AAAI 2026 Papers — Page 38
AAAI Conference on Artificial Intelligence · 4149 papers
The Correspondence Between Bounded Graph Neural Networks and Fragments of First-Order Logic
Bernardo Cuenca Grau (University of Oxford), Przemysław Andrzej Wałęga
Explainability and InterpretabilityRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Formally study bounded graph neural networks (bounded GNN), proving their expressive power exactly corresponds to multiple fragments of first-order logic (e.g., modal logic, counting modal logic, and two-variable first-order logic).
The Curious Case of Analogies: Investigating Analogical Reasoning in Large Language Models
Taewhoo Lee (Korea University), Jaewoo Kang (Korea University)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Analyze the internal mechanisms of large language models (LLMs) in proportion analogy and story analogy, revealing how they encode, recognize, and apply relationships.
The Emotional Baby Is Truly Deadly: Does Your Multimodal Large Reasoning Model Have Emotional Flattery Towards Humans?
Yuan Xun (Institute of Information Engineering, Chinese Academy of Sciences), Hua Zhang (Institute of Information Engineering, Chinese Academy of Sciences)
Safty and PrivacyAdversarial AttackLarge Language ModelAgentic AIPrompt EngineeringMultimodality
🎯 What it does: This paper systematically evaluates the security of multimodal large reasoning models (MLRMs), revealing their vulnerability to emotional prompts during deep reasoning phases, and proposes EmoAgent, an automated adversarial agent that induces safety biases in models by generating emotionally charged prompts.
The Finer the Better: Towards Granular-aware Open-set Domain Generalization
Yunyun Wang (University of Posts and Telecommunications), Songcan Chen (Nanjing University of Aeronautics and Astronautics)
Domain AdaptationTransformerPrompt EngineeringDiffusion modelContrastive LearningImageBenchmark
🎯 What it does: Propose the SeeCLIP framework for fine-grained open-domain generalization (OSDG).
The GATTACA Framework: Graph Neural Network-Based Reinforcement Learning for Controlling Biological Networks
Andrzej Mizera (University of Warsaw), Jakub Zarzycki (University of Warsaw)
Graph Neural NetworkReinforcement LearningBiomedical Data
🎯 What it does: Propose a deep reinforcement learning framework called GATTACA based on graph neural networks for controlling asynchronous Boolean networks to achieve cellular reprogramming;
The Last Byte: Learning Just Enough for Machine-Oriented Image Compression
Wuyuan Xie (Shenzhen University), Miaohui Wang (Shenzhen University)
CompressionConvolutional Neural NetworkImage
🎯 What it does: Proposed a no-reference framework called MVRNet based on machine vision, which can predict the entire frame image's JRD (Just Recognizable Distortion) quantization map in one go, thereby achieving efficient machine-oriented image compression; simultaneously improved the JRD data annotation standards and constructed a large-scale MVRSet dataset;
The Limitations and Power of NP-Oracle Based Functional Synthesis Techniques
Brendan Juba, Kuldeep S. Meel (Washington University in St. Louis)
🎯 What it does: Investigate the theoretical limits of function synthesis techniques and the role of NP oracles, exploring the fundamental limitations of traditional learning and interpolation methods under multi-output specifications;
The Other Mind: How Language Models Exhibit Human Temporal Cognition
Lingyu Li (Shanghai Artificial Intelligence Laboratory), Yingchun Wang (Shanghai Artificial Intelligence Laboratory)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Assessed large language models' (LLM) understanding of years through similarity judgment tasks, revealing that models spontaneously generate a subjective temporal reference point (≈2025) and follow the Weber-Fechner Law (logarithmic compression of time distance) as their scale increases; further analyzed mechanisms of temporal cognition at neural, representational, and information exposure levels through neuron screening, activation patterns, linear probing, and semantic embedding; proposed an 'empiricism' perspective, viewing LLM cognition as a subjective construction of the external world through internal representation systems.
The Power of Initial Investigation in Audit Games
Ren Liu (Renmin University of China), Weiran Shen (Renmin University of China)
OptimizationTabularFinance Related
🎯 What it does: This paper proposes a two-stage audit game model, adding an investigation phase before auditing, using clues obtained from the investigation to guide subsequent resource allocation;
The Publication Choice Problem
Haichuan Wang (Harvard University), Haifeng Xu (University of Chicago)
🎯 What it does: Proposed the Publication Choice Problem game model, analyzing the interaction between researchers' submission strategies to different conferences/journals and venue impact factors, and proved the uniqueness of pure strategy equilibrium and the threshold effect of spotlight marking under binary types.
The River Voting Method
Michelle Döring, Jobst Heitzig (Potsdam Institute for Climate Impact Research)
Graph
🎯 What it does: Proposed the River voting method, constructing a surjective tree rooted at the winner (River graph), sorting edges by weight in the majority edge graph while avoiding cycles and branches, ensuring each candidate has only one incoming edge.
The Semantic Architect: How FEAML Bridges Structured Data and LLMs for Multi-Label Tasks
Wanfu Gao (Jilin University), Jun Gao (Jilin University)
ClassificationExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTabularFinance Related
🎯 What it does: Designed and implemented FEAML, a closed-loop feature engineering framework that automatically generates interpretable features for multi-label learning tasks using large language models.
The Silent Amplifier: In-Context Examples Fuel Bias in Large Language Models
Xinwei Guo (Southern University of Science and Technology), Xuetao Wei (Southern University of Science and Technology)
Explainability and InterpretabilityTransformerLarge Language ModelContrastive LearningText
🎯 What it does: This paper studies the impact of example selection in in-context learning (ICL) of large language models on gender and racial bias, and proposes a Prompt tuning method called ReBE that obtains bias-aware embeddings through contrastive learning to alleviate bias without significantly reducing accuracy;
The Strong Lottery Ticket Hypothesis for Multi-Head Attention Mechanisms
Hikari Otsuka (Institute of Science Tokyo), Masato Motomura (Institute of Science Tokyo)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Prove the existence of strong lottery ticket subnetworks (SLT) in randomly initialized multi-head attention mechanisms (MHA) and Transformers without normalization layers, and provide an upper bound on the approximation error for the target model; experiments verify that the error decreases exponentially with increasing hidden dimensions, is insensitive to input length, and propose an initialization scaling strategy for query/key weights.
The Structure-Equivalent Prior: Unifying Temporal Dynamics and 3D Evolution in 4D Latent Space
Jingyuan Gao (Beijing University of Chemical Technology), Kunfeng Wang (Beijing University of Chemical Technology)
GenerationAutonomous DrivingAuto EncoderVideo
🎯 What it does: Propose a continuous 4D latent space representation (SEP-4D) based on structural equivalence prior for high-quality occupancy map reconstruction in dynamic 3D scenes.
The Visual Prism: Refracting Images into Parallel Multilingual Descriptions with Structured Visual Guidance
Chengpeng Fu (Harbin Institute of Technology), Ting Liu (Harbin Institute of Technology)
GenerationTransformerSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelMultimodality
🎯 What it does: This paper proposes the PRISMS framework, which utilizes visual scene graphs as structured visual guidance to generate multilingual parallel image descriptions, addressing the issue of inconsistent focus across languages.
Themis: Automated Constraint-Aware Test Synthesis Framework for Code Reinforcement Learning
Shengyu Ye, Zhenya Huang (University Of Science And Technology Of China)
Data SynthesisData-Centric LearningLarge Language ModelTextBenchmark
🎯 What it does: Themis automatically generates high-quality test sets that meet constraints to enhance the reliability of reward signals in Code Reinforcement Learning (Code RL).
Theoretical and Empirical Analysis of Lehmer Codes to Search Permutation Spaces with Evolutionary Algorithms
Yuxuan Ma, Carsten Witt (Southern University of Science and Technology)
OptimizationBenchmark
🎯 What it does: Investigate the representation advantages of Lehmer codes (inversion vectors) in evolutionary algorithms, conduct rigorous runtime analysis of RLS and (1+1)-EA in Lehmer space, and compare with classical linear permutation encoding; subsequently validate experimentally on theoretical benchmark functions and real-world linear ordering and quadratic assignment problems.
Thermal-Physics Guided Infrared Image Super-Resolution with Dynamic High-Frequency Amplification
Mingxuan Zhou (Beijing Institute Of Technology), Shuigen Wang (Iray Technology Co Ltd)
Super ResolutionDiffusion modelImagePhysics Related
🎯 What it does: Propose a ThesIS framework specifically designed for infrared image super-resolution, combining thermal physics regularization and dynamic high-frequency enhancement to achieve more accurate thermal radiation distribution and more refined visual details.
THGB: A Comprehensive Benchmark for Text-attributed Heterogeneous Graphs
Lixin Zhou (Zhejiang University), Jing Ying (Singapore Management University)
Representation LearningGraph Neural NetworkLarge Language ModelTextGraphBenchmark
🎯 What it does: Proposed the first publicly available benchmark (THGB) for Text-Attributed Heterogeneous Graphs (TAHG), and evaluated multiple graph learning models on it;
Think Before You Segment: An Object-aware Reasoning Agent for Referring Audio-Visual Segmentation
Jinxing Zhou, Rao Muhammad Anwer (Mohamed Bin Zayed University Of Artificial Intelligence)
SegmentationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper proposes TGS-Agent, which follows a three-step process of 'thinking-localization-segmentation'. It first explicitly reasons about the target object using the multimodal large language model Ref-Thinker, then generates bounding boxes with Grounding-DINO, followed by pixel-level segmentation with SAM2. Meanwhile, the paper constructs a more challenging benchmark called R2-AVSBench.
Think How Your Teammates Think: Active Inference Can Benefit Decentralized Execution
Hao Wu (Beijing Jiaotong University), Kai Lv (Beijing Jiaotong University)
Reinforcement LearningAuto EncoderBenchmark
🎯 What it does: Propose the AIM framework, replacing communication with thinking, constructing three portraits (perception-belief-action) of teammates based on active inference, and adaptively selecting partners through dual filters to achieve communication-free multi-agent collaboration.
Think Then Rewrite: Reasoning Enhanced Query Rewriting for Domain Specific Retrieval
Ang Li (Zhejiang University), Kun Kuang (Ant Group)
RetrievalLarge Language ModelReinforcement LearningContrastive LearningTextBiomedical Data
🎯 What it does: This paper proposes the Think-Then-Rewrite (TTR) framework, which leverages the reasoning capabilities of large language models to enhance query rewriting in specialized domains.
Think Wise, Collaborate Effectively: A Rationale-Aware LLM-Based Recommender with Reinforcement Learning from Collaborative Signals
Chung Park (SK Telecom), Jaegul Choo (Korea Advanced Institute of Science and Technology)
Recommendation SystemTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Propose a recommendation framework TWiCE-Rec that integrates large language models with collaborative filtering. It first generates high-quality reasoning explanations using self-annotated prompting learning, then aligns collaborative signals through multi-task instruction fine-tuning and reinforcement learning, ultimately outputting both recommended items and interpretable reasons during recommendations.
Think-J: Learning to Think for Generative LLM-as-a-Judge
Hui Huang (Alibaba Group), Wenbo Su (Alibaba Group)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: Propose Think-J, which first initializes the generative LLM discriminator using a small set of labeled thinking trajectories, and then optimizes its thinking process through offline/online reinforcement learning to improve preference evaluation of generated text;
Think, Speak, Decide: Language-Augmented Multi-Agent Reinforcement Learning for Economic Decision-Making
Heyang Ma (Chinese Academy of Sciences), Haifeng Zhang (Peking University)
Large Language ModelReinforcement LearningTextFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed the LAMP framework, integrating the reasoning and reflection of large language models (LLMs) with multi-agent reinforcement learning (MARL) to achieve language-enhanced economic decision-making.
Thinker: Training LLMs in Hierarchical Thinking for Deep Search via Multi-Turn Interaction
Jun Xu (Ant Group), Jun Zhou (Ant Group)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Hierarchical thinking training for large language models, achieving deep retrieval through multi-round interaction.
Thinking Aesthetics Assessment of Image Color Temperature: Models, Datasets and Benchmarks
Jinguang Cheng (Beijing University of Posts and Telecommunications), Anlong Ming (Beijing University of Posts and Telecommunications)
ClassificationTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes the Image Color Temperature Aesthetics Assessment task (ICTAA) and constructs a large-scale color temperature aesthetics dataset named ICTAA240K;
Thinking Forward and Backward: Multi-Objective Reinforcement Learning for Retrieval-Augmented Reasoning
Wenda Wei (Chinese Academy Of Sciences), Xueqi Cheng (Baidu Inc)
RetrievalReinforcement LearningTextRetrieval-Augmented Generation
🎯 What it does: Designed and implemented a retrieval-augmented reasoning framework called Bi-RAR based on bidirectional information distance, which uses multi-objective reinforcement learning to perform forward and backward supervision at each reasoning step, optimizing the synergy between retrieval and reasoning.
Through the Water: Refractive Gaussian Splatting for Water Surface Scenes
Yeonghun Yoon (Chung-Ang University), Jongwon Choi (Chung-Ang University)
GenerationData SynthesisGaussian SplattingImage
🎯 What it does: Designed a Gaussian Splatting framework tailored for water scenes, capable of simultaneously modeling water reflection and refraction, and achieving high-quality view synthesis.
TiCAL:Typicality-Based Consistency-Aware Learning for Multimodal Emotion Recognition
Wen Yin (University of Electronic Science and Technology of China), Tao He (Monash University)
ClassificationRecognitionMultimodality
🎯 What it does: Proposed a multimodal emotion recognition framework TiCAL based on typicality and consistency assessment.
TIDE: Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generation
Victor Shea-Jay Huang (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationExplainability and InterpretabilityTransformerDiffusion modelAuto EncoderImage
🎯 What it does: Propose a temporal-aware sparse autoencoder (TIDE) for Diffusion Transformer, which can extract interpretable sparse activation features across multiple time steps.
Tighter Truncated Rectangular Prism Approximation for RNN Robustness Verification
Xingqi Lin, Zhenbing Zeng (East China Normal University)
OptimizationExplainability and InterpretabilityRecurrent Neural NetworkImageTextAudio
🎯 What it does: Proposed a dual-plane compact approximation method based on truncated rectangular prisms for robustness verification of RNNs (such as LSTM), and implemented the DeepPrism verification tool.
TileGS: Adaptive Gaussian Densification Through Tile-Guided Perceptual Analysis
Yiwen Wang (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
OptimizationComputational EfficiencyGaussian SplattingImagePoint Cloud
🎯 What it does: By integrating tile-based perceptual feedback into the 3D Gaussian scattering rendering pipeline, dynamically densifying and pruning Gaussians to enhance detail and boundary performance;
TIM++: Transductive Information Maximization for Few-Shot CLIP
Yingping Li (Xidian University), Shuiping Gou (Xidian University)
ClassificationRepresentation LearningMeta LearningVision Language ModelContrastive LearningMultimodality
🎯 What it does: In transductive few-shot classification, joint learning is performed using textual information and visual features from the pre-trained vision-language model CLIP to enhance few-shot recognition performance.
TIMA: Text-Image Mutual Awareness for Balancing Zero-Shot Adversarial Robustness and Generalization Ability
Fengji Ma, Li Liu (Tencent AI Lab)
Adversarial AttackTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose the TIMA framework, achieving a balance between zero-shot robustness and generalization through text-image mutual awareness
Time Series Class-Incremental Learning via Confidence-guided Mask Distillation and Prototype-guided Contrastive Learning
Yu Liu (Dalian University of Technology), Qi Jia (Dalian University of Technology)
ClassificationKnowledge DistillationContrastive LearningTime Series
🎯 What it does: Proposes two novel techniques for incremental learning on time series: Confidence-Guided Mask Distillation (CMD) and Prototype-Guided Contrastive Learning (PCL), and jointly applies them to an incremental learning framework without rehearsal.
Time Series Forecasting via Direct Per-Step Probability Distribution Modeling
Linghao Kong (Harbin Institute of Technology), Xiaopeng Hong (Harbin Institute of Technology)
Convolutional Neural NetworkTime SeriesBenchmark
🎯 What it does: Propose the interPDN model, which directly outputs a discrete probability distribution at each time step and takes the expectation for prediction, thereby better capturing uncertainty in time series forecasting.
Time Shuffle: A Transferability-Booster for Multiple Audio Adversarial Tasks
JiaCheng Deng (Wuhan University), Haoran Duan (Wuhan University)
Explainability and InterpretabilityAdversarial AttackAudio
🎯 What it does: This paper proposes enhancing the cross-model transferability of audio adversarial examples through time shuffling (TS/TS-p), forcing attacks to exploit local temporal features rather than source model-specific patterns;
Time-Frequency Augmented Multi-level Contrastive Clustering for Time Series
Congyu Wang (Jiangsu Normal University), Xiang Jiang (Jiangsu Normal University)
Representation LearningConvolutional Neural NetworkContrastive LearningTime Series
🎯 What it does: Proposed an end-to-end deep time series clustering framework called TFMCC, which improves clustering performance by jointly learning representations and clustering.
Time-Frequency Token Advantage Clipping for Training Efficient Large Reasoning Model
Rong Bao (Fudan University), Dacheng Tao (Nanyang Technological University)
Computational EfficiencyTransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: Designed and verified a reinforcement learning framework called TFAC based on time-frequency advantage clipping, aimed at training large language models to efficiently reason in long-chain thinking (Chain-of-Thought).
TIME: Temporal-Sensitive Multi-Dimensional Instruction Tuning and Robust Benchmarking for Video-LLMs
Yunxiao Wang (Shandong University), Liqiang Nie (Kuaishou Technology)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: This paper proposes the TIME instruction tuning dataset, the MTP multi-task prompting fine-tuning method, and the TIMEBench evaluation benchmark to enhance the temporal understanding capabilities of video large language models.
TimeBill: Time-Budgeted Inference for Large Language Models
Qi Fan (Shanghai Jiao Tong University), Yehan Ma (Shanghai Jiao Tong University)
Computational EfficiencyLarge Language ModelTextBenchmark
🎯 What it does: Proposes the TimeBill framework to achieve inference completion of large language models within a time budget while maintaining response quality.
TimeCAP: A Channel-Aware Pre-Training Framework for Multivariate Time Series Forecasting
Chuanru Ren, Lu Liu (Tongji University)
Knowledge DistillationRepresentation LearningTransformerSupervised Fine-TuningTime Series
🎯 What it does: Proposed a channel-aware pre-training framework called TimeCAP for multivariate time series forecasting.
TimeMosaic: Temporal Heterogeneity Guided Time Series Forecasting via Adaptive Granularity Patch and Segment-wise Decoding
Kuiye Ding (Institute of Computing Technology, Chinese Academy of Sciences), Jianfeng Zhan (Institute of Computing Technology, Chinese Academy of Sciences)
TransformerPrompt EngineeringTime Series
🎯 What it does: Proposed a time series forecasting framework called TimeMosaic that addresses the issues of encoding heterogeneity and decoding heterogeneity.
Timestep-Compressed Attack on Spiking Neural Networks Through Timestep-Level Backpropagation
Donghwa Kang, Brent ByungHoon Kang (Korea Advanced Institute Of Science And Technology)
ClassificationAdversarial AttackSpiking Neural NetworkImage
🎯 What it does: This paper proposes a Time-Step Compression Attack (TCA) framework that significantly reduces the latency of adversarial attacks on spiking neural networks while maintaining high attack success rates.
TinyChemVL: Advancing Chemical Vision-Language Models via Efficient Visual Token Reduction and Complex Reaction Tasks
Xuanle Zhao (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences), Bo Xu (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences)
Computational EfficiencyDrug DiscoveryTransformerVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes TinyChemVL, an efficient cheminformatics vision-language model, and designs adaptive visual token merging/cropping strategies and a new reaction-level benchmark, ChemRxn-V, achieving unified processing for molecular and reaction tasks;
TIV: Thought Injection via Vectors for Efficient Reasoning in Large Reasoning Models
Yi Cao (Soochow University), Jiajie Xu (Soochow University)
Computational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: Propose and implement the TIV framework, which significantly compresses the inference output length by injecting a learnable vector into the post-attention hidden states of the final token in Transformers, replacing the traditional token-by-token reasoning process.
TLAGC: Taylor Linear Attention-Guided Graph Convolutions for Revealing Spatial Domains in Spatial Multi-Omics Data
Aoyun Geng (Hainan University), Zilong Zhang (Hainan University)
Graph Neural NetworkContrastive LearningBiomedical Data
🎯 What it does: This paper proposes the TLAGC framework for integrating spatial multi-omics data and partitioning spatial domains.
TMAE:Learning Targeted Multi-Agent Exploration via Causal Inference
Chuxiong Sun, Jiangmeng Li (Chinese Academy of Sciences)
Reinforcement Learning
🎯 What it does: This paper proposes a causal inference-based multi-agent sparse reward task exploration framework called TMAE, which automatically extracts causal relationships between subspaces and rewards from historical data to guide targeted exploration;
TMDC: A Two-Stage Modality Denoising and Complementation Framework for Multimodal Sentiment Analysis with Missing and Noisy Modalities
Yan Zhuang (University of Electronic Science and Technology of China), Fuji Ren (University of Electronic Science and Technology of China)
ClassificationComputational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerMultimodality
🎯 What it does: Proposed a two-stage modal denoising and compensation framework (TMDC) for simultaneously handling missing and noisy modalities in multimodal sentiment analysis.
To Align or Not to Align: Strategic Multimodal Representation Alignment for Optimal Performance
Wanlong Fang (Nanyang Technological University), Alvin Chan (Nanyang Technological University)
Representation LearningTransformerContrastive LearningMultimodality
🎯 What it does: Investigated the impact of explicit multimodal alignment on monomodal model performance under different information structures, and proposed a contrastive learning framework with controllable alignment strength;
TO-GATE: Clarifying Questions and Summarizing Responses with Trajectory Optimization for Eliciting Human Preference
Yulin Dou (Yunnan University), Jiangming Liu (Yunnan University)
Reinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningMixture of ExpertsContrastive LearningText
🎯 What it does: This work proposes a training framework called TO-GATE, which enhances the ability of LLMs to raise clarifying questions and generate final personalized answers in multi-turn dialogues through trajectory optimization.
ToC: Tree-of-Claims Search with Multi-Agent Language Models
Shuyang Yu (Columbia University), Hui Hu (AIGROW)
OptimizationTransformerLarge Language ModelAgentic AITextMultimodalityChain-of-Thought
🎯 What it does: Proposed the Tree of Claims framework, modeling patent claim optimization as a structured search problem based on multi-agent LLM and Monte Carlo Tree Search.
TOFA: Training-Free One-Shot Federated Adaptation for Vision-Language Models
Li Zhang (Zhejiang University), Chaochao Chen (Ant Group)
Federated LearningLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose a training-free, single-round federated learning framework called TOFA for rapidly adapting vision-language models while preserving privacy.
Token Painter: Training-Free Text-Guided Image Inpainting via Mask Autoregressive Models
Longtao Jiang (University Of Science And Technology Of China), Zhihui Li (Mbzuai)
RestorationGenerationTransformerVision Language ModelImageTextBenchmark
🎯 What it does: Achieve training-free text-guided image inpainting using the Mask AutoRegressive (MAR) model, preserving the background while generating content consistent with the text prompt.
Token-Context Attention for NLI: An Alternative to Self-Attention
Xin Zhang (Texas Tech University), Victor S. Sheng (Texas Tech University)
ClassificationExplainability and InterpretabilityComputational EfficiencyAdversarial AttackTransformerText
🎯 What it does: Proposes a complex vector-based word-context attention model, replacing traditional self-attention to enhance parameter efficiency of small-scale models in natural language inference tasks.
Tokenize Once, Recommend Anywhere: Unified Item Tokenization for Multi-domain LLM-based Recommendation
Yu Hou (Yonsei University), Won-Yong Shin (Yonsei University)
Recommendation SystemTransformerLarge Language ModelMixture of ExpertsAuto EncoderText
🎯 What it does: Propose a unified item tokenization framework called UniTok, enabling LLM generative recommendation systems to share the same tokenizer across multiple domains.
TokenPowerBench: Benchmarking the Power Consumption of LLM Inference
Chenxu Niu (Texas Tech University), Yong Chen (Texas Tech University)
Computational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposes TokenPowerBench, a lightweight and scalable power benchmark framework for large language model (LLM) inference.
TongUI: Internet-Scale Trajectories from Multimodal Web Tutorials for Generalized GUI Agents
Bofei Zhang (Beijing Institute Of Technology), Qing Li (Bigai)
Data SynthesisTransformerLarge Language ModelVision-Language-Action ModelVideoTextMultimodality
🎯 What it does: Built a framework named TongUI, converting millions of multimodal tutorials from the internet into 1 million GUI interaction trajectories, and used these trajectories to train GUI agents.
Too Sure for Our Own Good: A User Study on AI Confidence and Human Reliance
Caterina Fregosi (University of Milano-Bicocca), Federico Cabitza (University of Milano-Bicocca)
Explainability and InterpretabilityText
🎯 What it does: This study examined the impact of AI confidence calibration on human decision accuracy, appropriate reliance, automation bias, and conservative bias through a logical reasoning task involving 184 participants.
ToolACE-R: Model-aware Iterative Training and Adaptive Refinement for Tool learning
Xingshan Zeng (Huawei Technologies Co Ltd), Qun Liu (Huawei Technologies Co Ltd)
Computational EfficiencyAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose the ToolACE-R framework, combining model-aware iterative training and adaptive self-refinement to enhance the tool calling capability of large language models (LLMs).
Top-Down Semantic Refinement for Image Captioning
Jusheng Zhang (Sun Yat Sen University), Keze Wang (Sun Yat Sen University)
GenerationComputational EfficiencyTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposed the Top-Down Semantic Refinement (TDSR) framework, treating image caption generation as a coarse-to-fine hierarchical planning problem, progressively refining descriptions through VLM-based MCTS.
TOP-RL: Task-Optimized Progressive Token Pruning with Reinforcement Learning for Vision Language Models
Hengyi Wang (Xidian University), Leyuan Fang (Xidian University)
Computational EfficiencyTransformerReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposed a task-optimized layer-wise visual token pruning framework TOP-RL based on reinforcement learning for efficient inference in large-scale vision-language models.
TOPOGRAPH: Topology-Preserving Graph Reduction with Adaptive Structure for Persistent Homology
Zonghao Chen (South China Normal University), Gang Li (Deakin University)
Computational EfficiencyPoint CloudGraphBenchmark
🎯 What it does: Proposes the TOPOGRAPH framework, integrating DTM outlier removal, adaptive kNN graph construction, and spectral graph coarsening, to achieve efficient and robust intrinsic persistent homology (IPH) computation, while significantly compressing data while preserving topological structures;
Topological Federated Clustering via Gravitational Potential Fields Under Local Differential Privacy
Yunbo Long (University of Cambridge), Alexandra Brintrup (University of Cambridge)
Federated LearningSafty and PrivacyImageTabular
🎯 What it does: Developed a one-shot federated clustering method called GFC, which can cluster non-IID data with heterogeneous distributions under local differential privacy (LDP).
Topology-aware Knowledge Preservation for Class-Incremental Learning
Han Zang (Hebei University of Technology), Yu Wang (Tianjin University)
Knowledge DistillationRepresentation LearningImage
🎯 What it does: Proposes a topology-preserving class-incremental learning framework called TaKP, integrating topological distillation based on persistent homology with a dual-branch adaptive rebalancing mechanism to better preserve old class knowledge during continual learning.
Topology-Aware Vision Transformers for Enhanced Scene Recognition
Yunxi Wang (University of Electronic Science and Technology of China), Xiaorong Pu (University of Electronic Science and Technology of China)
RecognitionGraph Neural NetworkTransformerImageGraph
🎯 What it does: This paper proposes a network called TANSR that integrates the topological information of superpixel graphs into Vision Transformers for scene recognition.
Topology-Enhanced and Label Correlation-Aware Model for Protein-Protein Interaction Prediction
Bin Deng (Northwest Normal University), Rui Bing (Northwest Normal University)
Drug DiscoveryGraph Neural NetworkGraphBiomedical Data
🎯 What it does: Propose the TELC-PPI model for multi-label protein-protein interaction (PPI) prediction.
Topology-Inspired Backward-Free Framework for Test-Time Adaptation in Medical Detection
Bin Pu (Hunan University), Kenli Li (Hunan University)
Object DetectionDomain AdaptationImageBiomedical DataUltrasound
🎯 What it does: Designed a test-time adaptation framework T3A without backpropagation, based on topological consistency, integrating structure-aware modeling and box regression adaptation for medical detection.
TOSC: Task-Oriented Shape Completion for Open-World Dexterous Grasp Generation from Partial Point Clouds
Weishang Wu (National University of Defense Technology), Zhiping Cai (National University of Defense Technology)
GenerationPose EstimationRobotic IntelligenceTransformerLarge Language ModelVision Language ModelDiffusion modelFlow-based ModelAuto EncoderPoint Cloud
🎯 What it does: This paper proposes a task-oriented shape completion framework to generate complete shapes from partial point clouds that are compatible with downstream grasping tasks, and further generates executable multi-finger grasping poses.
TouchFormer: A Robust Transformer-based Framework for Multimodal Material Perception
Kailin Lyu (Institute of Automation, Chinese Academy of Sciences), Jie Hao (Institute of Automation, Chinese Academy of Sciences)
ClassificationRecognitionTransformerMultimodality
🎯 What it does: A Transformer-based multimodal material perception framework called TouchFormer is proposed for vision-restricted or noisy environments, addressing the issue of significant performance degradation of traditional methods under missing modalities and noise.
Toward Better EHR Reasoning in LLMs: Reinforcement Learning with Expert Attention Guidance
Yue Fang (Peking University), Yasha Wang (Peking University)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningBiomedical DataElectronic Health Records
🎯 What it does: Proposed the EAG-RL two-stage training framework, which significantly enhances LLM's reasoning capabilities on electronic health records (EHR) through expert-guided Monte Carlo Tree Search initialization and attention alignment-based reinforcement learning.
Toward Multimodal Fake News Detection by Multi-perspective Rationale Generation and Verification
Junyang Chen (Shenzhen University), Liang-Jie Zhang (Shenzhen University)
ClassificationTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningMultimodality
🎯 What it does: This paper proposes the MMRGV framework, which utilizes multi-perspective reasoning generation and cross-validation for multi-modal fake news detection.
Toward Real-World High-Precision Image Matting and Segmentation
Haipeng Zhou (Hong Kong University of Science and Technology (Guangzhou)), Lei Zhu (Hong Kong Polytechnic University)
SegmentationDomain AdaptationKnowledge DistillationTransformerGenerative Adversarial NetworkImageMultimodality
🎯 What it does: Designed a foreground-consistent learning model, FCLM, for high-precision image matting and binary segmentation, addressing domain differences in synthetic data, foreground consistency learning, and interactive multi-instance prediction.
Toward the Frontiers of Reliable Diffusion Sampling via Adversarial Sinkhorn Attention Guidance
Kwanyoung Kim (Samsung Research)
GenerationAdversarial AttackTransformerDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a guidance method that introduces adversarial Sinkhorn transformation into the self-attention mechanism of diffusion models, improving the quality and diversity of generated images without requiring additional training.
Toward Time-Continuous Data Inference in Sparse Urban CrowdSensing
Hao Du (Jilin University), Yuanbo Xu (Jilin University)
Computational EfficiencyData-Centric LearningTime Series
🎯 What it does: Proposes the TIME-DMF framework for temporally continuous sparse urban sensing data completion, achieving high-precision inference at any arbitrary time point.
Towards 3D Object-Centric Feature Learning for Semantic Scene Completion
Weihua Wang (Northeastern University), Zheng Fang (University of Hong Kong)
SegmentationAutonomous DrivingTransformerDiffusion modelImagePoint CloudBenchmark
🎯 What it does: Propose the Ocean framework, which accomplishes the monocular visual semantic scene completion task by leveraging instance-based object-centric feature learning.
Towards a Rigorous Understanding of the Population Dynamics of the NSGA-III: Tight Runtime Bounds
Andre Opris (University of Passau)
OptimizationBenchmark
🎯 What it does: This paper conducts a rigorous theoretical analysis of the runtime of the multi-objective evolutionary algorithm NSGA-III on the two-objective OneMinMax (2-OMM) problem, providing matching lower and upper bounds.
Towards Accurate 3D Object Detection in Adverse Weather by Leveraging 4D Radar for LiDAR Geometry Enhancement
Tianxu Tong (University of Science and Technology Beijing), Bin Fan (University of Science and Technology Beijing)
Object DetectionAutonomous DrivingConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: In adverse weather conditions, generating virtual LiDAR points through 4D radar guidance enhances LiDAR point cloud geometric density and structure, thereby improving 3D object detection;
Towards Acyclic Preference Evaluation of Language Models via Multiple Evaluators
Zhengyu Hu (Hong Kong University of Science and Technology), Ranjay Krishna (University of Washington)
Reinforcement Learning from Human FeedbackLarge Language ModelTextBenchmark
🎯 What it does: Propose the PGED framework, which constructs a preference graph using multiple evaluators and performs denoising to eliminate cyclic preferences, achieving more reliable evaluation and ranking of LLM outputs.
Towards Adaptive Humanoid Control via Multi-Behavior Distillation and Reinforced Fine-Tuning
Yingnan Zhao (Harbin Engineering University), Chenjia Bai (Harbin Institute of Technology)
Knowledge DistillationRobotic IntelligenceReinforcement LearningMixture of ExpertsTime Series
🎯 What it does: Proposed the Adaptive Humanoid Control (AHC) two-stage framework, first obtaining a multi-behavior controller through behavior distillation, then fine-tuning with reinforcement learning on diverse terrains to achieve adaptive standing, walking, and other skills for humanoid robots;
Towards Affordance-Aware Robotic Dexterous Grasping with Human-like Priors
Haoyu Zhao (Wuhan University), Hua Zou (Wuhan University)
SegmentationKnowledge DistillationRobotic IntelligenceTransformerReinforcement LearningVision Language ModelImageMultimodalityPoint Cloud
🎯 What it does: Propose the AffordDex framework, leveraging human hand trajectory pre-training and negative affordance segmentation to achieve universal, anthropomorphic, and functionally correct multi-finger grasping.
Towards Authentic Movie Dubbing with Retrieve-Augmented Director-Actor Interaction Learning
Rui Liu (Inner Mongolia University), Zhenqi Jia (Inner Mongolia University)
GenerationRetrievalGraph Neural NetworkLarge Language ModelMultimodalityRetrieval-Augmented Generation
🎯 What it does: Implemented a retrieval-enhanced director-actor interactive framework called Authentic-Dubber for generating emotion-expressive movie dubbing.
Towards Autonomous UAV Visual Object Search in City Space: Benchmark and Agentic Methodology
Yatai Ji (National University of Defense Technology), Quanjun Yin (National University of Defense Technology)
Autonomous DrivingTransformerLarge Language ModelPrompt EngineeringVision Language ModelWorld ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose the CityAVOS benchmark for urban space UAV visual target search (AVOS), and design a PRPSearcher agent based on a multi-modal large language model, achieving perception, reasoning, and planning through three specialized maps.
Towards Better Code Understanding in Decoder-Only Models with Contrastive Learning
Jiayi Lin (International Digital Economy Academy), Yutao Xie (Nanjing University)
Representation LearningAI Code AssistantTransformerContrastive LearningTextBenchmark
🎯 What it does: Improving performance on code understanding tasks such as code search and code clone detection by continuing contrastive learning training on a pre-trained decoder-only model.
Towards Better Correctness and Efficiency in Code Generation
Yunlong Feng (Alibaba Group), Junyang Lin (Alibaba Group)
OptimizationAI Code AssistantLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes a two-stage training framework, first using DPO to optimize the model's code correctness, and then using RLOO reinforcement learning to improve the runtime efficiency of generated code.
Towards Better IncomLDL: We Are Unaware of Hidden Labels in Advance
Jiecheng Jiang (Southeast University), Yuheng Jia (Southeast University)
OptimizationRepresentation LearningBenchmark
🎯 What it does: Propose the Hidden Label Distribution Learning (HidLDL) problem and design a recovery method leveraging proportion constraints, graph similarity, and low-rank structure.
Towards Closed-Loop Embodied Empathy Evolution: Probing LLM-Centric Lifelong Empathic Motion Generation in Unseen Scenarios
Jiawen Wang (Soochow University), Guodong Zhou (Soochow University)
GenerationData SynthesisTransformerLarge Language ModelMixture of ExpertsAuto EncoderVideoTextMultimodality
🎯 What it does: Proposes the LLM-driven lifelong empathetic motion generation (L-EMG) task, aiming to enable large language models to continuously learn in emerging action scenarios while maintaining the association between learned emotions and actions, addressing issues of generalization degradation and catastrophic forgetting for unseen scenarios.
Towards Distance-Invariant Radio Frequency Fingerprinting via Augmented Unsupervised Learning
Shiyue Huang (University of Electronic Science and Technology of China), Haitao Jia (University of Electronic Science and Technology of China)
RecognitionDomain AdaptationContrastive Learning
🎯 What it does: Proposed a completely unsupervised cross-distance radio spectrum fingerprint recognition framework that maintains high recognition accuracy under different transmission distances.
Towards Effective and Efficient Context-aware Nucleus Detection in Histopathology Whole Slide Images
Zhongyi Shui (Zhejiang University), Lin Yang (Westlake University)
Object DetectionSegmentationComputational EfficiencyConvolutional Neural NetworkTransformerBiomedical Data
🎯 What it does: Proposed a context-aware method for nucleus detection on whole-slide images, enhancing detection accuracy and speed by aggregating features from historical sliding windows.
Towards Effective Code-Integrated Reasoning
Fei Bai (Renmin University of China), Hongteng Xu (DataCanvas Alaya NeW)
AI Code AssistantReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper studies Code-Integrated Reasoning (CIR) and proposes the ETIR method to improve the training stability and performance of tool-enhanced reinforcement learning in mathematical reasoning tasks.
Towards Effective Offensive Security LLM Agents: Hyperparameter Tuning, LLM as a Judge, and a Lightweight CTF Benchmark
Minghao Shao (New York University), Muhammad Shafique (New York University)
Hyperparameter SearchTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: This paper systematically evaluates the performance of large language models in Capture the Flag (CTF) attack agents and investigates the impact of hyperparameters such as temperature, top-p, maximum token length, and multi-agent collaboration on problem-solving success rates.
Towards Effective, Stealthy, and Persistent Backdoor Attacks Targeting Graph Foundation Models
Jiayi Luo (Beihang University), Jianxin Li (Guangxi Normal University)
Adversarial AttackGraph Neural NetworkGraph
🎯 What it does: Propose a backdoor attack framework for Graph Foundation Models (GFM) called GFM-BA, which can inject backdoors during the pre-training phase and maintain effectiveness, stealthiness, and persistence across different downstream tasks.
Towards Efficient and Effective Interactive 3D Segmentation
Wei Cong (Shenyang Institute of Automation, Chinese Academy of Sciences), Gan Sun (South China University of Technology)
SegmentationComputational EfficiencyPoint Cloud
🎯 What it does: This paper proposes an interactive 3D segmentation model called E2I3D, which adopts a two-stage efficiency-to-effectiveness framework. In the first stage, model compression is achieved through heterogeneous pruning, while the second stage enhances segmentation accuracy by utilizing hierarchical click-aware attention.
Towards Efficient and Robust Manipulation via Multi-Frame Vision-Language-Action Modeling
Hao Li (University of Science and Technology of China), Jiangmiao Pang (Shanghai Artificial Intelligence Laboratory)
Robotic IntelligenceTransformerVision-Language-Action ModelDiffusion modelVideoTextMultimodalityBenchmark
🎯 What it does: Propose the CronusVLA framework, extending single-frame Vision-Language-Action (VLA) models to multi-frame learning through single-frame pre-training + multi-frame fine-tuning, leveraging feature tiling and multi-frame regularization for efficient inference and strong robustness.
Towards Efficient Low-rate Image Compression with Frequency-aware Diffusion Prior Refinement
Yichong Xia (Tsinghua University), Bin Chen (Harbin Institute of Technology)
CompressionDiffusion modelAuto EncoderImage
🎯 What it does: Propose DiffCR, a low-bitrate image compression framework that leverages pre-trained diffusion models with frequency-aware skip estimation (FaSE) and frequency-decoupled attention (FDA), achieving fast decoding in just two steps.
Towards Explainable Video Camouflaged Object Detection: SAM2 with Eventstream-Inspired Data
Hong Zhang (Beihang University), Yifan Yang (Beihang University)
Object DetectionExplainability and InterpretabilitySpiking Neural NetworkTransformerPrompt EngineeringVideoBenchmark
🎯 What it does: Propose an event-stream inspired dual-branch framework that automatically generates prompts and integrates temporal memory to achieve interpretable video camouflage object detection.
Towards Federated Clustering: A Client-wise Private Graph Aggregation Framework
Guanxiong He (Northwestern Polytechnical University), Feiping Nie (Northwestern Polytechnical University)
Federated LearningSafty and PrivacyGraph Neural NetworkAuto EncoderImageTabular
🎯 What it does: Propose a federated clustering framework based on local structure graphs and differential privacy: SPP-FGC (single-round) and SPP-FGC+ (iterative)
Towards High-Fidelity 3D Portrait Generation with Rich Details by Cross-View Prior-Aware Diffusion
Haoran Wei (University of Macau), Jianbing Shen (Wuhan University)
GenerationDiffusion modelScore-based ModelGenerative Adversarial NetworkImageMesh
🎯 What it does: Proposed the Portrait Diffusion pipeline, leveraging Hybrid Priors Diffusion Model and Multi-View Noise Resampling Strategy to achieve high-fidelity 3D portrait generation from a single image.