AAAI

PA-FAS: Towards Interpretable and Generalizable Multimodal Face Anti-Spoofing via Path-Augmented Reinforcement Learning

PA-FAS: 向可解释和通用的多模态人脸反伪装迈进：通过路径增强强化学习

Yingjie Ma, Xun Lin, Yong Xu, Weicheng Xie, Zitong Yu

AI总结 PA-FAS通过路径增强强化学习提升多模态人脸反伪装的可解释性和泛化能力。

Comments Accepted by AAAI 2026 (Oral)

Cost-Sensitive Conformal Training with Provably Controllable Learning Bounds

具有可证明可控学习界限的成本敏感符合训练

Xuesong Jia, Yuanjie Shi, Ziquan Liu, Yi Xu, Yan Yan

AI总结 本文提出一种成本敏感符合训练算法，通过排名加权策略有效控制学习界限，实验表明其在预测效率上优于现有方法，平均预测集大小减少21.38%。

Comments Accepted for Publication at Association for the Advancement of Artificial Intelligence (AAAI), 2026

A superpersuasive autonomous policy debating system

一种超级有说服力的自主辩论系统

Allen Roush, Devin Gonier, John Hines, Judah Goldfeder, Philippe Martin Wyder, Sanjay Basu, Ravid Shwartz Ziv

AI总结 DeepDebater是一种能够自主参与并赢得完整政策辩论的AI系统，通过多智能体协作生成高质量论证并战胜人类对手。

Comments Accepted to CLIP workshop at AAAI 2026

A Hybrid Classical-Quantum Fine Tuned BERT for Text Classification

一种混合的经典-量子微调BERT用于文本分类

Abu Kaisar Mohammad Masum, Naveed Mahmud, M. Hassan Najafi, Sercan Aygun

AI总结 本文提出一种结合量子电路和经典BERT的混合模型，用于提升文本分类性能，实验表明其在多个基准数据集上表现优异。

Comments This paper has been accepted by First AAAI Symposium on Quantum Information & Machine Learning (QIML): Bridging Quantum Computing and Artificial Intelligence at AAAI 2025 Fall Symposium

TTF: A Trapezoidal Temporal Fusion Framework for LTV Forecasting in Douyin

TTF: 一种用于抖音LTV预测的梯形时间融合框架

Yibing Wan, Zhengxiong Guan, Chaoli Zhang, Xiaoyang Li, Lai Xu, Beibei Jia, Zhenzhe Zheng, Fan Wu

机构 * Douyin Group（抖音集团） 纠错

AI总结 TTF框架通过梯形多时间序列模块和MT-FusionNet结构，解决LTV预测中的数据不匹配、SILO挑战和序列非平稳问题，提升预测精度。

Comments Accepted by AAAI IAAI Track

Journal ref AAAI 2026 IAAI Track

BD-Net: Has Depth-Wise Convolution Ever Been Applied in Binary Neural Networks?

BD-Net: 卷积深度学习中是否曾应用于二进制神经网络？

DoYoung Kim, Jin-Seop Lee, Noo-ri Kim, SungJoon Lee, Jee-Hyong Lee

AI总结 BD-Net通过1.58位卷积和预BN残差连接，首次实现了二进制神经网络中深度卷积的二值化，在多个数据集上取得显著准确率提升。

Comments Paper accepted to AAAI 2026

Enhanced Federated Deep Multi-View Clustering under Uncertainty Scenario

不确定场景下的增强联邦深度多视图聚类

Bingjun Wei, Xuemei Cao, Jiafen Liu, Haoyang Liang, Xin Yang

AI总结 本文提出EFDMVC框架，通过语义对齐、视图自适应漂移和平衡聚合机制，解决联邦多视图聚类中的视图不确定性和聚合不确定性问题，实验验证其在异构不确定视图下的鲁棒性。

Journal ref AAAI 2026

CompTrack: Information Bottleneck-Guided Low-Rank Dynamic Token Compression for Point Cloud Tracking

CompTrack: 基于信息瓶颈的点云跟踪低秩动态令牌压缩

Sifan Zhou, Yichao Cao, Jiahao Nie, Yuqian Fu, Ziyu Zhao, Xiaobo Lu, Shuo Wang

AI总结 CompTrack通过信息瓶颈引导的低秩动态令牌压缩技术，有效解决点云中空间和信息冗余问题，实现高效实时的三维单目标跟踪。

Comments Accepted by AAAI 2026 (Oral)

Multi-Aspect Cross-modal Quantization for Generative Recommendation

多方面跨模态量化用于生成性推荐

Fuwei Zhang, Xiaoyu Liu, Dongbo Xi, Jishen Yin, Huan Chen, Peng Yan, Fuzhen Zhuang, Zhao Zhang

AI总结 MACRec通过多方面跨模态量化方法，提升生成性推荐中多模态信息利用和语义ID学习的性能。

Comments Accepted by AAAI 2026 (Oral)

Graph Out-of-Distribution Detection via Test-Time Calibration with Dual Dynamic Dictionaries

基于测试时间校准的双动态字典图域外分布检测

Yue Hou, Ruomei Liu, Yingke Su, Junran Wu, Ke Xu

AI总结 BaCa通过双动态字典校准测试时间图域外分布检测，无需微调预训练模型，有效提升OOD检测性能。

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

ComLQ: Benchmarking Complex Logical Queries in Information Retrieval

ComLQ：信息检索中复杂逻辑查询的基准测试

Ganlin Xu, Zhitao Yin, Linghao Zhang, Jiaqing Liang, Weijia Lu, Xiaodong Zhang, Zhifei Yang, Sihang Jiang, Deqing Yang

AI总结 ComLQ通过构建包含2909个查询和11251个候选段落的信息检索数据集，提出Log-Scaled Negation Consistency指标，评估检索模型在复杂逻辑查询，特别是含否定查询上的性能。

Comments Accepted by AAAI 2026

Beyond Superficial Forgetting: Thorough Unlearning through Knowledge Density Estimation and Block Re-insertion

超越表面遗忘：通过知识密度估计和块重新插入实现彻底遗忘

Feng Guo, Yuntao Wen, Shen Gao, Junshuo Zhang, Shuo Shang

AI总结 KUnBR通过知识密度估计和块重新插入策略，有效解决大语言模型中有害知识的彻底移除问题，实现高效的遗忘与模型实用性平衡。

Comments Accepted by AAAI 2026

DeepRWCap: Neural-Guided Random-Walk Capacitance Solver for IC Design

DeepRWCap：基于神经网络的随机游走电容求解器用于集成电路设计

Hector R. Rodriguez, Jiechen Huang, Wenjian Yu

AI总结 DeepRWCap通过神经网络引导随机游走求解器，提高集成电路设计中电容计算的精度与效率。

Comments Accepted to AAAI-26

Red Teaming Multimodal Language Models: Evaluating Harm Across Prompt Modalities and Models

针对多模态语言模型的红队测试：评估不同提示模态和模型的有害性

Madison Van Doren, Casey Ford

AI总结 本研究通过红队测试评估多模态语言模型在不同提示模态下的安全性，发现Pixtral 12B的有害响应率最高，而Claude Sonnet 3.5最安全，凸显了建立多模态安全基准的必要性。

Journal ref AAAI 2026 AIGOV Workshop and EurIPS 2025 Workshop on Unifying Perspectives on Learning Biases

SpecDiff: Accelerating Diffusion Model Inference with Self-Speculation

SpecDiff: 通过自推测加速扩散模型推理

Jiayi Pan, Jiaming Xu, Yongkang Zhou, Guohao Dai

AI总结 SpecDiff通过自推测引入未来信息，提出无需训练的多级特征缓存策略，实现扩散模型推理的高效加速。

Comments Accepted by AAAI 2026 Oral

Tapas Are Free! Training-Free Adaptation of Programmatic Agents via LLM-Guided Program Synthesis in Dynamic Environments

Tapas Are Free! 通过LLM引导的程序合成实现动态环境下的免训练程序代理适应

Jinwei Hu, Yi Dong, Youcheng Sun, Xiaowei Huang

AI总结 TAPA通过LLM引导的程序合成，在动态环境中实现免训练的程序代理适应，提升网络安全和群体智能任务的性能与可靠性。

Comments Extended version of the paper accepted at AAAI-26 Oral to comply with the AAAI camera-ready requirements with minor revisions

Vision-Only Gaussian Splatting for Collaborative Semantic Occupancy Prediction

仅视觉的高斯散射用于协作语义占用预测

Cheng Chen, Hao Huang, Saurabh Bagchi

AI总结 本研究提出基于稀疏3D语义高斯散射的协作语义占用预测方法，通过高效融合和减少通信开销，提升了协作感知的性能和鲁棒性。

Comments Accepted by AAAI 2026 (Oral)

Whispering Agents: An Event-driven Covert Communication Protocol For the Internet of Agents

低语代理：一种基于事件的隐秘通信协议用于代理互联网

Kaibo Huang, Yukun Wei, Wansheng Wu, Tianhua Zhang, Zhongliang Yang, Linna Zhou

AI总结 本文提出了一种基于事件的隐秘通信协议ΠCCAP，利用代理对话的特性实现安全隐秘通信，为代理互联网的安全监控和防御提供基础支持。

Comments Accepted to AAAI-26 (Main, Oral)

Backdoors in Conditional Diffusion: Threats to Responsible Synthetic Data Pipelines

条件扩散中的后门：对负责任的合成数据管道的威胁

Raz Lapid, Almog Dubin

机构 * Raz Lapid ； Almog Dubin 纠错

AI总结 本文提出了一种针对ControlNet的模型污染攻击，通过后门生成攻击者指定内容，并提出CFT方法作为防御措施。

Comments Accepted at RDS @ AAAI 2026

Journal ref AAAI 2026 Workshop on Shaping Responsible Synthetic Data in the Era of Foundation Models

ReCode: Updating Code API Knowledge with Reinforcement Learning

ReCode: 通过强化学习更新代码API知识

Haoze Wu, Yunzhi Yao, Wenhao Yu, Ningyu Zhang

AI总结 ReCode通过强化学习方法提升LLMs在动态API环境下的代码生成能力，尤其在CodeUpdateArena任务中表现优异。

Comments AAAI 2026

Graph of Verification: Structured Verification of LLM Reasoning with Directed Acyclic Graphs

验证图：基于有向无环图的结构化LLM推理验证

Jiwei Fang, Bin Zhang, Changwei Wang, Jin Wan, Zhiwei Xu

AI总结 本文提出验证图（GoV），通过灵活的节点块架构实现适应性多粒度验证，有效解决LLM推理验证中的精度与鲁棒性平衡问题。

Comments Accepted to AAAI 2026

LoKI: Low-damage Knowledge Implanting of Large Language Models

LoKI: 大型语言模型的低损伤知识植入

Runyu Wang, Peng Ping, Zhengyu Guo, Xiaoye Zhang, Quan Shi, Liting Zhou, Tianbo Ji

AI总结 LoKI通过利用transformer架构的知识存储机制，实现了在保持通用能力的同时提升任务特定性能的参数高效微调方法。

Comments AAAI-26 Oral

A Roadmap to Guide the Integration of LLMs in Hierarchical Planning

引导大型语言模型在分层规划中整合的路线图

Israel Puerta-Merino, Carlos Núñez-Molina, Pablo Mesejo, Juan Fernández-Olivares

AI总结 本文提出了一条引导大型语言模型在分层规划中整合的路线图，通过分类整合方法和建立基准测试，探索LLMs在分层规划中的应用潜力。

Comments 5 pages, 0 figures, to be published in the AAAI Workshop on Planning in the Era of LLMs ( https://llmforplanning.github.io )

Learn from Global Correlations: Enhancing Evolutionary Algorithm via Spectral GNN

从全局相关性学习：通过谱图神经网络增强进化算法

Kaichen Ouyang, Zong Ke, Shengwei Fu, Lingjie Liu, Puning Zhao, Dayu Hu

AI总结 GNE通过谱图神经网络增强进化算法，利用频率过滤策略提升全局信息利用，解决探索与利用平衡问题，实现更高效的优化性能。

Comments Accepted by the 40th Annual AAAI Conference on Artificial Intelligence

Inductive Generative Recommendation via Retrieval-based Speculation

基于检索的生成推荐

Yijie Ding, Jiacheng Li, Julian McAuley, Yupeng Hou

AI总结 SpecGR通过引入检索生成机制，提升生成推荐模型在归纳设置下的推荐能力与整体性能。

Comments Accepted to AAAI 2026 (oral)

PriorDrive: Enhancing Online HD Mapping with Unified Vector Priors

PriorDrive: 通过统一向量先验增强在线高精度地图

Shuang Zeng, Xinyuan Chang, Xinran Liu, Yujian Yuan, Shiyi Liang, Zheng Pan, Mu Xu, Xing Wei

机构 * Shuang Zeng 1,2*† , Xinyuan Chang 2† , Xinran Liu 2 , Yujian Yuan 3 , Shiyi Liang 1,2 , Zheng Pan 2 , Mu Xu 2 , Xing Wei 1‡（未知） 纠错

AI总结 PriorDrive通过整合多种向量化先验地图，提升在线高精度地图构建的鲁棒性和准确性。

Comments AAAI 2026; Code: https://github.com/MIV-XJTU/PriorDrive

TrackGS: Optimizing COLMAP-Free 3D Gaussian Splatting with Global Track Constraints

TrackGS: 通过全局轨迹约束优化无COLMAP的3D高斯点分布

Dongbo Shi, Shen Cao, Lubin Fan, Bojian Wu, Jinhui Guo, Ligang Liu, Renjie Chen

AI总结 TrackGS通过引入全局轨迹约束优化3D高斯点分布，实现无COLMAP的视点合成，提升渲染质量和降低姿态误差。

Journal ref AAAI 2026

Hallucinate Less by Thinking More: Aspect-Based Causal Abstention for Large Language Models

通过更多思考减少幻觉：基于方面的因果回避用于大语言模型

Vy Nguyen, Ziqi Xu, Jeffrey Chan, Estrid He, Feng Xia, Xiuzhen Zhang

AI总结 本文提出基于方面的因果回避方法，通过分析LLM知识的内部多样性，实现早期回避以减少幻觉。

Comments Accepted to AAAI 2026 (Main Technical Track)

DiffRefiner: Coarse to Fine Trajectory Planning via Diffusion Refinement with Semantic Interaction for End to End Autonomous Driving

DiffRefiner: 通过扩散细化与语义交互实现从粗到细的轨迹规划用于端到端自动驾驶

Liuhan Yin, Runkun Ju, Guodong Guo, Erkang Cheng

机构 * Nullmax 纠错

AI总结 DiffRefiner通过两阶段框架结合扩散细化与语义交互，实现更精确的端到端自动驾驶轨迹规划，取得新纪录。

Comments Accepted to AAAI 2026

OmniPT: Unleashing the Potential of Large Vision Language Models for Pedestrian Tracking and Understanding

OmniPT：释放大型视觉语言模型在行人跟踪与理解中的潜力

Teng Fu, Mengyang Zhao, Ke Niu, Kaixin Peng, Bin Li

AI总结 OmniPT 通过结合强化学习和微调方法，提升大型视觉语言模型在行人跟踪和语义理解中的性能。

Comments AAAI 2026