AAAI 2025 Papers — Page 26
AAAI Conference on Artificial Intelligence · 3028 papers
SIGraph: Saliency Image-Graph Network for Retinal Disease Classification in Fundus Image
Peng Zhang (Zhejiang University), Hongguang Cui
ClassificationConvolutional Neural NetworkGraph Neural NetworkTransformerImageBiomedical Data
🎯 What it does: A Saliency Image-Graph network (SIGraph) that combines saliency features with lesion spatial distribution is proposed. It constructs a complete graph through pixel-level lesion nodes and uses mixed graph pooling to obtain global distribution information, aiming to improve the classification accuracy of retinal lesions.
SigStyle: Signature Style Transfer via Personalized Text-to-Image Models
Ye Wang (Jilin University), Rui Ma (Jilin University)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: The SigStyle framework is proposed, achieving signature style transfer using only a single style image, supporting global, local, texture, style fusion, and text-guided image generation.
Sim4Rec: Data-Free Model Extraction Attack on Sequential Recommendation
Yihao Wang (Zhejiang University), Jun Wang (OPPO Research Institute)
Recommendation SystemKnowledge DistillationAdversarial AttackRecurrent Neural NetworkTransformerReinforcement LearningGenerative Adversarial NetworkSequential
🎯 What it does: A data-independent model extraction attack framework called Sim4Rec is proposed to steal knowledge from black-box sequential recommendation models and construct approximate models.
Similar Modality Enhancement and Action Consistency Learning for Weakly Supervised Temporal Action Localization
Maodong Li (Wuhan University), Bing Li (Hubei Luojia Laboratory)
RecognitionObject DetectionOptical FlowVideo
🎯 What it does: Proposes the SEAL method, which combines the SME module and the ACL module to achieve temporal action localization using weakly supervised video-level labels.
Simplifying Control Mechanism in Text-to-Image Diffusion Models
Zhida Feng (Wuhan University of Science and Technology), Shikun Feng (Baidu Inc.)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImageText
🎯 What it does: Proposes SimpleControlNet, a simplified control mechanism for text-to-image diffusion models;
SimProF: A Simple Probabilistic Framework for Unsupervised Domain Adaptation
Meng-zhu Wang
Domain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a simple probabilistic framework called SimProF for unsupervised domain adaptation.
SimRP: Syntactic and Semantic Similarity Retrieval Prompting Enhances Aspect Sentiment Quad Prediction
Zhongquan Jian (Xiamen University), Qingqiang Wu (Xiamen University)
GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: By constructing prompts that include syntactic and semantic similar examples, the generation performance of Aspect Sentiment Quad Prediction (ASQP) is improved.
Simulate and Eliminate: Revoke Backdoors for Generative Large Language Models
Haoran Li (Hong Kong University of Science and Technology), Yangqiu Song (Hong Kong University of Science and Technology)
GenerationAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: A backdoor removal method named SANDE is proposed for generative large language models, which can directly repair LLMs that have been implanted with backdoors without relying on clean models.
Simulation-Free Hierarchical Latent Policy Planning for Proactive Dialogues
Tao He (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes a non-simulated, offline hierarchical reinforcement learning framework called LDPP, which can automatically mine fine-grained latent strategy vectors from real dialogue records and perform strategy planning, thereby enhancing the effectiveness of proactive dialogue systems.
Single Exposure Quantitative Phase Imaging with a Conventional Microscope Using Diffusion Models
Gabriel della Maggiora (Helmholtz Zentrum Dresden Rossendorf), Artur Yakimovich (Helmholtz Zentrum Dresden Rossendorf)
RestorationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes single-exposure color phase imaging induced by chromatic aberration and uses a zero-mean diffusion model to achieve quantitative phase recovery.
Single Image Rolling Shutter Removal with Diffusion Models
Zhanglei Yang (University of Electronic Science and Technology of China), Shuaicheng Liu (Megvii Technology)
RestorationDiffusion modelOptical FlowImage
🎯 What it does: This paper proposes RS-Diffusion, which utilizes a diffusion model to achieve single-frame rolling shutter correction and constructs a real annotated RS-Real dataset.
Single-Loop Federated Actor-Critic across Heterogeneous Environments
Ye Zhu (Auburn University), Xiaowen Gong (Auburn University)
Federated LearningReinforcement Learning
🎯 What it does: A single-loop federated actor-critic (SFAC) framework is proposed in heterogeneous environments, combining value evaluation and policy improvement of multiple agents to learn a globally shared policy.
Single-View Graph Contrastive Learning with Soft Neighborhood Awareness
Qingqiang Sun (Great Bay University), Kai Wang (Central South University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes a single-view graph contrastive learning framework called SIGNA, which learns node representations using 'soft neighborhood awareness' without relying on cross-view augmentation.
Single-view Image to Novel-view Generation for Hand-Object Interactions
Zhongqun Zhang (University of Birmingham), Jifei Song (Huawei)
GenerationData SynthesisPose EstimationDiffusion modelGaussian SplattingImage
🎯 What it does: Proposes the HO123 method, which generates arbitrary viewpoint images of hand-object interactions from a single RGB image and can reconstruct 3D object meshes.
Singular Value Scaling: Efficient Generative Model Compression via Pruned Weights Refinement
Hyeonjin Kim (Ulsan National Institute of Science and Technology), Jaejun Yoo (Ulsan National Institute of Science and Technology)
GenerationCompressionDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the Singular Value Scaling (SVS) technique, which improves the fine-tuning process by scaling the singular values of pruned weights, thereby achieving general compression for GANs and diffusion models while preserving the model's pre-trained knowledge.
Situation Calculus Temporally Lifted Abstractions for Generalized Planning
Giuseppe de Giacomo, Matteo Mancanelli (York University)
🎯 What it does: A general planning framework based on temporally lifted abstraction is proposed, which combines specific situation calculus with high-level non-deterministic action theory through LTL constraints. A complete proof is provided on how to generate strategies at the abstract level using LTL synthesis and then refine them to the concrete level; the feasibility of this method is demonstrated using the case of finding the minimum value in a linked list.
Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion
Haipeng Chen (Jilin University), Yingda Lyu (Jilin University)
ClassificationRecognitionKnowledge DistillationGraph Neural NetworkVideo
🎯 What it does: This paper proposes a skeleton dependency refinement method based on Gaussian correlation functions and a classification framework utilizing the Hilbert-Schmidt independence criterion to address the challenges of nonlinear long-range joint dependencies and high-dimensional feature probability estimation in skeleton action recognition.
SKI Models: Skeleton Induced Vision-Language Embeddings for Understanding Activities of Daily Living
Arkaprava Sinha (University of North Carolina at Charlotte), Srijan Das (University of North Carolina at Charlotte)
RecognitionKnowledge DistillationRepresentation LearningVision Language ModelVideoText
🎯 What it does: The SKI model is proposed, which achieves zero-shot recognition and video description of daily life actions by injecting 3D skeletal information into the visual-language embedding space.
Skill Disentanglement in Reproducing Kernel Hilbert Space
Vedant Dave (Montanuniversitat Leoben), Elmar Rueckert (Montanuniversitat Leoben)
Robotic IntelligenceReinforcement LearningContrastive LearningSequentialBenchmark
🎯 What it does: A framework for unsupervised skill learning that combines Maximum Mean Discrepancy (MMD) with entropy rewards—HUSD—is proposed to encourage both behavioral diversity and skill distinguishability.
SkillTree: Explainable Skill-Based Deep Reinforcement Learning for Long-Horizon Control Tasks
Yongyan Wen (Harbin Institute of Technology), Peng Liu (Kuaishou Technology)
Explainability and InterpretabilityKnowledge DistillationRobotic IntelligenceReinforcement LearningTabular
🎯 What it does: A hierarchical interpretable deep reinforcement learning framework called SkillTree is constructed, which maps continuous action spaces to discrete skill spaces and implements high-level policies using differentiable decision trees, making the decision-making process at the skill level interpretable.
Skip Mamba Diffusion for Monocular 3D Semantic Scene Completion
Li Liang (University of Western Australia), Ajmal Saeed Mian (University of Western Australia)
SegmentationGenerationDiffusion modelAuto EncoderPoint Cloud
🎯 What it does: This paper proposes a monocular 3D semantic scene completion method based on a variational autoencoder conditional latent space and Skimba diffusion network.
SkipPool: Improved Sparse Hierarchical Graph Pooling with Differentiable Exploration
Sarith Imaduwage (Independent Researcher)
Graph Neural NetworkPoint CloudGraph
🎯 What it does: An improved sparse hierarchical graph pooling method called SKIPPOOL is proposed, which skips over-represented areas through differentiable exploration to better preserve graph-level information.
SLACE: A Monotone and Balance-Sensitive Loss Function for Ordinal Regression
Inbar Nachmani (Technion Israel Institute of Technology), Avigdor Gal (Technion Israel Institute of Technology)
ClassificationOptimizationTabular
🎯 What it does: The SLACE loss function is proposed, which combines monotonicity, balanced sensitivity, and convexity for ordinal regression.
SlerpFace: Face Template Protection via Spherical Linear Interpolation
Zhizhou Zhong (Fudan University), Shuigeng Zhou (Tencent)
RecognitionSafty and PrivacyAdversarial AttackDiffusion modelImage
🎯 What it does: This paper studies a new facial template protection method called SlerpFace, designed to resist inversion attacks based on diffusion models.
Slice-and-Pack: Tailoring Deep Models for Customized Requirements
Ruice Rao (Nanjing University), Ming Li (Nanjing University)
OptimizationKnowledge DistillationConvolutional Neural NetworkRecurrent Neural NetworkImageText
🎯 What it does: Proposes the Slice-and-Pack framework, which first splits existing models into fragments by category and then assembles a minimized usable model according to user needs.
SLIP: Spoof-Aware One-Class Face Anti-Spoofing with Language Image Pretraining
Pei-Kai Huang (National Tsing Hua University), Chiou-Ting Hsu (National Tsing Hua University)
ClassificationRecognitionTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper proposes the SLIP framework through a CLIP-based visual-language pre-training model and prompt learning, utilizing language-guided pseudo-spoof prompts to generate spoof cue maps, achieve feature decoupling, and integrate pseudo-spoof image features, addressing the performance decline caused by the lack of spoof data in single-class FAS.
SLR-MVTC: Smooth Low-Rank Multi-View Tensor Clustering
Zhen Long (University of Electronic Science and Technology of China), Ce Zhu (University of Electronic Science and Technology of China)
Representation LearningImage
🎯 What it does: This paper proposes a multi-view clustering method based on smooth low-rank tensors (SLR-MVTC), which directly learns shared latent features from multi-view data and achieves a unified modeling of local smoothness and global high-order associations across views through low-rank approximation of low-frequency components.
SLRL: Semi-Supervised Local Community Detection Based on Reinforcement Learning
Li Ni (Anhui University), Victor S. Sheng (Texas Tech University)
Graph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper proposes a semi-supervised local community detection method based on reinforcement learning, called SLRL, which utilizes known communities to guide local structure extraction and community expansion.
Small Language Model Makes an Effective Long Text Extractor
Yelin Chen (Xinjiang University), Jie Tang (Tsinghua University)
RecognitionTransformerLarge Language ModelText
🎯 What it does: A lightweight span-based method SeNER is proposed for entity recognition in long texts, capable of identifying long entity blocks under GPU memory-friendly conditions.
SMamba: Sparse Mamba for Event-based Object Detection
Nan Yang (Chang'an University), Xiangmo Zhao (Tsinghua University)
Object DetectionTransformerImage
🎯 What it does: This paper proposes Sparse Mamba, a sparse Transformer based on the Mamba architecture for object detection using event cameras.
SMMF: Square-Matricized Momentum Factorization for Memory-Efficient Optimization
Kwangryeol Park (Ulsan National Institute of Science and Technology), Seulki Lee (Ulsan National Institute of Science and Technology)
OptimizationConvolutional Neural NetworkTransformerImageText
🎯 What it does: The paper presents SMMF, an adaptive learning rate optimizer that compresses arbitrary-order momentum tensors through square matricization and one-time matrix decomposition, significantly reducing memory usage.
Smoothness Really Matters: A Simple Yet Effective Approach for Unsupervised Graph Domain Adaptation
Wei Chen (Beihang University), Fuzhen Zhuang (Independent Researcher)
Domain AdaptationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a UGDA method for direct structural smoothing on the target graph—TDSS, which combines neighborhood sampling and Laplacian smoothing to enhance the robustness of node representations.
SMoSE: Sparse Mixture of Shallow Experts for Interpretable Reinforcement Learning in Continuous Control Tasks
Mátyás Vincze (University of Trento), Giovanni Iacca (Fondazione Bruno Kessler)
Explainability and InterpretabilityKnowledge DistillationReinforcement LearningMixture of ExpertsSequential
🎯 What it does: This paper proposes a Sparse Mixture of Experts (SMOSE) model that combines a single linear expert and an interpretable router to achieve reinforcement learning for continuous control tasks.
SMR-Net: Semantic-Guided Mutually Reinforcing Network for Cross-Modal Image Fusion and Salient Object Detection
Guobao Xiao (Tongji University), Rui Ming (Minjiang University)
Image TranslationObject DetectionConvolutional Neural NetworkImage
🎯 What it does: A lightweight Semantic-guided Mutually Reinforcing network (SMR-Net) is proposed, which simultaneously performs cross-modal image fusion and salient object detection.
Social Recommendation via Graph-Level Counterfactual Augmentation
Yinxuan Huang (National University of Defense Technology), Bin Zhou (National University of Defense Technology)
Recommendation SystemGraph Neural NetworkGraph
🎯 What it does: In the social recommendation task, the SR-GCA model is proposed, which constructs and learns enhanced user embeddings through graph-level counterfactual links, thereby improving recommendation accuracy.
SocialSim: Towards Socialized Simulation of Emotional Support Conversation
Zhuang Chen (Central South University), Minlie Huang (Tsinghua University)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Proposes the SocialSim framework, which combines a diverse seeker personality library and a supporter cognitive reasoning chain to generate high-quality emotional support dialogues, and based on this, constructs the SSConv dataset.
SOLA-GCL: Subgraph-Oriented Learnable Augmentation Method for Graph Contrastive Learning
Tianhao Peng (Beihang University), Haoyi Xiong (Baidu Inc.)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A learnable data augmentation method based on subgraphs (SOLA-GCL) is proposed, which divides the graph into multiple dense subgraphs and learns the optimal augmentation strategy for each subgraph to generate diverse views, enhancing the effectiveness of graph contrastive learning.
SoLA: Leveraging Soft Activation Sparsity and Low-Rank Decomposition for Large Language Model Compression
Xinhao Huang (Hong Kong University of Science and Technology), Zeyi Wen (Hong Kong University of Science and Technology)
CompressionTransformerLarge Language ModelText
🎯 What it does: A training-free LLM compression method called SoLA is proposed, which utilizes soft activation sparsity and low-rank decomposition for fine-grained model compression, retaining a small number of highly activated neurons and performing SVD decomposition on the remaining parts, achieving significant compression rates without the need for post-training.
Solving Epistemic Logic Programs Using Generate-and-Test with Propagation
Jorge Fandinno (University of Nebraska Omaha), Lute Lillo (University of Vermont)
Benchmark
🎯 What it does: A general generation-testing framework is proposed, and a new epistemic logic program solver is implemented within this framework.
Solving Higher-Order Quantified Boolean Satisfiability via Higher-Order Model Checking
Hiroshi Unno (Tohoku University), Jie-Hong Roland Jiang (National Taiwan University)
Benchmark
🎯 What it does: The first higher-order quantified Boolean satisfiability (HOQBF) solver HOMCSAT is proposed and implemented by transforming HOQBF into a higher-order model checking problem and utilizing the HORSAT2 solver to complete the solution.
Solving Multiagent Path Finding on Highly Centralized Networks
Foivos Fioravantes (Czech Technical University in Prague), Tung Anh Vu (Charles University)
OptimizationGraph
🎯 What it does: A fixed-parameter tractable (FPT) algorithm for the multi-agent pathfinding (MAPF) problem in highly centralized networks is proposed, and it is proven that the problem remains NP-hard under star topologies or low-degree tree structures.
Solving Robust Markov Decision Processes: Generic, Reliable, Efficient
Tobias Meggendorfer (Lancaster University), Patrick Wienhöft (Dresden University of Technology)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: A general, reliable, and efficient solving framework for Robust Markov Decision Processes (RMDP) is proposed, supporting various uncertainty sets and objectives, including total reward, long-term average reward, and random shortest path.
SongEditor: Adapting Zero-Shot Song Generation Language Model as a Multi-Task Editor
Chenyu Yang (Chinese University of Hong Kong), Haizhou Li
GenerationTransformerLarge Language ModelDiffusion modelAuto EncoderTextAudio
🎯 What it does: This paper presents SongEditor, a multi-task editor that extends a zero-shot song generation language model, supporting segment-wise and track-wise editing as well as generating songs from scratch.
SongGLM: Lyric-to-Melody Generation with 2D Alignment Encoding and Multi-Task Pre-Training
Jiaxing Yu (Zhejiang University), Kejun Zhang (Zhejiang University)
GenerationTransformerTextAudio
🎯 What it does: This paper presents SongGLM, an automatic lyric-to-melody generation system based on unified two-dimensional alignment encoding and multi-task pre-training;
SongSong: A Time Phonograph for Chinese SongCi Music from Thousand of Years Away
Jiliang Hu (Wuhan University), Lefei Zhang (Wuhan University)
GenerationTransformerDiffusion modelTextAudio
🎯 What it does: Designed and implemented the SongSong model, which can automatically generate complete classical music (including melody, vocals, and accompaniment) based on ancient poetry (Song Ci), and released the first classical Song Ci music dataset OpenSongSong (29.9 hours).
SORREL: Suboptimal-Demonstration-Guided Reinforcement Learning for Learning to Branch
Shengyu Feng (Carnegie Mellon University), Yiming Yang (Carnegie Mellon University)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: A reinforcement learning framework called SORREL based on suboptimal demonstrations is proposed to automatically learn branching heuristics for mixed-integer linear programming.
Sound Over-Approximation of Equational Reasoning with Variable-Preserving Rules Parameterized by Derivation Depth
Mateus de Oliveira Oliveira (Stockholm University)
🎯 What it does: This study investigates a decidable and feasible algorithm for equivalent reasoning under the constraint of a given depth d while maintaining variable equality.
SoundBrush: Sound as a Brush for Visual Scene Editing
Kim Sung-Bin (POSTECH), Tae-Hyun Oh (Institute for Convergence Research and Education in Advanced Technology)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningImageMultimodalityAudio
🎯 What it does: A model called SoundBrush is proposed, which uses sound as a 'brush' to edit visual scenes. It can accurately insert sound-producing objects or adjust the scene to match the sound semantics based on the input natural sound, while keeping the original structure of the image unchanged; this method is also extended to 3D scene editing.
SOVGaussian: Sparse-View 3D Gaussian Splatting for Open-Vocabulary Scene Understanding
Peng Ling (Tsinghua University), Wenming Yang (Tsinghua University)
SegmentationGenerationDepth EstimationAnomaly DetectionGaussian SplattingPoint Cloud
🎯 What it does: By training a 3D Gaussian scene representation with only three perspective images, a depth-constrained neural language field is constructed, enabling open vocabulary 3D scene understanding and new perspective language querying and synthesis under sparse viewpoints.
Sp3ctralMamba: Physics-Driven Joint State Space Model for Hyperspectral Image Reconstruction
Ge Meng (Xiamen University), Xinghao Ding (Xiamen University)
RestorationImagePhysics Related
🎯 What it does: For hyperspectral image reconstruction, we propose Sp3ctralMamba—a joint state space model that combines physical priors;
SPAC: Sparse Partitioning and Adaptive Core Tensor Pruning Model for Knowledge Graph Completion
Chuhong Yang (Beijing Institute of Technology), Nan Wu (Beijing Institute of Technology)
OptimizationKnowledge DistillationGraph Neural NetworkGraph
🎯 What it does: The SPAC model is proposed, utilizing sparse partitioning, gated intermediate variables, and an adaptive pruning tensor decomposition method to complete knowledge graph completion.
SPACETIME: Causal Discovery from Non-Stationary Time Series
Sarah Mameche (CISPA Helmholtz Center for Information Security), Jilles Vreeken (CISPA Helmholtz Center for Information Security)
Anomaly DetectionOptimizationTime Series
🎯 What it does: A unified framework named SPACETIME is proposed, capable of simultaneously performing causal graph discovery for non-stationary multi-dataset time series, regime change point detection, and spatial context grouping.
Sparis: Neural Implicit Surface Reconstruction of Indoor Scenes from Sparse Views
Yulun Wu (Tsinghua University), Yu-Shen Liu (Tsinghua University)
Depth EstimationOptimizationNeural Radiance FieldImagePoint Cloud
🎯 What it does: A neural implicit surface reconstruction method named Sparis is proposed, which utilizes matching information between images from sparse viewpoints and triangulation to obtain more accurate depth priors, and combines cross-view reprojection loss to achieve high-quality 3D reconstruction of indoor scenes.
Sparse Transfer Learning Accelerates and Enhances Certified Robustness: A Comprehensive Study
Zhangheng Li (University of Texas), Zhangyang Wang (University of Texas)
OptimizationComputational EfficiencyAdversarial AttackDiffusion modelImage
🎯 What it does: A method combining sparse transfer learning with sparse differential verification is proposed to accelerate and enhance the robustness of L2-norm certification.
SparX: A Sparse Cross-Layer Connection Mechanism for Hierarchical Vision Mamba and Transformer Networks
Meng Lou (University of Hong Kong), Yizhou Yu (University of Hong Kong)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a sparse cross-layer connection mechanism called SparX, and constructs two visual backbone networks, SparX-Mamba and SparX-Swin, based on this mechanism to achieve cross-layer feature fusion and reuse.
Spatial Annealing for Efficient Few-shot Neural Rendering
Yuru Xiao (Harbin Institute of Technology), Xianming Liu (Harbin Institute of Technology)
GenerationData SynthesisComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: A method called SANeRF based on Spatial Annealing is proposed, achieving efficient few-shot view synthesis in the TriMipRF (Tri-plane Neural Radiance Field) with mixed pre-filtering.
Spatial-Temporal Heterogenous Graph Contrastive Learning for Microservice Workload Prediction
Mohan Gao (Shanghai Jiao Tong University), Haoyuan Ge (Ant Group)
Graph Neural NetworkContrastive LearningGraphTime Series
🎯 What it does: A model for microservice workload prediction called STEAM is proposed, which combines spatiotemporal heterogeneous graph contrastive learning and multi-scale learning;
Spatial-Temporal Knowledge Distillation for Takeaway Recommendation
Shuyuan Zhao (Beijing Jiaotong University), Huaiyu Wan (Beijing Jiaotong University)
Recommendation SystemKnowledge DistillationGraph Neural NetworkTransformerSequential
🎯 What it does: This paper proposes a delivery recommendation framework based on spatial-temporal knowledge distillation, STKDRec, which first extracts high-order dependencies using a spatial-temporal knowledge graph (STKG), then models user dynamic preferences with a spatial-temporal Transformer, and integrates knowledge from both perspectives through knowledge distillation.
Spatiotemporal Blind-Spot Network with Calibrated Flow Alignment for Self-Supervised Video Denoising
Zikang Chen (Tsinghua University), Haoqian Wang (Tsinghua University)
RestorationKnowledge DistillationOptical FlowVideo
🎯 What it does: This paper proposes a self-supervised video denoising spatiotemporal blind spot network (STBN), which combines bidirectional optical flow alignment, spatial receptive field expansion, and unsupervised optical flow distillation to achieve global spatiotemporal information aggregation.
SpatioTemporal Learning for Human Pose Estimation in Sparsely-Labeled Videos
Yingying Jiao (Jilin University), Zheqi Wu (Zhejiang Gongshang University)
Pose EstimationTransformerVideo
🎯 What it does: In sparsely annotated videos, the STDPose framework is proposed to achieve pose propagation and video pose estimation.
Spatiotemporal-Aware Neural Fields for Dynamic CT Reconstruction
Qingyang Zhou (National University of Defense Technology), Zhiping Cai (National University of Defense Technology)
RestorationTransformerNeural Radiance FieldBiomedical DataComputed Tomography
🎯 What it does: Designed the STNF4D framework to achieve high-resolution dynamic CT reconstruction;
Spatiotemporal-aware Trend-Seasonality Decomposition Network for Traffic Flow Forecasting
Lingxiao Cao (Ocean University of China), Junyu Dong (Ocean University of China)
Graph Neural NetworkTransformerTime Series
🎯 What it does: This paper proposes a Spatio-Temporal Aware Trend-Season Decomposition Network (STDN) that achieves traffic flow prediction through dynamic relationship graph learning, spatio-temporal embedding, and a trend-season decomposition module.
Specifying What You Know or Not for Multi-Label Class-Incremental Learning
Aoting Zhang (Institute of Information Engineering, Chinese Academy of Sciences), Yu Zhou (Nankai University)
ClassificationObject DetectionTransformerImage
🎯 What it does: This paper studies multi-label incremental learning and proposes the HCP framework, which addresses the contradiction of learning objectives through dynamic feature purification, memory enhancement, and unknown knowledge mining, significantly alleviating catastrophic forgetting under no-replay conditions.
Spectra of Cardinality Queries over Description Logic Knowledge Bases
Quentin Manière (Leipzig University), Marcin Przybyłko (University of Warsaw)
🎯 What it does: The paper studies the spectra of answers to cardinality queries on description logic (DL) knowledge bases—specifically, the set of integers or ∞ that cardinality queries can yield across all models—and provides an effective representation of this spectrum.
Spectral Motion Alignment for Video Motion Transfer Using Diffusion Models
Geon Yeong Park (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: This paper proposes the Spectral Motion Alignment (SMA) framework, which aligns video motion vectors globally and locally in the frequency domain (wavelet transform and Fourier transform) to more accurately extract and transfer motion patterns in diffusion models.
Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration
Ziyang Ma (Shanghai Jiao Tong University), Xie Chen (Shanghai Jiao Tong University)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningBenchmarkAudio
🎯 What it does: This work systematically explores the key design and optimal configuration of combining large language models (LLMs) with speech encoders for automatic speech recognition (ASR), and provides a complete set of benchmark experiments.
Speech Watermarking with Discrete Intermediate Representations
Shengpeng Ji (Zhejiang University), Zhou Zhao (Zhejiang University)
Data SynthesisCompressionAuto EncoderGenerative Adversarial NetworkAudio
🎯 What it does: A speech watermarking framework called DiscreteWM based on discrete intermediate representation is proposed, which embeds watermark information by altering the modulus relationship of VQ-VAE discrete codes into single-second speech, supporting variable coding capacities from 1 to 150 bits.
Speed Master: Quick or Slow Play to Attack Speaker Recognition
Zhe Ye (Sun Yat-sen University), Shiqi Wang (City University of Hong Kong)
RecognitionAdversarial AttackAudio
🎯 What it does: A backdoor attack method called Speed Master, which utilizes speech rate transformation to covertly implant backdoors in speech recognition models, has been designed and implemented.
Speeding Up the NSGA-II with a Simple Tie-Breaking Rule
Benjamin Doerr (Institut Polytechnique de Paris), Martin S. Krejca (Institut Polytechnique de Paris)
Optimization
🎯 What it does: A balance tie-breaking rule based on 'rarity' was added to the selection phase of the classic NSGA-II, and a rigorous runtime analysis of the modified algorithm was conducted.
Speedup Techniques for Switchable Temporal Plan Graph Optimization
He Jiang (Carnegie Mellon University), Jiaoyang Li (Carnegie Mellon University)
OptimizationGraph
🎯 What it does: To address the delay issues that arise during the execution of Multi-Agent Path Finding (MAPF), this paper proposes an optimal optimization method based on the Switchable Temporal Plan Graph (STPG) and achieves significant acceleration on the original GSES search algorithm through four acceleration techniques.
SpeHeaTal: A Cluster-Enhanced Segmentation Method for Sperm Morphology Analysis
Yi Shi (Nanjing University), Rong Zeng (Nanjing University)
SegmentationImage
🎯 What it does: This paper addresses the issues of overlapping tails and staining impurities in clinical sperm images by proposing an unsupervised sperm complete segmentation method called SPEHEATAL, which achieves complete segmentation of sperm heads and tails.
Spike2Former: Efficient Spiking Transformer for High-performance Image Segmentation
Zhenxin Lei (University of Chinese Academy of Sciences), Guoqi Li (Institute of Automation, Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology, Chinese Academy of Sciences)
SegmentationSpiking Neural NetworkTransformerImage
🎯 What it does: This study rewrites Mask2Former into an efficient SNN architecture called Spike2Former for low-power image segmentation.
SpikeGS: Reconstruct 3D Scene Captured by a Fast-Moving Bio-Inspired Camera
Yijia Guo (Peking University), Tiejun Huang (Peking University)
RestorationData SynthesisDepth EstimationGaussian SplattingImagePoint Cloud
🎯 What it does: Developed the SpikeGS framework, which combines Bayer-pattern spike flow with 3D Gaussian Splatting to achieve rapid 3D reconstruction of short-term motion scenes captured by high-speed color Spike cameras.
Spiking Point Transformer for Point Cloud Classification
Peixi Wu (University of Science and Technology of China), Xiaoyan Sun (University of Science and Technology of China)
ClassificationSpiking Neural NetworkTransformerPoint Cloud
🎯 What it does: This paper proposes a Transformer-based spiking neural network framework—Spiking Point Transformer (SPT)—for 3D point cloud classification.
SpikingSSMs: Learning Long Sequences with Sparse and Parallel Spiking State Space Models
Shuaijie Shen (Southern University of Science and Technology), Luziwei Leng (Huawei Technologies)
Spiking Neural NetworkTextMultimodalitySequential
🎯 What it does: This paper proposes SpikingSSMs, which combines sparse parallel temporal neural networks with state space models for learning long sequence tasks.
SpikingYOLOX: Improved YOLOX Object Detection with Fast Fourier Convolution and Spiking Neural Networks
Wei Miao (Dalian University of Technology), Fengyu Cong (Dalian University of Technology)
Object DetectionSpiking Neural NetworkImage
🎯 What it does: Designed and implemented SpikingYOLOX, which combines spiking neural networks with the YOLOX framework to achieve a low-energy target detection model.
Spin: Diffusion-based Semantic Image Painting Through Independent Information Injection
Dantong Wu (Tsinghua University), Kai Zhang (Tsinghua University)
Image HarmonizationDiffusion modelImage
🎯 What it does: A semantic image painting framework named Spin is proposed, which achieves precise semantic injection and style consistency in masked areas through saliency-focused loss, attention region guidance, and style guidance.
SpotActor: Training-Free Layout-Controlled Consistent Image Generation
Jiahao Wang (Xi'an Jiaotong University), Jingdong Wang
GenerationData SynthesisDiffusion modelImageTextBenchmark
🎯 What it does: A training-free image generation pipeline called SpotActor has been developed, which can generate multiple visually consistent images based on a given layout box and text prompt.
SpotDiff: Spatial Gene Expression Imputation Diffusion with Single-Cell RNA Sequencing Data Integration
Tianyi Chen (City University of Hong Kong), Hau-San Wong (City University of Hong Kong)
TransformerDiffusion modelBiomedical Data
🎯 What it does: SpotDiff is proposed, a multimodal conditional diffusion model that combines prompt learning of spatial gene expression and the fusion of single-cell RNA-seq data to accurately fill in sparse missing values in spatial transcriptomics.
SPU-IMR: Self-supervised Arbitrary-scale Point Cloud Upsampling via Iterative Mask-recovery Network
Ziming Nie (Northwestern Polytechnical University), Jiaqi Yang (Northwestern Polytechnical University)
RestorationGenerationTransformerPoint Cloud
🎯 What it does: This paper proposes a self-supervised iterative mask recovery network (SPU-IMR), treating point cloud upsampling as a global shape completion task, achieving arbitrary point cloud densification through mask recovery.
Spurious Feature Eraser: Stabilizing Test-Time Adaptation for Vision-Language Foundation Model
Huan Ma (Tianjin University), Bingzhe Wu (Tencent)
Domain AdaptationRepresentation LearningTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a testing-time prompt tuning method called Spurious Feature Eraser (SEraser), which utilizes auxiliary images to eliminate spurious features (decision shortcuts) in visual-language foundation models (such as CLIP), forcing the model to focus on causally invariant features during inference.
SQLFixAgent: Towards Semantic-Accurate Text-to-SQL Parsing via Consistency-Enhanced Multi-Agent Collaboration
Jipeng Cen (Soochow University), Jingjing Wang (Soochow University)
Large Language ModelAgentic AITextRetrieval-Augmented Generation
🎯 What it does: This paper presents SQLFixAgent, a collaborative framework consisting of three LLM agents (SQLReviewer, QueryCrafter, SQLRefiner) for detecting and fixing syntax and semantic errors in SQL generated by fine-tuned LLMs.
SR-FoT: A Syllogistic-Reasoning Framework of Thought for Large Language Models Tackling Knowledge-based Reasoning Tasks
Wentao Wan (Sun Yat-sen University), Keze Wang (Sun Yat-sen University)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: A multi-stage reasoning framework SR-FoT is proposed, enabling large language models to solve knowledge-based question-answering tasks through deductive reasoning.
SrSv: Integrating Sequential Rollouts with Sequential Value Estimation for Multi-agent Reinforcement Learning
Xu Wan (Zhejiang University), Mingyang Sun (Alibaba DAMO Academy)
TransformerReinforcement LearningSequential
🎯 What it does: A multi-agent reinforcement learning framework SrSv is proposed, which combines serialized actions and serialized value estimation using Transformer to address credit allocation and scalability issues in large-scale systems.
SS-GEN: A Social Story Generation Framework with Large Language Models
Yi Feng (Beijing Jiaotong University), Jian Yu (Beijing Jiaotong University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: A Social Story dataset suitable for children with autism has been constructed, and the SS-GEN framework has been proposed.
SSAN: A Symbol Spatial-Aware Network for Handwritten Mathematical Expression Recognition
Haoran Zhang (Inner Mongolia University), Guanglai Gao (Inner Mongolia University)
RecognitionImage
🎯 What it does: An auxiliary task is proposed - predicting the spatial distribution map of handwritten mathematical expressions, and a Symbol Space Awareness Network (SSAN) is designed to be jointly trained with the HMER model;
SSC-VAE: Structured Sparse Coding Based Variational Autoencoder for Detail Preserved Image Reconstruction
Hao Wang (Tongji University), Ye Luo (Tongji University)
RestorationConvolutional Neural NetworkAuto EncoderImageUltrasound
🎯 What it does: This paper proposes a Structured Sparse Coding Variational Autoencoder (SSC-VAE), which enhances image reconstruction and denoising detail preservation by introducing a learnable threshold inference and refinement module into sparse coding.
SSE-SAM: Balancing Head and Tail Classes Gradually Through Stage-Wise SAM
Xingyu Lyu (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
ClassificationOptimizationImage
🎯 What it does: A two-stage variant of Sharpness-Aware Minimization (SAM) called SSE-SAM is proposed, which first uses SAM to help the main class escape saddle points, and then switches to ImbSAM to focus on the tail class for balanced training under long-tail distributions.
SSL-STMFormer Self-Supervised Learning Spatio-Temporal Entanglement Transformer for Traffic Flow Prediction
Zetao Li (University of Electronic Science and Technology of China), Shimin Cai (University of Electronic Science and Technology of China)
Autonomous DrivingOptimizationTransformerGraphTime Series
🎯 What it does: A model called SSL-STMFormer based on self-supervised learning and spatiotemporal attention is proposed for traffic flow prediction.
SSLFusion: Scale and Space Aligned Latent Fusion Model for Multimodal 3D Object Detection
Bonan Ding (Chongqing University), Jiale Cao (Tianjin University)
Object DetectionAutonomous DrivingGraph Neural NetworkImageMultimodalityPoint Cloud
🎯 What it does: A multi-modal 3D object detection framework called SSLFusion is designed and implemented, which enhances detection accuracy by aligning and fusing features from LiDAR point clouds and camera images at different scale levels.
SSUN-Net: Spatial-Spectral Prior-Aware Unfolding Network for Pan-Sharpening
Shijie Fang (Northwestern Polytechnical University), Hongping Gan (Northwestern Polytechnical University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a deep unfolding network based on spatial-spectral priors, SSUN-Net, for image pyramid fusion (pan-sharpening). By explicitly introducing physical prior constraints of PAN and MS and constructing a multi-scale prior structure, it significantly enhances the recovery of spatial and spectral details in images.
ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data
Zhenyu Lei (University of Virginia), Chen Chen (University of Virginia)
Graph Neural NetworkAuto EncoderGraphTime Series
🎯 What it does: Proposes the ST-FiT framework to address the inductive prediction problem in spatial-temporal graphs when training nodes lack time series.
ST-ReP: Learning Predictive Representations Efficiently for Spatial-Temporal Forecasting
Qi Zheng (Tongji University), Yaying Zhang (Tongji University)
Computational EfficiencyRepresentation LearningTransformerContrastive LearningTime Series
🎯 What it does: Proposes ST-ReP, a lightweight self-supervised spatio-temporal representation learning model that integrates current value reconstruction with future value prediction and incorporates multi-scale temporal supervision;
ST3: Accelerating Multimodal Large Language Model by Spatial-Temporal Visual Token Trimming
Jiedong Zhuang (Zhejiang University), Haoji Hu (Zhejiang University)
OptimizationComputational EfficiencyTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Visual token pruning for multimodal large language models (MLLM) to accelerate inference.
Stability and Generalization of Zeroth-Order Decentralized Stochastic Gradient Descent with Changing Topology
Xiaolin Hu (Renmin University of China), Yong Liu (Renmin University of China)
Optimization
🎯 What it does: This paper presents a generalization error analysis of zero-order decentralized stochastic gradient descent (ZO-DSGD) under variable topology, systematically deriving generalization upper bounds for convex, strongly convex, and non-convex problems, and providing topology-related generalization errors for local models.
Stability-based Generalization Analysis of Randomized Coordinate Descent for Pairwise Learning
Liang Wu (Southwestern University of Finance and Economics), Yunwen Lei (University of Hong Kong)
OptimizationTabular
🎯 What it does: This study investigates the generalization performance of Random Coordinate Descent (RCD) within the pairwise learning framework and provides a generalization error upper bound based on on-average parameter stability.
Stable Mean Teacher for Semi-supervised Video Action Detection
Akash Kumar (University of Central Florida), Yogesh Singh Rawat (University of Central Florida)
Object DetectionSegmentationKnowledge DistillationConvolutional Neural NetworkVideo
🎯 What it does: The Stable Mean Teacher framework is proposed, which combines student-teacher consistency, the EoR error recovery module, and DoP temporal consistency constraints in semi-supervised learning for video action detection.
Stable-Hair: Real-World Hair Transfer via Diffusion Model
Yuxuan Zhang (Shanghai Jiao Tong University), Jiaming Liu
Image TranslationGenerationLarge Language ModelDiffusion modelImageVideo
🎯 What it does: A hair transfer framework based on diffusion models, called Stable-Hair, is proposed, employing a two-stage pipeline: first, the Bald Converter transforms the user-input portrait into a bald proxy image, and then the Hair Extractor and Latent IdentityNet finely and high-fidelity transfer the target hairstyle onto the bald image, achieving virtual hair try-on.
StableVC: Style Controllable Zero-Shot Voice Conversion with Conditional Flow Matching
Jixun Yao (Northwestern Polytechnical University), Lei Xie (Northwestern Polytechnical University)
GenerationData SynthesisComputational EfficiencyFlow-based ModelOrdinary Differential EquationAudio
🎯 What it does: This paper proposes StableVC, which can independently control the timbre of the target speaker and the expressive style of the reference speech in zero-shot voice conversion, generating natural and high-quality speech.
STAIR: Manipulating Collaborative and Multimodal Information for E-Commerce Recommendation
Cong Xu (East China Normal University), Wei Zhang (East China Normal University)
Recommendation SystemGraph Neural NetworkContrastive LearningTextMultimodality
🎯 What it does: The research explores how to simultaneously utilize collaborative information and multimodal information in e-commerce scenarios, proposing the STAIR model and implementing multimodal initialization and step graph convolution fusion.