AAAI

Vehicle-centric Perception via Multimodal Structured Pre-training

基于多模态结构预训练的车辆感知

Wentao Wu, Xiao Wang, Chenglong Li, Jin Tang, Bin Luo

机构 * Information Materials and Intelligent Sensing Laboratory of Anhui Province（安徽省信息材料与智能感知实验室） ； Anhui Provincial Key Laboratory of Multimodal Cognitive Computation（安徽省多模态认知计算重点实验室） ； the School of Artificial Intelligence, Anhui University（安徽大学人工智能学院） ； School of Computer Science and Technology, Anhui University（安徽大学计算机科学与技术学院） ； Institute of Artificial Intelligence, Hefei Comprehensive National Science Center（合肥综合性国家科学中心人工智能研究院） 纠错

AI总结 本文提出VehicleMAE-V2，通过多模态结构先验知识提升车辆感知的预训练能力，采用SMM、CRM和SRM模块增强模型对车辆结构和语义的理解。

Comments Journal extension of VehicleMAE (AAAI 2024)

How Many Experts Are Enough? Towards Optimal Semantic Specialization for Mixture-of-Experts

需要多少专家？迈向混合专家架构的最优语义专业化

Sumin Park, Noseong Park

AI总结 MASS通过自适应语义专业化和动态路由策略，优化混合专家架构的语义分化，提升模型性能和领域鲁棒性。

Comments Accepted to AAAI 2026 (Main Track)

HeadHunt-VAD: Hunting Robust Anomaly-Sensitive Heads in MLLM for Tuning-Free Video Anomaly Detection

HeadHunt-VAD: 在MLLM中寻找鲁棒的异常敏感头部以实现无调优视频异常检测

Zhaolin Cai, Fan Li, Ziwei Zheng, Haixia Bi, Lijun He

AI总结 HeadHunt-VAD通过直接在冻结的MLLM中寻找鲁棒的异常敏感头部，实现高效的无调优视频异常检测。

Comments AAAI 2026 Oral

Continuous Vision-Language-Action Co-Learning with Semantic-Physical Alignment for Behavioral Cloning

具有语义-物理对齐的连续视觉-语言-动作共学用于行为克隆

Xiuxiu Qi, Yu Yang, Jiannong Cao, Luyao Bai, Chongshan Fan, Chengtai Cao, Hongpeng Wang

AI总结 本文提出CCoL框架，通过连续共学视觉、语言和动作，解决行为克隆中的语义-物理不一致问题，提升动作执行的准确性和鲁棒性。

Comments Accepted at AAAI 2026, the Project website is available at https://qhemu.github.io/CCoL/

Enhancing Uncertainty Estimation in LLMs with Expectation of Aggregated Internal Belief

通过聚合内部信念期望增强大语言模型的不确定性估计

Zeguan Xiao, Diyang Dou, Boya Xiong, Yun Chen, Guanhua Chen

AI总结 本文提出EAGLE方法，通过聚合内部信念期望提升大语言模型的不确定性估计，改进校准性能。

Comments Accepted by AAAI 2026

Persistent Instability in LLM's Personality Measurements: Effects of Scale, Reasoning, and Conversation History

大语言模型人格测量中的持久不稳定性：规模、推理和对话历史的影响

Tommaso Tosato, Saskia Helbling, Yorguin-Jose Mantilla-Ramos, Mahmood Hegazy, Alberto Tosato, David John Lemay, Irina Rish, Guillaume Dumas

AI总结 研究发现大语言模型在人格测量中存在持续不稳定性，规模扩大、推理和对话历史等干预措施未能有效提升稳定性。

Comments Accepted at AAAI 2026, Track on AI Alignment

C3RL: Rethinking the Combination of Channel-independence and Channel-mixing from Representation Learning

C3RL: 重新思考从表示学习角度的通道独立性和通道混合的结合

Shusen Ma, Yun-Bo Zhao, Yu Kang

AI总结 C3RL通过联合建模通道混合与通道独立策略，提升多变量时间序列预测的性能和泛化能力。

Comments Accepted by AAAI 2026

AI-based Traffic Modeling for Network Security and Privacy: Challenges Ahead

基于AI的网络流量建模用于网络安全与隐私：面临的挑战

Dinil Mon Divakaran

AI总结 本文探讨了基于AI的网络流量分析在网络安全和隐私保护中的应用及面临的挑战。

Comments Accepted at the AAAI 2026 Workshop on Artificial Intelligence for Cyber Security (AICS)

From Points to Coalitions: Hierarchical Contrastive Shapley Values for Prioritizing Data Samples

从点到联盟：用于优先级数据样本的分层对比Shapley值

Canran Xiao, Jiabao Dou, Zhiming Lin, Zong Ke, Liwei Hou

AI总结 HCDV通过分层对比方法高效评估数据样本价值，提升模型准确性并优化计算效率。

Comments AAAI'26 Oral

First-Order Representation Languages for Goal-Conditioned RL

面向目标的强化学习中的第一阶表示语言

Simon Ståhlberg, Hector Geffner

AI总结 本文提出利用第一阶表示语言在面向目标的强化学习中学习通用策略，通过自动创建目标课程提升性能。

Comments In Proceedings of the 40th AAAI Conference on Artificial Intelligence (AAAI 2026)

WorldRFT: Latent World Model Planning with Reinforcement Fine-Tuning for Autonomous Driving

WorldRFT: 通过强化微调的潜在世界模型进行自动驾驶的规划

Pengxuan Yang, Ben Lu, Zhongpu Xia, Chao Han, Yinfeng Gao, Teng Zhang, Kun Zhan, XianPeng Lang, Yupeng Zheng, Qichao Zhang

AI总结 WorldRFT通过强化学习微调提升自动驾驶规划性能，实现安全性和效率的双重优化。

Comments AAAI 2026, first version

Retrieving Objects from 3D Scenes with Box-Guided Open-Vocabulary Instance Segmentation

通过框引导的开放词汇实例分割从3D场景中检索物体

Khanh Nguyen, Dasith de Silva Edirimuni, Ghulam Mubashar Hassan, Ajmal Mian

AI总结 本文提出了一种基于2D开放词汇检测器引导的3D实例分割方法，用于从RGB图像中快速准确检索稀有物体实例。

Comments Accepted to AAAI 2026 Workshop on New Frontiers in Information Retrieval

A Study of Finetuning Video Transformers for Multi-view Geometry Tasks

对多视角几何任务进行视频变换器微调的研究

Huimin Wu, Kwang-Ting Cheng, Stephen Lin, Zhirong Wu

AI总结 本文通过微调视频基础模型，提出了一种简单有效的方法，实现了多视角几何任务如光流估计的高性能表现。

Comments AAAI 20206, Project website: geovit-aaai26.github.io

LLM-CAS: Dynamic Neuron Perturbation for Real-Time Hallucination Correction

LLM-CAS: 动态神经元扰动用于实时幻觉修正

Jensen Zhang, Ningyuan Liu, Yijia Fan, Zihao Huang, Qinglin Zeng, Kaitong Cai, Jian Wang, Keze Wang

机构 * Sun Yat-sen University（中山大学） ； Snap Inc（Snap公司） 纠错

AI总结 LLM-CAS通过动态神经元扰动实现实时幻觉修正，提升大型语言模型的事实准确性。

Comments Accepted at AAAI 2026

Coarse-to-Fine Open-Set Graph Node Classification with Large Language Models

粗到细的开放式图节点分类与大语言模型

Xueqi Ma, Xingjun Ma, Sarah Monazam Erfani, Danilo Mandic, James Bailey

AI总结 本文提出CFC框架，利用大语言模型实现图节点的粗到细开放式分类，提升OOD检测和分类性能。

Comments Accepted to AAAI 2026

BRIC: Bridging Kinematic Plans and Physical Control at Test Time

BRIC: 在测试时弥合运动计划与物理控制之间的差距

Dohun Lim, Minji Kim, Jaewoon Lim, Sungchan Kim

AI总结 BRIC通过结合测试时适应与轻量级引导机制，实现长期人体运动生成的物理合理性和一致性。

Comments Accepted to AAAI'26

HVAdam: A Full-Dimension Adaptive Optimizer

HVAdam：一种全维度自适应优化器

Yiheng Zhang, Shaowu Wu, Yuanzhuo Xu, Jiajun Wu, Shang Xu, Steve Drew, Xiaoguang Niu

AI总结 HVAdam提出一种具有连续可调自适应性的新型优化器，通过增量延迟更新机制提升收敛性与鲁棒性，在多种任务中超越现有优化器。

Comments Accepted at AAAI2025

Journal ref In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pp. 22623-22631, 2025

GateRA: Token-Aware Modulation for Parameter-Efficient Fine-Tuning

GateRA：基于令牌的调制用于参数高效微调

Jie Ou, Shuaihong Jiang, Yingjun Du, Cees G. M. Snoek

AI总结 GateRA通过引入令牌感知调制，实现参数高效微调中的动态更新调整，提升模型适应性和效率。

Comments Accepted by AAAI 2026

Journal ref AAAI 2026

Towards 3D Object-Centric Feature Learning for Semantic Scene Completion

面向语义场景补全的3D物体感知特征学习

Weihua Wang, Yubo Cui, Xiangru Lin, Zhiheng Li, Zheng Fang

AI总结 本文提出Ocean框架，通过分解场景为独立物体实例，提升语义场景补全的精度和性能。

Comments Accepted to AAAI-2026

BiCA: Effective Biomedical Dense Retrieval with Citation-Aware Hard Negatives

BiCA: 基于引用感知的生物医学密集检索

Aarush Sinha, Pavan Kumar S, Roshan Balaji, Nirav Pravinbhai Bhatt

机构 * University of Copenhagen（哥本哈根大学） 纠错

AI总结 BiCA通过利用引用链接生成引用感知硬负样本，提升生物医学领域密集检索性能，实现零样本检索和长尾主题的优越表现。

Comments Accepted for oral presentation at AAAI 2026

Preserving Topological and Geometric Embeddings for Point Cloud Recovery

保持拓扑和几何嵌入以恢复点云

Kaiyue Zhou, Zelong Tan, Hongxiao Wang, Ya-Li Li, Shengjin Wang

AI总结 本文提出TopGeoFormer架构，通过融合拓扑和几何嵌入提升点云恢复性能。

Comments Accepted to AAAI 202

Temporal Causal Reasoning with (Non-Recursive) Structural Equation Models

基于（非递归）结构方程模型的时序因果推理

Maksim Gladyshev, Natasha Alechina, Mehdi Dastani, Dragan Doder, Brian Logan

AI总结 本文提出了一种基于非递归结构方程模型的时序因果推理方法，引入CPLTL逻辑以处理反馈回路和相互依赖过程，并展示了其高效的模型检查能力。

Comments This is an extended version of the same title paper published in the proceeding of AAAI-25 conference (Gladyshev et al. 2025). This version contains an additional section, in which we introduce TSEMs with arbitrary long temporal delays between causal dependencies and prove that these models are equivalent to TSEMs with 1-step delays only introduced in the original version of the paper

Structured Language Generation Model: Loss Calibration and Formatted Decoding for Robust Structure Prediction and Knowledge Retrieval

结构语言生成模型：损失校准与格式解码以实现稳健的结构预测与知识检索

Minho Lee, Junghyun Min, Yerang Kim, Woochul Lee, Yeonsoo Lee

机构 * KT Gen AI Lab（KT生成人工智能实验室） ； Georgetown University（乔治城大学） ； Korea University（韩国大学） ； NC AI（NC人工智能） 纠错

AI总结 SLGM通过损失校准和格式解码提升结构预测和知识检索性能，无需额外参数或数据集特定工程。

Comments 20 pages, 4 figures. FrontierIR at AAAI 2026

Enhancing Medical Large Vision-Language Models via Alignment Distillation

通过对齐蒸馏增强医学大视觉-语言模型

Aofei Chang, Ting Wang, Fenglong Ma

AI总结 本文提出MEDALIGN方法，通过蒸馏CLIP模型的知识提升医学大视觉-语言模型的视觉对齐和生成质量。

Comments Accepted to AAAI'2026 (Main track)

Intelligent Human-Machine Partnership for Manufacturing: Enhancing Warehouse Planning through Simulation-Driven Knowledge Graphs and LLM Collaboration

智能人机协作制造：通过仿真驱动的知识图谱与大语言模型协作提升仓库规划

Himabindu Thogaru, Saisubramaniam Gopalakrishnan, Zishan Ahmad, Anirudh Deodhar

AI总结 本研究提出通过仿真驱动的知识图谱与大语言模型协作，提升制造业仓库规划的智能化水平，实现人机高效协作以优化运营决策。

Comments 13 pages, 2 figures, accepted for oral presentation at AAAI Human Machine Collaboration Workshop 2026

ReGal: A First Look at PPO-based Legal AI for Judgment Prediction and Summarization in India

ReGal：基于PPO的法律AI在印度判决预测和摘要中的初步探索

Shubham Kumar Nigam, Tanuj Tyagi, Siddharth Shukla, Aditya Kumar Guru, Balaramamahanthi Deepak Patnaik, Danush Khanna, Noel Shallum, Kripabandhu Ghosh, Arnab Bhattacharya

AI总结 ReGal通过PPO结合多任务指令微调与强化学习，探索法律AI在印度判决预测和摘要中的应用，揭示了强化学习在法律文本中的挑战与潜力。

Comments Accepted in AILaw @ AAAI 2026 conference

Seeing Justice Clearly: Handwritten Legal Document Translation with OCR and Vision-Language Models

清晰地看见正义：结合OCR和视觉-语言模型的 handwritten 法律文件翻译

Shubham Kumar Nigam, Parjanya Aditya Shukla, Noel Shallum, Arnab Bhattacharya

AI总结 本文提出利用视觉大语言模型统一OCR与机器翻译流程，以提升低资源环境下手写法律文件的翻译效率和准确性。

Comments Accepted in AILaw @ AAAI 2026 Conference

Let the Model Learn to Feel: Mode-Guided Tonality Injection for Symbolic Music Emotion Recognition

让模型学会感受：基于模式的音调注入用于符号音乐情感识别

Haiying Xia, Zhongyi Huang, Yumei Tan, Shuxiang Song

AI总结 本文提出模式引导的音调注入方法，通过引入心理模式见解提升符号音乐情感识别的准确率。

Comments Accepted by AAAI 2026

SSR: Semantic and Spatial Rectification for CLIP-based Weakly Supervised Segmentation

SSR：基于CLIP的弱监督分割的语义与空间校正

Xiuli Bi, Die Xiao, Junchao Fan, Bin Xiao

AI总结 本文提出SSR方法，通过语义与空间校正提升CLIP在弱监督分割中的性能，实现更高的mIoU分数。

Comments Accepted in AAAI 2026

FreeAskWorld: An Interactive and Closed-Loop Simulator for Human-Centric Embodied AI

FreeAskWorld: 一种面向人类的交互式闭环模拟器用于具身人工智能

Yuhang Peng, Yizhou Pan, Xinning He, Jihaoyu Yang, Xinyu Yin, Han Wang, Xiaoji Zheng, Chao Gao, Jiangtao Gong

AI总结 FreeAskWorld通过整合大语言模型和交互式闭环模拟，提升具身人工智能在复杂社会行为和自然交互中的表现。

Comments 9 pages, 4 figures

Journal ref AAAI 2026 Oral