AAAI

TermGPT: Multi-Level Contrastive Fine-Tuning for Terminology Adaptation in Legal and Financial Domain

TermGPT：面向法律和金融领域术语适应的多级对比微调

Yidan Sun, Mengying Zhu, Feiyue Chen, Yangyang Wu, Xiaolei Dan, Mengyuan Yang, Xiaolin Zheng, Shenglin Ben

AI总结 TermGPT通过多级对比微调提升法律和金融领域术语区分能力，构建首个金融术语数据集并优于现有基线。

Comments 13 pages, 4 figures, AAAI'26

Verb Mirage: Unveiling and Assessing Verb Concept Hallucinations in Multimodal Large Language Models

动词幻觉：揭示和评估多模态大语言模型中的动词概念幻觉

Zehao Wang, Xinpeng Liu, Yudonglin Zhang, Xiaoqian Wu, Zhou Fang, Yifan Fang, Junfu Pu, Cewu Lu, Yong-Lu Li

AI总结 本文首次揭示多模态大语言模型中的动词概念幻觉问题，并提出基于丰富动词知识的调优方法以有效缓解该问题。

Comments Accepted by AAAI-26

Assessing Automated Fact-Checking for Medical LLM Responses with Knowledge Graphs

利用知识图谱评估医疗大语言模型响应的自动事实核查

Shasha Zhou, Mingyu Huang, Jack Cole, Charles Britton, Ming Yin, Jan Wolber, Ke Li

AI总结 本文提出FAITH框架，利用医学知识图谱评估医疗大语言模型响应的事实性，证明其在医疗领域具有较高的可靠性与实用性。

Comments Accepted as a conference paper at AAAI'26

Logical Characterizations of GNNs with Mean Aggregation

具有均值聚合的GNNs的逻辑刻画

Moritz Schönherr, Carsten Lutz

AI总结 本文研究了具有均值聚合的GNNs的表达能力，发现其在非均匀和均匀设置下分别与比值模态逻辑和无交替模态逻辑等价，且在实践中受连续性和阈值假设限制。

Comments 26 pages, extended version of paper to appear in AAAI 2026

Beyond Semantic Features: Pixel-level Mapping for Generalized AI-Generated Image Detection

超越语义特征：用于通用AI生成图像检测的像素级映射

Chenming Zhou, Jiaan Wang, Yu Li, Lei Li, Juan Cao, Sheng Tang

AI总结 本文提出通过像素级映射预处理破坏图像语义特征，提升AI生成图像检测器的跨模型泛化能力。

Comments Accepted by AAAI 2026

A Solver-in-the-Loop Framework for Improving LLMs on Answer Set Programming for Logic Puzzle Solving

一种用于改进逻辑谜题求解的LLM在Answer Set Programming中的求解器循环框架

Timo Pierre Schrader, Lukas Lange, Tobias Kaminski, Simon Razniewski, Annemarie Friedrich

AI总结 本文提出了一种求解器循环框架，通过引导求解器进行指令微调，提升LLM在Answer Set Programming中生成逻辑谜题解答的性能。

Comments 15 pages, 7 figures, accepted at AAAI'26

Realist and Pluralist Conceptions of Intelligence and Their Implications on AI Research

现实主义与多元主义对智能的观念及其对人工智能研究的影响

Ninell Oldenburg, Ruchira Dhar, Anders Søgaard

AI总结 本文探讨了智能的现实主义与多元主义观念对人工智能研究的影响，指出不同观念导致方法、解释和风险评估的根本差异。

Comments The 40th Annual AAAI Conference on Artificial Intelligence, 8 pages (excl. references), 1 table

CD-DPE: Dual-Prompt Expert Network Based on Convolutional Dictionary Feature Decoupling for Multi-Contrast MRI Super-Resolution

CD-DPE：基于卷积字典特征解耦的双提示专家网络用于多对比MRI超分辨率

Xianming Gu, Lihui Wang, Ying Cao, Zeyu Deng, Yingfeng Ou, Guodong Hu, Yi Chen

AI总结 CD-DPE通过双提示专家网络和卷积字典特征解耦策略，提升多对比MRI超分辨率的重建质量与泛化能力。

Comments Accepted to AAAI-2026

Phantom Menace: Exploring and Enhancing the Robustness of VLA Models Against Physical Sensor Attacks

幻影威胁：探索和增强VLA模型对物理传感器攻击的鲁棒性

Xuancun Lu, Jiaxiang Chen, Shilin Xiao, Zizhi Jin, Zhangrui Chen, Hanwen Yu, Bohan Qian, Ruochen Zhou, Xiaoyu Ji, Wenyuan Xu

AI总结 本文提出了一种新型框架，用于研究和增强VLA模型对物理传感器攻击的鲁棒性，并开发了对抗训练方法以提升模型安全性。

Comments Accepted by AAAI 2026 main track

OptScale: Probabilistic Optimality for Inference-time Scaling

OptScale: 推断时间缩放的概率最优性

Youkang Wang, Jian Wang, Rubing Chen, Xiao-Yong Wei

AI总结 OptScale通过概率框架和动态样本决策，提升LLMs推理性能并减少采样开销。

Comments Accepted by AAAI-2026

On the Edge of Core (Non-)Emptiness: An Automated Reasoning Approach to Approval-Based Multi-Winner Voting

在核心（非）空集边缘：一种基于自动推理的批准型多胜者投票方法

Ratip Emin Berker, Emanuel Tewolde, Vincent Conitzer, Mingyu Guo, Marijn Heule, Lirong Xia

AI总结 本文提出了一种基于混合整数线性规划的方法，用于确定多胜者投票中核心稳定委员会的存在性，并揭示了核心稳定性与其他属性之间的关系。

Comments 29 pages, 1 figure, main body to be published in Proceedings of the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), Singapore, 2026

KineST: A Kinematics-guided Spatiotemporal State Space Model for Human Motion Tracking from Sparse Signals

KineST: 一种基于运动学的时空状态空间模型用于从稀疏信号中的人体动作跟踪

Shuting Zhao, Zeyu Xiao, Xinrong Chen

AI总结 KineST通过运动学引导的状态空间模型，有效提取时空依赖性，提升稀疏信号下的人体动作跟踪精度与时间一致性。

Comments Accepted by AAAI 2026

Causal-Tune: Mining Causal Factors from Vision Foundation Models for Domain Generalized Semantic Segmentation

Causal-Tune: 从视觉基础模型中挖掘因果因素以实现领域通用的语义分割

Yin Zhang, Yongqiang Zhang, Yaoyue Zheng, Bogdan Raducanu, Dan Liu

AI总结 Causal-Tune通过频谱分析和因果因素分离提升领域通用语义分割的鲁棒性

Comments Accepted by AAAI 2026

Hierarchical Semantic Alignment for Image Clustering

层次化语义对齐用于图像聚类

Xingyu Zhu, Beier Zhu, Yunfan Li, Junfeng Fang, Shuo Wang, Kesen Zhao, Hanwang Zhang

机构 * AAAI Press Staff（AAAI出版社员工） 纠错

AI总结 CAE通过层次化语义对齐方法提升图像聚类性能，无训练即可实现4.2%的准确率提升。

Comments AAAI 2026

ManiLong-Shot: Interaction-Aware One-Shot Imitation Learning for Long-Horizon Manipulation

ManiLong-Shot：交互感知的一次学习方法用于长周期操控

Zixuan Chen, Chongkai Gao, Lin Shao, Jieqi Shi, Jing Huo, Yang Gao

机构 * National University of Singapore（新加坡国立大学） 纠错

AI总结 ManiLong-Shot通过交互感知原语分解实现长周期操控任务的一次学习，有效提升机器人任务执行能力。

Comments Accepted by AAAI 2026

Tuning for Two Adversaries: Enhancing the Robustness Against Transfer and Query-Based Attacks using Hyperparameter Tuning

针对两个对抗者的调优：通过超参数调优增强对转移和查询攻击的鲁棒性

Pascal Zimmer, Ghassan Karame

AI总结 本文通过超参数调优提升模型对转移和查询攻击的鲁棒性，发现学习率调整对不同攻击类型有不同影响，分布式模型表现最优。

Comments To appear in the Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) 2026

PerTouch: VLM-Driven Agent for Personalized and Semantic Image Retouching

PerTouch：基于VLM的个性化和语义图像润色代理

Zewei Chang, Zheng-Peng Duan, Jianxing Zhang, Chun-Le Guo, Siyu Liu, Hyungju Chun, Hyunhee Park, Zikun Liu, Chongyi Li

AI总结 PerTouch通过VLM驱动代理实现个性化和语义图像润色，结合语义替换与参数扰动机制，提升用户意图匹配与长期偏好捕捉能力。

Comments To appear at AAAI 2026

MAVIS: A Benchmark for Multimodal Source Attribution in Long-form Visual Question Answering

MAVIS：多模态来源归因的长形式视觉问答基准

Seokwon Song, Minsu Park, Gunhee Kim

AI总结 MAVIS基准旨在评估多模态来源归因系统，通过多模态文档检索和长形式答案生成，揭示多模态设置下信息量、 groundedness 和流畅性之间的权衡与改进方向。

Comments AAAI 2026; code is available at https://github.com/seokwon99/MAVIS

Alternative Fairness and Accuracy Optimization in Criminal Justice

刑事司法中的替代公平与准确性优化

Shaolong Wu, James Blume, Geshi Yeung

机构 * Harvard University（哈佛大学） ； Massachusetts Institute of Technology（麻省理工学院） 纠错

AI总结 本文提出在刑事司法中通过优化误差损失和控制假阴性率差异来提升公平与准确性的方法，并提供部署框架以确保决策系统的合法性。

Comments AAAI 2026, Singapore (AIGOV)

Journal ref AAAI 2026, Singapore (AIGOV)

LaF-GRPO: In-Situ Navigation Instruction Generation for the Visually Impaired via GRPO with LLM-as-Follower Reward

LaF-GRPO：通过GRPO与LLM-as-Follower生成视障人士的现场导航指令

Yi Zhao, Siqi Wang, Jing Li

AI总结 LaF-GRPO通过GRPO与LLM-as-Follower生成更直观安全的视障人士现场导航指令，提升指令准确性和实用性。

Comments Accepted at AAAI-26

Behavior Tokens Speak Louder: Disentangled Explainable Recommendation with Behavior Vocabulary

行为令牌发声更大：解耦可解释推荐与行为词汇

Xinshun Feng, Mingzhe Liu, Yi Qiao, Tongyu Zhu, Leilei Sun, Shuai Wang

AI总结 BEAT通过解耦行为词汇与语义，提升推荐系统的可解释性和零样本性能。

Comments accepted by AAAI 2026

CLIP-FTI: Fine-Grained Face Template Inversion via CLIP-Driven Attribute Conditioning

CLIP-FTI: 通过CLIP驱动的属性条件化实现细粒度面部模板逆向

Longchen Dai, Zixuan Shen, Zhiheng Zhou, Peipeng Yu, Zhihua Xia

AI总结 CLIP-FTI通过CLIP驱动的属性条件化实现细粒度面部模板逆向，提升识别准确性和属性相似性，增强跨模型攻击的可转移性。

Comments Accepted by AAAI 2026

RPM-MCTS: Knowledge-Retrieval as Process Reward Model with Monte Carlo Tree Search for Code Generation

RPM-MCTS：基于蒙特卡洛树搜索的知识检索作为过程奖励模型用于代码生成

Yuanyuan Lin, Xiangyu Ouyang, Teng Zhang, Kaixin Sui

AI总结 RPM-MCTS通过结合知识检索与蒙特卡洛树搜索，有效评估代码生成过程中的中间步骤，减少错误并提升效率

Comments Accepted at AAAI 2026

MovSemCL: Movement-Semantics Contrastive Learning for Trajectory Similarity (Extension)

MovSemCL：用于轨迹相似性的运动-语义对比学习（扩展）

Zhichen Lai, Hua Lu, Huan Li, Jialiang Li, Christian S. Jensen

AI总结 MovSemCL通过运动-语义对比学习框架提升轨迹相似性计算的精度与效率，有效解决现有方法在语义建模、计算成本和增强合理性方面的不足。

Comments 8 pages, 6 figures; accepted by AAAI 2026 as an Oral paper

Explainable AI for Classifying UTI Risk Groups Using a Real-World Linked EHR and Pathology Lab Dataset

利用真实世界链接的电子健康记录和病理实验室数据集进行UTI风险分组分类的可解释AI

Yujie Dai, Brian Sullivan, Axel Montout, Amy Dillon, Chris Waller, Peter Acs, Rachel Denholm, Philip Williams, Alastair D Hay, Raul Santos-Rodriguez, Andrew Dowsey

机构 * University of Bristol（布里斯托大学） ； Kettering General Hospital NHS Foundation Trust（凯特林医院国家健康服务基金会信托） ； NHS England（英格兰国家健康服务） ； University Hospitals Bristol and Weston NHS Foundation Trust（布里斯托和韦斯特医院国家健康服务基金会信托） 纠错

AI总结 本研究利用真实世界链接的EHR和病理实验室数据集，通过可解释AI技术对UTI风险分组进行分类，以提高临床决策的透明性和可解释性。

Comments Peer-reviewed; accepted at Health Intelligence (W3PHIAI-25) Workshop, AAAI Conference 2025 (to appear in Studies in Computational Intelligence, Springer/Nature)

Well Begun, Half Done: Reinforcement Learning with Prefix Optimization for LLM Reasoning

开篇即胜，半途而废：基于前缀优化的强化学习用于大语言模型推理

Yiliu Sun, Zicheng Zhao, Yang Wei, Yanfang Zhang, Chen Gong

AI总结 PPPO通过优化LLM推理的前缀部分，提升推理能力，实验显示其在推理任务中表现更优。

Comments Accepted by AAAI 2026

RFKG-CoT: Relation-Driven Adaptive Hop-count Selection and Few-Shot Path Guidance for Knowledge-Aware QA

RFKG-CoT: 基于关系的自适应步数选择与少样本路径引导用于知识感知问答

Chao Zhang, Minghan Li, Tianrui Lv, Guodong Zhou

机构 * Chao Zhang, Minghan Li, Tianrui Lv, Guodong Zhou（张超，李明翰，吕天睿，周国栋） 纠错

AI总结 RFKG-CoT通过关系驱动的自适应步数选择和少样本路径引导，提升知识感知问答的准确性与可靠性。

Comments 9pages, 5 figures, accepted by AAAI 2026

Feature-Centric Unsupervised Node Representation Learning Without Homophily Assumption

以特征为中心的无监督节点表示学习无需同质性假设

Sunwoo Kim, Soo Yong Lee, Kyungho Kim, Hyunjin Hwang, Jaemin Yoo, Kijung Shin

AI总结 FUEL通过自适应学习图卷积的使用程度，提升嵌入空间中的类间分离性和类内相似性，从而在无监督节点表示学习中取得最佳性能。

Comments Published in AAAI 2026

Factor(U,T): Controlling Untrusted AI by Monitoring their Plans

Factor(U,T): 通过监控其计划来控制不可信的AI

Edward Lue Chee Lip, Anthony Channg, Diana Kim, Aaron Sandoval, Kevin Zhu

AI总结 Factor(U,T)通过监控AI分解计划来控制不可信AI，实验表明在仅观察自然语言指令时，监控恶意活动效果有限，而结合子任务监控则能实现高区分度和安全性。

Comments Accepted to AAAI 2026 Workshop on Trust and Control in Agentic AI (TrustAgent). 6 pages body, 8 pages total, 3 figures

SoMe: A Realistic Benchmark for LLM-based Social Media Agents

SoMe：一种用于基于大语言模型的社会媒体代理的现实基准

Dizhan Xue, Jing Cui, Shengsheng Qian, Chuanrui Hu, Changsheng Xu

AI总结 SoMe是一个用于评估基于大语言模型的社会媒体代理能力的现实基准，通过多样化任务和大量数据揭示了现有LLM在处理社交媒体任务中的局限性。

Comments Accepted by AAAI 2026