arXivDaily arXiv每日学术速递 周一至周五更新

高校专区

Shanghai Jiao Tong University(上海交通大学)

2026-07-27 至 2026-07-27 共收录 5
2605.04624 2026-07-27 cs.AI cs.SE 版本更新

AuditRepairBench: A Paired-Execution Trace Corpus for Evaluator-Channel Ranking Instability in Agent Repair

AuditRepairBench:用于评估智能体修复中评估器通道排名不稳定性的配对执行跟踪语料库

Yuelin Hu, Zhenbo Yu, Zhengxue Cheng, Wei Liu, Li Song

机构 * Shanghai Jiao Tong University(上海交通大学) Shanghai Maritime University(上海 Maritime University)

AI总结 研究智能体修复排行榜因评估器重新配置而重新排序的问题,通过AuditRepairBench语料库及模块化筛选架构实现评估器通道阻塞排名不稳定性评估,经多种验证和实验,证明筛选引导盲补丁能有效减少排名位移。

Comments The authors have identified substantial issues in the experimental design and evaluation that affect the validity of the main conclusions. Therefore, the authors have decided to withdraw this manuscript

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.22494 2026-07-27 cs.MM cs.CV 新提交

CARA: Concept-Aware Risk Attention for Interpretable Collision Anticipation

CARA:用于可解释碰撞预测的概念感知风险注意力

Zhishan Tao, Ruoyu Wang, Yucheng Wu, Enjun Du, Yilei Yuan, Sherwin Ho, Yue Su, Jinbo Su, Yi Hong

机构 * Shanghai Jiao Tong University(上海交通大学) The University of Hong Kong(香港大学) Fudan University(复旦大学) The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)) University of Pennsylvania(宾夕法尼亚大学) Renmin University of China(中国人民大学)

AI总结 研究自动驾驶碰撞预测问题,提出CARA框架,从事故叙述中获取风险概念并与视频帧对齐成轨迹,将语义风险因素作为动态证据,结合可解释性与预测过程,实验证明其能提高预测准确性和预警及时性。

Comments Accepted to ACM Multimedia 2026(Oral)

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.10135 2026-07-27 cs.CV cs.AI 版本更新

BiWM: Advancing Open-Source Interactive Video World Models with Bidirectional Autoregression

BiWM:利用双向自回归推进开源交互式视频世界模型

Shaohao Rui, Xiaofeng Mao, Zhanyu Zhang, Peijia Lin, Yansong Zhu, Yibo Zhang, Haibin Wan, Zhangrui Zhao, Weijie Ma

机构 * LynnReal AI Shanghai Innovation Institute(上海创新研究院) Shanghai Jiao Tong University(上海交通大学) Fudan University(复旦大学)

AI总结 提出BiWM框架,通过双向自回归范式将预训练视频骨干转化为交互式世界模型,仅需两阶段训练(微调+分布匹配蒸馏),支持多尺度模型和长程生成,优于现有因果流水线。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.17770 2026-07-27 cs.AI cs.CL 版本更新

Entropy-Gradient Inversion: Moving Toward Internal Mechanism of Large Reasoning Models

熵梯度反转:迈向大型推理模型的内部机制

Junyao Yang, Chen Qian, Kun Wang, Linfeng Zhang, Quanshi Zhang, Yong Liu, Dongrui Liu

机构 * National University of Singapore(新加坡国立大学) Renmin University of China(中国人民大学) Shanghai Jiao Tong University(上海交通大学) Nanyang Technological University(南洋理工大学)

AI总结 本文发现大型推理模型中令牌熵与logit梯度之间的稳健负相关(熵梯度反转),并提出相关性正则化组策略优化(CorR-PO)将其嵌入强化学习奖励正则化,从而提升推理性能。

Comments 15 pages, 5 figures, 8 tables

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.22407 2026-07-27 cs.LG cs.AI 版本更新

Hidden Failure Modes of Gradient Modification under Adam in Continual Learning, and Adaptive Decoupled Moment Routing as a Repair

Adam在持续学习中梯度修改的隐藏失败模式,以及自适应解耦动量路由作为修复

Yuelin Hu, Zhenbo Yu, Zhengxue Cheng, Wei Liu, Li Song

机构 * Shanghai Jiao Tong University(上海交通大学) Shanghai Maritime University(上海海洋大学)

AI总结 研究揭示了Adam在持续学习中梯度修改的隐藏失败模式,提出自适应解耦动量路由作为修复方法,通过调整动量路径提升模型稳定性与性能。

Comments The authors have identified substantial issues in the experimental design and evaluation that affect the validity of the main conclusions. Therefore, the authors have decided to withdraw this manuscript

详情

展开后加载摘要…

URL PDF HTML 收藏