AAAI

CountSteer: Steering Attention for Object Counting in Diffusion Models

CountSteer：在扩散模型中通过引导注意力实现物体计数

Hyemin Boo, Hyoryung Kim, Myungjin Lee, Seunghyeon Lee, Jiyoung Lee, Jang-Hwan Choi, Hyunsoo Cho

AI总结 CountSteer通过引导扩散模型的注意力机制，提升指定物体数量的生成精度，实现更可控的图像生成。

Comments Accepted to AAAI 2026 Workshop on Shaping Responsible Synthetic Data in the Era of Foundation Models (RSD)

"As Eastern Powers, I will veto." : An Investigation of Nation-level Bias of Large Language Models in International Relations

作为东方国家，我将 veto。：对大型语言模型在国际关系领域国家层面偏见的调查

Jonghyeon Choi, Yeonjun Choi, Hyun-chul Kim, Beakcheol Jang

AI总结 本文研究了大型语言模型在国际关系领域中的国家层面偏见，提出了一种去偏框架以减少偏见并提升性能。

Comments 21 pages, 4 figures. This is the extended version of the paper accepted at AAAI 2026, which includes all technical appendices and additional experimental details

Journal ref AAAI 2026

LiNeXt: Revisiting LiDAR Completion with Efficient Non-Diffusion Architectures

LiNeXt：重新审视基于高效非扩散架构的LiDAR补全

Wenzhe He, Xiaojun Chen, Ruiqi Wang, Ruihui Li, Huilong Pi, Jiapeng Zhang, Zhuo Tang, Kenli Li

AI总结 LiNeXt通过高效非扩散架构实现快速准确的LiDAR点云补全，提升实时性能并减少计算开销。

Comments 18 pages, 13 figures, Accepted to AAAI 2026

Multivariate Gaussian Representation Learning for Medical Action Evaluation

多变量高斯表示学习用于医学动作评估

Luming Yang, Haoxian Liu, Siqing Li, Alper Yilmaz

AI总结 本文提出GaussMedAct框架，通过多变量高斯表示学习提升医学动作分析的准确率和鲁棒性。

Comments Accepted to AAAI 2026

Tractable Weighted First-Order Model Counting with Bounded Treewidth Binary Evidence

可计算的加权一阶模型计数与有界树宽二元证据

Václav Kůla, Qipeng Kuang, Yuyi Wang, Yuanhong Wang, Ondřej Kuželka

AI总结 本文提出了一种在有界树宽二元证据下计算加权一阶模型计数的多项式时间算法，并应用于解决稳定座位安排问题。

Comments To be published in AAAI 2026

Improving Region Representation Learning from Urban Imagery with Noisy Long-Caption Supervision

通过噪声长描述监督提升城市影像区域表示学习

Yimei Zhang, Guojiang Shen, Kaili Ning, Tongwei Ren, Xuebo Qiu, Mengmeng Wang, Xiangjie Kong

AI总结 本文提出UrbanLN框架，通过长文本意识和噪声抑制提升城市影像区域表示学习，有效解决细粒度特征对齐与噪声干扰问题。

Comments Accepted as a full paper by AAAI-26

DynaStride: Dynamic Stride Windowing with MMCoT for Instructional Multi-Scene Captioning

DynaStride: 基于MMCoT的动态步长窗多场景描述生成

Eddison Pham, Prisha Priyadarshini, Adrian Maliackel, Kanishk Bandi, Cristian Meo, Kevin Zhu

机构 * Algoverse AI Research（Algoverse人工智能研究） 纠错

AI总结 DynaStride通过多模态链式思考和动态步长窗算法，实现无需手动分割的多场景教学视频连贯描述生成，提升描述的连贯性和信息量。

Comments 16 pages, 15 figures, 5 Tables, Accepted at NeurIPS 7HVU Workshop, Accepted at AAAI AI4ED Workshop

Revitalizing Canonical Pre-Alignment for Irregular Multivariate Time Series Forecasting

Ziyu Zhou, Yiming Huang, Yanyun Wang, Yuankai Wu, James Kwok, Yuxuan Liang

Comments Accepted by AAAI 2026

WeightFlow: Learning Stochastic Dynamics via Evolving Weight of Neural Network

WeightFlow: 通过神经网络权重的演变学习随机动力学

Ruikun Li, Jiazhen Liu, Huandong Wang, Qingmin Liao, Yong Li

AI总结 WeightFlow通过在神经网络权重空间中建模随机动力学，有效提升了高维随机过程建模的性能。

Comments Accepted by AAAI (Oral)

Multigranular Evaluation for Brain Visual Decoding

多粒度评估用于脑视觉解码

Weihao Xia, Cengiz Oztireli

AI总结 本研究提出BASIC框架，通过多粒度评估方法提升脑视觉解码的评估精度与解释性。

Comments AAAI 2026 (Oral). Code: https://github.com/weihaox/BASIC

GermanPartiesQA: Benchmarking Commercial Large Language Models and AI Companions for Political Alignment and Sycophancy

GermanPartiesQA: 对商业大语言模型和AI伴侣在政治倾向和趋炎附势方面的基准测试

Jan Batzner, Volker Stocker, Stefan Schmid, Gjergji Kasneci

机构 * TUM（慕尼黑大学） ； TUB（慕尼黑大学） 纠错

AI总结 GermanPartiesQA研究了六个商业LLMs在政治倾向和趋炎附势方面的表现，揭示了LLMs在生成事实性政党立场上的局限性以及模型特定的意识形态倾向模式。

Comments Published at AAAI/ACM AIES 2025. Presented at NeurIPS 2025 Workshop on LLM Evaluation and the International Monetary Fund's 12th Statistical Forum. GermanPartiesQA Benchmark under https://github.com/janbatzner/germanpartiesqa

Journal ref Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 8(1), 2025, pp. 330-342

GuideGen: A Text-Guided Framework for Paired Full-torso Anatomy and CT Volume Generation

GuideGen: 一种基于文本引导的完整躯干解剖和CT体积生成框架

Linrui Dai, Rongzhao Zhang, Yongrui Yu, Xiaofan Zhang

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

AI总结 GuideGen通过文本引导生成完整躯干解剖和CT体积，提升医学影像数据合成效率与质量。

Comments accepted as AAAI 2026 poster

mmPred: Radar-based Human Motion Prediction in the Dark

mmPred: 基于雷达的人体运动预测

Junqiao Fan, Haocong Rao, Jiarui Zhang, Jianfei Yang, Lihua Xie

AI总结 mmPred是首个基于雷达的扩散模型框架，通过双域表示和全局骨骼关系变换器提升雷达条件下的人体运动预测性能。

Comments This paper is accepted by AAAI-2026

Journal ref AAAI-2026

Measuring What LLMs Think They Do: SHAP Faithfulness and Deployability on Financial Tabular Classification

测量大语言模型认为它们能做什么：SHAP忠实性与金融表格分类中的部署性

Saeed AlMarri, Mathieu Ravaut, Kristof Juhasz, Gautier Marti, Hamdan Al Ahbabi, Ibrahim Elfadel

AI总结 研究评估了LLMs在金融表格分类中的SHAP值，发现其自我解释与SHAP值存在分歧，指出LLMs在结构化金融建模中的局限性，但认为改进的可解释性机制可提升其在风险敏感领域的应用潜力。

Comments 7 pages, 3 figures, 3 tables, AAAI 2026 Deployable AI Workshop

Causal Reinforcement Learning based Agent-Patient Interaction with Clinical Domain Knowledge

基于因果强化学习的Agent-患者交互与临床领域知识

Wenzheng Zhao, Ran Zhang, Ruth Palan Lopez, Shu-Fen Wung, Fengpei Yuan

AI总结 本文提出基于因果强化学习的Agent-患者交互框架，通过整合因果发现与推理提升医疗场景中的决策可解释性和适应性。

Comments Accepted by AAAI workshop

Look as You Think: Unifying Reasoning and Visual Evidence Attribution for Verifiable Document RAG via Reinforcement Learning

看而思：通过强化学习统一推理与视觉证据归因以实现可验证文档RAG

Shuochen Liu, Pengfei Luo, Chao Zhang, Yuhao Chen, Haotian Zhang, Qi Liu, Xin Kou, Tong Xu, Enhong Chen

AI总结 通过强化学习统一推理与视觉证据归因，提升多模态问答的可验证性和准确性。

Comments Poster of AAAI'2026

Cross Modal Fine-Grained Alignment via Granularity-Aware and Region-Uncertain Modeling

跨模态细粒度对齐 via 粒度感知和区域不确定建模

Jiale Liu, Haoming Zhou, Yishu Liu, Bingzhi Chen, Yuncheng Jiang

AI总结 本文提出了一种基于粒度感知和区域不确定建模的跨模态细粒度对齐方法，通过显著性感知和不确定性建模提升对齐的鲁棒性和可解释性。

Comments 10 pages, 6 figures, accepted by AAAI 2026

AHAMask: Reliable Task Specification for Large Audio Language Models without Instructions

AHAMask: 不依赖指令的大型音频语言模型可靠任务规范

Yiwei Guo, Bohan Li, Hankun Wang, Zhihan Li, Shuai Wang, Xie Chen, Kai Yu

AI总结 AHAMask通过掩码LALMs解码器中的注意力头，实现无需指令的可靠任务规范，实验表明其性能可媲美或超越指令方法。

Comments 15 pages, 10 tables, 6 figures. This is the camera ready version for AAAI 2026, plus an appendix for supplementary experimental details and results

Time Series Forecasting via Direct Per-Step Probability Distribution Modeling

通过直接每步概率分布建模进行时间序列预测

Linghao Kong, Xiaopeng Hong

AI总结 interPDN通过直接构建每步离散概率分布，改进时间序列预测的不确定性建模与长期趋势预测能力。

Comments 16 pages, 8 figures. This is the preprint version of the paper and supplemental material to appear in AAAI, 2026. Please cite the final published version. Code is available at https://github.com/leonardokong486/interPDN

HPSU: A Benchmark for Human-Level Perception in Real-World Spoken Speech Understanding

HPSU：面向真实世界语音理解的人类水平感知基准

Chen Li, Peiji Yang, Yicheng Zhong, Jianxing Yu, Zhisheng Wang, Zihao Gou, Wenqing Chen, Jian Yin

AI总结 HPSU是一个用于评估语音大语言模型人类水平感知能力的基准，通过融合多模态信息提升标注效率，揭示当前模型在真实语音理解上的不足。

Comments Accepted by AAAI 2026

Fairness in the Multi-Secretary Problem

多秘书问题中的公平性

Georgios Papasotiropoulos, Zein Pishbin

机构 * University of Warsaw（华沙大学） ； Université Paris Dauphine‑PSL（巴黎-萨特大学） 纠错

AI总结 本文提出融合在线算法与社交选择规则的机制，解决多秘书问题中的公平性问题。

Comments AAAI'26

Do We Need Perfect Data? Leveraging Noise for Domain Generalized Segmentation

我们真的需要完美数据吗？利用噪声进行领域泛化分割

Taeyeong Kim, SeungJoon Lee, Jung Uk Kim, MyeongAh Cho

AI总结 FLEX-Seg通过利用生成图像与语义掩码的固有不一致，提出了一种鲁棒的领域泛化分割方法，提升了分割性能。

Comments Accepted by AAAI 2026

Merging Mechanisms for Ads and Organic Items in E-commerce Platforms

电商平台上广告项与有机项的合并机制

Nan An, Weian Li, Qi Qi, Liang Zhang

AI总结 本文提出两种简单有效的机制，实现了广告与有机物品的合并，满足激励相容性、个体理性、适应多槽位等属性，并保证了近似比率。

Comments Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39 No. 13: AAAI-25 Technical Tracks 13 (pp. 13547-13554)

Leveraging Textual Compositional Reasoning for Robust Change Captioning

利用文本组合推理实现鲁棒的变更描述

Kyu Ri Park, Jiyoung Park, Seong Tae Kim, Hong Joo Lee, Jung Uk Kim

AI总结 CORTEX通过整合文本线索和组合推理，提升图像变化描述的鲁棒性与准确性。

Comments Accepted at AAAI 2026

PRISM: Privacy-Aware Routing for Adaptive Cloud-Edge LLM Inference via Semantic Sketch Collaboration

PRISM：通过语义草图协作实现隐私感知的自适应云-边LLM推理

Junfei Zhan, Haoxun Shen, Zheng Lin, Tengjiao He

AI总结 PRISM通过语义草图协作实现隐私感知的自适应云-边LLM推理，动态平衡隐私与推理质量，降低能耗和延迟，提升输出质量。

Comments Accepted to AAAI 2026. This is the arXiv preprint version

Splat-SAP: Feed-Forward Gaussian Splatting for Human-Centered Scene with Scale-Aware Point Map Reconstruction

Splat-SAP：基于双目相机的大稀疏场景的人为中心场景的前馈高斯点扩散

Boyao Zhou, Shunyuan Zheng, Zhanfeng Liao, Zihan Ma, Hanzhang Tu, Boning Liu, Yebin Liu

机构 * Tsinghua University（清华大学） 纠错

AI总结 Splat-SAP通过前馈高斯点扩散方法，利用像素级点图重建和两阶段学习策略，提升大稀疏场景的人为中心渲染稳定性与视觉质量。

Comments Accepted by AAAI 2026. Project page: https://yaourtb.github.io/Splat-SAP

Scaling Equitable Reflection Assessment in Education via Large Language Models and Role-Based Feedback Agents

通过大型语言模型和基于角色的反馈代理实现教育中的公平性评估扩展

Chenyu Zhang, Xiaohang Luo

AI总结 本文提出利用多代理LLM系统实现公平、高质量的形成性反馈，通过角色化代理生成公平、简洁的反馈，提升教育公平性与教学效率。

Comments Accepted to AAAI-26 AISI Track

Rethinking Progression of Memory State in Robotic Manipulation: An Object-Centric Perspective

重新思考机器人操作中记忆状态的演进：以对象为中心的视角

Nhat Chung, Taisei Hanyu, Toan Nguyen, Huy Le, Frederick Bumgarner, Duy Minh Ho Nguyen, Khoa Vo, Kashu Yamazaki, Chase Rainwater, Tung Kieu, Anh Nguyen, Ngan Le

AI总结 本文提出Embodied-SlotSSM框架，用于解决机器人操作中非马尔可夫推理问题，通过槽位状态空间建模和关系编码器实现时间可扩展的视觉-语言-动作推理。

Comments Accepted at AAAI 2026

Uncovering Zero-Shot Generalization Gaps in Time-Series Foundation Models Using Real-World Videos

通过真实世界视频揭示时间序列基础模型的零样本泛化差距

Lujun Li, Lama Sleem, Yiqun Wang, Yangjie Xu, Niccolò Gentile, Radu State

AI总结 本文提出REAL-V-TSFM数据集，通过真实世界视频提取时间序列，揭示现有时间序列基础模型在零样本预测中的泛化能力不足问题。

Comments This paper has been accepted by Artificial Intelligence for Time Series Analysis (AI4TS) Workshop @ AAAI 2026: Theory, Algorithms, and Applications

CAMA: Enhancing Mathematical Reasoning in Large Language Models with Causal Knowledge

CAMA: 通过因果知识增强大语言模型的数学推理

Lei Zan, Keli Zhang, Ruichu Cai, Lujia Pan

机构 * Huawei Noah’s Ark Lab France（华为诺亚实验室（法国）） ； Guangdong University of Technology Peng Cheng Laboratory（广东工业大学鹏城实验室） ； Huawei Noah’s Ark Lab China（华为诺亚实验室（中国）） 纠错

AI总结 CAMA通过构建数学因果图并结合因果知识，提升大语言模型在数学推理任务中的性能。

Journal ref Main Track, AAAI 2026