魔幻庄园式评测，专测LLM代理能否在陌生工具场景下自主推理与修正操作。
arXiv:2605.07926v2 Announce Type: replace Abstract: As LLM-based agents increasingly rely on external tools, it is important to evaluate their ability…
被ICML 2026收录，提出针对低资源语言的LLM鲁棒多语言适配方法，增强跨语言泛化能力。
arXiv:2510.14466v3 Announce Type: replace Abstract: Large language models (LLMs) continue to struggle with low-resource languages, primarily due to li…
首次定义具身动作表征的ImageNet基准，揭示人类视频数据可驱动机器人泛化学习。
LARYBench （Latent Action Representation Yielding Benchmark），一个指引从大规模的视觉数据学习到通用的隐式动作表征的系统化评测基准。实验结果表明：在动作泛化和控制精度上，通用视觉模型的表现均显著优于专门为具身智能设计的动作专家模型，具身动作表征…
用国际象棋棋局拆解大模型的记忆与推理边界，揭示模型何时是在背诵、何时真的在推演。
arXiv:2601.16823v2 Announce Type: replace Abstract: Large Language Models (LLMs) exhibit remarkable capabilities, yet it remains unclear to what exten…
用脆性测试揭秘语言模型下象棋：是真正理解规则还是死记硬背？
arXiv:2605.17565v1 Announce Type: cross Abstract: Recent work has fine-tuned language models on chess data and reported high benchmark scores as evide…
从理论层面揭示Transformer在噪声与任务级流形上的学习能力，近似与泛化分析带来新洞察
arXiv:2505.03205v3 Announce Type: replace Abstract: Transformers serve as the foundational architecture for large language and video generation models…
首个泛化感知基准评估GPU内核优化代理，推动自动调优技术落地。
arXiv:2605.16819v1 Announce Type: cross Abstract: GPU kernel optimization is increasingly critical for efficient deep learning systems, but writing hi…
深度神经网络泛化理论新视角，揭示点态泛化机制，突破传统平均泛化分析局限。
arXiv:2605.18598v1 Announce Type: new Abstract: We address the fundamental question of why deep neural networks generalize by establishing a pointwise…
大模型面对同一概念的不同语言或编码形式时，会表现出“分隔化”现象，揭示其泛化能力的缺陷。
arXiv:2605.19284v1 Announce Type: cross Abstract: In the training data used by large language models (LLMs), the same latent concept is often presente…
机器人基础模型新突破：通用姿态预训练让视觉-语言-动作策略泛化能力飙升，已被RSS 2026接收。
arXiv:2602.19710v2 Announce Type: replace-cross Abstract: Existing Vision-Language-Action (VLA) models often suffer from feature collapse and low trai…
提出ARROW增强回放框架，显著提升世界模型在分布外场景的鲁棒性。
arXiv:2603.11395v2 Announce Type: replace Abstract: Continual reinforcement learning challenges agents to acquire new skills while retaining previousl…
超越正确性：通过强化学习调和过程与结果奖励，为模型训练提供新视角
arXiv:2509.03403v2 Announce Type: replace Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) improves final-answer accuracy on reasoning …
深入探讨神经正切核在分类任务中的理论分析，为神经网络泛化研究提供新视角。
arXiv:2605.17606v1 Announce Type: new Abstract: In wide neural networks, the Neural Tangent Kernel (NTK) remains approximately constant during trainin…
新框架PIQL利用特权信息同时加速表格基础模型训练并提升泛化能力。
arXiv:2605.07799v2 Announce Type: replace-cross Abstract: Training foundation models is computationally intensive and often slow to converge. We intro…
用矩阵空间抽象局部转移几何，强化学习实现组合泛化新范式。
arXiv:2605.14304v1 Announce Type: cross Abstract: Compositional generalization in sequential decision-making requires identifying which parts of prior…
OpenAI发布强化学习泛化新基准，加速AI在复杂环境中的快速适应能力。
策略性过参数化提升低秩适应泛化能力，为参数高效微调提供新理论。
arXiv:2605.16470v1 Announce Type: new Abstract: Adapting large language models (LLMs) to downstream tasks via full fine-tuning is increasingly impract…
无需修改现有模型即可融合多种药物信息，大幅提升对未见过药物的相互作用预测能力，AI+生物医药交叉的前沿研究。
arXiv:2605.14327v1 Announce Type: cross Abstract: Drug-drug interaction (DDI) prediction is a critical task in computational biomedicine, as adverse i…
探究任务感知剪枝如何提升模型在分布外数据上的表现，揭示内在机制
arXiv:2605.14738v1 Announce Type: cross Abstract: Recent work has promoted task-aware layer pruning as a way to improve model performance on particula…
提出ReAlign方法，通过推理对齐表征实现高泛化性的AI生成图像伪造检测
arXiv:2605.16080v1 Announce Type: new Abstract: The rise of AI-generated images (AIGIs) poses growing challenges for digital authenticity, prompting t…