AAAI

Self-Supervised Inductive Logic Programming

自监督归纳逻辑编程

Stassa Patsantzis

AI总结 Poker通过自监督学习方法，在无负例情况下自动生成正负例，提升归纳逻辑编程的性能。

Comments To appear in AAAI-26 conference proceedings. Updated 04 March 2026 to include appendices, update affiliation, contact details

Conversational Learning Diagnosis via Reasoning Multi-Turn Interactive Learning

通过推理的多轮交互学习诊断

Fangzhou Yao, Sheng Chang, Weibo Gao, Qi Liu

AI总结 本文提出ParLD框架，通过多智能体协作实现对话学习中的认知诊断，提升诊断的可靠性和洞察力。

Comments AAAI 2026

I-CAM-UV: Integrating Causal Graphs over Non-Identical Variable Sets Using Causal Additive Models with Unobserved Variables

I-CAM-UV: 在非相同变量集上整合因果图使用带有未观测变量的因果加法模型

Hirofumi Suzuki, Kentaro Kanamori, Takuya Takagi, Thong Pham, Takashi Nicholas Maeda, Shohei Shimizu

AI总结 I-CAM-UV通过整合带有未观测变量的因果加法模型，解决多数据集非相同变量集下的因果图整合问题。

Comments 16 pages, 22 figures, to appear in the 40th AAAI Conference on Artificial Intelligence (AAAI 2026)

LLandMark: A Multi-Agent Framework for Landmark-Aware Multimodal Interactive Video Retrieval

LLandMark：一种用于地标感知多模态交互视频检索的多智能体框架

Minh-Chi Phung, Thien-Bao Le, Cam-Tu Tran-Thi, Thu-Dieu Nguyen-Thi, Vu-Hung Dao

机构 * AI VIETNAM（AI越南） 纠错

AI总结 LLandMark通过多智能体框架实现地标感知的多模态视频检索，结合LLM和OCR技术提升文化场景下的检索能力与可解释性。

Comments Accepted by AAAI 2026 Workshop on New Frontiers in Information Retrieval

Journal ref AAAI 2026 Workshop on New Frontiers in Information Retrieval

MoToRec: Sparse-Regularized Multimodal Tokenization for Cold-Start Recommendation

MoToRec：基于稀疏正则化的多模态分词用于冷启动推荐

Jialin Liu, Zhaorui Zhang, Ray C. C. Cheung

AI总结 MoToRec通过稀疏正则化的多模态分词方法，解决冷启动推荐中的数据稀疏性和新物品表示问题，提升推荐系统性能。

Comments Accepted to AAAI 2026 (Main Track)

MedLA: A Logic-Driven Multi-Agent Framework for Complex Medical Reasoning with Large Language Models

MedLA：基于大语言模型的复杂医学推理多智能体框架

Siqi Ma, Jiajie Huang, Fan Zhang, Yue Shen, Jinlin Wu, Guohui Fan, Zhu Zhang, Zelin Zang

AI总结 MedLA是一种基于大语言模型的逻辑驱动多智能体框架，通过多轮图引导讨论实现透明推理和共识达成，有效提升复杂医学推理性能。

Comments accepted by AAAI-26 (ORAL)

SPARE: Single-Pass Annotation with Reference-Guided Evaluation for Automatic Process Supervision and Reward Modelling

SPARE：基于参考引导评估的单次标注用于自动过程监督与奖励建模

Md Imbesat Hassan Rizvi, Xiaodan Zhu, Iryna Gurevych

AI总结 SPARE通过单次生成与参考引导评估，实现高效过程标注，提升LLM多步推理中的奖励建模与微调效果。

Comments Accepted to AAAI 2026 (Oral)

SToLa: Self-Adaptive Touch-Language Framework with Tactile Commonsense Reasoning in Open-Ended Scenarios

SToLa：具有触觉常识推理的自适应触觉-语言框架在开放性场景中

Ning Cheng, Jinan Xu, Jialing Chen, Bin Fang, Wenjuan Han

AI总结 SToLa通过专家混合架构实现触觉与语言模态的自适应统一，解决开放性场景中触觉常识推理的挑战，提升多模态推理性能。

Comments Accepted by AAAI 2026

Constrained Particle Seeking: Solving Diffusion Inverse Problems with Just Forward Passes

受约束的粒子寻找：仅通过正向传递解决扩散逆问题

Hongkun Dou, Zike Chen, Zeyu Li, Hongjue Li, Lijun Yang, Yue Deng

AI总结 受约束的粒子寻找方法通过无梯度优化高效解决图像和科学逆问题，性能优于传统方法。

Comments Accepted by AAAI 2026

Voices, Faces, and Feelings: Multi-modal Emotion-Cognition Captioning for Mental Health Understanding

声音、面孔与情感：多模态情绪-认知描述用于心理健康理解

Zhiyuan Zhou, Yanrong Guo, Shijie Hao

AI总结 ECMC通过多模态数据生成情绪-认知描述，提升心理健康评估的准确性和可解释性。

Comments Accepted at AAAI 2026

ALTER: Asymmetric LoRA for Token-Entropy-Guided Unlearning of LLMs

ALTER: 用于LLM中基于令牌熵引导的不对称LoRA去学习

Xunlei Chen, Jinyu Guo, Yuang Li, Zhaokun Wang, Yi Gong, Jie Zou, Jiwei Wei, Wenhong Tian

AI总结 ALTER通过不对称LoRA架构实现基于令牌熵引导的LLM高效去学习，取得SOTA性能并保留高模型效用。

Comments Accepted at The 40th Annual AAAI Conference on Artificial Intelligence (AAAI 2026)

LexChronos: An Agentic Framework for Structured Event Timeline Extraction in Indian Jurisprudence

LexChronos：一个用于印度法理学中结构化事件时间线提取的代理框架

Anka Chandrahas Tummepalli, Preethu Rose Anish

AI总结 LexChronos通过双代理架构提取印度最高法院判决中的结构化事件时间线，提升法律文本理解和推理能力。

Comments Published in AILaw @ AAAI 2026 Conference

Journal ref AILaw @ AAAI 2026

Quantifying Conversational Reliability of Large Language Models under Multi-Turn Interaction

量化大型语言模型在多轮交互中的对话可靠性

Jiyoon Myung

AI总结 研究评估了大型语言模型在多轮对话中的可靠性，发现其在复杂交互中存在显著下降，揭示了指令漂移、意图混淆等失败模式，强调了对LLM进行压力测试和改进评估方法的重要性。

Comments Accepted at the Workshop on Assessing and Improving Reliability of Foundation Models in the Real World (AAAI 2026)

ReFeed: Retrieval Feedback-Guided Dataset Construction for Style-Aware Query Rewriting

ReFeed: 基于检索反馈的领域感知查询重写数据集构建

Jiyoon Myung, Jungki Son, Kyungro Lee, Jihyeon Park, Joohyung Han

AI总结 ReFeed通过检索反馈驱动数据集构建，提升查询重写模型对文档风格的感知能力，改进领域特定场景下的检索效果。

Comments Accepted at the Workshop on New Frontiers in Information Retrieval (AAAI 2026)

SubstratumGraphEnv: Reinforcement Learning Environment (RLE) for Modeling System Attack Paths

SubstratumGraphEnv：用于系统攻击路径建模的强化学习环境（RLE）

Bahirah Adewunmi, Edward Raff, Sanjay Purushotham

AI总结 SubstratumGraphEnv通过图卷积网络建模系统事件序列，自动化识别攻击路径，为网络安全分析提供深度强化学习环境。

Comments Presented at the AI for Cyber Security Workshop at AAAI-26

Gender Bias in Emotion Recognition by Large Language Models

大语言模型在情绪认知中的性别偏见

Maureen Herbert, Katie Sun, Angelica Lim, Yasaman Etesam

AI总结 本研究探讨了大语言模型在情绪认知任务中的性别偏见问题，并提出基于训练的干预方法以减少偏见。

Comments Accepted at AAAI 2026 Workshop (WS37)

HierarchicalPrune: Position-Aware Compression for Large-Scale Diffusion Models

分层剪枝：面向大规模扩散模型的位置感知压缩

Young D. Kwon, Rui Li, Sijia Li, Da Li, Sourav Bhattacharya, Stylianos I. Venieris

AI总结 HierarchicalPrune通过分层位置剪枝、位置权重保护和敏感度引导的知识蒸馏，实现大规模扩散模型的高效压缩，减少内存占用和延迟，同时保持图像质量。

Comments Accepted at AAAI 2026 (Main Technical Track)

UniCUE: Unified Recognition and Generation Framework for Chinese Cued Speech Video-to-Speech Generation

UniCUE: 中国手语提示语音视频到语音生成统一识别与生成框架

Jinting Wang, Shan Yang, Chenxing Li, Dong Yu, Li Liu

机构 * HKUST-GZ（香港科技大学-珠海校区） 纠错

AI总结 UniCUE提出首个统一框架，直接从CS视频生成语音，无需中间文本，通过整合姿态感知视觉处理器、语义对齐池和VisioPhonetic适配器，提升语音生成精度。

Comments 13 pages, 12 figures

Journal ref AAAI 2026 Oral

Distributional Priors Guided Diffusion for Generating 3D Molecules in Low Data Regimes

基于分布先验的扩散模型生成低数据条件下3D分子

Haokai Hong, Wanyu Lin, Ming Yang, Kay Chen Tan

AI总结 本文提出几何OOD扩散模型，通过捕捉分布结构先验生成低数据条件下3D分子，提升生成成功率12.6%。

Comments 24 pages. Accepted by AAAI 2026

STMI: Segmentation-Guided Token Modulation with Cross-Modal Hypergraph Interaction for Multi-Modal Object Re-Identification

STMI: 基于跨模态超图交互的多模态目标重识别分割引导令牌调节

Xingguo Xu, Zhanyu Liu, Weixiang Zhou, Yuansheng Gao, Junjie Cao, Yuhao Wang, Jixiang Luo, Dell Zhang

AI总结 STMI通过分割引导特征调节、语义令牌重新分配和跨模态超图交互，提升多模态目标重识别的性能与鲁棒性。

Comments Accepted to AAAI 2026

Learning to Explore: Policy-Guided Outlier Synthesis for Graph Out-of-Distribution Detection

学习探索：基于策略的异常合成用于图分布外检测

Li Sun, Lanxu Yang, Jiayu Tian, Bowen Fang, Xiaoyan Yu, Junda Ye, Peng Tang, Hao Peng, Philip S. Yu

AI总结 PGOS通过强化学习策略生成高质量伪OOD图，提升图分布外检测的鲁棒性与性能

Comments Accepted by AAAI'26, 9 pages

Mamba-CAD: State Space Model For 3D Computer-Aided Design Generative Modeling

Mamba-CAD：用于3D计算机辅助设计生成建模的状态空间模型

Xueyang Li, Yunzhong Lou, Yu Song, Xiangdong Zhou

AI总结 Mamba-CAD通过自监督生成建模方法，利用更长的参数化CAD序列生成复杂的工业CAD模型。

Comments Accepted to AAAI 2025

Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment

重新审视在跨模态不匹配下的LVLMs视觉标记减少

Rui Xu, Yunke Wang, Yong Luo, Bo Du

AI总结 本文提出VisionDrop方法，通过视觉-only修剪框架减少LVLMs中的视觉标记，无需额外训练，提升推理效率并保持性能。

Comments AAAI 2026

AStar: Boosting Multimodal Reasoning with Automated Structured Thinking

AStar: 通过自动化结构化思维提升多模态推理

Jinyang Wu, Mingkuan Feng, Guocheng Zhai, Shuai Zhang, Zheng Lian, Fangrui Lv, Pengpeng Shao, Ruihan Jin, Zhengqi Wen, Jianhua Tao

AI总结 AStar通过自动化结构化思维提升多模态推理效率，实现更高准确率和更强迁移能力。

Comments Accepted by AAAI 2026 Oral

ACPBench: Reasoning about Action, Change, and Planning

ACPBench: 关于行动、变化和规划的推理

Harsha Kokel, Michael Katz, Kavitha Srinivas, Shirin Sohrabi

AI总结 ACPBench是一个用于评估规划领域推理能力的基准，包含7个推理任务和13个规划领域，评估了多种LLM和推理模型的性能差异。

Comments In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI 2025)

SongSong: A Time Phonograph for Chinese SongCi Music from Thousand of Years Away

SongSong：从千年前远距离还原中国宋词音乐的时间 Phonograph

Jiajia Li, Jiliang Hu, Ziyi Pan, Chong Chen, Zuchao Li, Ping Wang, Lefei Zhang

AI总结 SongSong是一款能够还原古代宋词音乐的生成模型，通过预测旋律并生成演唱和伴奏，结合综合数据集提升古代音乐生成质量。

Comments 9 pages, 6 figures, accepted by AAAI 2025

Actor-Critic for Continuous Action Chunks: A Reinforcement Learning Framework for Long-Horizon Robotic Manipulation with Sparse Reward

用于连续动作片段的Actor-Critic：一种针对长周期机械臂操作的强化学习框架

Jiarui Yang, Bin Zhu, Jingjing Chen, Yu-Gang Jiang

AI总结 AC3是一种用于连续动作分块的强化学习框架，通过定向稳定机制和自监督模块，在稀疏奖励环境下实现高效稳定的动作序列学习。

Comments 14 pages, 13 figures, Accepted by AAAI 2026 (oral)

GenVidBench: A 6-Million Benchmark for AI-Generated Video Detection

GenVidBench：一个600万的AI生成视频检测基准数据集

Zhenliang Ni, Qiangyu Yan, Mouxiao Huang, Tianning Yuan, Yehui Tang, Hailin Hu, Xinghao Chen, Yunhe Wang

AI总结 GenVidBench是一个包含600万视频的AI生成视频检测基准数据集，通过大规模、跨源和跨生成器的视频收集，提升AI生成视频检测模型的泛化能力。

Comments AAAI 2026

Towards LLM-Empowered Knowledge Tracing via LLM-Student Hierarchical Behavior Alignment in Hyperbolic Space

基于大语言模型的知识追踪：通过超几何空间中的LLM-学生层次行为对齐

Xingcheng Fu, Shengpeng Wang, Yisen Gao, Xianxian Li, Chunpei Li, Qingyun Sun, Dongran Yu

AI总结 本文提出L-HAKT框架，通过超几何空间中的层次行为对齐，提升知识追踪对认知状态层次演化和个性化问题难度感知的建模能力。

Comments 9 pages, 6 figures, Accepted to AAAI 2026

A Mind Cannot Be Smeared Across Time

意识无法跨越时间被抹平

Michael Timothy Bennett

机构 * Michael Timothy Bennett（独立研究者） 纠错

AI总结 作者提出意识统一性需要客观同时实例化，而非时间顺序实现，指出软件意识在严格顺序子系统中无法实现，强调硬件对意识的重要性。

Comments Forthcoming in the proceedings of the AAAI 2026 Spring Symposium on Machine Consciousness: Integrating Theory, Technology, and Philosophy