AAAI

Bandit Learning in Housing Markets

住房市场中的多玩家多臂老虎机学习

Shiyun Lin

AI总结 本文提出了一种基于多玩家多臂老虎机框架的统计学习模型，用于在未知偏好下学习住房市场的核心稳定分配。

Comments Accepted to AAAI 2026 as oral

Uncovering Bias Paths with LLM-guided Causal Discovery: An Active Learning and Dynamic Scoring Approach

通过LLM引导的因果发现揭示偏见路径：一种主动学习和动态评分方法

Khadija Zanna, Akane Sano

AI总结 本文提出一种结合LLM引导的主动学习和动态评分方法，用于在存在噪声和混杂因素时发现公平性相关的因果路径。

Comments To be presented at AAAI 2026

TransMamba: A Sequence-Level Hybrid Transformer-Mamba Language Model

TransMamba: 一种序列级混合Transformer-Mamba语言模型

Yixing Li, Ruobing Xie, Zhen Yang, Xingwu Sun, Shuaipeng Li, Weidong Han, Zhanhui Kang, Yu Cheng, Chengzhong Xu, Di Wang, Jie Jiang

机构 * Tencent Hunyuan（腾讯文英） ； The Chinese University of Hong Kong（香港中文大学） ； University of Macau（澳门大学） 纠错

AI总结 TransMamba通过共享参数矩阵统一Transformer和Mamba，实现序列级动态切换注意力与SSM机制，提升训练效率和性能。

Comments Accepted by AAAI 2026. Code: https://github.com/Yixing-Li/TransMamba

Difficulty Controlled Diffusion Model for Synthesizing Effective Training Data

困难控制扩散模型用于合成有效训练数据

Zerun Wang, Jiafeng Mao, Xueting Wang, Toshihiko Yamasaki

机构 * CyberAgent 纠错

AI总结 本文提出了一种困难控制扩散模型，通过引入学习难度作为额外条件信号，高效生成具有显著性能提升的困难样本，从而在多个数据集上实现了更低的生成成本和更高的性能。

Comments AAAI 2026 accepted

X-MuTeST: A Multilingual Benchmark for Explainable Hate Speech Detection and A Novel LLM-consulted Explanation Framework

X-MuTeST：一个用于可解释仇恨言论检测的多语言基准及一种新颖的LLM咨询解释框架

Mohammad Zia Ur Rehman, Sai Kartheek Reddy Kasu, Shashivardhan Reddy Koppula, Sai Rithwik Reddy Chirra, Shwetank Shekhar Singh, Nagendra Kumar

机构 * Indian Institute of Technology Indore（印度理工学院Indore） ； Indian Institute of Information Technology Dharwad（印度信息科技学院Dharwad） ； Arizona State University（亚利桑那州立大学） ； Indian Institute of Technology Mandi（印度理工学院Mandi） 纠错

AI总结 X-MuTeST提出一种结合LLM和传统技术的多语言仇恨言论检测框架，通过人类注释的解释提升分类性能和可解释性。

Comments Accepted in the proceedings of AAAI 2026

Journal ref AAA 2026 (AISI)

DiffBench Meets DiffAgent: End-to-End LLM-Driven Diffusion Acceleration Code Generation

DiffBench 与 DiffAgent：端到端的 LLM 驱动扩散加速代码生成

Jiajun jiao, Haowei Zhu, Puyuan Yang, Jianghui Wang, Ji Liu, Ziqiong Liu, Dong Li, Yuejian Fang, Junhai Yong, Bin Wang, Emad Barsoum

AI总结 本文提出DiffBench和DiffAgent，通过LLM驱动的闭环流程，实现端到端的扩散模型加速代码生成，显著提升生成策略的有效性。

Comments Accepted to AAAI 2026

AbductiveMLLM: Boosting Visual Abductive Reasoning Within MLLMs

AbductiveMLLM: 提升多模态大语言模型中的视觉归纳推理

Boyu Chang, Qi Wang, Xi Guo, Zhixiong Nan, Yazhou Yao, Tianfei Zhou

AI总结 AbductiveMLLM通过结合REASONER和IMAGINER组件，提升多模态大语言模型在视觉归纳推理中的性能。

Comments Accepted by AAAI 2026 as Oral. Code:https://github.com/ChangPtR/AbdMLLM

Inferring Causal Graph Temporal Logic Formulas to Expedite Reinforcement Learning in Temporally Extended Tasks

推断因果图时序逻辑公式以加速时序扩展任务中的强化学习

Hadi Partovi Aria, Zhe Xu

AI总结 GTL-CIRL通过同时学习策略和挖掘因果图时序逻辑规范，加速时序扩展任务中的强化学习，提升学习效率和行为可验证性。

Comments Accepted to AAAI-26 Bridge Program B10: Making Embodied AI Reliable with Testing and Formal Verification

LendNova: Towards Automated Credit Risk Assessment with Language Models

LendNova：迈向利用语言模型进行自动信用风险评估

Kiarash Shamsi, Danijel Novokmet, Joshua Peters, Mao Lin Liu, Paul K Edwards, Vahab Khoshdel

AI总结 LendNova利用语言模型实现自动信用风险评估，通过直接处理原始信用记录文本，无需手动特征工程，提高准确性和可扩展性。

Journal ref AAAI 2026, Workshop on Agentic AI in Financial Services

Spiking Heterogeneous Graph Attention Networks

脉冲异构图注意力网络

Buqing Cao, Qian Peng, Xiang Xie, Liang Chen, Min Shi, Jianxun Liu

AI总结 SpikingHAN通过融合脉冲神经网络的节能特性，降低异构图学习的计算成本，同时保持性能。

Comments This paper has been accepted by AAAI 2026

Thucy: An LLM-based Multi-Agent System for Claim Verification across Relational Databases

Thucy：基于LLM的多智能体系统，用于跨关系数据库的声明验证

Michael Theologitis, Dan Suciu

AI总结 Thucy是一种基于LLM的多智能体系统，能够跨关系数据库验证声明，并提供支持裁定的SQL查询，准确率达94.3%。

Comments Accepted at AAAI 2026 Workshop on LLM-based Multi-Agent Systems (LaMAS)

Mitigating Error Accumulation in Co-Speech Motion Generation via Global Rotation Diffusion and Multi-Level Constraints

通过全局旋转扩散和多级约束缓解语音同步运动生成中的误差累积

Xiangyue Zhang, Jianfang Li, Jianqiang Ren, Jiaxu Zhang

AI总结 本文提出GlobalDiff，通过全局旋转扩散和多级约束缓解语音同步运动生成中的误差累积，提升生成运动的平滑度和准确性。

Comments AAAI 2026

Seeing the Unseen: Zooming in the Dark with Event Cameras

看见不可见的：利用事件相机在黑暗中放大

Dachun Kai, Zeyu Xiao, Huyue Zhu, Jiaxiao Wang, Yueyi Zhang, Xiaoyan Sun

AI总结 RetinexEVSR通过事件驱动和Retinex先验知识提升低光视频超分辨率性能，实现高对比度细节恢复与低光伪影抑制。

Comments Accepted to AAAI 2026

BiPrompt: Bilateral Prompt Optimization for Visual and Textual Debiasing in Vision-Language Models

BiPrompt：面向视觉与文本去偏的双重视觉语言模型双向提示优化

Sunny Gupta, Shounak Das, Amit Sethi

AI总结 BiPrompt通过双重视觉语言模型双向提示优化，同时减轻视觉和文本中的非因果特征依赖，提升模型在分布偏移下的鲁棒性和因果推理能力。

Comments Accepted at the AAAI 2026 Workshop AIR-FM, Assessing and Improving Reliability of Foundation Models in the Real World

XAI-MeD: Explainable Knowledge Guided Neuro-Symbolic Framework for Domain Generalization and Rare Class Detection in Medical Imaging

XAI-MeD: 可解释知识引导的神经符号框架用于医学影像中的领域泛化和稀有类别检测

Midhat Urooj, Ayan Banerjee, Sandeep Gupta

AI总结 XAI-MeD通过整合临床知识的神经符号框架，提升医学影像中领域泛化和稀有类别检测的性能，实现更鲁棒和可解释的多模态医学AI。

Comments Accepted at AAAI Bridge Program 2026

MacVQA: Adaptive Memory Allocation and Global Noise Filtering for Continual Visual Question Answering

MacVQA: 适应性内存分配与全局噪声过滤用于连续视觉问答

Zhifei Li, Yiran Wang, Chenyi Xiong, Yujing Xia, Xiaoju Hou, Yue Zhao, Miao Zhang, Kui Xiao, Bing Yang

AI总结 MacVQA通过适应性内存分配和全局噪声过滤提升持续视觉问答的性能，实现更高的准确率和更低的遗忘率。

Comments Accepted to AAAI 2026

Forget Less by Learning from Parents Through Hierarchical Relationships

通过层级关系从父母学习来减少遗忘

Arjun Ramesh Kaushik, Naresh Kumar Devulapally, Vishnu Suresh Lokhande, Nalini K. Ratha, Venu Govindaraju

AI总结 FLLP通过在超几何空间中引入父-子跨概念学习机制，减少生成模型在连续学习中的遗忘问题，提升模型的鲁棒性和泛化能力。

Comments Accepted at AAAI-26

Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach

基于视觉-语言推理的地理定位：一种强化学习方法

Biao Wu, Meng Fang, Ling Chen, Ke Xu, Tao Cheng, Jun Wang

AI总结 本文提出Geo-R，一种基于强化学习的无检索地理定位框架，通过链式区域生成可解释的监督，提升定位准确性与可解释性。

Comments Accepted to AAAI 2026. Project Page: https://github.com/aialt/geo-r

Improved Runtime Guarantees for the SPEA2 Multi-Objective Optimizer

SPEA2多目标优化器的改进运行时间保证

Benjamin Doerr, Martin S. Krejca, Milan Stanković

AI总结 SPEA2在多目标优化中通过改进的运行时间保证，更有效地处理特定基准问题。

Comments Accepted for AAAI 2026

TimeMosaic: Temporal Heterogeneity Guided Time Series Forecasting via Adaptive Granularity Patch and Segment-wise Decoding

TimeMosaic: 通过自适应粒度补丁和分段解码指导的时间序列预测

Kuiye Ding, Fanda Fan, Chunyi Hou, Zheya Wang, Lei Wang, Zhengxin Yang, Jianfeng Zhan

AI总结 TimeMosaic通过自适应粒度补丁和分段解码，解决时间序列预测中的时间异质性问题，提升预测精度与效率。

Comments This paper has been accepted by AAAI

Tackling Resource-Constrained and Data-Heterogeneity in Federated Learning with Double-Weight Sparse Pack

针对联邦学习中资源受限和数据异质性的双重权重稀疏打包方法

Qiantao Yang, Liquan Chen, Mingfu Xue, Songze Li

AI总结 FedCSPACK通过余弦稀疏打包和双权重聚合方法，有效解决联邦学习中资源受限和数据异质性问题，提升通信和计算效率的同时保持高模型精度。

Comments Accepted in AAAI 2026

Multi-granularity Interactive Attention Framework for Residual Hierarchical Pronunciation Assessment

多粒度交互注意力框架用于残差分层发音评估

Hong Han, Hao-Chen Pei, Zhao-Zheng Nie, Xin Luo, Xin-Shun Xu

AI总结 本文提出了一种多粒度交互注意力框架，通过双向建模和残差分层结构提升发音评估性能。

Comments 9 pages, 4 figures, 5 tables, accepted by AAAI 2026

From Theory of Mind to Theory of Environment: Counterfactual Simulation of Latent Environmental Dynamics

从心智理论到环境理论：潜在环境动态的反事实模拟

Ryutaro Uchiyama

机构 * Ryutaro Uchiyama（独立研究者） 纠错

AI总结 本文提出'环境理论'，通过社会线索推断隐藏环境动态，利用与心智理论共享的计算机制，扩展运动探索维度以促进行为创新。

Comments Accepted to the AAAI 2026 Workshop on Theory of Mind for Artificial Intelligence (ToM4AI). Extended abstract, 2 pages

Towards Long-window Anchoring in Vision-Language Model Distillation

迈向视觉-语言模型蒸馏中的长窗口锚定

Haoyi Zhou, Shuo Li, Tianyu Chen, Qi Song, Chonghan Gao, Jianxin Li

AI总结 本文提出LAid方法，通过改进注意力机制和RoPE响应调节，提升视觉-语言模型的长上下文理解能力，使蒸馏模型的有效上下文窗口长度增加3.2倍。

Comments Accepted by AAAI 2026

DISCODE: Distribution-Aware Score Decoder for Robust Automatic Evaluation of Image Captioning

DISCODE：面向分布的评分解码器，用于鲁棒的图像描述评估

Nakamasa Inoue, Kanoko Goto, Masanari Oi, Martyna Gruszka, Mahiro Ukai, Takumi Hirose, Yusuke Sekikawa

AI总结 DISCODE是一种无需微调的鲁棒图像描述评估方法，通过自适应测试时损失和多领域基准提升评估鲁棒性。

Comments Paper accepted to AAAI 2026

InfoDecom: Decomposing Information for Defending Against Privacy Leakage in Split Inference

InfoDecom：分解信息以防止在分割推断中的隐私泄露

Ruijun Deng, Zhihui Lu, Qiang Duan

AI总结 InfoDecom通过分解和去除冗余信息并注入校准噪声，实现了在分割推断中更优的隐私保护与效用平衡。

Comments 11pages, 6figures. Accepted by AAAI 2026

Rethinking Saliency Maps: A Cognitive Human Aligned Taxonomy and Evaluation Framework for Explanations

重新思考显著性图：一种与人类认知对齐的分类和评估框架用于解释

Yehonatan Elisha, Seffi Cohen, Oren Barkan, Noam Koenigstein

AI总结 本文提出RFxG分类法，通过区分点对点与对比性解释及不同粒度，改进显著性图的评估框架，以更符合人类认知需求。

Journal ref AAAI 2026, Paper page: https://yonisgit.github.io/rfxg1/

Optimal Look-back Horizon for Time Series Forecasting in Federated Learning

联邦学习中时间序列预测的最优回溯范围

Dahao Tang, Nan Yang, Yanli Li, Zhiyu Zhu, Zhibo Jin, Dong Yuan

AI总结 本文提出了一种在联邦学习中通过内在空间公式进行时间序列预测回溯范围选择的原理性框架，通过合成数据生成器和损失分解理论，为适应性回溯范围选择提供了理论支持。

Comments Accepted by AAAI-26 as Oral Presentation

Behaviour Policy Optimization: Provably Lower Variance Return Estimates for Off-Policy Reinforcement Learning

行为策略优化：为离线策略强化学习提供可证明的更低方差回报估计

Alexander W. Goodall, Edwin Hamel-De le Court, Francesco Belardinelli

AI总结 本文提出了一种通过行为策略收集数据以降低回报估计方差的方法，提升离线强化学习的样本效率和性能。

Comments Main Track at AAAI 2026

MPJudge: Towards Perceptual Assessment of Music-Induced Paintings

MPJudge：迈向音乐诱导绘画的感知评估

Shiqi Jiang, Tianyi Liang, Huayuan Ye, Changbo Wang, Chenhui Li

AI总结 MPJudge通过调制融合机制和直接偏好优化，提升音乐诱导绘画的感知评估精度。

Journal ref AAAI 2026