AAAI

Bipartite Mode Matching for Vision Training Set Search from a Hierarchical Data Server

双域模式匹配用于从分层数据服务器中寻找视觉训练集

Yue Yao, Ruining Yang, Tom Gedeon

AI总结 本文提出双域模式匹配算法，通过分层数据服务器优化训练集搜索，提升模型在目标领域上的准确性和领域适应性。

Comments Accepted to AAAI 2026

Improving Symbolic Translation of Language Models for Logical Reasoning

提升语言模型的符号翻译以增强逻辑推理

Ramya Keerthy Thatikonda, Jiuzhou Han, Wray Buntine, Ehsan Shareghi

AI总结 本文通过微调和增量推理框架，提升小型语言模型在逻辑推理中的符号翻译能力，减少错误率并提高推理性能。

Comments The Third workshop of NeusymBridge @AAAI 2026 (Bridging Neurons and Symbols for NLP and Knowledge Graph Reasoning)

ReflexDiffusion: Reflection-Enhanced Trajectory Planning for High-lateral-acceleration Scenarios in Autonomous Driving

ReflexDiffusion：面向自动驾驶高侧向加速度场景的反射增强轨迹规划

Xuemei Yao, Xiao Yang, Jianbin Sun, Liuwei Xie, Xuebin Shao, Xiyu Fang, Hang Su, Kewei Yang

AI总结 ReflexDiffusion通过反射调整机制提升自动驾驶在高侧向加速度场景下的轨迹规划安全性与可靠性。

Comments Accepted by AAAI 2026

Blue Teaming Function-Calling Agents

蓝队功能调用代理

Greta Dolcetti, Giulio Zizzo, Sergio Maffeis

AI总结 该研究评估了四种开源LLM在功能调用能力下的鲁棒性，并测试了八种防御措施的有效性，发现模型存在安全隐患，防御措施尚不成熟。

Comments This work has been accepted to appear at the AAAI 2026 Workshop on Trust and Control in Agentic AI (TrustAgent)

Annealed Relaxation of Speculative Decoding for Faster Autoregressive Image Generation

退火放松的投机解码用于更快的自回归图像生成

Xingyao Li, Fengzhuo Zhang, Cunxiao Du, Hui Ji

AI总结 本文提出COOL-SD，通过退火放松的投机解码方法提升自回归图像生成的速度与质量。

Comments Accepted to AAAI 2026

Pairing-free Group-level Knowledge Distillation for Robust Gastrointestinal Lesion Classification in White-Light Endoscopy

无需配对的群体级知识蒸馏用于白光内镜中胃肠道病变的鲁棒分类

Qiang Hu, Qimei Wang, Yingjie Guo, Qiang Li, Zhiwei Wang

AI总结 本文提出PaGKD，一种无需配对的群体级知识蒸馏框架，通过未配对的WLI和NBI数据实现跨模态学习，显著提升胃肠道病变分类性能。

Comments Accepted to AAAI 2026

Position on LLM-Assisted Peer Review: Addressing Reviewer Gap through Mentoring and Feedback

LLM辅助同行评审的位置：通过指导与反馈解决评审缺口

JungMin Yun, JuneHyoung Kwon, MiHyeon Kim, YoungBin Kim

AI总结 本文提出通过LLM辅助指导和反馈系统解决评审缺口问题，旨在提升评审质量并促进学术生态可持续发展。

Comments Accepted to AAAI 2026 Workshop on AI for Scientific Research (AI4Research)

KTCF: Actionable Recourse in Knowledge Tracing via Counterfactual Explanations for Education

KTCF：通过反事实解释为教育中的知识追踪提供可操作的救济

Woojin Kim, Changkwon Lee, Hyeoncheol Kim

AI总结 KTCF通过反事实解释为教育中的知识追踪提供可操作的救济，提升教育AI的实用性和责任性。

Comments Accepted to AAAI-26 Special Track AI for Social Impact (oral presentation)

Small but Mighty: Dynamic Wavelet Expert-Guided Fine-Tuning of Large-Scale Models for Optical Remote Sensing Object Segmentation

小而强大：动态小波专家引导的大型模型微调用于光学遥感目标分割

Yanguang Sun, Chao Wang, Jian Yang, Lei Luo

AI总结 本文提出WEFT方法，通过动态小波专家引导微调，提升光学遥感图像目标分割性能，优于21种SOTA方法。

Comments Accepted at AAAI 2026

Gaming the Answer Matcher: Examining the Impact of Text Manipulation on Automated Judgment

利用答案匹配器：考察文本操纵对自动判断的影响

Manas Khatore, Sumana Sridharan, Kevork Sulahian, Benjamin J. Smith, Shi Feng

AI总结 研究通过测试文本操纵对自动答案匹配的影响，发现其对模型评分鲁棒，且在有参考答案时可作为替代评估方法。

Comments Accepted to the AAAI 2026 Workshop on AI Governance (AIGOV)

BeDKD: Backdoor Defense Based on Directional Mapping Module and Adversarial Knowledge Distillation

BeDKD: 基于方向映射模块和对抗知识蒸馏的后门防御

Zhengxian Wu, Juan Wen, Wanli Peng, Yinghan Zhou, Changtong dou, Yiming Xue

AI总结 BeDKD通过方向映射模块和对抗知识蒸馏有效防御后门攻击，显著降低攻击成功率。

Journal ref AAAI 2026

Divergence-Based Similarity Function for Multi-View Contrastive Learning

基于分歧的多视图对比学习相似度函数

Jae Hyoung Jeon, Cheolsu Lim, Myungjoo Kang

机构 * Department of Mathematics, Seoul National University（首尔国立大学数学系） ； Research Institute of Mathematics, Seoul National University（首尔国立大学数学研究所） 纠错

AI总结 本文提出基于分歧的多视图对比学习相似度函数，通过分布分歧建模多视图联合结构，提升多种任务性能并提高效率。

Comments 9 pages, 5 figures. Code and Pretrained Model: https://github.com/Jeon789/DSF, v4: Removed erroneous AAAI copyright notice

Quiet Feature Learning in Algorithmic Tasks

算法任务中的安静特征学习

Prudhviraj Naidu, Zixian Wang, Leon Bergen, Ramamohan Paturi

AI总结 该研究发现，在算法任务中，模型在损失曲线平坦区域学习了对任务性能至关重要的中间特征，挑战了传统损失作为学习代理的假设。

Comments Accepted as Oral presentation @ AAAI 2026 Special Track on AI Alignment

"Hiding in Plain Sight": Designing Synthetic Dialog Generation for Uncovering Socially Situated Norms

隐于 plain sight：设计合成对话生成以揭示社会情境规范

Chengfei Wu, Dan Goldwasser

AI总结 本文提出 NormHint 数据集，通过多步骤框架生成合成对话，自动揭示社会规范，提升模型对对话中潜在规范违规的检测能力。

Comments Accepted at AAAI Workshop on Shaping Responsible Synthetic Data in the Era of Foundation Models

Can Editing LLMs Inject Harm?

能否通过编辑LLM注入危害？

Canyu Chen, Baixiang Huang, Zekun Li, Zhaorun Chen, Shiyang Lai, Xiongxiao Xu, Jia-Chen Gu, Jindong Gu, Huaxiu Yao, Chaowei Xiao, Xifeng Yan, William Yang Wang, Philip Torr, Dawn Song, Kai Shu

机构 * Northwestern University（西北大学） 纠错

AI总结 本文提出编辑攻击作为LLM安全威胁，揭示了通过编辑注入虚假信息和偏见的风险及高隐蔽性。

Comments Accepted to Proceedings of AAAI 2026. The first two authors contributed equally. 7 pages for main paper, 31 pages including appendix. The code, results, dataset for this paper and more resources are on the project website: https://llm-editing.github.io