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~~

期刊&会议

AAAI Conference on Artificial Intelligence · 会议 · Artificial Intelligence

至 收录 41
2508.03177 2026-07-07 cs.CV 版本更新

SAVER: Mitigating Hallucinations in Large Vision-Language Models via Style-Aware Visual Early Revision

SAVER:通过风格感知视觉早期修正减轻大型视觉语言模型中的幻觉

Zhaoxu Li, Chenqi Kong, Yi Yu, Qiangqiang Wu, Xinghao Jiang, Ngai-Man Cheung, Bihan Wen, Alex Kot, Xudong Jiang

机构 * ROSE Lab, Interdisciplinary Graduate Programme, Nanyang Technological University, Singapore(南洋理工大学罗思实验室,跨学科研究生项目,新加坡) ROSE Lab, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore(南洋理工大学罗思实验室,电子与电气工程学院,新加坡) City University of Hong Kong, Hong Kong SAR(香港城市大学,香港特别行政区) Shanghai Jiao Tong University, China(上海交通大学,中国) Singapore University of Technology and Design, Singapore(新加坡科技设计大学,新加坡) VinUniversity, Hanoi, Vietnam(越南文大学,河内,越南)

AI总结 研究大型视觉语言模型幻觉问题,构建含照片及风格化图像数据集并对比,提出SAVER机制,利用早期层反馈基于视觉注意力模式动态调整输出,减轻风格化图像引起的幻觉,实验证明其性能先进。

Comments Accepted at AAAI 2026. 24 pages, 10 figures. Code: https://github.com/llizhaoxu/SAVER

URL PDF HTML 收藏
2511.10841 2026-07-02 cs.LG cs.AI 版本更新

FlowPath: Learning Data-Driven Manifolds with Invertible Flows for Robust Irregularly-sampled Time Series Classification

FlowPath: 通过可逆流学习数据驱动的流形以实现鲁棒的不规则采样时间序列分类

YongKyung Oh, Dong-Young Lim, Sungil Kim

机构 * Medical & Imaging Informatics (MII) Group, University of California, Los Angeles (UCLA), CA, USA(加州大学洛杉矶分校医学与影像信息学组) Department of Industrial Engineering, Ulsan National Institute of Science and Technology (UNIST), Republic of Korea(韩国蔚山科学技术院工业工程系) Artificial Intelligence Graduate School, Ulsan National Institute of Science and Technology (UNIST), Republic of Korea(韩国蔚山科学技术院人工智能研究生院)

AI总结 FlowPath通过可逆神经流学习控制路径的几何结构,提升不规则采样时间序列分类的鲁棒性,实验表明其在18个基准数据集和实际案例中均优于传统方法。

Comments Published at the 40th Annual AAAI Conference on Artificial Intelligence (AAAI 2026). https://ojs.aaai.org/index.php/AAAI/article/view/39643

URL PDF HTML 收藏
2507.05740 2026-07-02 cs.CL 版本更新

GPTKB v1.5: A Massive Knowledge Base for Exploring Factual LLM Knowledge

GPTKB v1.5:用于探索事实性LLM知识的大规模知识库

Yujia Hu, Tuan-Phong Nguyen, Shrestha Ghosh, Moritz Müller, Simon Razniewski

机构 * ScaDS.AI Dresden/Leipzig & TU Dresden, Germany(德国德累斯顿/莱比锡ScaDS.AI & 德累斯顿工业大学) VNU University of Engineering and Technology, Hanoi, Vietnam(越南河内国立大学工程与技术大学)

AI总结 提出GPTKB v1.5,一个通过GPT-4.1构建的1亿三元组知识库,支持链接遍历、SPARQL查询和LLM知识优劣比较,用于系统分析LLM知识。

Comments 3 pages, 1 figure, 1 table

Journal ref AAAI 2026: Demo track

URL PDF HTML 收藏
2511.11421 2026-06-25 cs.CV cs.LG 版本更新

BOFA: Bridge-Layer Orthogonal Low-Rank Fusion for CLIP-Based Class-Incremental Learning

BOFA: 基于CLIP的类增量学习中的桥接层正交低秩融合

Lan Li, Tao Hu, Da-Wei Zhou, Jia-Qi Yang, Han-Jia Ye, De-Chuan Zhan

AI总结 提出BOFA框架,通过将适应限制在CLIP的跨模态桥接层并使用正交低秩融合防止遗忘,无需额外参数或数据回放,实现高效类增量学习。

Comments Accepted by AAAI 2026

Journal ref Proceedings of the AAAI Conference on Artificial Intelligence, 40(27): 22967-22975, 2026

URL PDF HTML 收藏
2511.08378 2026-06-19 cs.IR cs.AI 版本更新

Bid Farewell to Seesaw: Towards Accurate Long-tail Session-based Recommendation via Dual Constraints of Hybrid Intents

告别跷跷板:通过混合意图的双重约束实现准确的长期会话推荐

Xiao Wang, Ke Qin, Dongyang Zhang, Xiurui Xie, Shuang Liang

机构 * University of Electronic Science and Technology of China(电子科技大学)

AI总结 针对会话推荐中长尾分布导致准确性与多样性冲突的跷跷板问题,提出混合意图双重约束框架HID,通过属性感知谱聚类重构意图映射并区分噪声意图,结合多样性与准确性约束损失,实现长尾与准确性的双赢。

Comments accepted by AAAI 2026 Oral

URL PDF HTML 收藏
2508.04266 2026-06-19 cs.CL 版本更新

ShoppingBench: A Real-World Intent-Grounded Shopping Benchmark for LLM-based Agents

ShoppingBench:面向LLM智能体的真实世界意图导向购物基准

Jiangyuan Wang, Kejun Xiao, Qi Sun, Huaipeng Zhao, Tao Luo, Jian Dong Zhang, Xiaoyi Zeng

机构 * Alibaba International Digital Commercial Group(阿里巴巴国际数字商业集团)

AI总结 提出ShoppingBench基准,包含多层级真实购物意图任务,通过模拟环境和250万商品评估LLM智能体,发现GPT-4.1成功率低于50%,并提出轨迹蒸馏策略提升小模型性能。

Comments Accepted for oral presentation at AAAI 2026

URL PDF HTML 收藏
2510.27568 2026-06-19 cs.AI cs.CL 版本更新

SIGMA: Search-Augmented On-Demand Knowledge Integration for Agentic Mathematical Reasoning

SIGMA: 搜索增强的按需知识集成用于智能体数学推理

Ali Asgarov, Umid Suleymanov, Aadyant Khatri

AI总结 提出SIGMA框架,通过多智能体独立推理、定向搜索和协调机制,实现上下文敏感的知识集成,在MATH500等基准上提升7.4%的绝对性能。

Comments AAAI 2026 LMReasoning

URL PDF HTML 收藏
2511.09882 2026-06-16 cs.GT cs.MA 版本更新

Truth, Justice, and Secrecy: Cake Cutting Under Privacy Constraints

真理、公正与保密:隐私约束下的蛋糕分割

Yaron Salman, Tamir Tassa, Omer Lev, Roie Zivan

AI总结 本文提出首个隐私保护的蛋糕分割协议,在保证无嫉妒和策略证明性的同时,通过密码学技术保护代理的偏好隐私。

Comments This is the full version of our paper published in the Proceedings of AAAI 2026

URL PDF HTML 收藏
2509.00135 2026-06-16 cs.AI 版本更新

Optimizing Health Coverage in Ethiopia: A Learning-augmented Approach and Persistent Proportionality Under an Online Budget

优化埃塞俄比亚的健康覆盖:一种学习增强的方法与在线预算下的持续比例性

Davin Choo, Yohai Trabelsi, Fentabil Getnet, Samson Warkaye Lamma, Wondesen Nigatu, Kasahun Sime, Lisa Matay, Milind Tambe, Stéphane Verguet

机构 * John A. Paulson School of Engineering and Applied Sciences(约翰·A·保罗森工程与应用科学学院) Harvard University(哈佛大学) National Data Management and Analytics Center for Health(健康国家数据管理与分析中心) Ethiopian Public Health Institute(埃塞俄比亚公共卫生研究所) Ministry of Health, Ethiopia(埃塞俄比亚卫生部) Department of Global Health and Population(全球卫生与人口部门) Harvard T.H. Chan School of Public Health(哈佛T.H. Chan公共卫生学院)

AI总结 针对埃塞俄比亚卫生系统强化中的预算不确定性和区域比例目标,提出基于学习增强和贪心算法的顺序设施规划框架,最大化人口覆盖。

Comments Published in the AISI track at AAAI 2026

URL PDF HTML 收藏
2601.11128 2026-06-11 cs.SI cs.HC cs.IR 版本更新

The Big Ban Theory: A Pre- and Post-Intervention Dataset of Online Content Moderation Actions

大封禁理论:在线内容审核行为的前后干预数据集

Aldo Cerulli, Lorenzo Cima, Benedetta Tessa, Serena Tardelli, Stefano Cresci

AI总结 针对在线平台审核干预研究缺乏综合数据集的问题,构建了包含Reddit和Voat上25种干预措施、超33.9万用户和近3900万条消息的数据集,提供标准化元数据和匿名化用户活动数据,支持干预效果的可比分析。

Comments Article published in ICWSM'26 - 20th AAAI Conference on Web and Social Media. Please, cite the published version

URL PDF HTML 收藏
2601.05675 2026-06-08 cs.AI 版本更新

CHDP: Cooperative Hybrid Diffusion Policies for Reinforcement Learning in Parameterized Action Space

CHDP:参数化动作空间中强化学习的协同混合扩散策略

Bingyi Liu, Jinbo He, Haiyong Shi, Enshu Wang, Weizhen Han, Jingxiang Hao, Peixi Wang, Zhuangzhuang Zhang

机构 * National University of Singapore(新加坡国立大学) University of Science and Technology of China(中国科学技术大学)

AI总结 针对混合动作空间中的策略表达力不足和高维扩展性差问题,提出协同混合扩散策略框架,通过离散和连续扩散策略的协作与顺序更新,结合码本嵌入和Q函数引导,在基准测试中成功率提升高达19.3%。

Comments Accepted by AAAI 2026

URL PDF HTML 收藏