AAAI 2026 Papers — Page 39
AAAI Conference on Artificial Intelligence · 4149 papers
Towards High-Resolution 3D Anomaly Detection: A Scalable Dataset and Real-Time Framework for Subtle Industrial Defects
Yuqi Cheng (Huazhong University of Science and Technology), Yunkang Cao (Hunan University)
Anomaly DetectionComputational EfficiencyPoint CloudBenchmark
🎯 What it does: Proposed the MiniShift high-resolution fine-grained 3D anomaly detection dataset and the Simple3D real-time detection framework, addressing the localization and identification of minute defects in industrial point clouds.
Towards Illumination-Aware Restoration of Metalens-Captured Images: A New Dataset and a Strong Baseline
Fen Fang, Zhengguo Li (Institute for Infocomm Research)
RestorationConvolutional Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: Study the restoration of metal lens images, propose the IlluMeta dataset, a spatially aware attention scheduler, and a reinforcement learning-based untrained illumination adapter, and integrate them into existing restoration models.
Towards LLM-Empowered Knowledge Tracing via LLM-Student Hierarchical Behavior Alignment in Hyperbolic Space
Xingcheng Fu (Guangxi Normal University), Dongran Yu (Guangxi Normal University)
Data SynthesisGraph Neural NetworkLarge Language ModelAgentic AIContrastive LearningMultimodalitySequential
🎯 What it does: Propose a framework (L-HAKT) that utilizes large language models and a dual-agent system (teacher + student) to generate hierarchical knowledge graphs and synthetic learning data, achieving knowledge tracing through contrastive alignment in hyperbolic space
Towards Long-window Anchoring in Vision-Language Model Distillation
Haoyi Zhou (Beihang University), Jianxin Li (Beihang University)
Knowledge DistillationTransformerVision Language ModelMultimodality
🎯 What it does: Studied a long-window anchoring distillation method for vision-language models to enhance the performance of small models in long-text and multi-image environments.
Towards Multimodal Continual Knowledge Embedding with Modality Forgetting Modulation
Xiaowen Jiang, Jieming Yang (Zhengzhou University)
Representation LearningConvolutional Neural NetworkGraph Neural NetworkTransformerVision Language ModelMultimodality
🎯 What it does: Proposes the MoFot framework, which mitigates catastrophic forgetting in multi-modal knowledge graph continual learning through multi-modal collaborative modulation, avoiding issues of retraining and knowledge loss.
Towards Multiple Missing Values-resistant Unsupervised Graph Anomaly Detection
Jiazhen Chen (University of Waterloo), Weihua Ou (University of Waterloo)
Anomaly DetectionGraph Neural NetworkAuto EncoderContrastive LearningGraph
🎯 What it does: Proposes an unsupervised graph anomaly detection framework named MV-UGAD for scenarios where both node attributes and graph structure are simultaneously missing.
Towards Non-Stationary Time Series Forecasting with Temporal Stabilization and Frequency Differencing
Junkai Lu (East China Normal University), Bin Yang (East China Normal University)
TransformerMixture of ExpertsTime SeriesElectronic Health Records
🎯 What it does: Propose a dual-branch framework DTAF, which extracts and suppresses non-stationary components in the time domain using a hybrid expert filter, highlights significant frequency changes through differentiation in the frequency domain, and subsequently generates robust long-term predictions by fusing time-frequency features via dual-branch attention.
Towards Nonlinear Sparse AUC Maximization via Compositional Stochastic Hard Thresholding
Wenkang Wang (Jilin University), Bin Gu (Jilin University)
ClassificationAdversarial AttackImageTabular
🎯 What it does: This paper proposes a Compositional Stochastic Hard-Thresholding (CSHT) algorithm that combines stochastic variance reduction gradient (SVRG) with hard thresholding projection for nonlinear sparse AUC maximization, and applies this framework to generate sparse general adversarial perturbations.
Towards OOD Generalization in Dynamic Graphs via Causal Invariant Learning
Xinxun Zhang (Hangzhou Dianzi University), Xuan Guo (Tianjin University)
Domain AdaptationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Proposes the DyCIL framework for achieving out-of-distribution (OOD) generalization in dynamic graphs through causal invariant learning.
Towards Privacy-Protected Generalized Gaze Estimation Using Diffusion Models and Domain Stability Adaptation Framework
Ziyi Wang (East China Normal University), Haichuan Song (East China Normal University)
Pose EstimationDomain AdaptationSafty and PrivacyDiffusion modelImage
🎯 What it does: Generate diverse, privacy-free gaze data using diffusion models and enhance cross-domain generalization of gaze estimation through a domain-stable adaptation (DSA) framework.
Towards Provably Secure and Highly Robust Generative Image Steganography Leveraging Latent Diffusion Model
Chengsheng Yuan (Nanjing University of Information Science and Technology), Zhihua Xia (Jinan University)
GenerationData SynthesisSafty and PrivacyDiffusion modelScore-based ModelAuto EncoderImage
🎯 What it does: Propose a provably secure and highly robust generative image steganography framework based on latent diffusion models (LDM), which embeds secret messages into latent vectors following a standard normal distribution using adaptive distribution preserving mapping (ADPM), and generates steganographic images through these vectors;
Towards Provably Unlearnable Examples via Bayes Error Optimization
Ruihan Zhang (Singapore Management University), Peixin Zhang (Singapore Management University)
OptimizationSafty and PrivacyData-Centric LearningImage
🎯 What it does: This paper proposes a framework for generating non-learnable samples based on maximizing Bayes error.
Towards Real-Time Neutral Atom Array Assembly via Unsupervised Hologram Generation and Path Optimization
Ge Yan (Nanyang Technological University), Junchi Yan (Shanghai Jiao Tong University)
OptimizationConvolutional Neural NetworkImagePhysics Related
🎯 What it does: Proposes a real-time neutral atom array assembly framework based on min-max path matching and Fourier U-Net.
Towards Reinforcement Learning from Neural Feedback: Mapping fNIRS Signals to Agent Performance
Julia Santaniello (Tufts University), Jivko Sinapov (Tufts University)
Reinforcement Learning from Human FeedbackTime SeriesBiomedical Data
🎯 What it does: Proposed a reinforcement learning feedback framework (RLNF) based on fNIRS neural signals, achieving implicit human feedback by mapping brain signals to agent performance;
Towards Robust Edge Model Adaptation via Elastic Architecture Search
Xianhang Chu (Xidian University), Xi Wang (Xidian University)
Domain AdaptationKnowledge DistillationNeural Architecture SearchSupervised Fine-TuningImage
🎯 What it does: Propose an elastic continuous test-time adaptation framework for edge devices (Elastic Edge CTTA), which generates personalized sub-models through resource-aware architecture search before deployment, followed by lightweight online self-supervised fine-tuning using low-rank adapters on the device, and feeds adaptation experiences back to the super network via structure-aware knowledge backflow to improve subsequent search quality.
Towards Robust Event-Based Depth Estimation: Bridging Synthetic and Real Domains with Motion Adaptation
Yuzhe Ji (Hong Kong University of Science and Technology), Xinhu Zheng (Hong Kong University of Science and Technology)
Depth EstimationDomain Adaptation
🎯 What it does: Proposes MA-Mamba, a dual-track framework combining a lightweight spatiotemporal correlation module and an adaptive memory balancing module, along with event-specific data augmentation, to achieve robust depth estimation for real-world motion patterns trained on synthetic data.
Towards Robust Text-Attributed Federated Graph Learning: Multimodal Threats and Defense
Zitong Shi (Wuhan University), Mang Ye (Renmin University of China)
Federated LearningSafty and PrivacyAdversarial AttackGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityGraph
🎯 What it does: In federated text attribute graph learning, the GTAE attack framework is proposed to simultaneously attack the structural and text modalities, along with the modality-aware robust training and weighted aggregation defense STRUM against it.
Towards Single Exponential Time for Temporal and Spatial Reasoning: A Study via Redundancy and Dynamic Programming
Victor Lagerkvist (Linkoping University), Leif Eriksson (Linkoping University)
OptimizationComputational Efficiency
🎯 What it does: This paper studies two classic infinite-domain constraint satisfaction problems—Allen's Interval Algebra (IA) and Region Connection Calculus (RCC)—and proposes new algorithms and theoretical analysis based on these.
Towards Spatially Consistent Image Generation: On Incorporating Intrinsic Scene Properties into Diffusion Models
Hyundo Lee (Seoul National University), Byoung-Tak Zhang (Seoul National University)
GenerationTransformerDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a diffusion model that simultaneously generates images and their corresponding intrinsic scene attributes (depth maps, normal vectors, segmentation maps, and line drawings) to enhance the spatial consistency of generated images.
Towards Synthesizing High-Dimensional Tabular Data with Limited Samples
Zuqing Li (University of Melbourne), Jianzhong Qi (University of Melbourne)
GenerationData SynthesisDiffusion modelTabular
🎯 What it does: This study proposes a conditional control diffusion model named CtrTab for synthesizing high-dimensional sparse (low-sample) tabular data;
Towards Temporal Fusion Beyond the Field of View for Camera-based Semantic Scene Completion
Jongseong Bae (Yonsei University), Ha Young Kim (Yonsei University)
SegmentationDepth EstimationAutonomous DrivingConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: This paper proposes a camera-based 3D semantic scene completion method that can utilize historical frame information to fill in spatial gaps outside the current viewpoint.
Towards Test-time Efficient Visual Place Recognition via Asymmetric Query Processing
Jaeyoon Kim (Korea Advanced Institute of Science and Technology), Sung-Eui Yoon (Korea Advanced Institute of Science and Technology)
RetrievalComputational EfficiencyConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper proposes an efficient asymmetric visual place recognition framework, which uses a high-capacity model for feature extraction on the gallery in the offline phase, and a lightweight network for query processing in the online phase;
Towards Training-Free and Accurate ANN-to-SNN Conversion via Activation-Aware Redistribution
Honglin Cao, Haizhou Li (Shenzhen Loop Area Institute)
OptimizationComputational EfficiencyHyperparameter SearchSpiking Neural NetworkTransformerImageText
🎯 What it does: Propose a training-free ANN-to-SNN conversion method that utilizes adaptive integrate-and-fire (AIF) neurons to achieve activation distribution-aware redistribution, significantly reducing conversion error.
Towards Ultrasound-based Reliable Disease Diagnosis Using Causal Inference
Bolei Chen (Central South University), Jianxin Wang (Tongji University)
ClassificationConvolutional Neural NetworkTransformerImageMultimodalityUltrasound
🎯 What it does: This paper proposes a multimodal ultrasound diagnostic framework based on causal inference, utilizing two causal adjustment methods, BACL and FACL, to eliminate explicit and implicit confounding biases, achieving consistency between model decisions and expert diagnostic logic.
Towards Understanding Generalization in DP-GD: A Case Study in Training Two-Layer CNNs
Zhongjie Shi (Georgia Institute of Technology), Yuan Cao (RPTU Kaiserslautern-Landau)
ClassificationSafty and PrivacyConvolutional Neural Network
🎯 What it does: This paper studies the performance of two-layer Huberized ReLU convolutional neural networks in binary classification tasks, comparing the generalization and privacy performance of non-private gradient descent (GD) and differential privacy gradient descent (DP-GD), and proves that DP-GD can achieve better test accuracy within a specific signal-to-noise ratio (SNR) range.
Towards Understanding In-Context Learning of Transformers Under Non-I.I.D. Scenarios
Qilu Shen (Key Laboratory of Smart Farming for Agricultural Animals), Jinhai Xiang (Key Laboratory of Smart Farming for Agricultural Animals)
Meta LearningTransformerText
🎯 What it does: This paper proposes a general theory of Transformer self-learning (ICL) under non-i.i.d. scenarios, providing upper bounds on gradient stability and generalization error, and extending the analysis to multi-layer Transformers, further quantifying the impact of random features (PRF) on attention kernel approximation error.
Towards Unified Vision-Language Models with Incomplete Multi-Modal Inputs
Xiang Fang (Huazhong University of Science and Technology), Wei Ji (Guangzhou University)
RetrievalKnowledge DistillationRepresentation LearningTransformerLarge Language ModelVision Language ModelVideoTextMultimodality
🎯 What it does: To address the modality missing problem in video-language models, this paper proposes a unified incomplete video-language alignment model, which enhances robustness for various downstream tasks (video retrieval, video question answering, video sentence localization) by simultaneously handling incomplete videos and texts during both training and inference phases.
Towards Zero-Shot Diabetic Retinopathy Grading: Learning Generalized Knowledge via Prompt-Driven Matching and Emulating
Huan Wang (University of Wollongong), Jun Shen (University of Wollongong)
ClassificationTransformerPrompt EngineeringContrastive LearningBiomedical Data
🎯 What it does: Propose a two-stage prompt-driven framework called ProME-DR based on CLIP for zero-shot diabetic retinopathy (DR) grading on multi-view retinal images.
TowerMind: A Tower Defence Game Learning Environment and Benchmark for LLM as Agents
Dawei Wang (Newcastle University), Richard Davison (Newcastle University)
TransformerLarge Language ModelReinforcement LearningAgentic AIPrompt EngineeringImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes TowerMind, a lightweight tower defense game environment designed to evaluate the long-term planning and decision-making capabilities of large language models (LLMs), supporting text, pixel, and structured observations.
TR-DQ: Time-Rotation Diffusion Quantization
Yihua Shao (Peking University), Jingcai Guo (Hong Kong Polytechnic University)
GenerationComputational EfficiencyTransformerDiffusion modelImageVideo
🎯 What it does: Propose the Time-Rotation Diffusion Quantization (TR-DQ) method, which applies time-step-aware rotation matrix transformation to the activations and weights of diffusion models, achieving efficient low-bit quantization;
TRACE: A Generalizable Drift Detector for Streaming Data-Driven Optimization
Yuan-Ting Zhong (South China University Of Technology), Yue-Jiao Gong (South China University Of Technology)
Anomaly DetectionOptimizationTransformerTabularTime Series
🎯 What it does: Proposed a transferable concept drift detector called TRACE, and integrated it as a plugin into a streaming data-driven optimization algorithm (TRACE-EA).
TRACE: Trajectory-based Activation Change Estimation for Task-specific Data Selection
Ye He (Harbin Institute of Technology), Qi Shi (Tsinghua University)
OptimizationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes the TRACE algorithm, which selects task-specific data subsets by using neural activation changes induced by a single fine-tuning step on the validation set as the selection signal.
TRACE: Transformation-Aware Graph Refinement for Reaction Condition Prediction
Yujie Chen (Hunan University), Xiangxiang Zeng (Hunan University)
Drug DiscoveryGraph Neural NetworkGraph
🎯 What it does: Proposed the TRACE framework, which constructs a joint graph of reactants and products and performs dynamic interaction refinement to achieve conditional prediction;
TraceTrans: Translation and Spatial Tracing for Surgical Prediction
Xiyu Luo (Southern University of Science and Technology), Tianyang Zhang (University of Birmingham)
Image TranslationConvolutional Neural NetworkGenerative Adversarial NetworkOptical FlowBiomedical Data
🎯 What it does: Propose TraceTrans, an end-to-end image-to-image translation model for postoperative prediction, capable of generating trackable spatial deformations from preoperative images while maintaining target distribution matching.
Tracing the Heart’s Pathways: ECG Representation Learning from a Cardiac Conduction Perspective
Tan Pan (Fudan University), Kaiyu Guo (Shanghai Academy of Artificial Intelligence for Science)
Representation LearningTransformerAuto EncoderContrastive LearningBiomedical DataElectrocardiogram
🎯 What it does: This paper proposes the CLEAR-HUG two-stage framework, which uses ECG self-supervised learning to capture subtle differences in cardiac conduction and simulate clinical diagnostic processes;
TrackGS: Optimizing COLMAP-Free 3D Gaussian Splatting with Global Track Constraints
Dongbo Shi (University Of Science And Technology Of China), Renjie Chen (Independent Researcher)
Pose EstimationDepth EstimationOptimizationGaussian SplattingSimultaneous Localization and MappingImageVideo
🎯 What it does: Propose the TrackGS method, which jointly optimizes camera parameters and scene representation for 3D Gaussian Splatting using global feature trajectories without relying on COLMAP;
Tracking and Segmenting Anything in Any Modality
Tianlu Zhang (Tsinghua University), Jungong Han (Tsinghua University)
Object TrackingSegmentationTransformerMixture of ExpertsImageVideoMultimodality
🎯 What it does: Proposed a unified tracking and segmentation framework named SATA, which can simultaneously accomplish four tasks (SOT, VOS, MOT, MOTS) under any modality (RGB, RGB-T, RGB-D, RGB-E), achieving fully shared model architecture and parameters.
Tracking the Unstable: Appearance-Guided Motion Modeling for Robust Multi-Object Tracking in UAV-Captured Videos
Jianbo Ma (State Key Laboratory of Optical Field Manipulation Science and Technology, Institute of Optics and Electronics, Chinese Academy of Sciences), Ming-Hsuan Yang (University of California, Merced)
Object TrackingConvolutional Neural NetworkVideo
🎯 What it does: Proposes a multi-object tracking framework AMOT based on appearance-motion consistency, achieving robust identity association and trajectory recovery in UAV videos through the AMC matrix and MTC module.
Tractable Sharpness-Aware Learning of Probabilistic Circuits
Hrithik Suresh (Indian Institute of Technology Palakkad), Narayanan Chatapuram Krishnan (University of Texas at Dallas)
OptimizationRepresentation LearningData-Centric LearningFlow-based ModelTabularBenchmark
🎯 What it does: Studied the overfitting problem of probabilistic circuits (PC) under limited data, proposed a sharpness regularization method based on the Hessian trace, and implemented closed-form updates in EM and gradient optimization;
Tractable Weighted First-Order Model Counting with Bounded Treewidth Binary Evidence
Václav Kůla (Czech Technical University in Prague), Ondřej Kuželka (Czech Technical University in Prague)
Computational EfficiencyGraph
🎯 What it does: This paper proposes a dynamic programming algorithm based on tree decomposition, which can solve weighted first-order model counting (WFOMC) for FO²/C² in polynomial time with respect to domain size when the tree width of the Gaifman graph for binary predicate evidence is limited.
Trade-offs in Large Reasoning Models: An Empirical Analysis of Deliberative and Adaptive Reasoning over Foundational Capabilities
Weixiang Zhao (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: This paper systematically evaluates the impact of integrating deep reasoning capabilities into large reasoning models (LRMs) on fundamental capabilities (usefulness and safety), and explores how adaptive reasoning modes can alleviate negative effects.
Trainable EEG Interpolation and Structure-Sharing Dual-Path Encoders for Brain-Assisted Target Speaker Extraction
Zhao Lv (Anhui University), Cunhang Fan (Anhui University)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkBiomedical DataAudio
🎯 What it does: Propose a brain-augmented target speaker extraction framework (TIDENet) that utilizes trainable EEG interpolation and a structure-shared dual-path encoder, enabling the simultaneous extraction of target-related and irrelevant features from EEG and mixed speech, and achieving feature complementary fusion within a single model.
Training and Inference Within 1 Second – Tackle Cross-Sensor Degradation of Real-World Pansharpening with Efficient Residual Feature Tailoring
Tianyu Xin (University of Electronic Science and Technology of China), Liang-Jian Deng (University of Electronic Science and Technology of China)
RestorationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose an unsupervised, no-additional-training-data "Efficient Residual Feature Tailoring (ERFT)" framework, achieving sub-second training and inference in cross-sensor pan-sharpening tasks.
Training-Free ANN-to-SNN Conversion for High-Performance Spiking Transformers
Jingya Wang (University of Electronic Science and Technology of China), Malu Zhang (University of Electronic Science and Technology of China)
ClassificationGenerationComputational EfficiencySpiking Neural NetworkTransformerImageText
🎯 What it does: Proposed an untrained ANN-to-SNN conversion framework, using Multi-Basis Exponential Decay (MBE) neurons to approximate non-linear operations in Transformer models;
Training-free Boosting for Few-shot Segmentation via Generalizing Semantic Mining
Kangyu Xiao (University Of Science And Technology Of China), Junjie Li (Hefei Comprehensive National Science Center)
SegmentationTransformerImage
🎯 What it does: Proposes a training-free Generalizing Semantic Mining (GSM) framework to enhance affinity computation during the inference phase of few-shot semantic segmentation (FSS), comprising three modules: Target Semantic Reusing (TSR), Query-specific Semantic Modeling (QSM), and Representative Re-weighting (RR).
Training-Free Multi-Character Audio-Driven Animation via Diffusion Transformer with Reward Feedback
Xingpei Ma (Guangzhou Quwan Network Technology), Shunsi Zhang (Guangzhou Quwan Network Technology)
GenerationTransformerReinforcement LearningDiffusion modelFlow-based ModelVideoMultimodalityAudio
🎯 What it does: Proposed an audio-driven multi-character animation framework based on Diffusion Transformer (DiT), capable of supporting arbitrary-length video generation and achieving untrained multi-character animation.
Training-Free Spatio-temporal Decoupled Reasoning Video Segmentation with Adaptive Object Memory
Zhengtong Zhu (Soochow University), Fanzhang Li (Soochow University)
Object TrackingSegmentationTransformerVision Language ModelVideo
🎯 What it does: Proposed a training-agnostic, spatiotemporally decoupled inference video segmentation framework SDAM, achieving stable segmentation through adaptive keyframe selection and memory modules.
TrajAgg: Dual-Scale Feature Aggregation with Hybrid Training for Trajectory Similarity Computation in Free Space
Xiao Zhang (Ocean University of China), Yanwei Yu (Ocean University of China)
RetrievalAutonomous DrivingRepresentation LearningRecurrent Neural NetworkTransformerSupervised Fine-TuningTime SeriesSequential
🎯 What it does: Proposed the TrajAgg framework, which achieves efficient and accurate trajectory similarity computation using aggregation Transformer and hybrid training.
TrajEvo: Trajectory Prediction Heuristics Design via LLM-driven Evolution
Zhikai Zhao (KAIST), Jinkyoo Park (KAIST)
Autonomous DrivingOptimizationTransformerLarge Language ModelTime SeriesSequential
🎯 What it does: Automatically generate trajectory prediction heuristic rules using large language model-driven evolutionary algorithms
Transferability of Adversarial Attacks in Video-based MLLMs: A Cross-modal Image-to-Video Approach
Linhao Huang (Tsinghua University), Feng Zheng (China Electronics Corporation)
Adversarial AttackVision Language ModelImageVideoMultimodality
🎯 What it does: Studied black-box transferable adversarial attacks on video multi-modal large language models (V-MLLM), and proposed a cross-modal I2V-MLLM method based on image large models
Transferable Backdoor Attacks for Code Models via Sharpness-Aware Adversarial Perturbation
Shuyu Chang (Nanjing University of Posts and Telecommunications), Leo Yu Zhang (Griffith University)
Adversarial AttackTransformerText
🎯 What it does: Proposed a transferable backdoor attack method named STAB for backdoor implantation in code models.
Transferable Graph Condensation from the Causal Perspective
Huaming Du (Southwestern University of Finance and Economics), Gang Kou (Emory University)
CompressionDomain AdaptationGraph Neural NetworkContrastive LearningGraphFinance Related
🎯 What it does: A transferable graph data compression method called TGCC based on causal invariance is studied.
Transferable Hypergraph Attack via Injecting Nodes into Pivotal Hyperedges
Meixia He (Northwestern Polytechnical University), Keke Tang (Inner Mongolia University)
Adversarial AttackGraph Neural NetworkGraph
🎯 What it does: Propose a transferable hypergraph node injection attack (TH-Attack), which identifies critical hyperedges in aggregation paths and generates malicious node injections using a feature inverter to disrupt feature propagation in hypergraph neural networks.
Transferable Model-agnostic Vision-Language Model Adaptation for Efficient Weak-to-Strong Generalization
Jihwan Park (Korea University), Hyunwoo J. Kim (KAIST)
Knowledge DistillationSupervised Fine-TuningVision Language ModelImage
🎯 What it does: Propose a lightweight adapter called TransMiter without gradient backpropagation, which transfers adaptation knowledge from weak models to stronger vision-language models.
Transferring Causal Driving Patterns for Generalizable Traffic Simulation with Diffusion-Based Distillation
Yuhang Chen (Tongji University), Jian Sun (Tongji University)
Autonomous DrivingKnowledge DistillationDiffusion modelContrastive LearningBenchmark
🎯 What it does: This paper proposes a two-stage knowledge distillation framework named CDPT, which utilizes diffusion models to distill causal driving patterns from traffic simulations and achieve cross-domain transfer.
Transform-Free Feature Coding via Entropy-Constrained Vector Quantization
Qiaoxi Chen, Dong Liu (University Of Science And Technology Of China)
ClassificationSegmentationCompressionConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose an entropy-constrained vector quantization (ECVQ)-based feature encoding framework that directly encodes deep features without using traditional transformation modules.
Transformer with Controlled Attention for Synchronous Motion Captioning
Karim Radouane (University of Montpellier), Andon Tchechmedjiev (University of Montpellier)
GenerationTransformerVision-Language-Action ModelMultimodality
🎯 What it does: This paper proposes a Transformer-based synchronized action caption generation method, which achieves synchronized output of actions and text through a control attention mechanism.
Transformers in Pseudo-Random Number Generation: A Dual Perspective on Theory and Practice
Ran Li (Northeast Normal University), Lingshu Zeng (Northeast Normal University)
GenerationTransformerLarge Language ModelSequentialChain-of-Thought
🎯 What it does: This paper studies the simulation of classical pseudo-random number generators (LCG and Mersenne Twister) using Transformer (decoder-only + Chain-of-Thought), and theoretically proves that it can achieve non-uniform AC0; subsequently, pseudo-random number sequences are generated using GPT-2 training, and their performance is verified through NIST tests and prediction attack experiments.
TransFR: Transferable Federated Recommendation with Adapter Tuning on Pre-trained Language Models
Honglei Zhang (Key Laboratory of Big Data and Artificial Intelligence in Transportation, Ministry of Education), Yidong Li (Peking University)
Recommendation SystemFederated LearningSafty and PrivacyKnowledge DistillationTransformerText
🎯 What it does: Proposed TransFR, a transferable federated recommendation framework that generates domain-agnostic text embeddings using pre-trained language models, and achieves cross-domain recommendations through federated adapter fine-tuning and post-adapter personalization;
TransMamba: A Sequence-Level Hybrid Transformer-Mamba Language Model
Yixing Li (Chinese University of Hong Kong), Yu Cheng (Tencent)
TransformerLarge Language ModelText
🎯 What it does: Proposed a sequence-level hybrid Transformer-Mamba language model called TransMamba, which dynamically switches between attention and state space models in the same layer with shared parameters, achieving efficient long-sequence modeling.
Transolver Is a Linear Transformer: Revisiting Physics-Attention Through the Lens of Linear Attention
Wenjie Hu (National University of Defense Technology), Yong Dou (National University of Defense Technology)
Computational EfficiencyTransformerMeshBenchmarkPhysics Related
🎯 What it does: This paper proposes LinearNO, which reinterprets Transolver's Physics-Attention as linear attention and achieves a more efficient model through two-step generalization and simplification;
TraveLLaMA: A Multimodal Travel Assistant with Large-Scale Dataset and Structured Reasoning
Meng Chu (Hong Kong University of Science and Technology), Jiaya Jia (Hong Kong University of Science and Technology)
Recommendation SystemExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningAgentic AIVision Language ModelVision-Language-Action ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Developed and evaluated TraveLLaMA—a multimodal language model specialized for travel planning, incorporating a large-scale travel QA dataset, a structured reasoning framework, and an interactive agent;
Treatment Stitching with Schrödinger Bridge for Enhancing Offline Reinforcement Learning in Adaptive Treatment Strategies
Dong-Hee Shin (Korea University), Tae-Eui Kam (Korea University)
Diffusion modelScore-based ModelBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: Propose a data augmentation framework named TreatStitch, which concatenates similar intermediate states from existing offline treatment trajectories and employs Schrödinger bridge to generate smooth bridging trajectories when similar states are absent, thereby generating clinically effective synthetic trajectories for offline reinforcement learning.
Tree-Based Stochastic Optimization for Solving Large-Scale Urban Network Security Games
Shuxin Zhuang (City University of Hong Kong), Youzhi Zhang (Hong Kong Institute of Science & Innovation)
OptimizationGraph
🎯 What it does: A tree-structured stochastic optimization framework (TSO) is studied for solving Nash equilibrium in large-scale urban cyber security games.
TriFusion-IDS: A Multimodal Graph-Tabular-Text Contrastive Framework for Cross-Dataset Intrusion Detection
Qinxin Zhao (Nanjing University), Sheng Zhong (Nanjing University)
Anomaly DetectionRepresentation LearningGraph Neural NetworkTransformerContrastive LearningTextGraphTabular
🎯 What it does: Propose a contrastive learning framework named TriFusion-IDS that integrates graph, tabular, and textual modalities for cross-dataset zero-shot intrusion detection.
Trimming the Fat: Redundancy-Aware Acceleration Framework for DGNNs
Renhong Huang (Zhejiang University), Yang Yang (MyBank AntGroup)
Computational EfficiencyGraph Neural NetworkGraph
🎯 What it does: Accelerating Dynamic Graph Neural Networks by Removing Static and Runtime Redundancies
TrinityDNA: A Bio-Inspired Foundational Model for Efficient Long-Sequence DNA Modeling
Qirong Yang (BioMap Research), Xiaoming Zhang (BioMap Research)
TransformerLarge Language ModelBiomedical DataBenchmark
🎯 What it does: Proposed TrinityDNA, a foundational model for DNA sequences, which captures DNA structural features and long-range dependencies using the Groove Fusion module, Gated Reverse Complement mechanism, and multi-scale sliding window attention.
TRIPLE: Theory-Driven Integration of Planned and Habitual Behaviors for LLM-based Personalization
Taehyung Noh (Hanyang University), Kyungsik Han (Hanyang University)
Recommendation SystemExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Propose the TRIPLE framework, leveraging the dual-process theory (habit and intention) to construct user profiles for LLMs and generate behavioral reasoning
TripleFDS: Triple Feature Disentanglement and Synthesis for Scene Text Editing
Yuchen Bao (Southern University of Science and Technology), Jianguo Zhang (Tencent)
Image TranslationGenerationData SynthesisTransformerAuto EncoderContrastive LearningImage
🎯 What it does: Propose the TripleFDS framework, which explicitly splits and reconstructs the ternary features (text content, style, and background) in scene text editing;
TRT: Harnessing Tensor Ring Transformer for Hyperspectral Image Super-Resolution
honghui xu, Jianwei Zheng (Zhejiang University of Technology)
Super ResolutionTransformerImage
🎯 What it does: The paper proposes reformulating the high-resolution super-resolution problem as a robust principal component analysis (RPCA)-style denoising task, and designs a Tensor Ring Transformer (TRT) as a prior network, achieving efficient reconstruction of hyperspectral images.
Truncated Counterfactual Learning for Anytime Multi-Agent Path Finding
Thomy Phan (University of Bayreuth), Sven Koenig (University of Bayreuth)
OptimizationGraphBenchmark
🎯 What it does: Propose an algorithm named TACKLE based on causal inference and truncated adaptive adversarial learning to improve the selection of seed agents in delay-driven MAPF-LNS, thereby enhancing the anytime performance of multi-agent path planning.
TRUST: Leveraging Text Robustness for Unsupervised Domain Adaptation
Mattia Litrico (University of Catania), Devis Tuia (École Polytechnique Fédérale de Lausanne)
Domain AdaptationTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: TRUST proposes a framework for unsupervised domain adaptation leveraging text robustness, enhancing the generalization of visual models in target domains through pseudo label generation, uncertainty estimation, and multimodal soft contrastive learning.
Trusted Multi-view Learning for Long-tailed Classification
Chuanqing Tang (Southwestern University of Finance and Economics), Long Shi (Southwestern University of Finance and Economics)
ClassificationImage
🎯 What it does: Proposed a trustworthy multi-perspective long-tailed classification framework called TMLC, which addresses imbalanced classification in multi-perspective scenarios through opinion aggregation and pseudo data generation.
Truth-Tracking Evaluation in Opinion-Based Argumentation
Juliete Rossie (CRIL, CNRS, Univ. Artois), Srdjan Vesic (CRIL, CNRS, Univ. Artois)
GraphBenchmark
🎯 What it does: Proposed the truth tracking problem under the OBA framework and constructed the VAST evaluation benchmark;
Truth, Justice, and Secrecy: Cake Cutting Under Privacy Constraints
Yaron Salman (Open University of Israel), Roie Zivan (Ben Gurion University of Negev)
OptimizationSafty and Privacy
🎯 What it does: Proposed the first privacy-preserving cake-cutting protocol (PP CC PUV) that simultaneously achieves fairness, envy-freeness, and strategy-proofness
TruthfulRAG: Resolving Factual-level Conflicts in Retrieval-Augmented Generation with Knowledge Graphs
Shuyi Liu (Beijing University of Posts and Telecommunications), Xi Zhang (Beijing University of Posts and Telecommunications)
GenerationGraph Neural NetworkTransformerLarge Language ModelGraphRetrieval-Augmented Generation
🎯 What it does: Propose the TruthfulRAG framework, using knowledge graphs to address factual knowledge conflicts in retrieval-augmented generation;
TSBOW – Traffic Surveillance Benchmark for Occluded Vehicles Under Various Weather Conditions
Ngoc Doan-Minh Huynh (Sungkyunkwan University), Jae Wook Jeon (Sungkyunkwan University)
Object DetectionConvolutional Neural NetworkTransformerVideoBenchmark
🎯 What it does: This paper proposes the TSBOW (Traffic Surveillance Benchmark for Occluded vehicles under various Weather conditions) dataset, which includes 32 hours of CCTV surveillance video, 3.2M frames, and annotations for 8 categories of traffic participants, with object detection experiments conducted on this dataset.
TSGDiff: Rethinking Synthetic Time Series Generation from a Pure Graph Perspective
Lifeng Shen (Chongqing University of Posts and Telecommunications), Lele Long (Chongqing University of Posts and Telecommunications)
GenerationData SynthesisGraph Neural NetworkDiffusion modelAuto EncoderGraphTime Series
🎯 What it does: Propose a graph-based temporal data generation framework called TSGDiff, which converts multivariate time series into dynamic graphs and generates synthetic time series through diffusion models in the latent graph space;
TSPE-GS: Probabilistic Depth Extraction for Semi-Transparent Surface Reconstruction via 3D Gaussian Splatting
Zhiyuan Xu (Southeast University), Tong Wei (Southeast University)
GenerationNeural Radiance FieldGaussian SplattingImagePoint Cloud
🎯 What it does: Propose the TSPE-GS method, which utilizes a probabilistic deep distribution for multi-modal depth extraction of semi-transparent surfaces in 3D Gaussian Splatting, and achieves multi-layer surface reconstruction through advanced TSDF fusion.
TSPO: Temporal Sampling Policy Optimization for Long-form Video Language Understanding
Canhui Tang (Xi'an Jiaotong University), Hao Sun (China Telecom)
TransformerLarge Language ModelReinforcement LearningVision Language ModelVideoMultimodality
🎯 What it does: Proposed the Temporal Sampling Policy Optimization (TSPO) framework, which employs reinforcement learning to end-to-end train an event-aware temporal sampler, achieving joint decision-making for sparse frame sampling and language generation in long videos;
TTA-Bench: A Comprehensive Benchmark for Evaluating Text-to-Audio Models
Hui Wang (Nankai University), Yong Qin (Nankai University)
GenerationLarge Language ModelPrompt EngineeringTextBenchmarkAudio
🎯 What it does: Proposed TTA-Bench—a comprehensive evaluation framework covering functional quality, reliability, and social responsibility, along with constructing 2,999 diverse text-audio prompts and a unified evaluation protocol.
TTF-VLA: Temporal Token Fusion via Pixel-Attention Integration for Vision-Language-Action Models
Chenghao Liu (Peking University), Huiling Duan (Peking University)
TransformerVision-Language-Action ModelMultimodality
🎯 What it does: Propose a training-agnostic temporal token fusion framework that intelligently merges historical and current visual tokens by leveraging gray-scale pixel differences and attention-based semantic relevance detection, thereby enhancing the inference quality of VLA models.
TubeRMC: Tube-conditioned Reconstruction with Mutual Constraints for Weakly-supervised Spatio-Temporal Video Grounding
Jinxuan Li (Sun Yat-sen University), Beihao Xia (Hong Kong University of Science and Technology)
Object DetectionObject TrackingSegmentationTransformerContrastive LearningGaussian SplattingVideoText
🎯 What it does: Propose a weakly supervised spatiotemporal video localization framework called TubeRMC, which generates and refines video tubes that conform to language descriptions through reconstruction and mutual constraints under textual conditions.
TuckA: Hierarchical Compact Tensor Experts for Efficient Fine-Tuning
Qifeng Lei (University of Chinese Academy of Sciences), Qingming Huang (Chinese Academy of Sciences)
Computational EfficiencyRepresentation LearningSupervised Fine-TuningMixture of ExpertsImageText
🎯 What it does: Construct a multi-expert adapter using Tucker decomposition to achieve parameter-efficient fine-tuning of large pre-trained models.
Tuning for Two Adversaries: Enhancing the Robustness Against Transfer and Query-Based Attacks Using Hyperparameter Tuning
Pascal Zimmer (Ruhr University Bochum), Ghassan Karame (Ruhr University Bochum)
Adversarial AttackHyperparameter SearchImage
🎯 What it does: This paper systematically studies the impact of training hyperparameters (learning rate, weight decay, momentum, batch size) on the robustness against black-box attacks, and proposes hyperparameter tuning strategies for transmission attacks and query attacks.
Tuning Medical Foundation Models for Inner Ear Temporal CT Analysis with Plug-and-play Domain Knowledge Aggregator
Weixun Wan, Cairong Zhao (Microsoft Research Asia)
ClassificationSegmentationAnomaly DetectionConvolutional Neural NetworkTransformerSupervised Fine-TuningImageBiomedical DataComputed TomographyBenchmark
🎯 What it does: Proposed a public pediatric inner ear CT dataset (CIED) and designed a pluggable Domain Knowledge Guided Tuning (DKGT) framework to enhance structural anomaly detection, post-operative hearing prediction, and anatomical segmentation of inner ear CT in data-scarce scenarios.
Tuning-Free Amodal Segmentation via the Occlusion-Free Bias of Inpainting Models
Jae Joong Lee (Purdue University), Raymond A. Yeh (Purdue University)
SegmentationDiffusion modelImage
🎯 What it does: Leveraging the unoccluded bias of pre-trained diffusion-based image inpainting models, we propose a fully parameter-free zero-shot method to achieve complete segmentation of occluded objects.
TuningIQA: Fine-Grained Blind Image Quality Assessment for Livestreaming Camera Tuning
Xiangfei Sheng (Xidian University), Leida Li (Xidian University)
OptimizationConvolutional Neural NetworkGraph Neural NetworkImageTabularBenchmark
🎯 What it does: Proposes TuningIQA, a fine-grained blind image quality assessment method specifically designed for live stream camera optimization, and constructs the FGLive-10K dataset.
Turbo-VAED: Fast and Stable Transfer of Video-VAEs to Mobile Devices
Ya Zou (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
GenerationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkVideo
🎯 What it does: To deploy large video generation models on mobile devices, we propose Turbo-VAED, a lightweight VAE decoder that reduces parameter redundancy using 3D depthwise separable convolutions, achieves mobile-friendly upsampling through decoupled 3D pixel shuffling, and rapidly transfers the teacher model via decoder-only distillation.
TVChain: Leveraging Textual-Visual Prompt Chains for Jailbreaking Large Vision-Language Models
Hao Yu (National University of Defense Technology), Xinwang Liu (National University of Defense Technology)
Safty and PrivacyAdversarial AttackLarge Language ModelPrompt EngineeringDiffusion modelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose a black-box jailbreak framework TVChain, which decomposes malicious queries into text-visual chains and bypasses security mechanisms through chain-of-thought text prompts.
TW-CRL: Time-Weighted Contrastive Reward Learning for Efficient Inverse Reinforcement Learning
Yuxuan Li (Northwestern University), Stephen Xia (Northwestern University)
Reinforcement LearningContrastive LearningBenchmark
🎯 What it does: Propose a time-weighted contrastive reward learning framework (TW-CRL) that utilizes successful and failed demonstrations to learn dense reward functions from sparse rewards in periodic tasks.
TweezeEdit: Consistent and Efficient Image Editing with Path Regularization
Jianda Mao (Hong Kong University of Science and Technology), Kani Chen (Hong Kong University of Science and Technology)
GenerationDiffusion modelImage
🎯 What it does: Proposes TweezeEdit, a parameter-free and inversion-free image editing framework based on consistency models, which can achieve target prompt alignment while preserving the semantics of the source image;
Two Constraint Compilation Methods for Lifted Planning
Periklis Mantenoglou (rebro University), Pedro Zuidberg Dos Martires (rebro University)
Computational EfficiencyBenchmark
🎯 What it does: Propose two lifted constraint compilation methods, LiftedTCORE and LCC, to handle qualitative state trajectory constraints in PDDL, avoiding global variable substitution across the problem domain.
Two Heads Are Better than One: Distilling Large Language Model Features into Small Models with Feature Decomposition and Mixture
Tianhao Fu (Peking University), Xixin Cao (Peking University)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelReinforcement LearningMixture of ExpertsTime SeriesFinance Related
🎯 What it does: Proposes a Cooperative Market Making (CMM) framework that decouples features of large language models (LLMs) in three dimensions—layer, task, and data—and leverages small models for collaborative learning. It further achieves efficient market making in real-time trading environments through H'ajek projection-based expert fusion.
U2B: Scale-unbiased Representation Converter for Graph Classification with Imbalanced and Balanced Scale Distributions
Guanjun Wang (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
ClassificationRepresentation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: Proposed the U2B framework to eliminate bias caused by graph scale imbalance in graph classification tasks, employing a two-stage distillation-refinement approach with node-level and graph-level transformers.
U2UData+: A Scalable Swarm UAVs Autonomous Flight Dataset for Embodied Long-horizon Tasks
Tongtong Feng (Tsinghua University), Wenwu Zhu (Tsinghua University)
Data SynthesisAutonomous DrivingImageMultimodalityPoint CloudTime SeriesSequentialBenchmark
🎯 What it does: This paper constructs the first large-scale drone swarm autonomous flight dataset for embedded long-haul tasks (ELH), named U2UData+, and introduces a scalable platform for online data collection and closed-loop verification;
UAV4D: Dynamic Neural Rendering of Human-Centric UAV Imagery Using Gaussian Splatting
Jaehoon Choi (University of Maryland), Heesung Kwon (University of Maryland)
GenerationData SynthesisPose EstimationNeural Radiance FieldGaussian SplattingVideoMesh
🎯 What it does: Propose the UAV4D framework, which utilizes the SMPL human mesh and 3D base model to separate and reconstruct dynamic humans from static backgrounds in drone monocular videos, and generates high-quality novel view renderings through Gaussian splatting.
uCLIP: Parameter-Efficient Multilingual Extension of Vision-Language Models with Unpaired Data
Dahyun Chung (Korea University), Byung-Jun Lee (Korea University)
RetrievalComputational EfficiencyRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A lightweight multilingual vision-language model called uCLIP was constructed, training only a 1.7M-parameter projection layer while freezing the CLIP visual encoder and multilingual text encoder. English was used as a semantic bridge to achieve unsupervised training, aligning images with multilingual text.
UCPO: A Universal Constrained Combinatorial Optimization Method via Preference Optimization
Zhanhong Fang (Sun Yat-sen University), Zizhen Zhang (National University of Singapore)
OptimizationReinforcement LearningGraphBenchmark
🎯 What it does: This work proposes the UCPO framework, which leverages preference optimization to directly embed constraints into neural combinatorial optimizers, achieving complex constraint solving without altering the network architecture.
UDA: Unsupervised Debiasing Alignment for Pair-wise LLM-as-a-Judge
Yang Zhang (ZhipuAI), Jie Tang (Westlake University)
TransformerLarge Language ModelContrastive LearningTextBenchmark
🎯 What it does: This paper proposes an unsupervised debiasing alignment (UDA) framework that dynamically corrects the Elo scores of LLM-as-a-judge to reduce evaluation discrepancies caused by self-preference.