AAAI

The Confidence Trap: Gender Bias and Predictive Certainty in LLMs

自信陷阱：大语言模型中的性别偏见与预测确定性

Ahmed Sabir, Markus Kängsepp, Rajesh Sharma

AI总结 本研究探讨大语言模型在性别偏见任务中的置信度校准问题，提出Gender-ECE指标以评估公平性差异，发现Gemma-2在性别偏见基准中校准最差。

Comments AAAI 2026 (AISI Track), Oral. Project page: https://bit.ly/4p8OKQD

Structural Approach to Guiding a Present-Biased Agent

结构化方法用于引导现时偏好代理

Tatiana Belova, Yuriy Dementiev, Artur Ignatiev, Danil Sagunov

AI总结 本文提出了一种结构化方法，用于引导现时偏好代理，通过参数化算法分析任务图的计算复杂性，并给出了固定参数可 tractable 算法。

Comments Accepted at AAAI 2026

Beyond Sharpness: A Flatness Decomposition Framework for Efficient Continual Learning

超越尖锐性：一种用于高效持续学习的平坦性分解框架

Yanan Chen, Tieliang Gong, Yunjiao Zhang, Wen Wen

AI总结 FLAD通过分解尖锐性感知扰动并保留噪声成分，提升持续学习的泛化能力与效率。

Comments Accepted by AAAI 2026

Task Prototype-Based Knowledge Retrieval for Multi-Task Learning from Partially Annotated Data

基于任务原型的知识检索多任务学习：从部分标注数据中学习

Youngmin Oh, Hyung-Il Kim, Jung Uk Kim

AI总结 本文提出基于原型的知识检索框架，通过任务原型嵌入和知识检索变压器实现稳健的多任务学习，适用于部分标注数据场景。

Comments Accepted at AAAI 2026

PulseMind: A Multi-Modal Medical Model for Real-World Clinical Diagnosis

PulseMind: 一种多模态医学模型用于真实世界临床诊断

Jiao Xu, Junwei Liu, Jiangwei Lao, Qi Zhu, Yunpeng Zhao, Congyun Jin, Shinan Liu, Zhihong Lu, Lihe Zhang, Xin Chen, Jian Wang, Ping Wang

机构 * Ant Group（蚂蚁集团） 纠错

AI总结 PulseMind通过整合多模态数据集和定制训练框架，提升真实世界临床诊断的准确性与实用性。

Comments Accepted to AAAI 2026

ActiShade: Activating Overshadowed Knowledge to Guide Multi-Hop Reasoning in Large Language Models

ActiShade: 通过激活被遮蔽的知识来指导大语言模型的多跳推理

Huipeng Ma, Luan Zhang, Dandan Song, Linmei Hu, Yuhang Tian, Jun Yang, Changzhi Zhou, Chenhao Li, Yizhou Jin, Xudong Li, Meng Lin, Mingxing Zhang, Shuhao Zhang

AI总结 ActiShade通过激活被遮蔽的知识来提升大语言模型在多跳推理中的表现。

Comments Accepted to AAAI 2026

Yes FLoReNce, I Will Do Better Next Time! Agentic Feedback Reasoning for Humorous Meme Detection

是的，Florne，我下次会做得更好！用于幽默表情包检测的代理反馈推理

Olivia Shanhong Liu, Pai Chet Ng, De Wen Soh, Konstantinos N. Plataniotis

AI总结 Florne通过闭环反馈机制提升表情包幽默检测的适应性和解释质量。

Comments LaMAS@AAAI 2026 (Oral)

Relink: Constructing Query-Driven Evidence Graph On-the-Fly for GraphRAG

Relink：为GraphRAG构建查询驱动的证据图谱

Manzong Huang, Chenyang Bu, Yi He, Xingrui Zhuo, Xindong Wu

AI总结 Relink通过动态构建查询驱动的证据图谱，解决GraphRAG中知识图谱不完整和干扰事实的问题，提升问答性能。

Comments Accepted by AAAI 2026

MacPrompt: Maraconic-guided Jailbreak against Text-to-Image Models

MacPrompt：基于马卡罗尼的文本到图像模型对抗攻击

Xi Ye, Yiwen Liu, Lina Wang, Run Wang, Geying Yang, Yufei Hou, Jiayi Yu

AI总结 MacPrompt通过跨语言字符级重组有害术语，实现对文本到图像模型的安全机制的高效对抗，突破现有防御体系。

Comments Accepted by AAAI 2026

Explicit World Models for Reliable Human-Robot Collaboration

显式世界模型以实现可靠的人机协作

Kenneth Kwok, Basura Fernando, Qianli Xu, Vigneshwaran Subbaraju, Dongkyu Choi, Boon Kiat Quek

AI总结 本文提出通过构建显式世界模型来实现可靠的显式人机协作，强调动态模糊的人机交互需求。

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

REXO: Indoor Multi-View Radar Object Detection via 3D Bounding Box Diffusion

REXO：基于3D边界框扩散的多视角室内雷达目标检测

Ryoma Yataka, Pu Perry Wang, Petros Boufounos, Ryuhei Takahashi

机构 * MERL ； ITC 纠错

AI总结 REXO通过3D边界框扩散提升多视角室内雷达目标检测，利用先验知识减少参数并提升检测性能。

Comments 26 pages; Accepted to AAAI 2026; Code available at https://github.com/merlresearch/radar-bbox-diffusion

Intervention Efficiency and Perturbation Validation Framework: Capacity-Aware and Robust Clinical Model Selection under the Rashomon Effect

干预效率与扰动验证框架：在拉索莫恩效应下考虑容量的临床模型选择

Yuwen Zhang, Viet Tran, Paul Weng

AI总结 本文提出干预效率与扰动验证框架，用于在资源约束下稳健选择临床模型，解决拉索莫恩效应下的模型选择问题。

Comments Accepted to the Workshop on Navigating Model Uncertainty and the Rashomon Effect: From Theory and Tools to Applications and Impact (AAAI 2026)

Enhancing Binary Encoded Crime Linkage Analysis Using Siamese Network

利用孪生网络增强二进制编码的犯罪关联分析

Yicheng Zhan, Fahim Ahmed, Amy Burrell, Matthew J. Tonkin, Sarah Galambos, Jessica Woodhams, Dalal Alrajeh

AI总结 本文提出利用孪生网络增强二进制编码的犯罪关联分析，通过学习潜在表示和整合地理时间特征，提升关联准确性并提供可解释的见解。

Comments AAAI 2026, 7 pages, 4 figures

A Graph-Theoretical Perspective on Law Design for Multiagent Systems

多智能体系统的法律设计图论视角

Qi Shi, Pavel Naumov

AI总结 本文从图论角度研究多智能体系统中法律设计问题，提出两种法律类型并证明其最小化问题的计算复杂性，同时利用超图顶点覆盖的近似算法进行高效近似。

Comments The 40th AAAI Conference on Artificial Intelligence (AAAI-26)

SpecDetect: Simple, Fast, and Training-Free Detection of LLM-Generated Text via Spectral Analysis

SpecDetect: 一种简单、快速且无需训练的LLM生成文本检测方法通过频谱分析

Haitong Luo, Weiyao Zhang, Suhang Wang, Wenji Zou, Chungang Lin, Xuying Meng, Yujun Zhang

AI总结 SpecDetect通过频谱分析技术，利用DFT总能量检测LLM生成文本，具有高效、快速且无需训练的特点。

Comments AAAI'26 Oral

GDBA Revisited: Unleashing the Power of Guided Local Search for Distributed Constraint Optimization

GDBA重审：释放指导局部搜索在分布式约束优化中的潜力

Yanchen Deng, Xinrun Wang, Bo An

AI总结 本文提出DGLS框架，通过改进约束惩罚机制和协调更新方案，提升分布式约束优化的性能。

Comments Accepted by AAAI 2026

Put the Space of LoRA Initialization to the Extreme to Preserve Pre-trained Knowledge

将LoRA初始化空间推向极端以保留预训练知识

Pengwei Tang, Xiaolin Hu, Yong Liu, Lizhong Ding, Dongjie Zhang, Xing Wu, Debing Zhang

机构 * Pengwei Tang 1,2,3（唐鹏伟） ； Xiaolin Hu 4（胡晓林） ； Yong Liu 1,2,3（刘勇） ； Lizhong Ding 5（丁立忠） ； Dongjie Zhang 6（张东杰） ； Xing Wu 6,7（吴星） ； Debing Zhang 6（张德兵） 纠错

AI总结 LoRA-Null通过在激活的空域中初始化LoRA，有效保留预训练知识并提升微调性能。

Comments Accepted at AAAI 2026. We rediscovered why our approach works from the perspective of the LoRA initialization space. Accordingly, we added new experiments and also removed inappropriate experiments (those without catastrophic forgetting)

Reimagining Anomalies: What If Anomalies Were Normal?

重新定义异常：如果异常是正常的呢？

Philipp Liznerski, Saurabh Varshneya, Ece Calikus, Puyu Wang, Alexander Bartscher, Sebastian Josef Vollmer, Sophie Fellenz, Marius Kloft

AI总结 本文提出一种生成多个替代修改以解释异常检测机制的方法，通过语义解释帮助用户探索异常的"如果"场景。

Comments 38 pages, published as a conference paper at AAAI 2026

SecMoE: Communication-Efficient Secure MoE Inference via Select-Then-Compute

SecMoE: 通过选择-然后计算实现通信高效的安全MoE推理

Bowen Shen, Yuyue Chen, Peng Yang, Bin Zhang, Xi Zhang, Zoe L. Jiang

AI总结 SecMoE通过选择-然后计算方法，在保持隐私的同时实现高效安全的MoE推理，使模型规模扩大63倍，运行时间仅增加15.2倍。

Comments Accepted by AAAI 2026

Robust Evacuation for Multi-Drone Failure in Drone Light Shows

多无人机故障下的鲁棒疏散

Minhyuk Park, Aloysius K. Mok, Tsz-Chiu Au

AI总结 本研究提出了一种针对多无人机故障的鲁棒疏散算法，通过预测故障无人机轨迹并利用隐藏无人机实现快速恢复，提高无人机灯光秀的可靠性。

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

N2N-GQA: Noise-to-Narrative for Graph-Based Table-Text Question Answering Using LLMs

N2N-GQA：基于图的表格-文本问答中的噪声到叙事使用大语言模型

Mohamed Sharafath, Aravindh Annamalai, Ganesh Murugan, Aravindakumar Venugopalan

机构 * comcast India Engineering Center (CIEC)（comcast印度工程中心） 纠错

AI总结 N2N-GQA通过构建动态证据图提升基于图的表格-文本问答性能，实现零样本多跳问答的高准确率。

Comments Accepted at an AAAI 2026 Workshop

Improving Day-Ahead Grid Carbon Intensity Forecasting by Joint Modeling of Local-Temporal and Cross-Variable Dependencies Across Different Frequencies

通过联合建模本地时间依赖性和跨变量依赖性提升日提前电网碳强度预测

Bowen Zhang, Hongda Tian, Adam Berry, A. Craig Roussac

AI总结 本文提出了一种联合建模本地时间依赖性和跨变量依赖性的新型模型，用于提升电网碳强度预测的准确性。

Comments 2026 40th AAAI Conference on Artificial Intelligence

A Unified Shape-Aware Foundation Model for Time Series Classification

面向时间序列分类的统一形状感知基础模型

Zhen Liu, Yucheng Wang, Boyuan Li, Junhao Zheng, Emadeldeen Eldele, Min Wu, Qianli Ma

AI总结 UniShape是一种面向时间序列分类的统一形状感知基础模型，通过自适应聚合多尺度判别子序列提升模型可解释性，并在多个数据集上实现最先进的分类性能。

Comments Accepted in AAAI 2026

Representing Sounds as Neural Amplitude Fields: A Benchmark of Coordinate-MLPs and A Fourier Kolmogorov-Arnold Framework

将声音表示为神经振幅场：坐标MLP的基准测试及傅里叶科尔莫戈洛夫-阿诺德框架

Linfei Li, Lin Zhang, Zhong Wang, Fengyi Zhang, Zelin Li, Ying Shen

AI总结 本文提出傅里叶ASR框架，利用傅里叶级数和科尔莫戈洛夫-阿诺德定理，通过傅里叶-KAN网络高效表示音频信号，提升音频表示的鲁棒性和质量。

Comments Accepted by AAAI 2025. Code: https://github.com/lif314/Fourier-ASR

Human-in-the-Loop Interactive Report Generation for Chronic Disease Adherence

具有临床医生的交互式报告生成用于慢性病依从性

Xiaotian Zhang, Jinhong Yu, Pengwei Yan, Le Jiang, Xingyi Shen, Mumo Cheng, Xiaozhong Liu

AI总结 本文提出一种具有临床医生的交互式报告生成系统，通过AI生成与医生审查的分工，提高慢性病依从性报告的效率和准确性，同时确保问责制。

Comments 5 pages, 3 figures. Accepted at the AAAI 2026 Workshop on AI for Healthy Aging and Longevity

PCoKG: Personality-aware Commonsense Reasoning with Debate

PCoKG：具有辩论机制的个性感知常识推理

Weijie Li, Zhongqing Wang, Guodong Zhou

AI总结 PCoKG通过引入辩论机制构建个性感知常识知识图谱，提升对话生成的一致性与个性化水平。

Comments Accept by AAAI-2026

COVR:Collaborative Optimization of VLMs and RL Agent for Visual-Based Control

COVR：视觉控制中视觉语言模型与强化学习代理的协同优化

Canming Xia, Peixi Peng, Guang Tan, Zhan Su, Haoran Xu, Zhenxian Liu, Luntong Li

AI总结 COVR通过协同优化视觉语言模型与强化学习代理，利用RL生成数据提升VLM语义推理能力，并通过动作先验引导策略学习，实现视觉控制任务中的高效优化。

Comments The paper was accepted by the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26)

TEAS: Trusted Educational AI Standard: A Framework for Verifiable, Stable, Auditable, and Pedagogically Sound Learning Systems

TEAS: 可信教育AI标准：可验证、稳定、可审计和教学上可靠的系统框架

Abu Syed

机构 * Abu Syed（阿布·赛德） 纠错

AI总结 本文提出TEAS框架，通过可验证性、稳定性、可审计性和教学合理性四个支柱，为教育AI系统提供可信度标准，强调系统架构而非模型能力的重要性。

Comments 19 pages, 6 tables, accepted at AAAI-26 (DAI Workshop) in Singapore

SmartSplat: Feature-Smart Gaussians for Scalable Compression of Ultra-High-Resolution Images

SmartSplat: 基于特征的智能高斯用于超高清图像的可扩展压缩

Linfei Li, Lin Zhang, Zhong Wang, Ying Shen

AI总结 SmartSplat通过特征感知的高斯方法实现超高清图像的高效压缩，兼顾压缩比与重建质量，展现强可扩展性。

Comments Accepted by AAAI 2026