AAAI

Neighbor-aware Instance Refining with Noisy Labels for Cross-Modal Retrieval

具有噪声标签的邻居感知实例细化用于跨模态检索

Yizhi Liu, Ruitao Pu, Shilin Xu, Yingke Chen, Quan-Hui Liu, Yuan Sun

AI总结 本文提出了一种新的跨模态学习框架NIRNL，通过引入噪声标签和邻居感知机制，提升模型在高噪声环境下的检索性能。

Comments 9 pages, 4 figures, and AAAI-26 conference

Beyond Hallucinations: A Composite Score for Measuring Reliability in Open-Source Large Language Models

超越幻觉：一种衡量开源大语言模型可靠性的综合评分

Rohit Kumar Salla, Manoj Saravanan, Shrikar Reddy Kota

AI总结 本文提出综合可靠性评分（CRS）用于评估开源大语言模型的可靠性，通过实验发现最可靠的系统在准确率、鲁棒性和校准不确定性间取得平衡。

Comments 5 pages, 4 tables, accepted at AAAI 2026

CEC-Zero: Zero-Supervision Character Error Correction with Self-Generated Rewards

CEC-Zero: 无监督字符错误纠正与自生成奖励

Zhiming Lin, Kai Zhao, Sophie Zhang, Peilai Yu, Canran Xiao

AI总结 CEC-Zero通过无监督强化学习框架，利用自动生成的奖励在无需标注的情况下提升大语言模型的字符错误纠正能力，实现了在多个基准测试中的性能提升。

Comments AAAI'26 poster

GaussianImage++: Boosted Image Representation and Compression with 2D Gaussian Splatting

GaussianImage++: 通过2D高斯散射提升图像表示与压缩

Tiantian Li, Xinjie Zhang, Xingtong Ge, Tongda Xu, Dailan He, Jun Zhang, Yan Wang

AI总结 GaussianImage++通过有限的高斯原始体实现高效的图像表示与压缩，结合驱动扭曲的密集化机制和上下文感知滤波器，提升性能并降低内存消耗。

Comments Accepted to AAAI 2026. Code URL:https://github.com/Sweethyh/GaussianImage_plus.git

FDP: A Frequency-Decomposition Preprocessing Pipeline for Unsupervised Anomaly Detection in Brain MRI

FDP：一种用于脑部MRI无监督异常检测的频率分解预处理流程

Hao Li, Zhenfeng Zhuang, Jingyu Lin, Yu Liu, Yifei Chen, Qiong Peng, Lequan Yu, Liansheng Wang

AI总结 FDP通过频域分解提升脑部MRI无监督异常检测性能，实现病灶抑制与解剖保持的平衡。

Comments Accepted by AAAI 2026

Open-sci-ref-0.01: open and reproducible reference baselines for language model and dataset comparison

Open-sci-ref-0.01: 开放和可重复的参考基线用于语言模型和数据集比较

Marianna Nezhurina, Jörg Franke, Taishi Nakamura, Timur Carstensen, Niccolò Ajroldi, Ville Komulainen, David Salinas, Jenia Jitsev

AI总结 open-sci-ref通过开放和可重复的参考基线，为语言模型和数据集的比较提供标准化评估框架，支持训练过程的比较与未来研究。

Comments v.1.1. AAAI Workshop on Reproducible Artificial Intelligence (RAI, https://reproducibleai.github.io) 2026, camera ready version. Model weights and intermediate training checkpoints are available at https://huggingface.co/collections/open-sci/open-sci-ref-001; code for reproducing training, evaluation and raw experiments data at https://github.com/LAION-AI/open-sci-ref-0.01

RAST: A Retrieval Augmented Spatio-Temporal Framework for Traffic Prediction

RAST：一种用于交通预测的检索增强时空框架

Weilin Ruan, Xilin Dang, Ziyu Zhou, Sisuo Lyu, Yuxuan Liang

AI总结 RAST通过整合检索增强机制与时空建模，解决交通预测中的上下文容量和细粒度预测性挑战，实现高效且准确的交通预测。

Comments Accepted by AAAI 2026 (AI for Social Impact)

Generalising Traffic Forecasting to Regions without Traffic Observations

面向无交通观测区域的交通预测泛化

Xinyu Su, Majid Sarvi, Feng Liu, Egemen Tanin, Jianzhong Qi

机构 * The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州）） 纠错

AI总结 本文提出GenCast模型，通过整合物理信息神经网络和外部信号学习模块，提升无交通观测区域的交通预测泛化能力。

Comments Accepted by AAAI 2026

SplatSSC: Decoupled Depth-Guided Gaussian Splatting for Semantic Scene Completion

SplatSSC：解耦深度引导的高斯点云法用于语义场景补全

Rui Qian, Haozhi Cao, Tianchen Deng, Shenghai Yuan, Lihua Xie

AI总结 SplatSSC通过深度引导的初始化策略和解耦高斯聚合器，提升语义场景补全的效率与鲁棒性，实现性能和资源消耗的优化。

Comments Accepted for oral presentation in The 40th Annual AAAI Conference on Artificial Intelligence (AAAI 2026)

Learning Network Dismantling Without Handcrafted Inputs

无需人工输入的学习网络拆解

Haozhe Tian, Pietro Ferraro, Robert Shorten, Mahdi Jalili, Homayoun Hamedmoghadam

AI总结 MIND通过消息迭代机制和合成网络训练，无需人工输入高效解决网络拆解问题，适用于大规模真实网络。

Comments Accepted for Oral Presentation at the 40th AAAI Conference on Artificial Intelligence (AAAI-26), Main Technical Track

An Analysis of Hyper-Parameter Optimization Methods for Retrieval Augmented Generation

检索增强生成方法的超参数优化分析

Matan Orbach, Ohad Eytan, Benjamin Sznajder, Ariel Gera, Odellia Boni, Yoav Kantor, Gal Bloch, Omri Levy, Hadas Abraham, Nitzan Barzilay, Eyal Shnarch, Michael E. Factor, Shila Ofek-Koifman, Paula Ta-Shma, Assaf Toledo

AI总结 本文分析了RAG超参数优化方法，通过五种算法和五种数据集的实验，发现贪婪或随机搜索能高效提升RAG性能，且优先优化模型选择优于传统流程顺序。

Comments AAAI 2026 Workshop on New Frontiers in Information Retrieval. For associated results, see https://github.com/IBM/rag-hpo-bench

HINTS: Extraction of Human Insights from Time-Series Without External Sources

HINTS: 从时间序列中提取人类洞察力而不依赖外部数据源

Sheo Yon Jhin, Noseong Park

机构 * Sheo Yon Jhin, Noseong Park 纠错

AI总结 HINTS通过自监督学习从时间序列残差中提取人类洞察力，无需外部数据源，提升预测准确性并增强可解释性。

Comments AAAI 2026 AI4TS Workshop paper

BCWildfire: A Long-term Multi-factor Dataset and Deep Learning Benchmark for Boreal Wildfire Risk Prediction

BCWildfire: 一种长期多因素数据集和深度学习基准，用于北极火灾风险预测

Zhengsen Xu, Sibo Cheng, Lanying Wang, Hongjie He, Wentao Sun, Jonathan Li, Lincoln Linlin Xu

AI总结 BCWildfire数据集和基准为长期多因素火灾风险预测提供了深度学习评估框架，包含38个协变量和多种模型评估。

Comments This paper has been accepted by AAAI-26

ParetoHqD: Fast Offline Multiobjective Alignment of Large Language Models using Pareto High-quality Data

ParetoHqD: 利用帕累托高质量数据快速实现大语言模型的多目标对齐

Haoran Gu, Handing Wang, Yi Mei, Mengjie Zhang, Yaochu Jin

AI总结 ParetoHqD通过利用帕累托高质量数据实现大语言模型的多目标对齐，提升对齐效果和效率。

Comments Accepted as a main conference paper at AAAI 2026

Rethinking the Spatio-Temporal Alignment of End-to-End 3D Perception

重新思考端到端3D感知的时空对齐

Xiaoyu Li, Peidong Li, Xian Wu, Long Shi, Dedong Liu, Yitao Wu, Jiajia Fu, Dixiao Cui, Lijun Zhao, Lining Sun

AI总结 HAT通过自适应解码多假设生成最优时空对齐提案，提升自动驾驶中3D感知精度和鲁棒性。

Comments Accepted to AAAI 2026

FRoD: Full-Rank Efficient Fine-Tuning with Rotational Degrees for Fast Convergence

FRoD：全秩高效微调与旋转自由度以实现快速收敛

Guoan Wan, Tianyu Chen, Fangzheng Feng, Haoyi Zhou, Runhua Xu

AI总结 FRoD通过结合层次联合分解与旋转自由度，实现高效全秩微调，提升收敛速度和鲁棒性，同时在多个基准测试中保持与全模型微调相当的准确性。

Comments The 40th Annual AAAI Conference on Artificial Intelligence

EquaCode: A Multi-Strategy Jailbreak Approach for Large Language Models via Equation Solving and Code Completion

EquaCode：通过方程求解和代码补全的多策略对抗大语言模型

Zhen Liang, Hai Huang, Zhengkui Chen

机构 * School of Computer Science and Technology, Zhejiang Sci-Tech University, Hangzhou, China（计算机科学与技术学院，浙江科技学院，杭州，中国） 纠错

AI总结 EquaCode通过方程求解和代码补全提出多策略对抗方法，有效提升大语言模型的鲁棒性。

Comments This is a preprint. A revised version will appear in the Proceedings of AAAI 2026

Understanding Syllogistic Reasoning in LLMs from Formal and Natural Language Perspectives

从形式和自然语言角度理解大语言模型中的三段论推理

Aheli Poddar, Saptarshi Sahoo, Sujata Ghosh

AI总结 本文从形式和自然语言角度研究大语言模型的三段论推理能力，通过测试14种模型的符号推理和自然语言理解，探讨大语言模型是否正向形式化推理机制发展。

Comments 9 pages, 4 figures, 5 tables. Accepted at AAAI 2026 Bridge Program on Logic & AI. Code available at https://github.com/XAheli/Logic-in-LLMs

Stable Voting and the Splitting of Cycles

稳定投票与循环的分裂

Wesley H. Holliday, Milan Mossé, Chase Norman, Eric Pacuit, Cynthia Wang

AI总结 本文证明了在最多6个选项的情况下，简单稳定投票是Split Cycle的改进，但在超过6个选项时推翻了这一猜想。

Comments Final version forthcoming in Proceedings of the 40th Annual AAAI Conference on Artificial Intelligence (AAAI 2026)

RefineVAD: Semantic-Guided Feature Recalibration for Weakly Supervised Video Anomaly Detection

RefineVAD: 语义引导的特征重校准用于弱监督视频异常检测

Junhee Lee, ChaeBeen Bang, MyoungChul Kim, MyeongAh Cho

AI总结 RefineVAD通过结合时间动态和语义结构，利用双过程推理提升弱监督视频异常检测的性能。

Comments Accepted to AAAI 2026

Expressive Temporal Specifications for Reward Monitoring

用于奖励监控的表达性时间规范

Omar Adalat, Francesco Belardinelli

AI总结 本文提出基于定量线性时间逻辑的奖励监控方法，用于提升强化学习中奖励密度和训练效率。

Comments Accepted at AAAI-26

Beyond Fixed Tasks: Unsupervised Environment Design for Task-Level Pairs

超越固定任务：用于任务级配对的无监督环境设计

Daniel Furelos-Blanco, Charles Pert, Frederik Kelbel, Alex F. Spies, Alessandra Russo, Michael Dennis

AI总结 ATLAS通过自动生成任务与环境的联合课程，提升复杂任务在强化学习中的训练效果。

Comments Extended version of paper accepted for publication at the 40th AAAI Conference on Artificial Intelligence (AAAI)

Learning Spatial Decay for Vision Transformers

学习空间衰减以用于视觉变换器

Yuxin Mao, Zhen Qin, Jinxing Zhou, Bin Fan, Jing Zhang, Yiran Zhong, Yuchao Dai

AI总结 本文提出空间衰减变换器（SDT），通过引入上下文感知门控机制，学习动态数据依赖的空间衰减，提升视觉变换器在空间结构化任务中的性能。

Comments AAAI 2026

Forecasting Clinical Risk from Textual Time Series: Structuring Narratives for Temporal AI in Healthcare

从文本时间序列预测临床风险：为医疗健康领域的时序AI构建叙述

Shahriar Noroozizadeh, Sayantan Kumar, Jeremy C. Weiss

AI总结 本文提出从文本时间序列预测临床风险的方法，通过构建时间有序的临床数据，提升时序AI在医疗健康领域的应用效果。

Comments AAAI AI for Social Impact 2026. Shahriar Noroozizadeh, Sayantan Kumar (authors contributed equally)

Text-Routed Sparse Mixture-of-Experts Model with Explanation and Temporal Alignment for Multi-Modal Sentiment Analysis

基于解释和时序对齐的文本引导稀疏专家混合模型用于多模态情感分析

Dongning Rao, Yunbiao Zeng, Zhihua Jiang, Jujian Lv

AI总结 本文提出基于解释和时序对齐的文本引导稀疏专家混合模型，用于提升多模态情感分析的性能。

Comments 9 pages, 9 figures, accepted by AAAI 2026

Hierarchical Pedagogical Oversight: A Multi-Agent Adversarial Framework for Reliable AI Tutoring

层级教学监督：一种多智能体对抗框架用于可靠的AI辅导

Saisab Sadhu, Ashim Dhor

机构 * AAAI 2026 EGSAI Community Activity（AAAI 2026 EGSAI 社区活动） 纠错

AI总结 本文提出层级教学监督框架，通过多智能体对抗机制提升AI辅导的可靠性与效率，实验结果显示其在资源受限环境下表现优异。

Comments Accepted for presentation at the AAAI 2026 EGSAI Community Activity (AAAI 2026)

MoFu: Scale-Aware Modulation and Fourier Fusion for Multi-Subject Video Generation

MoFu: 多主体视频生成的尺度感知调制与傅里叶融合

Run Ling, Ke Cao, Jian Lu, Ao Ma, Haowei Liu, Runze He, Changwei Wang, Rongtao Xu, Yihua Shao, Zhanjie Zhang, Peng Wu, Guibing Guo, Wei Feng, Zheng Zhang, Jingjing Lv, Junjie Shen, Ching Law, Xingwei Wang

AI总结 MoFu通过尺度感知调制和傅里叶融合解决多主体视频生成中的尺度不一致和排列敏感性问题，提升生成质量。

Comments AAAI 2026

LLA: Enhancing Security and Privacy for Generative Models with Logic-Locked Accelerators

LLA：利用逻辑锁定加速器增强生成模型的安全性和隐私性

You Li, Guannan Zhao, Yuhao Ju, Yunqi He, Jie Gu, Hai Zhou

AI总结 LLA通过软硬件协同增强生成模型的安全性和隐私性，利用逻辑锁定加速器保护知识产权，有效防御供应链威胁并保持低计算开销。

Comments Accepted by AAAI'26 as a conference paper and selected for oral presentation

With Great Capabilities Come Great Responsibilities: Introducing the Agentic Risk & Capability Framework for Governing Agentic AI Systems

能力越大责任越大：引入代理风险与能力框架以治理代理AI系统

Shaun Khoo, Jessica Foo, Roy Ka-Wei Lee

机构 * GovTech Singapore（新加坡政府科技局） ； Singapore University of Technology and Design（新加坡科技与设计大学） 纠错

AI总结 本文提出代理风险与能力框架，用于治理代理AI系统，通过能力中心视角和风险来源分析，帮助组织有效管理AI系统风险。

Comments Accepted at IASEAI 2026 (Main Track) and AAAI 2026 3rd International AI Governance Workshop

Emergent Persuasion: Will LLMs Persuade Without Being Prompted?

涌现说服：大型语言模型是否会无提示地说服？

Vincent Chang, Thee Ho, Sunishchal Dev, Kevin Zhu, Shi Feng, Kellin Pelrine, Matthew Kowal

AI总结 本文研究了大型语言模型在无提示情况下说服人类的机制，发现监督微调（SFT）能提高模型在争议性话题上的说服倾向，表明需进一步研究无提示说服的风险。

Comments This paper was accepted to AAAI 2026 AIGOV Workshop