩溃：在密集人群中扩大神经导航的安全操作范围
链接：https://arxiv.org/abs/2603.06729
【61】Scaling Agentic Capabilities, Not Context: Efficient Reinforcement Finetuning for Large Toolspaces
标题：扩展统计能力，而不是上下文：大型工具空间的高效强化微调
链接：https://arxiv.org/abs/2603.06713
【62】On the Generalization Capacities of MLLMs for Spatial Intelligence
标题：论MLLM空间智能的概括能力
链接：https://arxiv.org/abs/2603.06704
备注：ICLR 2026 (Oral)
【63】One step further with Monte-Carlo sampler to guide diffusion better
标题：蒙特卡洛采样器更进一步，更好地引导扩散
链接：https://arxiv.org/abs/2603.06685
备注：16 pages, 7 figures, accepted at ICLR2026
【64】Unmixing microinfrared spectroscopic images of cross-sections of historical oil paintings
标题：分解历史油画横剖面的微红外光谱图像
链接：https://arxiv.org/abs/2603.06673
备注：5 pages
【65】SR-TTT: Surprisal-Aware Residual Test-Time Training
标题：SR-TTT：惊喜感知剩余测试时间训练
链接：https://arxiv.org/abs/2603.06642
备注：7 pages, 5 figures
【66】Multi-Agent DRL for V2X Resource Allocation: Disentangling Challenges and Benchmarking Solutions
标题：V2X资源分配的多代理DRL：解决挑战和基准解决方案
链接：https://arxiv.org/abs/2603.06607
【67】Scale Dependent Data Duplication
标题：规模相关数据重复
链接：https://arxiv.org/abs/2603.06603
【68】FuzzingRL: Reinforcement Fuzz-Testing for Revealing VLM Failures
标题：FuzzingRL：揭示VLM故障的强化模糊测试
链接：https://arxiv.org/abs/2603.06600
备注：18 pages, 4 figures. † These authors jointly supervised this work: Jiageng Mao and Yue Wang
【69】XInsight: Integrative Stage-Consistent Psychological Counseling Support Agents for Digital Well-Being
标题：XInsight：数字福祉的综合阶段一致心理咨询支持代理
链接：https://arxiv.org/abs/2603.06583
备注：Accepted by WWW 2026
【70】Unifying On- and Off-Policy Variance Reduction Methods
标题：统一政策内和政策外差异减少方法
链接：https://arxiv.org/abs/2603.08370
【71】Sign Identifiability of Causal Effects in Stationary Stochastic Dynamical Systems
标题：平稳随机动力系统因果效应的符号可识别性
链接：https://arxiv.org/abs/2603.08311
【72】Beyond ReinMax: Low-Variance Gradient Estimators for Discrete Latent Variables
标题：超越ReinMax：离散潜在变量的低方差梯度估计器
链接：https://arxiv.org/abs/2603.08257
【73】RL unknotter, hard unknots and unknotting number
标题：RL打结器、硬打结器和打结号码
链接：https://arxiv.org/abs/2603.07955
备注：15 pages, many figures, comments welcome
【74】Fast and Flexible Audio Bandwidth Extension via Vocos
标题：通过Vocos快速灵活的音频带宽扩展
链接：https://arxiv.org/abs/2603.07285
备注：5 pages, 2 figures, 5 tables. Submitted to INTERSPEECH 2026. Code available at https://github.com/ysharma3501/LavaSR.git
【75】TEA-Time: Transporting Effects Across Time
标题：TEA-Time：跨时间传输效应
链接：https://arxiv.org/abs/2603.07018
【76】Masked Unfairness: Hiding Causality within Zero ATE
标题：被掩盖的不公平：将因果关系隐藏在零ATE中
链接：https://arxiv.org/abs/2603.06984
【77】Bilateral Trade Under Heavy-Tailed Valuations: Minimax Regret with Infinite Variance
标题：重尾估值下的双边贸易：无限方差的极小极大遗憾
链接：https://arxiv.org/abs/2603.06851
备注：9 pages
【78】CREDO: Epistemic-Aware Conformalized Credal Envelopes for Regression
标题：CREDO：认知意识回归的适形Credal信封
链接：https://arxiv.org/abs/2603.06826
备注：26 pages, 5 figures
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/193789
本文地址：http://www.python88.com/topic/193789