arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

International Conference on Learning Representations · 会议 · Machine Learning

2026-02-02 至 2026-02-02 共收录 21
2601.23159 2026-02-02 cs.CV

Segment Any Events with Language

通过语言进行任意事件分割

Seungjun Lee, Gim Hee Lee

机构 * Department of Computer Science, National University of Singapore(新加坡国立大学计算机科学系)

AI总结 SEAL通过语义感知框架实现开放词汇事件实例分割,支持多粒度级别的事件分割与掩码分类,并在多个基准测试中表现出优越的性能和效率。

Comments ICLR 2026. Project Page: https://0nandon.github.io/SEAL

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.22905 2026-02-02 cs.LG

FlexLoRA: Entropy-Guided Flexible Low-Rank Adaptation

FlexLoRA: 基于熵的灵活低秩适应

Muqing Liu, Chongjie Si, Yuheng Jia

机构 * Chien-Shiung Wu College, Southeast University(陈希ung Wu学院,东南大学) MoE Key Lab of Artificial Intelligence, AI Institute, School of Computer Science, Shanghai Jiao Tong University(人工智能MOE实验室,人工智能研究院,上海交通大学计算机科学学院) School of Computer Science and Engineering, Southeast University(计算机科学与工程学院,东南大学) Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China(新一代人工智能技术及其跨学科应用关键实验室(东南大学),教育部,中国)

AI总结 FlexLoRA通过基于熵的灵活低秩适应框架,解决PEFT中粒度、灵活性和稳定性问题,实现性能提升。

Comments 2026 ICLR. Codes in https://github.com/Chongjie-Si/Subspace-Tuning

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.13837 2026-02-02 cs.CV

FastGHA: Generalized Few-Shot 3D Gaussian Head Avatars with Real-Time Animation

FastGHA: 基于实时动画的通用少样本3D高斯头身像

Xinya Ji, Sebastian Weiss, Manuel Kansy, Jacek Naruniec, Xun Cao, Barbara Solenthaler, Derek Bradley

机构 * Nanjing University(南京大学) ETH Zürich(苏黎世联邦理工学院) DisneyResearch|Studios(迪士尼研究与工作室)

AI总结 FastGHA通过少样本输入图像生成高质量3D头身像,并支持实时动画,提升了推理效率和渲染质量。

Comments Accepted to ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.05005 2026-02-02 cs.LG cs.AI cs.RO

Multi-agent Coordination via Flow Matching

通过流匹配实现多智能体协调

Dongsu Lee, Daehee Lee, Amy Zhang

机构 * University of Texas at Austin(德克萨斯大学奥斯汀分校) Sungkyunkwan University(松均大学)

AI总结 MAC-Flow通过流匹配方法实现多智能体协调,平衡了性能与计算效率,显著提升推理速度的同时保持良好表现。

Comments ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.26374 2026-02-02 cs.AI

BOTS: A Unified Framework for Bayesian Online Task Selection in LLM Reinforcement Finetuning

BOTS:一种用于LLM强化微调中贝叶斯在线任务选择的统一框架

Qianli Shen, Daoyuan Chen, Yilun Huang, Zhenqing Ling, Yaliang Li, Bolin Ding, Jingren Zhou

机构 * Alibaba Group(阿里巴巴集团)

AI总结 BOTS通过贝叶斯推断和汤普森采样,提高LLM强化微调中任务选择的效率和性能。

Comments Accepted as a conference paper at ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.25801 2026-02-02 cs.LG cs.AI cs.CL cs.CV

Metis-SPECS: Decoupling Multimodal Learning via Self-distilled Preference-based Cold Start

Metis-SPECS: 通过基于偏好自我蒸馏的冷启动解耦多模态学习

Kun Chen, Peng Shi, Haibo Qiu, Zhixiong Zeng, Siqi Yang, Wenji Mao, Lin Ma

机构 * School of Artificial Intelligence, University of Chinese Academy of Sciences(中国科学院大学人工智能学院) MAIS, Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所MAIS) Meituan(美团)

AI总结 Metis-SPECS通过基于偏好的自我蒸馏冷启动框架解耦多模态学习,提升泛化能力和下游RL表现。

Comments Published as a conference paper at ICLR 2026!

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.04347 2026-02-02 cs.CL cs.LG

Unmasking Backdoors: An Explainable Defense via Gradient-Attention Anomaly Scoring for Pre-trained Language Models

揭示后门:通过梯度-注意力异常评分对预训练语言模型进行可解释防御

Anindya Sundar Das, Kangjie Chen, Monowar Bhuyan

机构 * Umeå University(乌梅学院) Nanyang Technological University(南洋理工大学)

AI总结 本文提出通过梯度-注意力异常评分对预训练语言模型进行可解释防御,有效降低后门攻击成功率。

Comments 17 pages total (9 pages main text + 6 pages appendix + references), 16 figures. Preprint version; the final camera-ready version may differ. Accepted to ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2508.19236 2026-02-02 cs.RO cs.CV

MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation

MemoryVLA: 视觉-语言-动作模型中用于机器人操作的感知-认知记忆

Hao Shi, Bin Xie, Yingfei Liu, Lin Sun, Fengrong Liu, Tiancai Wang, Erjin Zhou, Haoqiang Fan, Xiangyu Zhang, Gao Huang

机构 * Department of Automation, BNRist, Tsinghua University(自动化系、BNRist、清华大学) Dexmal MEGVII Technology(MEGVII技术) Tianjin University(天津大学) Harbin Institute of Technology(哈尔滨工业大学) StepFun

AI总结 MemoryVLA通过结合感知与认知记忆机制,提升机器人操作中长时间跨度任务的性能,实现对时间依赖任务的高效处理。

Comments ICLR 2026 | The project is available at https://shihao1895.github.io/MemoryVLA

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.12260 2026-02-02 cs.IR cs.AI cs.CL

LightRetriever: A LLM-based Text Retrieval Architecture with Extremely Faster Query Inference

LightRetriever: 一种基于大语言模型的文本检索架构,具有极快的查询推理速度

Guangyuan Ma, Yongliang Ma, Xuanrui Gou, Zhenpeng Su, Ming Zhou, Songlin Hu

机构 * Institute of Information Engineering, Chinese Academy of Sciences(中国科学院信息工程研究所) School of Cyber Security, University of Chinese Academy of Sciences(中国科学院大学网络安全学院) Langboat Technology(兰舟科技)

AI总结 LightRetriever通过轻量级查询编码器显著提升检索效率,实现查询推理速度提升超1000倍,端到端吞吐量提升超10倍,且在多种任务中保持95%的检索性能。

Comments Accepted by ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.22647 2026-02-02 cs.AI

Test-Time Mixture of World Models for Embodied Agents in Dynamic Environments

动态环境中具身智能体的测试时间世界模型混合

Jinwoo Jang, Minjong Yoo, Sihyung Yoon, Honguk Woo

机构 * Department of Computer Science and Engineering(计算机科学与工程系) Sungkyunkwan University(庆尚大学)

AI总结 TMoW通过测试时间更新路由函数,提升具身智能体在动态环境中的适应性与持续学习能力。

Comments Accepted at ICLR 2026. 10 pages. Code available at https://github.com/doldam0/tmow

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.22570 2026-02-02 cs.CV cs.LG

Leveraging Data to Say No: Memory Augmented Plug-and-Play Selective Prediction

利用数据说不:基于记忆的插拔式选择预测

Aditya Sarkar, Yi Li, Jiacheng Cheng, Shlok Mishra, Nuno Vasconcelos

机构 * University of Maryland, College Park(马里兰大学) University of California, San Diego(加州大学圣地亚哥分校) Qualcomm AI(高通人工智能) Yale University(耶鲁大学) Meta AI

AI总结 本文提出基于记忆的插拔式选择预测方法,通过增强视觉语言嵌入来减少方差并改进分数校准,提升图像描述、匹配和细粒度分类性能。

Comments ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.22550 2026-02-02 cs.RO cs.GR cs.LG

Exo-Plore: Exploring Exoskeleton Control Space through Human-aligned Simulation

Exo-Plore:通过与人类对齐的模拟探索外骨骼控制空间

Geonho Leem, Jaedong Lee, Jehee Lee, Seungmoon Song, Jungdam Won

机构 * Seoul National University(首尔国立大学) Holiday Robotics(假日机器人) Northeastern University(东北大学)

AI总结 Exo-Plore通过神经机械模拟与深度强化学习优化髋部外骨骼辅助,无需真实人类实验,生成逼真步态数据并推广至病理性步态。

Comments 10 pages, 9 figures, ICLR 2026 accepted

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.21969 2026-02-02 cs.CL cs.AI

Token-Guard: Towards Token-Level Hallucination Control via Self-Checking Decoding

Token-Guard: 通过自检解码实现token级幻觉控制

Yifan Zhu, Huiqiang Rong, Haoran Luo

机构 * Beijing University of Posts and Telecommunications(北京邮电大学) Nanyang Technological University(南洋理工大学)

AI总结 Token-Guard通过自检解码技术,实现对大型语言模型中token级幻觉的有效控制,提升生成准确性与输出可靠性。

Comments Accepted by ICLR 2026 main conference

Journal ref ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.11391 2026-02-02 cs.LG

Mitigating the Safety Alignment Tax with Null-Space Constrained Policy Optimization

通过空域约束策略优化缓解安全对齐税

Yifan Niu, Han Xiao, Dongyi Liu, Nuo Chen, Jia Li

机构 * The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州))

AI总结 通过空域约束策略优化缓解安全对齐税,提出NSPO框架在保留LLM核心能力的同时提升安全性能。

Comments accepted by ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.07222 2026-02-02 cs.CV

Omni-View: Unlocking How Generation Facilitates Understanding in Unified 3D Model based on Multiview images

Omni-View: 通过多视角图像解锁生成如何促进理解的统一3D模型

JiaKui Hu, Shanshan Zhao, Qing-Guo Chen, Xuerui Qiu, Jialun Liu, Zhao Xu, Weihua Luo, Kaifu Zhang, Yanye Lu

机构 * Institute of Medical Technology(医学技术研究所) Alibaba International Digital Commerce Group(阿里巴巴国际数字商务集团) CASIA TeleAI

AI总结 Omni-View通过多视角图像实现3D场景的理解与生成,结合纹理和几何模块,提升3D场景建模性能。

Comments Accepted by ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.02767 2026-02-02 cs.CV

Dynamic Reflections: Probing Video Representations with Text Alignment

动态反射:通过文本对齐探测视频表示

Tyler Zhu, Tengda Han, Leonidas Guibas, Viorica Pătrăucean, Maks Ovsjanikov

机构 * Princeton University(普林斯顿大学) Google DeepMind(谷歌DeepMind)

AI总结 本研究通过视频-文本表示对齐探索现代视频和语言编码器的能力,揭示了跨模态对齐与数据丰富性、语义对齐与性能相关性以及时间推理与对齐的关系。

Comments To appear at ICLR 2026. 27 pages, 12 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.10197 2026-02-02 cs.AI

Don't Just Fine-tune the Agent, Tune the Environment

不要只微调智能体,而是微调环境

Siyuan Lu, Zechuan Wang, Hongxuan Zhang, Qintong Wu, Leilei Gan, Chenyi Zhuang, Jinjie Gu, Tao Lin

机构 * Zhejiang University(浙江大学) Shanghai Innovation Institute(上海创新研究院) Westlake University(西湖大学) Nanjing University(南京大学)

AI总结 本文提出环境微调方法,通过结构化课程和环境增强,使智能体无需专家轨迹即可高效学习复杂行为,实现更稳健的泛化能力。

Comments ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.03267 2026-02-02 cs.LG cs.AI

PT$^2$-LLM: Post-Training Ternarization for Large Language Models

PT²-LLM:大语言模型的后训练三元化

Xianglong Yan, Chengzhu Bao, Zhiteng Li, Tianao Zhang, Kaicheng Yang, Haotong Qin, Ruobing Xie, Xingwu Sun, Yulun Zhang

机构 * Shanghai Jiao Tong University(上海交通大学) ETH Zürich(苏黎世联邦理工学院) Tencent Hunyuan(腾讯文言)

AI总结 PT²-LLM通过后训练三元化技术,在降低内存成本的同时提升大语言模型的推理速度和效率。

Comments Accepted at ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.02295 2026-02-02 cs.CV cs.AI cs.LG

VideoNSA: Native Sparse Attention Scales Video Understanding

VideoNSA:原生稀疏注意力扩展视频理解

Enxin Song, Wenhao Chai, Shusheng Yang, Ethan Armand, Xiaojun Shan, Haiyang Xu, Jianwen Xie, Zhuowen Tu

AI总结 VideoNSA通过端到端训练提升视频理解,结合原生稀疏注意力实现长视频和时间推理的性能优化

Comments ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.21612 2026-02-02 cs.GT

Incentives in Federated Learning with Heterogeneous Agents

在异质智能体中的联邦学习激励

Ariel D. Procaccia, Han Shao, Itai Shapira

AI总结 本文提出了一种基于博弈论的联邦学习激励机制,通过线性规划和 pay what you contribute 规则,实现策略证明和最小化合作成本。

Comments ICLR 2026

Journal ref International Conference on Learning Representations (ICLR), 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.07656 2026-02-02 cs.LG cs.AI

Causal Imitation Learning under Expert-Observable and Expert-Unobservable Confounding

在专家可观测和专家不可观测混杂下的因果模仿学习

Daqian Shao, Thomas Kleine Buening, Marta Kwiatkowska

机构 * Department of Computer Science, University of Oxford, UK(计算机科学系,牛津大学) ETH Zurich, Switzerland(苏黎世联邦理工学院)

AI总结 本文提出DML-IL算法,通过工具变量回归解决因果模仿学习中的隐藏混杂问题,并在连续状态-动作环境中优于现有基线方法。

Comments Proceedings of the International Conference on Learning Representations (ICLR) 2026

详情

展开后加载摘要…

URL PDF HTML 收藏