AAAI 2024 Papers — Page 21
AAAI Conference on Artificial Intelligence · 2331 papers
SUF: Stabilized Unconstrained Fine-Tuning for Offline-to-Online Reinforcement Learning
Jiaheng Feng (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
Reinforcement LearningTabular
🎯 What it does: Proposes the SUF (Stabilized Unconstrained Fine-Tuning) framework, achieving unconstrained offline-to-online reinforcement learning, eliminating policy collapse while maintaining efficiency.
Summarizing Stream Data for Memory-Constrained Online Continual Learning
Jianyang Gu (Zhejiang University), Yang You (National University of Singapore)
SegmentationKnowledge DistillationData-Centric LearningImage
🎯 What it does: Proposes the SSD method, which stores more informative samples summarized from streaming data in memory to enhance the replay effect of online continual learning.
Sunshine to Rainstorm: Cross-Weather Knowledge Distillation for Robust 3D Object Detection
Xun Huang (Xiamen University), Cheng Wang (Xiamen University)
Object DetectionDomain AdaptationAutonomous DrivingKnowledge DistillationPoint Cloud
🎯 What it does: The DRET model is proposed to achieve realistic rain and fog LiDAR point cloud simulation, and the SRKD framework is designed to implement knowledge distillation for sunny and rainy conditions, enhancing the robustness of 3D detection in rainy weather.
SuperJunction: Learning-Based Junction Detection for Retinal Image Registration
Yu Wang (Institute for Infocomm Research), Jun Cheng (Institute for Infocomm Research)
RecognitionSegmentationConvolutional Neural NetworkGraph Neural NetworkImage
🎯 What it does: A learning-based framework for vascular junction detection and description, called SuperJunction, is proposed for retinal image registration.
Superposed Atomic Representation for Robust High-Dimensional Data Recovery of Multiple Low-Dimensional Structures
Yulong Wang (Huazhong Agricultural University)
RestorationSegmentationImageVideo
🎯 What it does: A unified Sparse Atomic Representation (SAR) framework and its weighted improvement (WSAR) are proposed for recovering multiple low-dimensional structures from high-dimensional data, along with algorithms and convergence proofs.
Supervision Interpolation via LossMix: Generalizing Mixup for Object Detection and Beyond
Thanh Vu (University of North Carolina at Chapel Hill), Jan-Michael Frahm
Object DetectionDomain AdaptationSupervised Fine-TuningImage
🎯 What it does: This paper proposes a loss mixing regularization method called LossMix to enhance the robustness of object detection models and cross-domain detection.
Suppressing Uncertainty in Gaze Estimation
Shijing Wang (Beijing Jiaotong University), Yaping Huang (Beijing Jiaotong University)
RecognitionPose EstimationImage
🎯 What it does: Proposes the SUGE method, which enhances model robustness by estimating and eliminating uncertainty in gaze estimation through three-label consistency.
SURER: Structure-Adaptive Unified Graph Neural Network for Multi-View Clustering
Jing Wang (Beijing Jiaotong University), Jiazheng Yuan (Beijing University of Technology)
Representation LearningGraph Neural NetworkAuto EncoderGraph
🎯 What it does: A structure-adaptive unified graph neural network (SURER) is proposed to simultaneously learn graph structure and aggregate representations in multi-view clustering.
SurgicalSAM: Efficient Class Promptable Surgical Instrument Segmentation
Wenxi Yue (University of Sydney), Zhiyong Wang (University of Sydney)
SegmentationTransformerContrastive LearningImage
🎯 What it does: In the surgical tool segmentation task, an end-to-end method was achieved through efficient fine-tuning of the Segment Anything Model (SAM), enabling segmentation with only category prompts and no explicit point/frame prompts.
Swift-Mapping: Online Neural Implicit Dense Mapping in Urban Scenes
Ke Wu (Harbin Institute of Technology), Wenchao Ding (Fudan University)
Autonomous DrivingOptimizationNeural Radiance FieldSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A framework called Swift-Mapping is proposed for online neural implicit dense mapping in urban scenarios, capable of constructing high-quality 3D maps in real-time.
SwiftPillars: High-Efficiency Pillar Encoder for Lidar-Based 3D Detection
Xin Jin (Chang'an University), Wei Wu (Tsinghua University)
Object DetectionAutonomous DrivingComputational EfficiencyPoint Cloud
🎯 What it does: This paper proposes an efficient 3D object detection framework based on LiDAR called SwiftPillars, which mainly includes the Swift Pillar Encoder (SPE) and Multi-scale Aggregation Decoder (MAD), achieving faster real-time inference.
SwitchTab: Switched Autoencoders Are Effective Tabular Learners
Jing Wu (Amazon), Hakan Brunzell (Amazon)
ClassificationRepresentation LearningAuto EncoderTabular
🎯 What it does: A self-supervised pre-training framework for tabular data called SwitchTab is proposed, which achieves feature decoupling and generation by swapping salient and mutual embeddings.
SyFormer: Structure-Guided Synergism Transformer for Large-Portion Image Inpainting
Jie Wu (Zhejiang University of Technology), Jianwei Zheng (Zhejiang University)
RestorationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A structure-guided collaborative Transformer (SyFormer) is proposed for high-quality inpainting of images with a large proportion of missing data.
Symbolic Cognitive Diagnosis via Hybrid Optimization for Intelligent Education Systems
Junhao Shen (Shanghai Institute of AI for Education), Aimin Zhou (Shanghai Institute of AI for Education)
OptimizationExplainability and InterpretabilityTabular
🎯 What it does: A Symbolic Cognitive Diagnosis (SCD) framework is proposed, utilizing a combination of symbolic trees and gradient optimization to diagnose students' knowledge attributes.
Symbolic Numeric Planning with Patterns
Matteo Cardellini (Politecnico di Torino), Marco Maratea (Università della Calabria)
OptimizationRobotic IntelligenceTabular
🎯 What it does: This paper proposes a new symbolic planning method called Pattern Planning, which encodes linear numerical planning problems by constructing a sequence of 'patterns'. This allows for a more concise SMT formulation for solving plans at a given number of steps compared to existing rolled-up and R2∃ encodings, and theoretically enables finding feasible plans at any number of steps.
Symbolic Regression Enhanced Decision Trees for Classification Tasks
Kei Sen Fong (National University of Singapore), Mehul Motani (National University of Singapore)
ClassificationExplainability and InterpretabilityComputational EfficiencyTabular
🎯 What it does: This paper proposes a method that combines Symbolic Regression (SR) with Decision Trees (DT) — SREDT, which uses SR to generate richer, non-axis-aligned splitting rules, thereby constructing smaller trees with stronger interpretability;
Symmetric Q-learning: Reducing Skewness of Bellman Error in Online Reinforcement Learning
Motoki Omura (University of Tokyo), Tatsuya Harada (University of Tokyo)
Reinforcement Learning
🎯 What it does: By adding learnable zero-mean noise to the target values, the Bellman error distribution becomes symmetric, thereby improving the performance of least-squares-based value function estimation in deep reinforcement learning.
Symmetric Self-Paced Learning for Domain Generalization
Di Zhao (University of Auckland), Philippe Fournier-Viger (Shenzhen University)
Domain AdaptationImage
🎯 What it does: This paper proposes Symmetric Self-Paced Learning (SSPL) for domain generalization, combining a Symmetric Self-Paced training scheduler and Gradient Difficulty Measurement (GDM);
Sync-NeRF: Generalizing Dynamic NeRFs to Unsynchronized Videos
Seoha Kim (Yonsei University), Youngjung Uh (Electronics and Telecommunications Research Institute)
GenerationOptimizationNeural Radiance FieldVideo
🎯 What it does: Training dynamic NeRF on unsynchronized multi-view videos to achieve automatic synchronization and improve reconstruction quality.
Synergistic Anchored Contrastive Pre-training for Few-Shot Relation Extraction
Da Luo (University of Electronic Science and Technology of China), Wannian Gao (University of Electronic Science and Technology of China)
Representation LearningTransformerContrastive LearningText
🎯 What it does: This paper proposes a Synergistic Anchored Contrastive pre-training framework (SaCon), which jointly trains a sentence encoder and a label encoder through symmetric contrastive learning of two views: sentences and labels, achieving a more robust representation for few-shot relation extraction under scarce labeled data.
Synergistic Multiscale Detail Refinement via Intrinsic Supervision for Underwater Image Enhancement
Dehuan Zhang (Dalian Maritime University), Chongyi Li (Nankai University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: The SMDR-IS model is proposed, which enhances underwater image details through multi-scale degradation of input and intrinsic supervision.
T-SciQ: Teaching Multimodal Chain-of-Thought Reasoning via Large Language Model Signals for Science Question Answering
Lei Wang (Beijing Forestry University), Heng Tao Shen (Beijing Rongda Technology)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextMultimodalityChain-of-Thought
🎯 What it does: The T-SciQ method is proposed, which generates high-quality chain-of-thought (CoT) and plan-based CoT (PCoT) as teaching signals through large language models, training smaller multimodal models to complete the ScienceQA task.
T2I-Adapter: Learning Adapters to Dig Out More Controllable Ability for Text-to-Image Diffusion Models
Chong Mou (Peking University), Ying Shan (Tencent)
SegmentationGenerationData SynthesisConvolutional Neural NetworkDiffusion modelImageText
🎯 What it does: A low-cost T2I-Adapter module is proposed to align external control signals (such as sketches, depth, color, key points, and semantic segmentation) with the internal knowledge of a pre-trained text-to-image diffusion model, enabling fine control over structure and color.
T2MAC: Targeted and Trusted Multi-Agent Communication through Selective Engagement and Evidence-Driven Integration
Chuxiong Sun (Institute of Software Chinese Academy of Sciences), Changwen Zheng (Institute of Software Chinese Academy of Sciences)
Robotic IntelligenceReinforcement Learning
🎯 What it does: A novel multi-agent communication framework T2MAC is proposed, which enables agents to extract evidence based on their own observations, select appropriate timing and trustworthy partners for directed message sending, and achieve more reliable decision-making at the receiving end through evidence-level fusion.
TA&AT: Enhancing Task-Oriented Dialog with Turn-Level Auxiliary Tasks and Action-Tree Based Scheduled Sampling
Longxiang Liu (Chinese Academy of Sciences), Yang Feng (Chinese Academy of Sciences)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This study proposes a dialogue turn-based auxiliary task and an action tree-based scheduling sampling method, improving the understanding and generation of end-to-end task-oriented dialogue systems;
TACIT: A Target-Agnostic Feature Disentanglement Framework for Cross-Domain Text Classification
Rui Song (Jilin University), Hao Xu (Jilin University)
ClassificationDomain AdaptationKnowledge DistillationTransformerAuto EncoderText
🎯 What it does: Proposes the TACIT framework for cross-domain text classification with unknown target domains, enhancing cross-domain generalization performance by decoupling features using only source domain data.
Tackling Vision Language Tasks through Learning Inner Monologues
Diji Yang (University of California), Yi Zhang (Mineral)
OptimizationExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: This paper proposes a multimodal optimization framework based on 'Inner Monologue' (IMMO), which utilizes a visual language model (Observer) and a large language model (Reasoner) to generate questions and answers and construct internal dialogues through natural language multi-turn conversations, addressing complex visual language reasoning tasks such as visual question answering and visual entailment.
TagCLIP: A Local-to-Global Framework to Enhance Open-Vocabulary Multi-Label Classification of CLIP without Training
Yuqi Lin (Zhejiang University), Deng Cai (Zhejiang University)
ClassificationSegmentationTransformerContrastive LearningImage
🎯 What it does: Without the need for any training, we propose the TagCLIP local-to-global three-step framework (patch-level classification → Dual-Mask Attention Refinement DMAR → Class-level Re-identification CWR), achieving open-vocabulary multi-label classification with CLIP and using the generated labels as pseudo-labels for weakly supervised semantic segmentation.
TagFog: Textual Anchor Guidance and Fake Outlier Generation for Visual Out-of-Distribution Detection
Jiankang Chen (Sun Yat-sen University), Ruixuan Wang (Sun Yat-sen University)
Anomaly DetectionRepresentation LearningLarge Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a new learning framework called TagFog, which generates pseudo-discrete samples using Jigsaw techniques and obtains semantic anchors through ID class descriptions generated by ChatGPT via the CLIP text encoder, guiding the visual encoder to learn more compact and semantically rich features, thereby improving visual OOD detection performance.
Tail-STEAK: Improve Friend Recommendation for Tail Users via Self-Training Enhanced Knowledge Distillation
Yijun Ma (Renmin University of China), Xiao Zhou (Renmin University of China)
Recommendation SystemKnowledge DistillationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: In response to the poor performance of graph neural networks for tail users (low degree) in friend recommendation tasks, the Tail-STEAK framework is proposed to achieve better representation learning for tail users.
Talk Funny! A Large-Scale Humor Response Dataset with Chain-of-Humor Interpretation
Yuyan Chen (Fudan University), Yanghua Xiao (Fudan University)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A large-scale explainable humor response dataset in Chinese, TalkFunny, is proposed and constructed, which includes chain humor explanations, humor mind maps, as well as two auxiliary tasks: humor sentiment style classification and humor rewriting.
Taming Binarized Neural Networks and Mixed-Integer Programs
Johannes Aspman (Czech Technical University), Jakub Marecek (Czech Technical University)
Optimization
🎯 What it does: This paper proposes to reformulate the training problem of Binary Neural Networks (BNN) as a Mixed Integer Programming (MIP) problem, and further constructs its sub-additive dual, proving that this dual problem is definable (tame) under o-minimal structures, thus enabling true backpropagation using nonsmooth implicit differentiation and the chain rule of conservative fields.
Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification
Andreas Grivas (University of Edinburgh), Adam Lopez (University of Edinburgh)
ClassificationBiomedical DataElectronic Health Records
🎯 What it does: This study investigates the bottleneck of low-rank sigmoid output layers in multi-label classification and proposes an output layer based on Discrete Fourier Transform (DFT) to ensure that all k-sparse label combinations can be correctly predicted.
TAPE: Leveraging Agent Topology for Cooperative Multi-Agent Policy Gradient
Xingzhou Lou (University of Chinese Academy of Sciences), Yali Du (King's College London)
Reinforcement Learning
🎯 What it does: This paper proposes a multi-agent policy gradient framework based on Agent Topology (TAPE) to enhance multi-agent collaboration while avoiding the centralized-decentralized mismatch (CDM) problem.
Targeted Activation Penalties Help CNNs Ignore Spurious Signals
Dekai Zhang (Imperial College London), Francesca Toni (Imperial College London)
ClassificationOptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A Targeted Activation Penalty (TAP) method is proposed for convolutional neural networks to suppress the model's reliance on spurious signals and enhance generalization performance.
Task Contamination: Language Models May Not Be Few-Shot Anymore
Changmao Li (University of California), Jeffrey Flanigan (University of California)
ClassificationTransformerLarge Language ModelTextTime Series
🎯 What it does: A systematic evaluation of the task contamination issue in large language models for zero-shot and few-shot tasks.
Task Planning for Object Rearrangement in Multi-Room Environments
Karan Mirakhor (TCS Research), Brojeshwar Bhowmick (TCS Research)
Robotic IntelligenceGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningGraph
🎯 What it does: This paper proposes an end-to-end task planning method for simultaneously discovering hidden objects and rearranging them in multi-room, partially observable environments, aiming to minimize the total travel distance and number of steps for the agent.
Task-Adaptive Prompted Transformer for Cross-Domain Few-Shot Learning
Jiamin Wu (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
Domain AdaptationMeta LearningTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes a task-adaptive prompt transformer (MetaPrompt) model for cross-domain few-shot learning (CD-FSL), utilizing attention to generate task prompts for rapid task adaptation.
Task-Agnostic Privacy-Preserving Representation Learning for Federated Learning against Attribute Inference Attacks
Caridad Arroyo Arevalo (Illinois Institute of Technology), Binghui Wang (Illinois Institute of Technology)
Federated LearningSafty and PrivacyRepresentation LearningImageTabularFinance Related
🎯 What it does: A task-agnostic federated learning privacy protection framework TAPPFL is proposed, which learns with information-theoretic objectives to reduce sensitive attribute information while maintaining model performance.
Task-Disruptive Background Suppression for Few-Shot Segmentation
Suho Park (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A Task-Disruptive Background Suppression (TBS) module is proposed, which utilizes specific adverse features in the support set background to suppress them, thereby enhancing few-shot semantic segmentation performance.
Task-Driven Causal Feature Distillation: Towards Trustworthy Risk Prediction
Zhixuan Chu (Ant Group), Sheng Li (University of Virginia)
Explainability and InterpretabilityKnowledge DistillationMixture of ExpertsTabular
🎯 What it does: This paper proposes a task-driven causal feature distillation (TDCFD) model that transforms original features into causal feature attribution and uses this for risk prediction.
Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster
Fei Ye (University of York), Adrian G. Bors (University of York)
GenerationRepresentation LearningAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A task-agnostic continual learning framework for generation and representation learning (Continual Variational Autoencoder, CAA) is proposed, which achieves task boundary-free generation and reconstruction of continuously changing data streams through a dual memory system (temporary memory and evolving memory) and a two-step optimization strategy.
Task-Free Dynamic Sparse Vision Transformer for Continual Learning
Fei Ye (University of York), Adrian G. Bors (University of York)
ClassificationRecognitionTransformerAuto EncoderImage
🎯 What it does: This paper proposes a Task-Free Dynamic Sparse Vision Transformer (TFDSViT) for training visual Transformers in continuous learning scenarios without task boundaries.
TaskLAMA: Probing the Complex Task Understanding of Language Models
Quan Yuan (Google Research), Deepak Ramachandran (Google Research)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper studies how to utilize large language models (LLMs) for structured complex task decomposition (SCTD), which involves breaking down a complex task into several steps and predicting the temporal dependencies between these steps.
Taxonomy Driven Fast Adversarial Training
Kun Tong (Southeast University), Yuan Cao (Ocean University of China)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Fast Adversarial Training based on Adversarial Sample Classification (TDAT), which systematically alleviates catastrophic overfitting in single-step adversarial training and significantly enhances model robustness through joint improvements in initialization, dynamic label relaxation, and classification-driven loss functions.
TC-LIF: A Two-Compartment Spiking Neuron Model for Long-Term Sequential Modelling
Shimin Zhang (Hong Kong Polytechnic University), Kay Chen Tan (Hong Kong Polytechnic University)
Spiking Neural NetworkTime SeriesSequentialAudio
🎯 What it does: A dual-chamber leakage integral firing (TC-LIF) synaptic neuron model is proposed to address the long-term temporal dependency learning problem.
TCI-Former: Thermal Conduction-Inspired Transformer for Infrared Small Target Detection
Tianxiang Chen (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
Object DetectionSegmentationTransformerImageOrdinary Differential Equation
🎯 What it does: A thermal conduction-inspired Transformer (TCI-Former) based on thermal conduction theory is proposed for infrared small target detection.
TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions
Hui Lu (Utrecht University), Ronald Poppe (Utrecht University)
RecognitionRecurrent Neural NetworkTransformerOptical FlowVideo
🎯 What it does: Designed the TCNet hybrid network, achieving continuous sign language recognition through trajectory modules and related modules.
TD²-Net: Toward Denoising and Debiasing for Video Scene Graph Generation
Xin Lin (Guangzhou University), Dacheng Tao (University of Sydney)
Object DetectionGenerationTransformerVideo
🎯 What it does: Designed and implemented TD-Net, a framework that enhances video scene graph generation through a denoising spatiotemporal transformer and an asymmetric re-weighted loss.
TDeLTA: A Light-Weight and Robust Table Detection Method Based on Learning Text Arrangement
Yang Fan (Harbin Institute of Technology), Qitian Wu (China Mobile Information Technology)
Object DetectionRecurrent Neural NetworkTransformerTabular
🎯 What it does: A lightweight table detection method TDeLTA based on text block arrangement learning is proposed, which locates tables using the positions of text blocks rather than image features.
Teacher as a Lenient Expert: Teacher-Agnostic Data-Free Knowledge Distillation
Hyunjune Shin (Inha University), Dong-Wan Choi (Inha University)
Knowledge DistillationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a teacher-agnostic data unsupervised knowledge distillation method (TA-DFKD), which removes class-prior and utilizes teacher-driven sample selection to generate high-quality diverse samples, achieving robust distillation from different teacher models.
Teaching Large Language Models to Translate with Comparison
Jiali Zeng (Tencent Inc), Jie Zhou (Tencent Inc)
TransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: By introducing output comparison and preference comparison within the framework of contrastive learning, we fine-tune open-source large language models to enhance machine translation performance.
TEILP: Time Prediction over Knowledge Graphs via Logical Reasoning
Siheng Xiong (Georgia Institute of Technology), Faramarz Fekri (Georgia Institute of Technology)
Graph Neural NetworkReinforcement LearningGraph
🎯 What it does: The TEILP framework is proposed, which converts temporal knowledge graphs into temporary event knowledge graphs and learns logical rules of conditional probability distributions for event time prediction.
Temporal Adaptive RGBT Tracking with Modality Prompt
Hongyu Wang (Xidian University), Jing Liu (Xidian University)
Object TrackingTransformerVideoMultimodality
🎯 What it does: This paper proposes a time-adaptive RGB-T tracking framework named TATrack, which can dynamically update the online template in each frame and integrate spatio-temporal and multi-modal information for target localization.
Temporal Correlation Vision Transformer for Video Person Re-Identification
Pengfei Wu (Xi'an Jiaotong University), Changyin Sun (Anhui University)
RecognitionRetrievalTransformerVideo
🎯 What it does: Proposes the Temporal Correlation Vision Transformer (TCViT), which enhances target pedestrian features through relative states and adapts frame weights using learnable temporal aggregation.
Temporal Graph Contrastive Learning for Sequential Recommendation
Shengzhe Zhang (University of Science and Technology of China), Hui Xiong (Hong Kong University of Science and Technology)
Recommendation SystemGraph Neural NetworkTransformerContrastive LearningGraphSequential
🎯 What it does: Proposes the Temporal Graph Contrastive Learning (TGCL) method, which combines global temporal item transition graphs (TITG) and local sequence encoding to achieve sequence recommendation.
Temporal-Distributed Backdoor Attack against Video Based Action Recognition
Xi Li (Pennsylvania State University), George Kesidis (Pennsylvania State University)
RecognitionAdversarial AttackVideo
🎯 What it does: This paper proposes a transform domain-based time-distributed backdoor attack that can implant imperceptible triggers in video action recognition models, inducing the model to misclassify as a category specified by the attacker.
Temporally and Distributionally Robust Optimization for Cold-Start Recommendation
Xinyu Lin (National University of Singapore), Tat-Seng Chua (National University of Singapore)
Recommendation SystemOptimizationTabular
🎯 What it does: To address the issue of temperature transfer feature drift in cold start recommendations, a Temporal Distribution Robust Optimization framework (TDRO) is proposed.
Tensorized Label Learning on Anchor Graph
Jing Li (Xidian University), Wei Xia (Xidian University)
OptimizationGraph
🎯 What it does: A tensor label learning method based on anchor graphs (TLL-AG) is proposed, which directly obtains soft labels from anchor graphs through orthogonal non-negative matrix factorization without the need for post-processing;
Ternary Spike: Learning Ternary Spikes for Spiking Neural Networks
Yufei Guo (Intelligent Science and Technology Academy of CASIC), Zhe Ma (Intelligent Science and Technology Academy of CASIC)
ClassificationSpiking Neural NetworkImage
🎯 What it does: This paper proposes a three-valued spiking neuron (-1, 0, 1) and learnable three-valued spikes to enhance the representational capacity of spiking neural networks while retaining the advantages of event-driven and additive operations.
Terrain Diffusion Network: Climatic-Aware Terrain Generation with Geological Sketch Guidance
Zexin Hu (University of Sydney), Zhiyong Wang (University of Sydney)
GenerationDiffusion modelAuto EncoderImage
🎯 What it does: The paper proposes a multi-level terrain generation network (TDN) based on diffusion models, capable of generating climate-aware high-quality terrain maps guided by user-provided sketches of rivers, ridges, basins, and peaks.
Test-Time Adaptation via Style and Structure Guidance for Histological Image Registration
Shenglong Zhou (University of Science and Technology of China), Feng Wu (Institute of Artificial Intelligence, Hefei Comprehensive National Science Center)
Domain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A test-time adaptation method for histological image registration is proposed, utilizing style guidance, structural guidance, and model continuity to enhance the generalization ability of learning methods.
Test-Time Domain Adaptation by Learning Domain-Aware Batch Normalization
Yanan Wu (Beijing Jiaotong University), Songhe Feng (Beijing Jiaotong University)
Domain AdaptationMeta LearningConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A testing time domain adaptation (TT-DA) framework is proposed, which utilizes the affine parameters of the batch normalization (BN) layers for domain knowledge learning, updating only the γ and β of BN; at the same time, a self-supervised branch is introduced to provide domain-related supervision for unlabeled data, and a meta-learning dual-layer optimization is employed to enable the affine parameters to quickly adapt to new domains; ultimately improving cross-domain performance without increasing inference costs.
Test-Time Personalization with Meta Prompt for Gaze Estimation
Huan Liu (Huawei), Yuanhao Yu (University of Toronto)
Pose EstimationDomain AdaptationMeta LearningConvolutional Neural NetworkPrompt EngineeringImage
🎯 What it does: This paper proposes a label-free personalized gaze estimation method called TPGaze, which achieves rapid adaptation by inserting trainable prompts (only 1% of the parameters of ResNet-18) into the model, and utilizes meta-learning pre-trained prompts to align the update direction with the reduction of gaze error.
Testing Self-Reducible Samplers
Rishiraj Bhattacharyya (University of Birmingham), Sayantan Sen (National University of Singapore)
Graph
🎯 What it does: This paper designs the first testing framework for self-reducible samplers, CubeProbeTester, and implements distance estimation and error detection for linear extension samplers through this framework.
TETRIS: Towards Exploring the Robustness of Interactive Segmentation
Andrey Moskalenko (Samsung Research), Konstantin Soshin (Samsung Research)
SegmentationAdversarial AttackImageBenchmark
🎯 What it does: This paper conducts a systematic study on the impact of user click positions on performance in interactive segmentation, and proposes white-box adversarial attacks to generate extreme clicks, robustness evaluation metrics, and a new benchmark called TETRIS.
TexFit: Text-Driven Fashion Image Editing with Diffusion Models
Tongxin Wang (Wuhan University), Mang Ye (Wuhan University)
Image TranslationGenerationDiffusion modelImageTextMultimodality
🎯 What it does: A Diffusion model framework called TexFit is proposed, which relies solely on text for local fashion image editing, using text to directly locate the editing area and perform fine editing in that area.
Text Diffusion with Reinforced Conditioning
Yuxuan Liu (Peking University), Qi Zhang (Microsoft Corporation)
GenerationData SynthesisTransformerReinforcement LearningDiffusion modelText
🎯 What it does: Proposes the TREC (Text Diffusion with Reinforced Conditioning) model, which combines reinforcement learning self-conditioning and time-aware variance scaling to improve the generation quality of text diffusion models.
Text Image Inpainting via Global Structure-Guided Diffusion Models
Shipeng Zhu (Southeast University), Hui Xue (Southeast University)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes the task of text image inpainting, constructs two real and synthetic datasets for scene text and handwritten text, and introduces a global structure-guided diffusion model (GSDM) to achieve complete restoration of text images.
Text-Based Occluded Person Re-identification via Multi-Granularity Contrastive Consistency Learning
Xinyi Wu (National University of Defense Technology), Zhiping Cai (HeFei University of Technology)
RecognitionRetrievalTransformerContrastive LearningImageText
🎯 What it does: A multi-granularity contrastive consistency learning framework MGCC is proposed for occluded text retrieval of portraits, and an occluded version of the T-ReID dataset is constructed using the occlusion generator OGor.
Text-Guided Molecule Generation with Diffusion Language Model
Haisong Gong (Chinese Academy of Sciences), Liang Wang (Chinese Academy of Sciences)
GenerationDrug DiscoveryTransformerDiffusion modelTextBiomedical Data
🎯 What it does: A text-guided molecular generation method based on diffusion language models (TGM-DLM) is proposed, which generates SMILES molecules that meet text descriptions from random noise through a two-stage diffusion process.
Text-to-Image Generation for Abstract Concepts
Jiayi Liao (University of Science and Technology of China), Dongmei Zhang (Microsoft)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageText
🎯 What it does: A framework called TIAC based on a three-layer theory of artistic creation is proposed, specifically addressing the problem of generating abstract concepts from text to image (T2I);
Text2Analysis: A Benchmark of Table Question Answering with Advanced Data Analysis and Unclear Queries
Xinyi He (Xi'an Jiaotong University), Dongmei Zhang (Microsoft)
Large Language ModelTabularBenchmark
🎯 What it does: Developed the Text2Analysis benchmark, collecting 2249 table-query-code-result quadruples, covering advanced analytical tasks (such as prediction, chart generation, basic insights, etc.) as well as ambiguous user queries.
Text2City: One-Stage Text-Driven Urban Layout Regeneration
Yiming Qin (Shanghai Jiao Tong University), Rynson W.H. Lau (City University of Hong Kong)
GenerationData SynthesisDiffusion modelText
🎯 What it does: A method for urban layout regeneration based on text descriptions and surrounding context, called Text2City, is proposed, which can jointly generate the road and building layouts of the target area in a single stage.
TextGT: A Double-View Graph Transformer on Text for Aspect-Based Sentiment Analysis
Shuo Yin (Ocean University of China), Guoqiang Zhong (Ocean University of China)
ClassificationGraph Neural NetworkTransformerText
🎯 What it does: This paper addresses the Aspect-based Sentiment Analysis (ABSA) task and proposes a dual-view graph Transformer (TextGT), which alternately stacks graph convolutional layers and Transformer layers, aiming to capture both syntactic structure and global semantics while addressing the over-smoothing problem commonly encountered in traditional GNNs and Transformers.
TF-CLIP: Learning Text-Free CLIP for Video-Based Person Re-identification
Chenyang Yu (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
RecognitionRetrievalTransformerContrastive LearningVideo
🎯 What it does: This paper proposes TF-CLIP, a one-time trained, text-free CLIP-based video person re-identification framework that utilizes identity-specific sequence features instead of text features, and enhances temporal information modeling through Sequence-Specific Prompt (SSP) and Temporal Memory Diffusion (TMD) modules.
The Causal Impact of Credit Lines on Spending Distributions
Yijun Li (City University of Hong Kong), Zhixiang Huang (JD Digits)
Recommendation SystemOptimizationExplainability and InterpretabilityComputational EfficiencyData-Centric LearningTabularFinance Related
🎯 What it does: This study investigates the causal impact of credit limits on the distribution of consumer spending and proposes a distributed causal estimation framework that views individual spending as a distribution.
The Choice of Noninformative Priors for Thompson Sampling in Multiparameter Bandit Models
Jongyeong Lee (University of Tokyo), Masashi Sugiyama (University of Tokyo)
Reinforcement Learning
🎯 What it does: This paper studies the impact of non-informative priors on the performance of Thompson Sampling (TS) in a multi-parameter bandwidth model and proposes a truncation strategy TS-T to achieve asymptotic optimality under reference priors and Jeffreys priors.
The Complexity of Computing Robust Mediated Equilibria in Ordinal Games
Vincent Conitzer (Carnegie Mellon University)
Optimization
🎯 What it does: This paper studies the construction of robust mediated equilibrium in ordinal games (where only the preference order of outcomes is given) and explores its existence, properties, and computational complexity.
The Complexity of Fair Division of Indivisible Items with Externalities
Argyrios Deligkas (Royal Holloway University of London), Šimon Schierreich (Czech Technical University in Prague)
🎯 What it does: This paper explores the computational complexity of fairly allocating indivisible goods under externality conditions.
The Complexity of Optimizing Atomic Congestion
Cornelius Brand (Regensburg University), Fionn Mc Inerney (Indian Institute of Technology Hyderabad)
Optimization
🎯 What it does: This paper systematically studies the problem of solving the system optimal (minimum average cost) path allocation in atomic congestion games. The authors first provide NP-hard reductions for various instances, proving that the problem remains non-polynomially solvable even on extremely simple structures (such as trees, star networks, or DAGs) or networks with constant tree width and depth; subsequently, within the framework of parameterized complexity, they identify for the first time the edge-separation parameter that makes the problem solvable—a combination of slim treecut width and maximum link capacity cmax—and present a dynamic programming FPT algorithm based on tree decomposition; finally, the research is extended to the min-max version (MSOAC) that allows some agents not to be routed, providing an FPT solution under the combination of tree width + maximum degree + cmax + α parameters.
The Expected Loss of Preconditioned Langevin Dynamics Reveals the Hessian Rank
Amitay Bar (Technion Israel Institute of Technology), Ronen Talmon (Technion Israel Institute of Technology)
OptimizationConvolutional Neural NetworkImageStochastic Differential Equation
🎯 What it does: Analyze the expected loss of preconditioned Langevin dynamics near the stationary point and derive its relationship with the order of the Hessian.
The Irrelevance of Influencers: Information Diffusion with Re-Activation and Immunity Lasts Exponentially Long on Social Network Models
Tobias Friedrich (Hasso Plattner Institute), Marcus Pappik (Hasso Plattner Institute)
Graph
🎯 What it does: A rigorous theoretical analysis of the survival time of the SIRS propagation model is conducted, providing thresholds and upper/lower bounds for different network structures (star, diffusion subgraph, random graph, hypergeometric random graph).
The Logic of Doxastic Strategies
Junli Jiang (Southwest University), Pavel Naumov (University of Southampton)
🎯 What it does: This paper introduces the concept of 'doxastic strategies' and provides a formal logical system for this concept.
The Moderating Effect of Instant Runoff Voting
Kiran Tomlinson (Cornell University), Jon Kleinberg (Cornell University)
🎯 What it does: This paper explores the moderating effect of Instant Runoff Voting (IRV) on the position of the median candidate under a one-dimensional Euclidean voter preference model, and proves that it has an exclusion zone that rejects extreme candidates.
Theoretical and Empirical Analysis of Cost-Function Merging for Implicit Hitting Set WCSP Solving
Javier Larrosa (Universitat Politècnica de Catalunya), Emma Rollon (Universitat Politècnica de Catalunya)
Optimization
🎯 What it does: This paper studies the application of the Implicit Hitting Set (HS) method in Weighted Constraint Satisfaction Problems (WCSP), theoretically analyzes the exponential iteration problem caused by core interchangeability, and proposes two reconstruction strategies, symbolic merging and numerical merging, to compress the core space.
Theoretical Aspects of Generating Instances with Unique Solutions: Pre-assignment Models for Unique Vertex Cover
Takashi Horiyama (Hokkaido University), Ryu Suzuki (Hokkaido University)
Optimization
🎯 What it does: This paper proposes a pre-allocation model for generating instances of combinatorial optimization problems with a unique solution, with a particular focus on the vertex cover problem.
Thompson Sampling for Real-Valued Combinatorial Pure Exploration of Multi-Armed Bandit
Shintaro Nakamura (University of Tokyo), Masashi Sugiyama (RIKEN AIP)
OptimizationReinforcement Learning
🎯 What it does: This paper studies the real-valued combinatorial pure exploration multi-armed bandit problem and proposes the GenTS-Explore algorithm, which can effectively identify the optimal action even when the action set is exponentially large.
Three Heads Are Better than One: Complementary Experts for Long-Tailed Semi-supervised Learning
Chengcheng Ma (Institute of Automation, Chinese Academy of Sciences), Changsheng Xu (Institute of Automation, Chinese Academy of Sciences)
ClassificationMixture of ExpertsImage
🎯 What it does: A multi-expert model CPE is proposed, which jointly trains three experts with different logit adjustment strengths, combined with class-level batch normalization, to generate high-quality pseudo-labels and improve model performance in long-tail semi-supervised learning.
Three Heads Are Better than One: Improving Cross-Domain NER with Progressive Decomposed Network
Xuming Hu (Hong Kong University of Science and Technology), Philip S. Yu (University of Illinois Chicago)
ClassificationDomain AdaptationTransformerSupervised Fine-TuningText
🎯 What it does: This paper designs a multi-source cross-domain named entity recognition framework, which splits NER into two steps: entity mention detection and entity type classification, and achieves multi-source knowledge transfer through explicit copying of potential entities and an implicit knowledge progression network.
Threshold-Based Responsive Simulated Annealing for Directed Feedback Vertex Set Problem
Qingyun Zhang (Huazhong University of Science and Technology), Zhipeng Lü
OptimizationReinforcement Learning
🎯 What it does: A threshold-based reactive simulated annealing algorithm (TRSA) is proposed to solve the Directed Feedback Vertex Set Problem (DFVSP).
TIKP: Text-to-Image Knowledge Preservation for Continual Semantic Segmentation
Zhidong Yu (University of Science and Technology of China), Zhenbo Shi (University of Science and Technology of China)
SegmentationKnowledge DistillationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a framework named TIKP, which utilizes text-to-image generation technology to automatically generate and maintain prompts during the process of Continual Semantic Segmentation (CSS). The generated images are used to retain old knowledge, thereby mitigating catastrophic forgetting.
Time-Aware Knowledge Representations of Dynamic Objects with Multidimensional Persistence
Baris Coskunuzer (University of Texas at Dallas), Yulia R. Gel (University of Texas at Dallas)
Computational EfficiencyRepresentation LearningGraph Neural NetworkTime SeriesSequentialBiomedical DataElectrocardiogram
🎯 What it does: This paper proposes the Temporal Multi-Persistence (TMP) method, which treats the time dimension as a one-dimensional multi-parameter persistence and combines it with zigzag persistence to generate time-aware multi-dimensional topological fingerprints, and constructs TMP-Nets based on TMP for temporal prediction.
TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning
Jiexi Liu (Nanjing University of Aeronautics and Astronautics), Songcan Chen (Nanjing University of Aeronautics and Astronautics)
Anomaly DetectionRepresentation LearningAuto EncoderContrastive LearningTime Series
🎯 What it does: This paper proposes a self-supervised contrastive learning framework called TimesURL, aimed at learning general time series representations that can be applied to various downstream tasks.
TiMix: Text-Aware Image Mixing for Effective Vision-Language Pre-training
Chaoya Jiang (National Engineering Research Center for Software Engineering), Shikun Zhang (Alibaba Group)
Data SynthesisRetrievalTransformerContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a text-aware image mixing method called TiMix, which applies data augmentation techniques such as CutMix to self-supervised multimodal contrastive learning (SMCL), thereby enhancing the data efficiency of visual-language pre-training (VLP).
TMFormer: Token Merging Transformer for Brain Tumor Segmentation with Missing Modalities
Zheyu Zhang (University of Science and Technology of China), Xiaoyan Sun (University of Science and Technology of China)
SegmentationTransformerImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposes the Token Merging Transformer (TMFormer) to address the missing modality problem in brain tumor segmentation, utilizing an attention network based on variable-length token sequences;
TMPNN: High-Order Polynomial Regression Based on Taylor Map Factorization
Andrei Ivanov (Independent Researcher), Stefan Ailuro (Independent Researcher)
TabularPhysics RelatedOrdinary Differential Equation
🎯 What it does: A high-order multi-objective polynomial regression model TMPNN based on Taylor mapping decomposition is proposed, which achieves high-order polynomial fitting by iteratively sharing low-order polynomial coefficients, avoiding the curse of dimensionality.
TNPAR: Topological Neural Poisson Auto-Regressive Model for Learning Granger Causal Structure from Event Sequences
Yuequn Liu (Guangdong University of Technology), Zhifeng Hao (Shantou University)
Generative Adversarial NetworkTime SeriesSequential
🎯 What it does: The Topological Neural Poisson Auto-Regressive (TNPAR) model is proposed for learning Granger causal structures from topological event sequences.
Token-Level Contrastive Learning with Modality-Aware Prompting for Multimodal Intent Recognition
Qianrui Zhou (Tsinghua University), Kai Gao (Hebei University of Science and Technology)
RecognitionTransformerPrompt EngineeringContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: This paper proposes a framework based on token-level contrastive learning and modality-aware prompts (TCL-MAP) to enhance multimodal intent recognition performance.
TOP-ReID: Multi-Spectral Object Re-identification with Token Permutation
Yuhao Wang (Dalian University of Technology), Huchuan Lu (Jiangsu University)
RecognitionRetrievalTransformerImage
🎯 What it does: This paper proposes a multi-spectral object re-identification framework called TOP-ReID, which extracts RGB, NIR, and TIR features using a visual Transformer. Based on this, a cyclic token permutation module and a complementary reconstruction module are designed to achieve spatial alignment and distribution alignment of multi-spectral features.