arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~

高校专区

University of Chinese Academy of Sciences(中国科学院大学)

2026-04-01 至 2026-04-01 共收录 4
2603.29557 2026-04-01 cs.AI cs.CL

FlowPIE: Test-Time Scientific Idea Evolution with Flow-Guided Literature Exploration

FlowPIE: 测试时的科学思想演变与流引导文献探索

Qiyao Wang, Hongbo Wang, Longze Chen, Zhihao Yang, Guhong Chen, Hamid Alinejad-Rokny, Hui Li, Yuan Lin, Min Yang

机构 * University of Chinese Academy of Sciences(中国科学院大学) Dalian University of Technology(大连理工大学) UNSW Sydney(新南威尔士大学悉尼分校) Shenzhen University of Advanced Technology(深圳理工大学) Xiamen University(厦门大学)

AI总结 本文提出FlowPIE框架,通过流引导的蒙特卡洛树搜索扩展文献轨迹,结合LLM生成奖励模型指导适应性检索,生成高质量且多样化的初始种群,进而通过选择、交叉和变异实现测试时的科学思想演变。

Comments 30 pages, 11 figures, 15 tables

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.01639 2026-04-01 cs.CV

ReCALL: Recalibrating Capability Degradation for MLLM-based Composed Image Retrieval

ReCALL: 为基于MLLM的组合图像检索 recalibrate 能力退化

Tianyu Yang, Chenwei He, Xiangzhao Hao, Tianyue Wang, Jiarui Guo, Haiyun Guo, Leigang Qu, Jinqiao Wang, Tat-Seng Chua

机构 * Foundation Model Research Center, Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所基础模型研究中心) School of Artificial Intelligence, University of Chinese Academy of Sciences(中国科学院大学人工智能学院) Southeast University(东南大学) Beijing University of Posts and Telecommunications(北京邮电大学) National University of Singapore(新加坡国立大学) Wuhan AI Research(武汉人工智能研究院) Guangdong Provincial Key Laboratory of Intellectual Property and Big Data, Guangdong Polytechnic Normal University(广东技术师范大学广东省知识产权大数据重点实验室)

AI总结 ReCALL通过诊断生成器盲点、生成修正指令和三元组、并持续训练来缓解基于生成式MLLM的检索能力退化问题,实验证明其在CIRR和FashionIQ上达到SOTA性能。

Comments Accepted to CVPR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.29262 2026-04-01 cs.AI

Grokking From Abstraction to Intelligence

从抽象到智能的领悟

Junjie Zhang, Zhen Shen, Gang Xiong, Xisong Dong

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

AI总结 研究通过模运算中的领悟现象,揭示模型泛化机制源于内部结构的自发简化,结合因果、谱和算法复杂性及奇异学习理论,提出冗余流形的物理坍缩是泛化关键。

Comments 22page and 5 figures,In this paper, we analyze the grokking phenomenon from the perspective of Singular Learning Theory (SLT). This work is currently under review for ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.29115 2026-04-01 astro-ph.GA cs.CV

Schrödinger's Seed: Purr-fect Initialization for an Impurr-fect Universe

薛定谔的种子:为一个不完美的宇宙寻找完美的初始化

Mi chen, Renhao Ye

机构 * Kapteyn Astronomical Institute, University of Groningen(格罗宁根大学卡普坦天文研究所) School of Astronomy and Space Science, University of Chinese Academy of Sciences(中国科学院大学天文与空间科学学院) Shanghai Astronomical Observatory, Chinese Academy of Sciences(中国科学院上海天文台)

AI总结 本文提出利用猫的特性生成随机种子,通过蒙特卡洛方法测试21只家猫的物理属性,结果表明猫驱动的种子在准确性上优于传统随机整数。

Comments 3 pages, 1 figure, 21 cats

详情

展开后加载摘要…

URL PDF HTML 收藏