AAAI 2024 Papers — Page 22
AAAI Conference on Artificial Intelligence · 2331 papers
Topic-VQ-VAE: Leveraging Latent Codebooks for Flexible Topic-Guided Document Generation
YoungJoon Yoo (NAVER Cloud), JongWon Choi
GenerationData SynthesisTransformerAuto EncoderImageText
🎯 What it does: Implement topic modeling using the discrete embedding codebook of VQ-VAE, and based on this, achieve scalable text and image generation.
TopoGCL: Topological Graph Contrastive Learning
Yuzhou Chen (Temple University), Yulia R. Gel (University of Texas at Dallas)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes an unsupervised graph contrastive learning framework called TopoGCL, which achieves stronger representation learning through topology-topology contrast between two graph views.
Toward Open-Set Human Object Interaction Detection
Mingrui Wu (Xiamen University), Rongrong Ji (Xiamen University)
RecognitionObject DetectionVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: A decoupled hierarchical method DHD based on open-source object detectors and visual language models has been developed, using discretized image-text contrastive learning to train lightweight unary and pairwise adapters, achieving open-set human-object interaction detection.
Towards a Theoretical Understanding of Why Local Search Works for Clustering with Fair-Center Representation
Zhen Zhang (Hunan University of Technology and Business), Qilong Feng (Central South University)
Optimization
🎯 What it does: Theoretical analysis of an O(1) scale local search (multi-exchange) algorithm for the representative k-median problem (fair center clustering) is conducted, proving that when the number of groups is constant, the algorithm can achieve a constant factor approximation, while if the number of groups exceeds a constant, the approximation ratio is unbounded; an example is provided to demonstrate the lower bound on the required exchange scale.
Towards Automated Chinese Ancient Character Restoration: A Diffusion-Based Method with a New Dataset
Haolong Li (Tongji University), Chen Ye (Tongji University)
RestorationDiffusion modelImage
🎯 What it does: A large-scale real ancient character erasure dataset ARMCD was constructed, and the DiffACR diffusion model was proposed to achieve automatic restoration of ancient Chinese characters.
Towards Automated RISC-V Microarchitecture Design with Reinforcement Learning
Chen Bai (Chinese University of Hong Kong), Martin D. F. Wong
OptimizationReinforcement Learning
🎯 What it does: A reinforcement learning-based automated design framework for RISC-V microarchitecture has been designed and implemented, capable of automatically generating microarchitecture configurations under given performance/power/area (PPA) targets.
Towards Balanced Alignment: Modal-Enhanced Semantic Modeling for Video Moment Retrieval
Zhihang Liu (University of Science and Technology of China), Guoqing Jin (University of Science and Technology of China)
RetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: A Modal-Enhanced Semantic Modeling (MESM) framework is proposed for Video Moment Retrieval (VMR), enhancing video and text features at both the frame-word and paragraph-sentence levels to address the modality imbalance issue.
Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders
Yaohua Zha (Tsinghua University), Shu-Tao Xia (Tsinghua University)
ClassificationSegmentationRepresentation LearningTransformerAuto EncoderPoint Cloud
🎯 What it does: A dual-branch autoencoder that combines global random masking and local block masking, with a local enhancement module (Patch-level convolution) added to the local branch to learn compact 3D representations.
Towards Continual Knowledge Graph Embedding via Incremental Distillation
Jiajun Liu (Southeast University), Yanhe Liu (Chinese Academy of Sciences)
Knowledge DistillationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A continuous knowledge graph embedding framework named IncDE is proposed, which can maintain old knowledge while incrementally learning new triples.
Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding
Depeng Li (Huazhong University of Science and Technology), Zhigang Zeng (Huazhong University of Science and Technology)
ClassificationRecognitionImage
🎯 What it does: A continuous learning framework CLDNet is proposed, which is free from replay and network expansion, utilizing HSIC-Bottleneck Orthogonalization and Equiangular Embedding to address catastrophic forgetting in continuous tasks.
Towards Detailed Text-to-Motion Synthesis via Basic-to-Advanced Hierarchical Diffusion Model
Zhenyu Xie (Sun Yat-sen University), Xiaodan Liang (Tencent)
GenerationData SynthesisDiffusion modelAuto EncoderVideoText
🎯 What it does: A hierarchical diffusion model (B2A‑HDM) is proposed, which first generates rough motion consistent with text in a low-dimensional latent space, and then enhances details through denoising in a high-dimensional latent space.
Towards Diverse Perspective Learning with Selection over Multiple Temporal Poolings
Jihyeon Seong (Korea Advanced Institute of Science and Technology), Jaesik Choi (Korea Advanced Institute of Science and Technology)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTime Series
🎯 What it does: This paper proposes SoM-TP, a learnable multi-temporal pooling method that dynamically selects the most suitable pooling strategy for each sample using an attention mechanism, and achieves multi-view learning through DPLN and perspective loss.
Towards Dynamic Spatial-Temporal Graph Learning: A Decoupled Perspective
Binwu Wang (University of Science and Technology of China), Yang Wang (Shanghai AI Laboratory)
Graph Neural NetworkTransformerAuto EncoderGraphTime Series
🎯 What it does: This paper proposes a Decoupled Learning Framework (DLF) that achieves efficient and sustainable traffic flow prediction by alternately updating seasonal and trend patterns in a dynamic space-time graph.
Towards Effective and General Graph Unlearning via Mutual Evolution
Xunkai Li (Beijing Institute of Technology), Guoren Wang (Beijing Institute of Technology)
Computational EfficiencyData-Centric LearningGraph Neural NetworkGraph
🎯 What it does: A learning-free framework for graph models named MEGU is proposed, which achieves efficient deletion of nodes, features, and edges while maintaining the original model's predictive performance through the mutual evolution of two modules (the prediction module and the forgetting module);
Towards Efficient and Effective Text-to-Video Retrieval with Coarse-to-Fine Visual Representation Learning
Kaibin Tian (Kuaishou Technology), Han Li (Kuaishou Technology)
RetrievalComputational EfficiencyRepresentation LearningTransformerContrastive LearningVideoText
🎯 What it does: The EERCF framework is proposed to achieve efficient and effective text-to-video retrieval;
Towards Efficient Diffusion-Based Image Editing with Instant Attention Masks
Siyu Zou (Xiamen University), Xiaoshuai Sun (NetEase Inc.)
Image TranslationRestorationSegmentationGenerationDiffusion modelImageBenchmark
🎯 What it does: This paper proposes InstDiffEdit, a no-training, no-manual-operation method that utilizes diffusion models and cross-modal attention to instantaneously generate masks for semantic image editing.
Towards Epistemic-Doxastic Planning with Observation and Revision
Thorsten Engesser (IRIT), Elise Perrotin (CRIL)
🎯 What it does: A lightweight multi-agent knowledge and belief planning framework is proposed, capable of handling observation, belief revision, and higher-order false belief tasks.
Towards Equipping Transformer with the Ability of Systematic Compositionality
Chen Huang (Sichuan University), Jiancheng Lv (Sichuan University)
TransformerLarge Language ModelText
🎯 What it does: A Transformer model capable of explicitly learning discrete primitive combinations (CAT) has been designed and implemented, along with two novel pre-training tasks to enhance systematic compositional ability.
Towards Evidential and Class Separable Open Set Object Detection
Ruofan Wang (Fudan University), Rui Feng (Fudan University)
Object DetectionContrastive LearningImage
🎯 What it does: Proposes the Evidential Object Detector (EOD), which estimates class uncertainty and identifies unknown objects in open-set object detection through an evidence deep learning framework.
Towards Explainable Joint Models via Information Theory for Multiple Intent Detection and Slot Filling
Xianwei Zhuang (Peking University), Yuexian Zou (Peking University)
OptimizationExplainability and InterpretabilityGraph Neural NetworkContrastive LearningText
🎯 What it does: This paper proposes a multi-stage iterative joint model called InfoJoint, based on information theory, for intent detection and slot filling in multi-intent speech understanding.
Towards Fair Graph Federated Learning via Incentive Mechanisms
Chenglu Pan (Zhejiang University), Yang Yang (FinVolution Group)
Federated LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a fair incentive mechanism for graph federated learning, which can simultaneously penalize harmful nodes, compensate for delayed contribution nodes, and achieve incentives through gradient sparse allocation and payment rewards.
Towards Fairness in Online Service with K Servers and Its Application on Fair Food Delivery
Daman Deep Singh (Indian Institute of Technology Delhi), Abhijnan Chakraborty (Indian Institute of Technology Delhi)
Flow-based ModelGraph
🎯 What it does: Two new online service models, k-Food and FAIR k-Food, are proposed to simulate real-world scenarios such as food delivery.
Towards Fine-Grained HBOE with Rendered Orientation Set and Laplace Smoothing
Ruisi Zhao (Zhejiang University), Boxi Wu (Zhejiang University)
Pose EstimationTransformerImage
🎯 What it does: This paper proposes a complete angle annotation dataset RMOS based on rendering, which combines Laplace smoothing labels and weighted Smooth-L1 loss, using the OEFormer model with local window self-attention, significantly improving the fine-grained accuracy of human body pose estimation.
Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation
Li Ren (University of Central Florida), Kien Hua (University of Central Florida)
RetrievalDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: A Data Augmentation Domain Adaptation (DADA) framework is proposed, which utilizes adversarial domain adaptation and a mixed feature intermediate domain to align the distributions of samples and proxies, enhancing the efficiency of proxy-based deep metric learning.
Towards Inductive Robustness: Distilling and Fostering Wave-Induced Resonance in Transductive GCNs against Graph Adversarial Attacks
Ao Liu (Sichuan University), Pan Zhou (Huazhong University of Science and Technology)
ClassificationAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: A graph neural network (GRN) based on the principle of wave resonance is constructed, utilizing the natural resonance mechanism of a three-layer GCN to achieve robustness against adversarial attacks on graph structures, and this mechanism is transferred to an inductive learning framework that can generalize to unseen nodes.
Towards Learning and Explaining Indirect Causal Effects in Neural Networks
Abbavaram Gowtham Reddy (Indian Institute of Technology Hyderabad), Satyanarayan Kar (Honeywell)
Explainability and InterpretabilityComputational EfficiencyTabularTime Series
🎯 What it does: A causal explanation method is proposed, which adds feedforward connections based on causal graphs to the input layer of neural networks, enabling the model to learn and explain the direct and indirect causal effects of input features on outputs during the training process.
Towards Making Learnware Specification and Market Evolvable
Jian-Dong Liu (Nanjing University), Zhi-Hua Zhou
ClassificationPose EstimationOptimizationComputational EfficiencyTabularRetrieval-Augmented Generation
🎯 What it does: In the learning model market, an Evolvable Learning Software Specification and Index (ELSI) framework is proposed, which constructs evolving specifications using Reduced Kernel Mean Embedding (RKME) and the loss of models on other specifications (Lf). It achieves fast and efficient learning software retrieval through a tree-structured RKME index and hash tables, thereby efficiently selecting the most helpful models for new tasks without exposing the user's original data.
Towards Model Extraction Attacks in GAN-Based Image Translation via Domain Shift Mitigation
Di Mi (Xiangtan University), Shirui Pan (Griffith University)
Image TranslationAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: By targeting model extraction attacks on GAN image translation models, a domain transfer mitigation method is proposed to enhance the effectiveness of the attack.
Towards Modeling Uncertainties of Self-Explaining Neural Networks via Conformal Prediction
Wei Qian (Iowa State University), Mengdi Huai (Pennsylvania State University)
Explainability and InterpretabilityAdversarial AttackImage
🎯 What it does: A distribution-independent self-explaining neural network uncertainty quantification framework (unSENN) is designed, generating confidence sets for the explanation layer and the final prediction layer through conformal prediction.
Towards More Faithful Natural Language Explanation Using Multi-Level Contrastive Learning in VQA
Chengen Lai (Xidian University), Guangneng Hu (Xidian University)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodalityChain-of-Thought
🎯 What it does: A multi-layer contrastive learning-based natural language explanation model (MCLE) is proposed, which generates explanations and answers through chain-of-thought (COT) and enhances the reliability and consistency of explanations using semantic-level, image-level, and instance-level contrastive learning.
Towards Multi-Intent Spoken Language Understanding via Hierarchical Attention and Optimal Transport
Xuxin Cheng (Peking University), Yuexian Zou (Peking University)
RecognitionOptimizationTransformerText
🎯 What it does: The HAOT framework is proposed to address the scope barrier and unidirectional guidance issues in multi-intent SLU.
Towards Multi-Mode Outlier Robust Tensor Ring Decomposition
Yuning Qiu (Guangdong University of Technology), Qibin Zhao (RIKEN Center for Advanced Intelligence Project)
Anomaly DetectionOptimizationVideoMultimodality
🎯 What it does: A multi-modal robust tensor ring decomposition (ORTRD) method is proposed to recover low-rank tensors from high-order tensors contaminated by multi-modal outliers and noise.
Towards Optimal Subsidy Bounds for Envy-Freeable Allocations
Yasushi Kawase (University of Tokyo), Makoto Yokoo (Kyushu University)
Optimization
🎯 What it does: The study presents a theoretical model for the fair distribution of indivisible goods under limited subsidies and provides an upper bound for optimal subsidies.
Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization
Yongyi Su (South China University of Technology), Kui Jia (Chinese University of Hong Kong)
Domain AdaptationSupervised Fine-TuningImage
🎯 What it does: The research addresses both global and local class imbalance as well as continuous domain shift in Test-Time Adaptation (TTA) in real-world testing, and proposes the TRIBE method.
Towards Robust Image Stitching: An Adaptive Resistance Learning against Compatible Attacks
Zhiying Jiang (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
Adversarial AttackConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a robust attack and adaptive adversarial training for image stitching, enhancing the resistance of deep stitching models to adversarial perturbations.
Towards Running Time Analysis of Interactive Multi-Objective Evolutionary Algorithms
Tianhao Lu (Nanjing University), Chao Qian (Nanjing University)
Optimization
🎯 What it does: The expected running time of the interactive NSGA-II (R-NSGA-II) on synthetic problems such as OneMinMax, OneJumpZeroJump, and OneMinMax* is theoretically analyzed, and its speed performance on different problems is experimentally validated.
Towards Safe Policy Learning under Partial Identifiability: A Causal Approach
Shalmali Joshi (Columbia University), Elias Bareinboim (Columbia University)
OptimizationSafty and PrivacyReinforcement Learning from Human FeedbackBiomedical Data
🎯 What it does: The study investigates how to safely learn personalized treatment strategies in non-identifiable environments with unobserved confounding or partially observable experimental data.
Towards Squeezing-Averse Virtual Try-On via Sequential Deformation
Sang-Heon Shim (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
Image TranslationGenerative Adversarial NetworkOptical FlowImage
🎯 What it does: This paper proposes a Sequential Deformation method to address visual defects of sleeve and waist shrinkage in high-resolution virtual fitting.
Towards Stability and Generalization Bounds in Decentralized Minibatch Stochastic Gradient Descent
Jiahuan Wang (Huazhong Agricultural University), Hong Chen (Huazhong Agricultural University)
Optimization
🎯 What it does: This paper conducts a theoretical analysis of the stability and generalization error of decentralized mini-batch stochastic gradient descent (DM-SGD), providing on-average stability upper bounds for three types of loss functions: convex, strongly convex, and non-convex, from which corresponding generalization error upper bounds are derived.
Towards the Disappearing Truth: Fine-Grained Joint Causal Influences Learning with Hidden Variable-Driven Causal Hypergraphs in Time Series
Kun Zhu (Zhejiang University), Chunhui Zhao (Zhejiang University)
Graph Neural NetworkTime Series
🎯 What it does: This paper proposes a causal hypergraph neural network (CHGNN) driven by hidden variables, which explicitly captures fine-grained joint causal influences to address the issue of weak causality disappearance in time series causal discovery.
Towards the Robustness of Differentially Private Federated Learning
Tao Qi (Tsinghua University), Yongfeng Huang (Tsinghua University)
Federated LearningSafty and PrivacyAdversarial AttackImage
🎯 What it does: This paper first proposes a novel poisoning attack method for differential privacy federated learning (DP-FL) called Attack-DPFL, and subsequently designs a robust aggregation algorithm called Robust-DPFL based on the detection of different distribution characteristics between poisoned gradients and clean gradients, aimed at enhancing the model's resistance to poisoning attacks while maintaining differential privacy.
Towards Transferable Adversarial Attacks with Centralized Perturbation
Shangbo Wu (Beijing Institute of Technology), Yuanzhang Li (Beijing Institute of Technology)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A transferable adversarial attack method for concentrated perturbation optimization in the frequency domain is proposed, utilizing DCT decomposition, quantization, and dynamic quantization matrix optimization to achieve concentrated perturbations, significantly improving the success rate of black-box attacks.
Towards Understanding Future: Consistency Guided Probabilistic Modeling for Action Anticipation
Zhao Xie (Hefei University of Technology), Dan Guo (Hefei University of Technology)
TransformerAuto EncoderVideo
🎯 What it does: A Consistency-Guided Probabilistic Model (CPM) is proposed, which utilizes three main modules—future semantic estimation, global semantic estimation, and global distribution estimation—within the Transformer architecture to smooth temporal continuity through past-future probabilistic consistency, achieving action prediction.
TR-DETR: Task-Reciprocal Transformer for Joint Moment Retrieval and Highlight Detection
Hao Sun (Central China Normal University), Wei Xie (Central China Normal University)
RetrievalTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes a task-complementary Transformer (TR-DETR) based on DETR, achieving joint inference for moment retrieval (MR) and highlight detection (HD).
TraceEvader: Making DeepFakes More Untraceable via Evading the Forgery Model Attribution
Mengjie Wu (Wuhan University), Lina Wang (Nanjing University of Aeronautics and Astronautics)
GenerationAdversarial AttackDiffusion modelGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper presents TraceEvader, a training-free, black-box DeepFake tracking evasion attack that disrupts model attribution by injecting Universal Imitation Traces (UIT) into high-frequency components and applying Gaussian blur to low-frequency components.
Traffic Flow Optimisation for Lifelong Multi-Agent Path Finding
Zhe Chen (Monash University), Peter J. Stuckey (Monash University)
OptimizationRobotic IntelligenceTabular
🎯 What it does: A traffic congestion-aware guiding path is proposed, improving the PIBT and LaCAM* algorithms, enabling low latency and high throughput in large-scale MAPF (Multi-Agent Path Finding) in high-density environments.
Training-Free Quantum Architecture Search
Zhimin He (Foshan University), Haozhen Situ (South China Agricultural University)
OptimizationNeural Architecture SearchTabularPhysics Related
🎯 What it does: A training-free quantum architecture search (TF-QAS) is proposed, which ranks quantum circuits using two zero-training proxies: path counting and expressibility, to automatically find high-performance parameterized quantum circuits.
Transfer and Alignment Network for Generalized Category Discovery
Wenbin An (Xi'an Jiaotong University), Ping Chen (University of Massachusetts Boston)
ClassificationTransformerSupervised Fine-TuningText
🎯 What it does: A framework called TAN is proposed, which combines knowledge transfer and feature alignment for the task of General Category Discovery (GCD);
Transferable Adversarial Attacks for Object Detection Using Object-Aware Significant Feature Distortion
Xinlong Ding (University of Science and Technology Beijing), Huimin Ma (Tsinghua University)
Object DetectionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an attack method based on 'salient feature suppression and neighborhood feature amplification' (OSFD). By using data augmentation (rotation, scaling, blurring) and a specially designed loss function, it perturbs the backbone features of the target detector, achieving high transferability in untargeted black-box attacks.
Transferable Video Moment Localization by Moment-Guided Query Prompting
Hao Jiang (Wangxuan Institute of Computer Technology, Peking University), Yadong Mu (Wangxuan Institute of Computer Technology, Peking University)
Object DetectionDomain AdaptationTransformerPrompt EngineeringVision Language ModelVideoMultimodality
🎯 What it does: This paper proposes a cross-domain transferable video moment localization framework and significantly enhances the model's generalization performance across different datasets through the Moment-Guided Query Prompting (MGQP) method.
Transformer as Linear Expansion of Learngene
Shiyu Xia (Southeast University), Xin Geng (Southeast University)
Computational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: The TLEG method is proposed, which generates different depth Transformers using two sets of shared parameters through a linear formula, and learns learngene through knowledge distillation.
Transformer-Based No-Reference Image Quality Assessment via Supervised Contrastive Learning
Jinsong Shi (Nanjing University of Aeronautics and Astronautics), Jie Qin (Nanjing University of Aeronautics and Astronautics)
Representation LearningConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A novel NR-IQA model SaTQA that combines supervised contrastive learning with Transformer is proposed, which uses pre-training to obtain distortion features and integrates perceptual information to predict image quality.
Transformer-Based Selective Super-resolution for Efficient Image Refinement
Tianyi Zhang (University of Minnesota), Yu Cao (University of Minnesota)
RestorationObject DetectionSuper ResolutionTransformerImage
🎯 What it does: A selective super-resolution (SSR) method is proposed, which divides low-resolution images into several blocks and uses the Tile Selection module of the transformer to select blocks containing the target, performing deep feature reconstruction only on these blocks, while background blocks are upsampled using a lightweight method.
Transformer-Based Video-Structure Multi-Instance Learning for Whole Slide Image Classification
Yingfan Ma (Fudan University), Manning Wang (Fudan University)
ClassificationTransformerImage
🎯 What it does: Divides whole slide images into multiple video segments and performs end-to-end multi-instance learning on these segments using a Transformer with shared parameters, achieving slice-level classification and positive region localization.
TransGOP: Transformer-Based Gaze Object Prediction
Binglu Wang (Xi'an University of Architecture and Technology), Nian Liu (Mohamed bin Zayed University of Artificial Intelligence)
Object DetectionPose EstimationTransformerImage
🎯 What it does: An end-to-end Transformer-based gaze object prediction method called TransGOP is proposed, which can simultaneously perform object detection, gaze estimation, and gaze object localization.
Transient Glimpses: Unveiling Occluded Backgrounds through the Spike Camera
Jiyuan Zhang (Peking University), Tiejun Huang (Peking University)
RestorationSpiking Neural NetworkTransformerVideo
🎯 What it does: Utilizing a single high-speed moving spike camera (Spike Camera) to achieve background de-occlusion tasks through continuous spike streams;
Transition-Informed Reinforcement Learning for Large-Scale Stackelberg Mean-Field Games
Pengdeng Li (Nanyang Technological University), Bo An (Nanyang Technological University)
OptimizationReinforcement Learning
🎯 What it does: A framework for solving large-scale average field games based on reinforcement learning (RL) is proposed, which enhances data efficiency and training stability through Transfer Information Reinforcement Learning (TIRL) and regularization methods.
Transitivity-Preserving Graph Representation Learning for Bridging Local Connectivity and Role-Based Similarity
Van Thuy Hoang (Catholic University of Korea), O-Joun Lee (Catholic University of Korea)
ClassificationRepresentation LearningGraph Neural NetworkTransformerContrastive LearningGraph
🎯 What it does: A Unified Graph Transformation Network (UGT) is proposed, which integrates local and global structural information in the graph attention mechanism by constructing features such as virtual edges, structural identity, structural distance, and transfer probabilities. It uses self-supervised pre-training to obtain universal graph embeddings that can be directly applied to downstream tasks such as node clustering, node classification, and graph-level classification.
Translate Meanings, Not Just Words: IdiomKB’s Role in Optimizing Idiomatic Translation with Language Models
Shuang Li (Fudan University), Yanghua Xiao (Huawei)
OptimizationKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: A multilingual idiom knowledge base (IDIOMKB) has been constructed, utilizing large language models (LLM) for knowledge distillation to generate metaphorical meanings of idioms, and combining this knowledge base with chain-of-thought (KB-CoT) prompts to enhance the performance of small language models in idiom translation tasks.
Transportable Representations for Domain Generalization
Kasra Jalaldoust (Columbia University), Elias Bareinboim (Columbia University)
Domain AdaptationTabular
🎯 What it does: This study addresses the problem of domain generalization by framing it as Transportable Representations, and provides graphical and data-driven criteria for determination.
Trash to Treasure: Low-Light Object Detection via Decomposition-and-Aggregation
Xiaohan Cui (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
Object DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A low-light object detection framework T2 is proposed, which utilizes Retinex decomposition to separate illumination from reflectance, further treating illumination as auxiliary information for detection, and designs a semantic aggregation module to enhance detection performance.
Tree Search-Based Evolutionary Bandits for Protein Sequence Optimization
Jiahao Qiu (Princeton University), Mengdi Wang (MLAB Biosciences Inc)
OptimizationProtein Structure PredictionReinforcement LearningBiomedical Data
🎯 What it does: A protein sequence optimization method based on tree search and bandit learning is proposed, utilizing local mutations and recursive expansion to efficiently explore the sequence space.
TREE-G: Decision Trees Contesting Graph Neural Networks
Maya Bechler-Speicher (Tel Aviv University), Ran Gilad-Bachrach (Tel Aviv University)
ClassificationExplainability and InterpretabilityGraph Neural NetworkGraphBenchmark
🎯 What it does: We propose TREE-G, a model that transforms the decision tree framework to be suitable for graph data. By using a specially designed splitting function, it integrates vertex features with graph topology information to achieve classification and regression predictions for graphs/vertices.
Tree-of-Reasoning Question Decomposition for Complex Question Answering with Large Language Models
Kun Zhang (Chinese Academy of Sciences), Jie Zhou (Chinese Academy of Sciences)
RetrievalOptimizationTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes a multi-hop question answering framework TRQA based on a reasoning tree structure, which decomposes complex questions into sub-questions, retrieves relevant documents, and interacts with large language models to aggregate answers layer by layer from root to leaf, completing multi-hop reasoning.
Trend-Aware Supervision: On Learning Invariance for Semi-supervised Facial Action Unit Intensity Estimation
Yingjie Chen (Peking University), Yun Liang (Peking University)
ClassificationRecognitionConvolutional Neural NetworkImageVideo
🎯 What it does: This paper proposes Trend-Aware Supervision (TAS) for semi-supervised estimation of facial action unit (AU) intensity, utilizing trend information from keyframe annotations to guide the model in learning AU features that are robust against spurious correlations through three types of trend awareness (intra-trend ranking, speed, inter-trend subject), thereby achieving more robust intensity predictions.
Triple Feature Disentanglement for One-Stage Adaptive Object Detection
Haoan Wang (East China Normal University), Zhi Li (East China Normal University)
Object DetectionDomain AdaptationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A three-stage feature separation framework (MDM, CDM, CFDM) is proposed to achieve domain adaptive object detection on a single-stage SSD detector.
TriSampler: A Better Negative Sampling Principle for Dense Retrieval
Zhen Yang (Tsinghua University), Jie Tang (Tsinghua University)
RetrievalTransformerContrastive LearningText
🎯 What it does: This paper proposes TriSampler, which utilizes the 'quasi-triangle principle' to sample negative samples with more informative content in dense retrieval, thereby improving model training effectiveness.
Trust Region Methods for Nonconvex Stochastic Optimization beyond Lipschitz Smoothness
Chenghan Xie (Fudan University), Yinyu Ye (Stanford University)
OptimizationImage
🎯 What it does: A trust-region-based stochastic non-convex optimization algorithm is proposed, achieving convergence to first-order and second-order approximate extremum points under general (L, L0, 1) smoothness and its second-order extension.
Tuning-Free Inversion-Enhanced Control for Consistent Image Editing
Xiaoyue Duan (Beihang University), Junshi Huang (Meituan)
Image TranslationRestorationGenerationDiffusion modelImage
🎯 What it does: A no fine-tuning DDIM inversion enhancement control (TIC) method is proposed to achieve consistent non-rigid editing of real images while maintaining object identity and background details.
TurboSVM-FL: Boosting Federated Learning through SVM Aggregation for Lazy Clients
Mengdi Wang (Technical University of Munich), Enkelejda Kasneci (Technical University of Munich)
OptimizationFederated LearningImage
🎯 What it does: TurboSVM-FL is proposed, an aggregation strategy that utilizes SVM for selective aggregation and maximum margin diffusion regularization in federated learning.
Turning Dust into Gold: Distilling Complex Reasoning Capabilities from LLMs by Leveraging Negative Data
Yiwei Li (Beijing Institute of Technology), Kan Li (Beijing Institute of Technology)
Knowledge DistillationTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper proposes the use of negative samples (incorrect reasoning chains) for knowledge distillation of small language models, thereby enhancing their performance on complex mathematical reasoning tasks.
Turning Waste into Wealth: Leveraging Low-Quality Samples for Enhancing Continuous Conditional Generative Adversarial Networks
Xin Ding (Nanjing University of Information Science and Technology), Zuheng Xu (University of British Columbia)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: In continuous conditional generation tasks, the Dual-NDA method is proposed, which utilizes two types of negative samples (real images with inconsistent labels and visually unrealistic pseudo-images) to improve the training of Continuous Conditional GAN (CcGAN), enhancing the visual quality and label consistency of generated images.
Twice Class Bias Correction for Imbalanced Semi-supervised Learning
Lan Li (Nanjing University), Han-jia Ye (Nanjing University)
ClassificationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposes the TCBC (Twice Class Bias Correction) method, which provides a unified correction for model bias and pseudo-label bias in class-imbalanced semi-supervised learning;
Two-Stage Evolutionary Reinforcement Learning for Enhancing Exploration and Exploitation
Qingling Zhu (Shenzhen University), Wei-Neng Chen (South China University of Technology)
OptimizationReinforcement LearningAgentic AISequential
🎯 What it does: A two-stage evolutionary reinforcement learning (TERL) framework is proposed, where a complete actor-critic network serves as the population individual. It first explores through joint training with RL and PSO, and then focuses on the best individuals for reinforcement learning.
U-Mixer: An Unet-Mixer Architecture with Stationarity Correction for Time Series Forecasting
Xiang Ma (Shandong University), Caiming Zhang (Shandong University)
Convolutional Neural NetworkRecurrent Neural NetworkTime Series
🎯 What it does: A U-Mixer framework that combines Unet and Mixer is proposed to address the non-stationarity problem in time series forecasting.
U-trustworthy Models. Reliability, Competence, and Confidence in Decision-Making
Ritwik Vashistha (University of Texas at Austin), Arya Farahi (University of Texas at Austin)
ClassificationOptimizationTabular
🎯 What it does: The U-Trustworthiness framework is proposed, which mathematically defines the reliability, capability, and confidence of models in decision-making tasks aimed at maximizing utility functions, and proves that the Bayes classifier is the optimal model under this framework. Furthermore, it is theoretically and experimentally demonstrated that calibration does not necessarily guarantee trustworthiness, and that 'properly-ranked' models are U-trustworthy under various cost-sensitive and fairness utility functions. Finally, the relationship between AUC and U-trustworthiness is established, showing that AUC better reflects utility maximization in model selection and hyperparameter tuning.
UCMCTrack: Multi-Object Tracking with Uniform Camera Motion Compensation
Kefu Yi (Changsha University of Science and Technology), Wei Hao (Changsha University of Science and Technology)
Object TrackingAutonomous DrivingVideo
🎯 What it does: A multi-object tracking framework called UCMCTrack is proposed, based on Uniform Camera Motion Compensation (UCMC), which utilizes Kalman filtering on the ground plane and Mapped Mahalanobis Distance to model and associate motion.
UFDA: Universal Federated Domain Adaptation with Practical Assumptions
Xinhui Liu (Xi'an Jiaotong University), Jizhong Zhao (Xi'an Jiaotong University)
Domain AdaptationFederated LearningContrastive LearningImage
🎯 What it does: A new scenario for Unsupervised Federated Domain Adaptation (UFDA) is proposed, which utilizes only the label sets of each source domain and the outputs of black-box models, with the target domain being completely unknown. Two techniques, HCLD (Hot-Learning with Contrastive Label Disambiguation) and MVD (Mutual-Voting Decision), are introduced to achieve robust pseudo-label generation and shared/unknown class discrimination.
UMIE: Unified Multimodal Information Extraction with Instruction Tuning
Lin Sun, Renze Lou (Pennsylvania State University)
GenerationTransformerSupervised Fine-TuningMultimodality
🎯 What it does: A unified multimodal information extraction model, UMIE, is proposed, which integrates MNER, MRE, and MEE tasks into a generative task and achieves cross-task reasoning through instruction fine-tuning.
Uncertainty Quantification for Data-Driven Change-Point Learning via Cross-Validation
Hui Chen (Jiangsu Normal University), Changliang Zou (Nankai University)
TabularTime Series
🎯 What it does: This paper proposes a change point learning method based on cross-validation and quantifies the uncertainty of overestimation through high-dimensional testing.
Uncertainty Quantification for Forward and Inverse Problems of PDEs via Latent Global Evolution
Tailin Wu (Westlake University), Jure Leskovec (Stanford University)
OptimizationConvolutional Neural NetworkAuto EncoderTime SeriesPhysics Related
🎯 What it does: The LE-PDE-UQ framework is proposed, which uses latent vectors for solving PDEs with time evolution and uncertainty quantification, supporting forward prediction and inverse optimization;
Uncertainty Quantification in Heterogeneous Treatment Effect Estimation with Gaussian-Process-Based Partially Linear Model
Shunsuke Horii (Waseda University), Yoichi Chikahara (NTT Communication Science Laboratories)
Tabular
🎯 What it does: A Gaussian process prior is proposed for the non-parametric components in partially linear models, resulting in a closed-form posterior distribution of CATE, enabling Bayesian quantification of heterogeneous treatment effects.
Uncertainty Regularized Evidential Regression
Kai Ye (University of Pittsburgh), Liang Zhan (University of Pittsburgh)
Depth EstimationOptimizationImageTabular
🎯 What it does: This study addresses and solves the problem of zero gradients in the Evidential Regression Network (ERN) in high uncertainty areas (HUA), which hinders learning, and proposes an uncertainty regularization method.
Uncertainty-Aware GAN for Single Image Super Resolution
Chenxi Ma (Fudan University)
RestorationSuper ResolutionGenerative Adversarial NetworkImage
🎯 What it does: A GAN based on uncertainty (UGAN) is proposed to enhance the perceptual quality and interpretability of single image super-resolution.
Uncertainty-Aware Yield Prediction with Multimodal Molecular Features
Jiayuan Chen (Ohio State University), Xiangliang Zhang (University of Notre Dame)
Drug DiscoveryGraph Neural NetworkTransformerMixture of ExpertsContrastive LearningMultimodalityGraph
🎯 What it does: A multimodal model for uncertainty-aware perception, UAM, is proposed to predict chemical reaction yields using multimodal molecular features.
Uncovering and Mitigating the Hidden Chasm: A Study on the Text-Text Domain Gap in Euphemism Identification
Yuxue Hu (Huazhong Agricultural University), Ying Sha (Huazhong Agricultural University)
RecognitionDomain AdaptationTransformerContrastive LearningText
🎯 What it does: This paper studies the text-text domain gap between the training set and the test set in the task of recognizing embellishing words, and proposes a feature alignment-based FA-Net model to enhance recognition performance.
Underspecification in Language Modeling Tasks: A Causality-Informed Study of Gendered Pronoun Resolution
Emily McMilin (Independent Researcher)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: This study investigates the under-specification problem in language model tasks, revealing how it induces spurious correlations with irrelevant features such as gender and time/location, and proposes two lightweight black-box evaluation methods to detect and quantify these associations.
Understanding and Improving Optimization in Predictive Coding Networks
Nicholas Alonso (University of California), Emre Neftci (Forschungszentrum Jülich)
OptimizationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: The inference learning algorithm of the predictive coding network has been studied and improved, proposing sequential inference and matrix update equilibrium (MQ) optimizer to reduce computational load and improve convergence.
Understanding and Leveraging the Learning Phases of Neural Networks
Johannes Schneider (University of Liechtenstein), Mohit Prabhushankar (Georgia Institute of Technology)
ClassificationOptimizationRepresentation LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: The researchers revealed three learning stages (approximately constant, decreasing, and increasing) during the training process through real-time monitoring of the reconstruction error of deep network layer outputs.
Understanding Distributed Representations of Concepts in Deep Neural Networks without Supervision
Wonjoon Chang (Korea Advanced Institute of Science and Technology), Jaesik Choi (Korea Advanced Institute of Science and Technology)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an unsupervised distributed concept representation method that constructs interpretable Relaxed Decision Regions (RDR) using the activation states of neurons within neural networks, thereby revealing the concepts learned by deep networks.
Understanding the Generalization of Pretrained Diffusion Models on Out-of-Distribution Data
Sai Niranjan Ramachandran (Indian Institute of Science), Vinay Namboodiri (University of Bath)
GenerationData SynthesisDiffusion modelAuto EncoderImageStochastic Differential Equation
🎯 What it does: Investigate the inversion, manipulation, and few-shot generation performance of diffusion models on out-of-distribution (OOD) images, and compare them with GANs.
Understanding the Role of the Projector in Knowledge Distillation
Roy Miles (Imperial College London), Krystian Mikolajczyk (Imperial College London)
ClassificationObject DetectionKnowledge DistillationTransformerImage
🎯 What it does: Revisiting knowledge distillation as a function matching and metric learning problem, a simple and efficient distillation scheme is proposed.
Underwater Organism Color Fine-Tuning via Decomposition and Guidance
Xiaofeng Cong (Southeast University), Junming Hou (Southeast University)
Image TranslationRestorationConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: An algorithm capable of finely adjusting biological colors in underwater images is proposed—CECF, which utilizes the decomposition of color and content encoding and guides the image to achieve colorful enhancement;
UNEX-RL: Reinforcing Long-Term Rewards in Multi-Stage Recommender Systems with UNidirectional EXecution
Gengrui Zhang (Kuaishou Technology), Ben Wang (Beihang University)
Recommendation SystemHyperparameter SearchReinforcement LearningTime SeriesSequential
🎯 What it does: This paper proposes a multi-agent reinforcement learning framework for multi-stage recommendation systems, UNEX-RL, which addresses the issue that traditional single-agent RL cannot optimize multiple stages simultaneously.
Uni-MIS: United Multiple Intent Spoken Language Understanding via Multi-View Intent-Slot Interaction
Shangjian Yin (South China Agricultural University), Yuhong Xu (South China Agricultural University)
RecognitionRecurrent Neural NetworkSupervised Fine-TuningText
🎯 What it does: The Uni-MIS model is proposed, unifying multi-intent speech language understanding into multi-perspective intent-slot interactions, enhancing the joint effect of intent detection and slot filling by combining sentence-level, chunk-level, and word-level intent information.
UniADS: Universal Architecture-Distiller Search for Distillation Gap
Liming Lu (Nanjing University of Science and Technology), Shuchao Pang (Nanjing University of Science and Technology)
OptimizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Designed a framework UniADS that jointly searches for student network architecture and distillation strategies.
UniAP: Towards Universal Animal Perception in Vision via Few-Shot Learning
Meiqi Sun (Zhejiang University), Gaoang Wang (Zhejiang University)
ClassificationSegmentationPose EstimationTransformerSupervised Fine-TuningImage
🎯 What it does: UniAP is proposed, a general model that achieves animal pose estimation, semantic segmentation, and classification using a small number of example prompts;
UniCATS: A Unified Context-Aware Text-to-Speech Framework with Contextual VQ-Diffusion and Vocoding
Chenpeng Du (Shanghai Jiao Tong University), Kai Yu (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerDiffusion modelAudio
🎯 What it does: A unified context-aware TTS framework called UniCATS is proposed, enabling speech continuation and editing.
UniCell: Universal Cell Nucleus Classification via Prompt Learning
Junjia Huang (Sun Yat-sen University), Guanbin Li (Chinese University of Hong Kong)
ClassificationObject DetectionSegmentationTransformerPrompt EngineeringContrastive LearningImageBiomedical Data
🎯 What it does: This paper presents UniCell, an end-to-end universal cell nucleus detection and classification framework capable of uniformly processing multiple heterogeneous datasets.
Unified Framework for Diffusion Generative Models in SO(3): Applications in Computer Vision and Astrophysics
Yesukhei Jagvaral (Carnegie Mellon University), Rachel Mandelbaum (Carnegie Mellon University)
GenerationData SynthesisPose EstimationDiffusion modelImagePhysics RelatedStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper extends diffusion generative models (SGM and DDPM) to the SO(3) rotation group and achieves efficient training and sampling; its effectiveness is validated through synthetic distributions, pose estimation, and galaxy orientation simulations.