arXivDaily arXiv每日学术速递 周一至周五更新

高校专区

Tsinghua University(清华大学)

共收录 4536
2511.07438 2026-06-02 cs.CV cs.NA math.NA stat.ME

Two Datasets Are Better Than One: Method of Double Moments for 3-D Reconstruction in Cryo-EM

两个数据集优于一个:冷冻电镜三维重建的双矩方法

Joe Kileel, Oscar Mickelin, Amit Singer, Sheng Xu

机构 * Department of Mathematics and Oden Institute, University of Texas at Austin(德克萨斯大学奥斯汀分校数学系与奥登研究所) Yau Mathematical Sciences Center, Tsinghua University(清华大学姚贝泰数学科学中心) Program in Applied and Computational Mathematics and Department of Mathematics, Princeton University(普林斯顿大学应用与计算数学项目及数学系) Program in Applied and Computational Mathematics, Princeton University(普林斯顿大学应用与计算数学项目)

AI总结 提出双矩方法(MoDM),利用均匀和非均匀两种取向分布下的二阶矩数据唯一确定分子结构,并开发基于凸松弛的算法实现高精度重建。

详情

展开后加载摘要…

URL PDF HTML 收藏
2410.02511 2026-06-02 cs.AI cs.MA

Stop Wandering, Find the Keys: LLMs Discriminate Key States for Efficient Multi-Agent Exploration

停止徘徊,找到关键:LLMs 辨别关键状态以实现高效多智能体探索

Yun Qu, Boyuan Wang, Yuhang Jiang, Jianzhun Shao, Yixiu Mao, Heming Zou, Chang Liu, Cheems Wang, Meiqin Liu, Xiangyang Ji

机构 * Department of Automation, Tsinghua University, Beijing 100084, China(清华大学自动化系) College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China(浙江大学电气工程学院) National Key Laboratory of Human-Machine Hybrid Augmented Intelligence, Xi’an Jiaotong University, Xi’an 710049, China(西安交通大学人机混合增强智能国家级重点实验室)

AI总结 提出 LEMAE 方法,利用大语言模型辨别关键状态并设计子空间内在奖励和关键状态记忆树,引导多智能体高效探索,在 SMAC 和 MPE 基准上显著超越现有方法,实现 10 倍加速。

Journal ref SCIENCE CHINA Information Sciences 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.31584 2026-06-01 cs.CL cs.AI cs.LG

LongTraceRL: Learning Long-Context Reasoning from Search Agent Trajectories with Rubric Rewards

LongTraceRL: 基于评分奖励从搜索智能体轨迹中学习长上下文推理

Nianyi Lin, Jiajie Zhang, Lei Hou, Juanzi Li

机构 * Tsinghua University(清华大学)

AI总结 提出LongTraceRL框架,通过知识图谱随机游走生成多跳问题并利用搜索智能体轨迹构建分层干扰物,结合基于实体链的评分奖励进行过程监督,提升大语言模型在长上下文推理中的表现。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.31476 2026-06-01 cs.RO

IDOL: Inverse-Dynamics-Guided Future Prediction for End-to-End Autonomous Driving

IDOL: 逆动力学引导的未来预测用于端到端自动驾驶

Chenghao Zhang, Timin Li, Dongmei Li

机构 * Department of Electronic Engineering, Tsinghua University(清华大学电子工程系)

AI总结 提出IDOL框架,通过逆动力学模型将BEV世界模型预测的未来潜在场景状态转化为规划相关的轨迹增量,实现未来预测与轨迹优化的紧密耦合,在NAVSIM基准上达到最优性能。

Comments 20 pages, 5 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.31314 2026-06-01 cs.RO

AR Forcing: Towards Long-Horizon Robot Navigation World Model

AR Forcing: 迈向长时域机器人导航世界模型

Yifei Yang, Zehua Fan, Huan Li, Aoqi Wang, Lida Huang, Haibao Yu, Haiyan Liu, Xuanyao Mao, Jason Bao, Liang Xu, Bingchuan Sun, Yan Wang

机构 * Institute for AI Industry Research, Tsinghua University(清华大学人工智能产业研究院) Shanghai Jiao Tong University(上海交通大学) School of Safety Science, Tsinghua University(清华大学安全科学学院) The University of Hong Kong(香港大学) Lenovo, Beijing, China(北京联想公司)

AI总结 提出AR Forcing自回归训练策略,通过将扩散损失集成到自回归训练循环中,解决训练与推理分布偏移问题,提升长时域导航中图像一致性和轨迹预测精度。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.31234 2026-06-01 cs.RO

HARP-VLA: Human-Robot Aligned Representation Learning for Vision-Language-Action Model

HARP-VLA:面向视觉-语言-动作模型的人机对齐表示学习

Xiang Zhu, Puzhen Yuan, Yichen Liu, Jianyu Chen

机构 * Institute for Interdisciplinary Information Sciences, Tsinghua University, China(清华大学交叉信息研究院) Shanghai Qi Zhi Institute, China(上海启智研究院)

AI总结 提出HARP框架,通过有限配对人机演示和未配对视频,学习对齐的人机视觉与潜在动作表示,提升VLA模型预训练效果,在CALVIN和真实世界任务中取得性能提升。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.31174 2026-06-01 cs.CV cs.LG

Detect in Any Scene: An Agentic Framework for Object Detection with Experience-Aware Reasoning

任意场景检测:一种具有经验感知推理的目标检测智能体框架

Wenlun Zhang, Jun Yin, Kentaro Yoshioka

机构 * Keio University(Keio大学) Tsinghua University(清华大学)

AI总结 提出DetAS/DetAS-X智能体框架,利用多模态大语言模型自适应组合恢复模块和专用检测器,通过自进化经验积累实现经验感知推理,在六个基准上平均F1提升28.36%。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.31148 2026-06-01 cs.CV cs.AI cs.CL

SpatialAct: Probing Spatial Reasoning-to-Action Capabilities of VLM Agents in 3D Scenes

SpatialAct:探测VLM智能体在3D场景中的空间推理到行动能力

Tianhui Liu, Jie Feng, Zhiheng Zheng, Shengyuan Wang, Yiming Guo, Yanxin Xi, Hangyu Fan, Yong Li, Pan Hui

机构 * The Hong Kong University of Science and Technology (Guangzhou)(香港科学与技术大学(广州)) Zhongguancun Academy(中关村学院) Tsinghua University(清华大学) Helsinki University(赫尔辛基大学)

AI总结 本文提出SpatialAct基准,通过多轮交互细化、单步错误检测与修复等任务,揭示当前视觉语言模型在3D场景中从空间推理到行动存在显著差距。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30963 2026-06-01 q-bio.BM cs.AI

AMix-2: Establishing Protein as a Native Modality in Large Language Models

AMix-2:将蛋白质确立为大语言模型的原生模态

Keyue Qiu, Yixin Wu, Lihao Wang, Yawen Ouyang, Jixiang Yu, Zihan Zhou, Changze Lv, Dongyu Xue, Yuxuan Song, Xinbo Zhang, Hao Wang, Jiangtao Feng, Zhiqiang Gao, Lijun Wu, Xiaoqing Zheng, Ka-Chun Wong, Lei Bai, Ya-Qin Zhang, Wei-Ying Ma, Dahua Lin, Bowen Zhou, Hao Zhou

机构 * Shanghai Artificial Intelligence Laboratory(上海人工智能实验室) Generative Symbolic Intelligence Lab (GenSI), Tsinghua University(生成符号智能实验室(GenSI),清华大学) Institute for AI Industry Research (AIR), Tsinghua University(人工智能产业研究院(AIR),清华大学)

AI总结 提出AMix-2,一种蛋白质-文本基础模型,通过统一蛋白质理解与序列设计,将蛋白质作为大语言模型的原生模态,并引入块状扩散语言建模骨干以更好地匹配蛋白质内在特性。

Comments 30 pages, 4 figures, 12 tables

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30904 2026-06-01 cs.CV

MergeTok: Unified Continuous and Discrete Visual Tokenization via Token Merging

MergeTok: 通过令牌合并实现统一连续和离散视觉令牌化

Luyuan Zhang, Siyuan Li, Zedong Wang, Qingsong Xie, Cheng Tan, Anna Wang, Yanhao Zhang, Chen Chen, Haonan Lu, Haoqian Wang

机构 * Tsinghua University(清华大学) Westlake University(西湖大学) Zhejiang University(浙江大学) Hong Kong University of Science and Technology(香港科学与技术大学) OPPO Shanghai AI Lab(上海人工智能实验室)

AI总结 提出MergeTok统一令牌化器,通过令牌合并技术联合优化连续VAE和离散VQ令牌化器,实现高保真重建与语义可控离散表示的兼顾。

Comments 11 pages (main text), 7 figures. Preprint. Under review at NeurIPS 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30846 2026-06-01 cs.CV

Count Anything

Count Anything

Mengqi Lei, Shuokun Cheng, Wei Bao, Shaoyi Du, Jun-Hai Yong, Siqi Li, Yue Gao

机构 * Tsinghua University(清华大学) China University of Geosciences, Wuhan(武汉地质大学) State Key Laboratory of Human-Machine Hybrid Augmented Intelligence(人机混合增强智能国家重点实验室) National Engineering Research Center for Visual Information and Applications(视觉信息与应用国家工程研究中心) Institute of Artificial Intelligence and Robotics(人工智能与机器人研究院)

AI总结 提出跨域文本引导的目标计数模型Count Anything,通过双粒度实例枚举和互补计数融合,在统一基准CLOC上实现多域泛化。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30611 2026-06-01 cs.CV cs.AI cs.CL

Crafter: A Multi-Agent Harness for Editable Scientific Figure Generation from Diverse Inputs

Crafter: 面向多样化输入的可编辑科学图表生成的多智能体框架

Haozhe Zhao, Shuzheng Si, Zhenhailong Wang, Zheng Wang, Liang Chen, Xiaotong Li, Zhixiang Liang, Maosong Sun, Minjia Zhang

机构 * University of Illinois at Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校) Tsinghua University(清华大学) Peking University(北京大学)

AI总结 提出Crafter多智能体框架,通过结构化组合离散语义组件,实现跨图表类型和输入条件的可编辑科学图表生成,并引入CraftEditor将栅格输出转换为可编辑SVG,在CraftBench基准上显著优于现有方法。

Comments 24 pages, 11 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30484 2026-06-01 cs.RO

ELAN4D: Embodiment-Centric 4D Supervision for Vision-Language-Action Models via Plug-and-Play Adaptation

ELAN4D:以具身为中心的4D监督用于视觉-语言-动作模型的即插即用适配

Zeyuan He, Bowen Yang, Zhirui Fang, Keru Zhou, Lei Jiang, Jingjing Qian, Fan Mo, Junchi Yan, Philip Torr, Xiu Li, Li Jiang, Jialin Yu

机构 * Torr Vision Group, University of Oxford(托尔视觉组,牛津大学) The Chinese University of Hong Kong, Shenzhen(香港大学(深圳)) Tsinghua University(清华大学) Shanghai Jiao Tong University(上海交通大学) University College London(伦敦大学学院) University of Cambridge(剑桥大学)

AI总结 提出ELAN4D框架,通过未来机器人关键点轨迹作为预测性时空监督,以即插即用方式增强VLA策略的鲁棒性和泛化能力。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30383 2026-06-01 cs.RO cs.AI

Structured interactions improve distributed coordination beyond model scaling in a real-world multi-robot system

结构化交互在真实世界多机器人系统中超越模型规模提升分布式协调能力

Junping Wang, Zhizhong Zhang, Yongqiang Tang, Geng Zheng, Jiaming Zhang, Shiji Song, Yanmei Li, Yushan Ma

机构 * State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences(多模态人工智能系统国家重点实验室,自动化研究所,中国科学院) School of Artificial Intelligence, University of Chinese Academy of Sciences(中国科学院大学人工智能学院) School of Computer Science and Technology, East China Normal University(华东师范大学计算机科学与技术学院) Department of Automation, Tsinghua University(清华大学自动化系) Liupanshan Laboratory, Ningxia University(宁夏大学鲁班实验室)

AI总结 通过真实多机器人实验,发现模块化层次化交互拓扑相比增加模型规模能更显著提升协调性能。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.29343 2026-06-01 cs.CL

Draft-OPD: On-Policy Distillation for Speculative Draft Models

Draft-OPD:用于推测草稿模型的在线策略蒸馏

Haodi Lei, Yafu Li, Haoran Zhang, Shunkai Zhang, Qianjia Cheng, Xiaoye Qu, Ganqu Cui, Bowen Zhou, Ning Ding, Yun Luo, Yu Cheng

机构 * Shanghai Jiao Tong University(上海交通大学) Shanghai AI Laboratory(上海人工智能实验室) Tsinghua University(清华大学) The Chinese University of Hong Kong(香港中文大学) Peking University(北京大学) Zhejiang University(浙江大学)

AI总结 针对推测解码中草稿模型因离线训练与在线推理不匹配导致性能瓶颈的问题,提出Draft-OPD方法,通过目标辅助展开和重放验证暴露的错误位置实现在线策略蒸馏,在多种任务上实现超过5倍的无损加速。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.25134 2026-06-01 cs.LG cs.AI

Theoretical Analysis of Sparse Optimization with Reparameterization, Weight Decay, and Adaptive Learning Rate

重参数化、权重衰减和自适应学习率下稀疏优化的理论分析

Huangyu Xu, Jingqin Yang, Qianqian Xu, Jiaye Teng

机构 * State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China(人工智能安全国家重点实验室,计算技术研究所,中国科学院,北京,中国) School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing, China(中国科学院大学计算机科学与技术学院,北京,中国) Beijing Academy of Artificial Intelligence (BAAI), Beijing, China(北京人工智能研究院(BAAI),北京,中国) IIIS, Tsinghua University, Beijing, China(清华大学人工智能院,北京,中国) School of Statistics and Management, Shanghai University of Finance and Economics, Shanghai, China(上海财经大学统计与管理学院,上海,中国) Institute of Data Science and Statistics, Shanghai University of Finance and Economics, Shanghai, China(上海财经大学数据科学与统计研究所,上海,中国)

AI总结 针对稀疏优化中的不稳定问题,提出基于重参数化、权重衰减和自适应学习率的ReWA方法,通过改善优化景观实现比ℓ1正则化更好的稀疏性,同时保持测试精度。

Comments 32 pages, 5 figures. Submitted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.22409 2026-06-01 cs.CV

SpaMEM: Benchmarking Dynamic Spatial Reasoning via Perception-Memory Integration in Embodied Environments

SpaMEM:具身环境中通过感知-记忆集成进行动态空间推理的基准测试

Chih-Ting Liao, Xi Xiao, Chunlei Meng, Zhangquan Chen, Yitong Qiao, Weilin Zhou, Tianyang Wang, Xu Zheng, Xin Cao

机构 * The University of New South Wales(新南威尔士大学) The University of Alabama at Birmingham(阿拉巴马大学伯明翰分校) Fudan University(复旦大学) Tsinghua University(清华大学) Zhejiang University(浙江大学) Xinjiang University(新疆大学) The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州))

AI总结 提出SpaMEM基准,通过动作条件场景变换和多模态数据,分层评估多模态大模型在具身环境中的空间信念演化能力,揭示坐标一致性和视觉记忆瓶颈。

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.20176 2026-06-01 q-bio.BM cs.LG

Cross-Chirality Generalization by Axial Vectors for Hetero-Chiral Protein-Peptide Interaction Design

通过轴向向量实现异手性蛋白质-肽相互作用设计的跨手性泛化

Ziyi Yang, Zitong Tian, Yinjun Jia, Tianyi Zhang, Jiqing Zheng, Hao Wang, Yubu Su, Juncai He, Lei Liu, Yanyan Lan

机构 * Department of Chemistry, Tsinghua University, Beijing, China(清华大学化学系) School of Life Sciences, Tsinghua University, Beijing, China(清华大学生命科学学院) Anew Labs, Shanghai, China(Anew实验室) Tsinghua-Peking Center for Life Sciences, Beijing, China(清华大学-北京大学生命科学中心) Ministry of Education Key Laboratory of Bioorganic Phosphorus Chemistry and Chemical Biology, Tsinghua University, Beijing, China(教育部生物有机磷化学与化学生物学重点实验室) Center for Synthetic and Systems Biology, Tsinghua University, Beijing, China(合成与系统生物学中心) Beijing Frontier Research Center for Biological Structure, Tsinghua University, Beijing, China(北京生物结构前沿研究中心) Qiuzhen College, Tsinghua University, Beijing, China(齐臻学院) Yau Mathematical Sciences Center, Tsinghua University, Beijing, China(叶德平数学科学中心) Institute for AI Industry Research (AIR), Tsinghua University, Beijing, China(人工智能产业研究院(AIR),清华大学) AI Industry Research Innovation Center,Wuxi Research Institute for Applied Technologies, Tsinghua University(人工智能产业研究创新中心,无锡应用技术研究院,清华大学)

AI总结 提出向E(3)等变(极)向量特征注入轴向特征的方法,结合潜在扩散模型实现从同手性训练数据到异手性设计任务的跨手性泛化,首次通过湿实验验证了生成式AI从头设计D-肽结合物的有效性。

Comments v3: Revised acknowledgements only. The paper has been accepted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.21190 2026-06-01 cs.LG cs.AI

Towards Foundation Models for Zero-Shot Time Series Anomaly Detection: Leveraging Synthetic Data and Relative Context Discrepancy

面向零样本时间序列异常检测的基础模型:利用合成数据和相对上下文差异

Tian Lan, Hao Duong Le, Jinbo Li, Wenjun He, Meng Wang, Chenghao Liu, Chen Zhang

机构 * Department of Industrial Engineering, Tsinghua University, Beijing, China(清华大学工业工程系) Datadog AI Research, Paris, France. This work was completed prior to joining Datadog(Datadog AI 研究院) Lab, Huawei Technologies, ShenZhen, China(华为技术2012实验室)

AI总结 提出基于相对上下文差异(RCD)的预训练范式,通过合成数据训练Transformer模型比较查询模式与上下文,实现零样本时间序列异常检测,在多个基准上超越现有基础模型。

Comments This manuscript is withdrawn, as the authors intend to further extend and develop the work beyond its current scope

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.17373 2026-06-01 cs.LG cs.AI

FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics

FML-bench:从搜索动力学视角对AI研究代理策略的受控研究

Qiran Zou, Hou Hei Lam, Wenhao Zhao, Tingting Chen, Yiming Tang, Samson Yu, Yingtao Zhu, Srinivas Anumasa, Zufeng Zhang, Tianyi Zhang, Chang Liu, Zhengyao Jiang, Anirudh Goyal, Dianbo Liu

机构 * National University of Singapore(国立新加坡大学) Tsinghua University(清华大学) University of Minnesota(明尼苏达大学) Weco Meta

AI总结 本文提出FML-Bench基准,通过分离策略与基础设施并定义过程级指标,评估六种代理策略,发现贪婪爬山法接近最优树搜索,且自适应策略基于搜索密度切换可超越其他代理。

Comments Our benchmark is available at: https://github.com/qrzou/FML-bench

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.00747 2026-06-01 cs.CL cs.AI

Decouple Searching from Training: Scaling Data Mixing via Model Merging for Large Language Model Pre-training

将搜索与训练解耦:通过模型合并实现大规模语言模型预训练的数据混合缩放

Shengrui Li, Fei Zhao, Kaiyan Zhao, Jieying Ye, Haifeng Liu, Fangcheng Shi, Zheyong Xie, Yao Hu, Shaosheng Cao

机构 * NLP Team, Xiaohongshu Inc., Shanghai, China(小红书自然语言处理团队,小红书公司,上海,中国) Tsinghua University, Beijing, China(清华大学,北京,中国) The University of Tokyo, Tokyo, Japan(东京大学,东京,日本)

AI总结 提出DeMix框架,通过模型合并预测最优数据配比,在降低搜索成本的同时提升基准性能。

Comments 18 pages, 5 figures, accepted at ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.18587 2026-06-01 cs.LG cs.AI cs.LO cs.PL

Compile to Compress: Boosting Formal Theorem Provers by Compiler Outputs

编译以压缩:通过编译器输出提升形式定理证明器

Guchan Li, Rui Tian, Hongning Wang

机构 * Department of Computer Science and Technology, Tsinghua University, Beijing, China(清华大学计算机科学与技术系)

AI总结 利用编译器将大量证明尝试压缩为结构化失败模式,提出一种学习-精炼框架,通过树搜索基于验证器反馈局部修正错误,在可比测试时预算下在PutnamBench上达到最先进性能。

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.16278 2026-06-01 cs.AI cs.CL cs.LG

Learning to Reason with Insight for Informal Theorem Proving

学习在非形式定理证明中进行洞察推理

Yunhe Li, Hao Shi, Bowen Deng, Wei Wang, Mengzhe Ruan, Hanxu Hou, Zhongxiang Dai, Siyang Gao, Chao Wang, Shuang Qiu, Linqi Song

机构 * City University of Hong Kong(香港城市大学) Tsinghua University(清华大学) Ke Holdings Inc.(Ke控股公司) Shenzhen University of Advanced Technology(深圳先进技术大学) Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳))

AI总结 针对非形式定理证明中缺乏洞察(识别核心技巧)的瓶颈,提出统一训练框架DeepInsight,通过分层数据集、渐进式多阶段SFT和基于洞察的策略优化方法,显著提升大语言模型的数学推理能力。

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.07751 2026-06-01 cs.CV cs.CL

3ViewSense: Spatial and Mental Perspective Reasoning from Orthographic Views in Vision-Language Models

3ViewSense: 视觉-语言模型中基于正交视图的空间与心理视角推理

Shaoxiong Zhan, Yanlin Lai, Zheng Liu, Hai Lin, Shen Li, Xiaodong Cai, Zijian Lin, Wen Huang, Hai-Tao Zheng

机构 * Shenzhen International Graduate School, Tsinghua University, Shenzhen, China(清华大学深圳国际研究生院) School of Software Engineering, Chongqing University, Chongqing, China(重庆大学软件学院)

AI总结 提出3ViewSense框架,通过正交视图的“模拟-推理”机制解决视觉-语言模型在空间推理中的视角一致性问题,显著提升遮挡计数和空间推理性能。

Comments Accepted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.07864 2026-06-01 cs.CV

Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces

在结构中思考:评估约束空间中的空间智能

Chen Yang, Guanxin Lin, Youquan He, Peiyao Chen, Guanghe Liu, Yufan Mo, Zhouyuan Xu, Linhao Wang, Guohui Zhang, Zihang Zhang, Shenxiang Zeng, Chen Wang, Jiansheng Fan

机构 * Tsinghua University(清华大学)

AI总结 提出SSI-Bench基准,通过结构约束下的空间推理任务评估视觉语言模型的空间智能,发现模型与人类存在巨大差距。

Comments ICML 2026, Project Page: https://ssi-bench.github.io

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.11683 2026-06-01 cs.LG cs.AI cs.CL

Boundary-Guided Policy Optimization for Memory-efficient RL of Diffusion Large Language Models

边界引导策略优化:面向扩散大语言模型的内存高效强化学习

Nianyi Lin, Jiajie Zhang, Lei Hou, Juanzi Li

机构 * Tsinghua University(清华大学)

AI总结 针对扩散大语言模型中似然函数难以处理导致强化学习内存开销大的问题,提出边界引导策略优化(BGPO),通过构造满足线性和等价性的下界实现内存高效训练,在数学求解、代码生成和规划任务中显著优于现有方法。

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.20795 2026-06-01 cs.RO

Learning Generalizable Robot Policy with Human Demonstration Video as a Prompt

以人类演示视频为提示学习可泛化的机器人策略

Xiang Zhu, Yichen Liu, Hezhong Li, Jianyu Chen

机构 * Tsinghua University, China(清华大学,中国) Shanghai Qi Zhi Institute, China(上海启智研究院,中国)

AI总结 提出两阶段框架,利用人类演示视频学习可泛化机器人策略,无需遥操作数据或微调即可执行新任务。

Comments Accepted to the IEEE International Conference on Robotics and Automation (ICRA), 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30247 2026-05-29 cs.LG cs.MM

OOD-GraphLLM: Graph Large Language Model for Out-of-Distribution Generalized Drug Synergy Prediction

OOD-GraphLLM:面向分布外泛化的药物协同预测图大语言模型

Xin Wang, Linxin Xiao, Yang Yao, Wenwu Zhu

机构 * DCST, BNRist, Tsinghua University(国防科技大学、北京理工大学、清华大学) DCST, Tsinghua University(国防科技大学、清华大学)

AI总结 针对药物协同预测中因新化合物导致的分布外偏移问题,提出OOD-GraphLLM框架,通过联合优化分子图表示与生物医学语义语言表示实现准确预测。

Comments 12 pages, 9 figures, ACM KDD 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30094 2026-05-29 cs.AI cs.GT

PokerSkill: LLMs Can Play Expert-Level Poker without Training or Solvers

PokerSkill: 无需训练或求解器,大语言模型可达到专家级扑克水平

Boning Li, Baoxiang Wang, Longbo Huang

机构 * IIIS, Tsinghua University(清华大学人工智能研究院) The Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳))

AI总结 提出PokerSkill框架,通过规则驱动的技能库约束大语言模型动作,无需训练或求解器即可在扑克中达到接近GTO水平的性能。

Comments 45 pages, 3 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.30065 2026-05-29 cs.CV

Boosting Zero-Shot 3D Style Transfer with 2D Pre-trained Priors

利用二维预训练先验提升零样本三维风格迁移

Xin Dong, Yunzhi Teng, Wenfeng Deng, Yansong Tang

机构 * Shenzhen International Graduate School, Tsinghua University(清华大学深圳国际研究生院) Pengcheng Laboratory(鹏城实验室)

AI总结 提出Data-Sufficient StyleGaussian模型,通过集成大规模2D图像数据集预训练的解码器,结合特征高斯溅射与延迟风格化,在数据稀缺条件下实现零样本3D风格迁移的高质量多视图一致渲染。

Comments Accepted by IEEE IVMSP2026

详情

展开后加载摘要…

URL PDF HTML 收藏