AAAI 2026 Papers — Page 34
AAAI Conference on Artificial Intelligence · 4149 papers
SD-PSFNet: Sequential and Dynamic Point Spread Function Network for Image Deraining
Jiayu Wang (University of Electronic Science and Technology of China), Shaoning Zeng (University of Electronic Science and Technology of China)
RestorationConvolutional Neural NetworkImage
🎯 What it does: Propose a multi-stage sequential network SD-PSFNet based on dynamic point spread function (PSF) for single raindrop image de-raining;
SDA: Steering-Driven Distribution Alignment for Open LLMs Without Fine-Tuning
Wei Xia (Peking University), Zhi-Hong Deng (Peking University)
GenerationTransformerLarge Language ModelText
🎯 What it does: Proposed a framework named SDA for distribution realignment of large language models during inference without training or parameter modifications, which can dynamically reallocate output probabilities during generation according to user instructions, thereby enhancing model consistency with human intentions.
SDE-HARL: Scalable Distributed Policy Execution for Heterogeneous-Agent Reinforcement Learning
Toan D. Gian (Northeastern University), Francesco Restuccia (Northeastern University)
CompressionComputational EfficiencyReinforcement LearningBenchmarkStochastic Differential Equation
🎯 What it does: Propose the SDE-HARL framework, which decomposes the policy network of heterogeneous multi-agents into local lightweight networks and edge global networks, achieving distributed execution through compressed hidden layers;
SDEval: Safety Dynamic Evaluation for Multimodal Large Language Models
Hanqing Wang (Shanghai Artificial Intelligence Laboratory), Xiangyang Zhu (ShanghaiTech University)
Safty and PrivacyExplainability and InterpretabilityLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed SDEval, a security dynamic evaluation framework for multimodal large language models;
SDNet: LiDAR Semantic Scene Completion with Sparse-Dense Fusion and Input-Aware Label Refinement
Tingming Bai, Eryun Liu (Zhejiang University)
SegmentationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: For the LiDAR semantic scene completion task, this paper proposes the SDNet framework and introduces an input-aware label correction strategy.
SE360: Semantic Edit in 360° Panoramas via Hierarchical Data Construction
Haoyi Zhong, Taehyun Rhee (University of Melbourne)
Image TranslationImage HarmonizationRestorationSegmentationGenerationData SynthesisTransformerDiffusion modelFlow-based ModelAuto EncoderImageTextMultimodality
🎯 What it does: Proposes the SE360 framework to achieve instruction-based 360° panorama semantic editing, incorporating a multi-level automated data generation pipeline with adaptive projection, and trains a Transformer-diffusion model on this data to support multi-condition editing (text, mask, or reference image).
SEA-PACE: Semi-Supervised Underwater Image Enhancement via Gaussian Process–Assisted Self-Paced Learning
Jingyang Wang (Ocean University of China), Junyu Dong (Ocean University of China)
RestorationConvolutional Neural NetworkImage
🎯 What it does: Proposes the SEA-PACE framework, which utilizes semi-supervised learning combined with model-aware reliability assessment and adaptive consistency learning to enhance underwater image enhancement.
SEAP: Sparse Expert Activation Pruning Unlocks the Brainpower of Large Language Models
Xun Liang (Renmin University of China), Zhiyu Li (Institute for Advanced Algorithms Research)
Computational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Proposes a training-free, task-adaptive sparse expert activation pruning (SEAP) method, which constructs task-specific pruning masks by analyzing hidden state clustering in multi-task calibration data, and dynamically selects corresponding computational paths during inference using a lightweight router, significantly reducing inference cost while maintaining most of the performance.
SEBSFormer: A Spectral-Enhanced Bi-Stream Transformer for Robust EEG Decoding
Lin Zhang (Shanghai Jiao Tong University), Lei Xu (Shanghai Jiao Tong University)
ClassificationRecognitionTransformerBiomedical Data
🎯 What it does: This paper proposes a Spectral Enhanced Dual-Stream Transformer (SEBSFormer), which enhances the spectral diversity and spatiotemporal representation capability of EEG signals through three mechanisms: a spectral compensation module, a multi-scale temporal attention module, and graph-guided dynamic fusion. Additionally, the paper formally proves the low-pass filtering characteristics of the Transformer's self-attention mechanism.
SecMoE: Communication-Efficient Secure MoE Inference via Select-Then-Compute
Bowen Shen (Harbin Institute of Technology), Zoe L. Jiang (Tianjin University)
Safty and PrivacyTransformerMixture of ExpertsText
🎯 What it does: Proposed a Select-Then-Compute based 2PC secure MoE inference framework called SecMoE, which achieves privacy preservation while maintaining sparse expert activation.
Security Games with Layered Defenses: Adaptive Adversaries and Gittins Indices
Chun Kai Ling (National University of Singapore), Christian Kroer (Columbia University)
OptimizationReinforcement LearningGraph
🎯 What it does: Proposes a security game model for hierarchical defense, studying the scenario where adaptive attackers dynamically switch between multiple attack paths, and provides corresponding optimal defense strategies;
See, Rank, and Filter: Important Word-Aware Clip Filtering via Scene Understanding for Moment Retrieval and Highlight Detection
YuEun Lee (Kyung Hee University), Jung Uk Kim (Kyung Hee University)
RetrievalTransformerLarge Language ModelVision Language ModelVideoTextMultimodality
🎯 What it does: This study proposes an important word-aware editing filtering framework, which identifies and ranks keywords in queries, and utilizes segment descriptions generated by multimodal large language models to achieve fine-grained editing filtering and precise localization in video retrieval and highlight detection.
SEED: Spectral Entropy-Guided Evaluation of Spatial-Temporal Dependencies for Multivariate Time Series Forecasting
Feng Xiong, Jianhong Lin (Tianjin University)
Graph Neural NetworkTransformerTime Series
🎯 What it does: This paper proposes the SEED framework, which uses spectral entropy to evaluate spatiotemporal dependencies in multivariate time series, dynamically balancing the independence and interdependence of variables, and achieving more accurate predictions through modules such as signed graph construction and context space extraction.
Seeing and Knowing in the Wild: Open-domain Visual Entity Recognition with Large-scale Knowledge Graphs via Contrastive Learning
Hongkuan Zhou (Robert Bosch GmbH), Steffen Staab (University of Stuttgart)
RecognitionTransformerVision Language ModelContrastive LearningImageTextGraph
🎯 What it does: Proposes the Knowledge-guided Contrastive Learning (KnowCoL) framework for open-domain visual entity recognition, mapping images, text descriptions, and structured information from the Wikidata knowledge graph into a shared semantic space to achieve zero-shot entity recognition.
Seeing Beyond Illusion: Generalized and Efficient Mirror Detection
Mingfeng Zha (University of Electronic Science and Technology of China), Yang Yang (University of Electronic Science and Technology of China)
SegmentationComputational EfficiencyTransformerSupervised Fine-TuningMixture of ExpertsContrastive LearningImage
🎯 What it does: Proposes a mirror detection framework called MirrorSAM based on the Segment Anything Model (SAM), achieving efficient and robust mirror segmentation through directional expert mixture, depth token calibration, and pixel-prototype contrast loss.
Seeing from Another Perspective: Evaluating Multi-View Understanding in MLLMs
Chun-Hsiao Yeh (University of California, Berkeley), Yi Ma (University of California, Berkeley)
Prompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper constructs a novel multi-perspective reasoning benchmark called All-Angles Bench and uses it to evaluate the multi-perspective understanding capabilities of 38 multimodal large language models (MLLMs);
Seeing in Double: Dual-Granularity BEV Segmentation via Mamba-Driven Alignment and Polar-Decoupled Experts
Jiaxin Cai (Fuzhou University), Wenxi Liu (Fuzhou University)
SegmentationAutonomous DrivingMixture of ExpertsPoint Cloud
🎯 What it does: Propose a dual-resolution polar BEV segmentation framework that preserves distant details and corrects geometric distortions through Mamba-driven cross-resolution alignment and polar-separated hybrid experts.
Seeing Is Believing: Grounding Long-Video Understanding in Spatio-Temporal Visual Evidence
Zhaoyang Wei (University of Chinese Academy of Sciences), Xuehui Yu (Tencent)
CompressionExplainability and InterpretabilityComputational EfficiencyReinforcement LearningVision Language ModelVideoBenchmark
🎯 What it does: Propose the EV 2-Bench long video understanding benchmark and design the DynamicSelect adaptive token compression method.
Seeing Is Believing: Rich-Context Hallucination Detection for MLLMs via Backward Visual Grounding
Pinxue Guo (Fudan University), Wenqiang Zhang (Fudan University)
SegmentationAnomaly DetectionExplainability and InterpretabilityVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposes VBackChecker, a reference-free multimodal large language model hallucination detection framework that verifies through pixel-level visual reverse attribution.
Seeing the Unseen: Zooming in the Dark with Event Cameras
Dachun Kai (University of Science and Technology of China), Xiaoyan Sun (University of Science and Technology of China)
Super ResolutionConvolutional Neural NetworkOptical FlowVideoMultimodality
🎯 What it does: Proposes RetinexEVSR, an end-to-end framework for low-light video super-resolution that leverages Retinex prior and event signals.
Seeing through the Conflict: Transparent Knowledge Conflict Handling in Retrieval-Augmented Generation
Hua Ye (Nanjing University), Fei Shen (University of Oklahoma)
RetrievalExplainability and InterpretabilityTransformerPrompt EngineeringContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: Propose the TCR framework, which decouples semantics and facts in retrieval-augmented generation (RAG) to real-time detect and resolve knowledge conflicts.
Seeing Through the Rain: Resolving High-Frequency Conflicts in Deraining and Super-Resolution via Diffusion Guidance
Wenjie Li (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes a joint de-raining and super-resolution framework called DHGM, which combines diffusion models with high-pass filtering and guided filtering to achieve high-frequency texture restoration and magnification of rainy images.
SEFEL: A Simple Yet Effective Framework for Fast Event Linking
Yinan Liu (Northeastern University), Xiaochun Yang (Northeastern University)
RetrievalComputational EfficiencyRepresentation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposed an end-to-end event linking framework called SEFEL, which constructs argument-aware event representations by leveraging event argument information and uniformly handles event linking both inside and outside knowledge bases.
SegDINO3D: 3D Instance Segmentation Empowered by Both Image-Level and Object-Level 2D Features
Jinyuan Qu (Tsinghua University), Lei Zhang (International Digital Economy Academy)
SegmentationConvolutional Neural NetworkTransformerImagePoint Cloud
🎯 What it does: Propose the SegDINO3D framework, which enhances 3D instance segmentation by leveraging 2D image-level and object-level features;
SegMem-RAG: Adaptive Memory for Retrieval-Augmented Generation in Open-Ended Knowledge Environments
Xuanbo Fan (Peking University), Yan Zhang (Microsoft Corporation)
GenerationRetrievalTransformerTextBiomedical DataBenchmarkFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes SEGMEM-RAG, constructing a retrieval-enhanced generation framework capable of self-planning, evaluation, and adaptation in dynamic, unlabeled, multi-source knowledge environments.
Segment and Matte Anything in a Unified Model
Zezhong Fan (Walmart Global Tech), Kannan Achan (Walmart Global Tech)
SegmentationTransformerImage
🎯 What it does: This paper proposes the Segment And Matte Anything (SAMA) model, extending Segment Anything (SAM) to achieve high-precision segmentation and interactive matting within the same framework.
Segment Anything Across Shots: A Method and Benchmark
Hengrui Hu (Fudan University), Henghui Ding (Fudan University)
SegmentationData SynthesisTransformerVideoBenchmark
🎯 What it does: This paper proposes a cross-camera semi-supervised video object segmentation framework called SAAS, and simulates multi-camera transitions on single-camera data through Transition Imitation Data Augmentation (TMA) to address the challenge of camera transitions in multi-camera video segmentation;
SELDON: Supernova Explosions Learned by Deep ODE Networks
Jiezhong Wu (NSF-Simons AI Institute for the Sky), Noelle I. Samia (NSF-Simons AI Institute for the Sky)
Recurrent Neural NetworkAuto EncoderTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: Proposed SELDON, a continuous-time variational autoencoder for real-time prediction and physical parameter inference in sparse, heteroscedastic, multi-band light curves.
Selection of LLM Fine-Tuning Data Based on Orthogonal Rules
Xiaomin Li (Harvard University), Hong Hu (Massachusetts Institute of Technology)
Data-Centric LearningLarge Language ModelText
🎯 What it does: This paper proposes a fully automated data selection framework that leverages LLM to automatically generate rules and employs DPP (Determinantal Point Process) for rule orthogonality screening, aiming to select high-quality subsets from large-scale corpora to enhance LLM fine-tuning performance.
Selective Diffusion Distillation for Real-World High-Scale Image Super-Resolution
Wenli Zheng (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
Super ResolutionKnowledge DistillationConvolutional Neural NetworkDiffusion modelAuto EncoderImageBenchmark
🎯 What it does: Proposes the Selective Diffusion Distillation (SDD) framework, which leverages knowledge from a low-scale diffusion teacher model to guide high-scale image super-resolution.
Self-Adaptive Graph Mixture of Models
Mohit Meena (Fujitsu Research of India), Mahesh Chandran (Fujitsu Research of India)
ClassificationComputational EfficiencyRepresentation LearningGraph Neural NetworkMixture of ExpertsGraph
🎯 What it does: Propose the Self-Adaptive Graph Mixture of Models (SAGMM) framework, which utilizes multiple GNNs as experts and adaptively selects and combines them through topology-aware attention gates to enhance performance in graph learning tasks.
Self-Correction Distillation for Structured Data Question Answering
Yushan Zhu (Zhejiang University), Junlan Feng (Ant Group)
Knowledge DistillationTransformerSupervised Fine-TuningPrompt EngineeringGraphTabularBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Designed and implemented a self-correcting distillation (SCD) framework for small-scale LLMs (≤8B parameters), significantly improving the accuracy of generating structured queries on structured data question answering (table, knowledge graph, temporal KG) tasks.
Self-Enhanced Image Clustering with Cross-Modal Semantic Consistency
Zihan Li (Shandong University), Jianlong Wu (Harbin Institute of Technology)
Representation LearningTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Propose a two-stage self-enhanced image clustering framework called SEIC, which first aligns the heads of the CLIP pre-trained encoder through cross-modal semantic consistency, and then fine-tunes the visual encoder using self-generated high-confidence pseudo labels with LoRA.
Self-Improving Sparse Retrieval Through Heuristic Representation Refinement and Representation-Focused Learning
Xiaojing Li, Meng Luo (Northeastern University)
RetrievalRepresentation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposes the self-improving sparse retrieval framework SIRE, which corrects model-generated representation hallucinations through iterative cycles of heuristic representation refinement and representation-focused learning.
Self-Indexing KVCache: Predicting Sparse Attention from Compressed Keys
Xu Yang (Hunan University), Zhuo Tang (Hunan University)
Computational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose Self-Indexing KVCache, which directly implements sparse attention by using compressed keys as a self-indexing structure;
Self-Interpretable Subgraph Neural Network with Deep Reinforcement Walk Exploration
Jianming Huang (WASEDA University), Hiroyuki Kasai (WASEDA University)
ClassificationExplainability and InterpretabilityGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Propose a self-explaining subgraph neural network that dynamically extracts discriminative substructures through reinforcement learning-based random walk exploration, achieving dual improvements in graph classification and interpretability.
Self-NPO: Data-Free Diffusion Model Enhancement via Truncated Diffusion Fine-Tuning
Fu-Yun Wang (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationComputational EfficiencyDiffusion modelImage
🎯 What it does: Propose Self-NPO, achieving diffusion model alignment without labels or reward models through self-supervised negative preference optimization (NPO), and significantly reduce training costs via truncated diffusion fine-tuning (TDFT).
Self-Supervised Contrastive Re-Learning for Multi-Graph Multi-Label Classification
Meixia Wang (Northeastern University), Xingwei Wang (Northeastern University)
ClassificationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose the PETAL framework, which in multi-graph multi-label learning learns bag-level and graph-level representations through unsupervised contrastive learning, achieving label prototype alignment and multi-granularity contrastive loss.
Self-Supervised Cross-City Trajectory Representation Learning Based on Meta-Learning
Yanwei Yu (Ocean University of China), Yuan Cao (Ocean University of China)
Representation LearningMeta LearningGraph Neural NetworkTransformerTime SeriesSequential
🎯 What it does: Propose MetaTRL, a self-supervised cross-city trajectory representation learning method based on meta-learning.
Self-Supervised Hypergraph Learning with Substructure Awareness for Hyperedge Prediction
Ming Li, Ke Lv (Central University Of Finance And Economics)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose a self-supervised hypergraph edge prediction framework named S Hyper 3, combining substructure context aggregation and tripartite contrastive learning.
Self-Supervised Inductive Logic Programming
Stassa Patsantzis (University of Surrey)
Explainability and InterpretabilityRepresentation LearningText
🎯 What it does: Propose a self-supervised ILP framework named SS-ILP, and design the Poker algorithm, which can automatically generate positive and negative samples and learn recursive logic programs by maximizing the second-order background theory (SONF) under conditions with only positive examples and unlabeled samples.
Self-Supervised Learning Based on Transformed Image Reconstruction for Equivariance-Coherent Feature Representation
Qin Wang (Institute for Advanced Simulation Data Analytics and Machine Learning Forschungszentrum Julich Gmbh), Kai Krajsek (Forschungszentrum Julich Gmbh)
Image TranslationObject DetectionSegmentationPose EstimationDepth EstimationRepresentation LearningTransformerAuto EncoderContrastive LearningImage
🎯 What it does: Propose a self-supervised learning framework based on intermediate transformation image reconstruction, which can achieve equivariant features while preserving invariance learning.
Self-supervised Multiplex Consensus Mamba for General Image Fusion
Yingying Wang (Xiamen University), Xinghao Ding (Xiamen University)
Mixture of ExpertsContrastive LearningImageMultimodalityMagnetic Resonance ImagingComputed TomographyPositron Emission Tomography
🎯 What it does: Proposed an SMc-Mamba framework for general image fusion, integrating Modality-Agnostic Feature Enhancement (MAFE), Multi-Channel Consensus Cross-Modal Mamba (MCCM), and Dual-Level Self-Supervised Contrastive Learning Loss (BSCL), achieving efficient multi-modal information fusion;
Self-Supervised One-Step Diffusion Refinement for Snapshot Compressive Imaging
Shaoguang Huang (China University of Geosciences), Hongyan Zhang (China University of Geosciences)
RestorationCompressionTransformerDiffusion modelImage
🎯 What it does: Proposed a self-supervised one-step diffusion refinement framework (OSD), which efficiently reconstructs multispectral images through an initial predictor and a single-step diffusion network.
Self-Supervised Representation Learning with Joint Embedding Predictive Architecture for Automotive LiDAR Object Detection
Haoran Zhu (New York University), Anna Ewa Choromanska (New York University)
Autonomous DrivingRepresentation LearningConvolutional Neural NetworkPoint Cloud
🎯 What it does: Propose AD-L-JEPA, a self-supervised pre-training framework that learns LiDAR object detection features by predicting masked region embeddings in the BEV space.
Semantic Document Derendering: SVG Reconstruction via Vision-Language Modeling
Adam Hazimeh (EPFL), Pascal Frossard (EPFL)
Image TranslationGenerationTransformerVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Propose a framework called SliDer based on Vision-Language Models (VLM) to restore color slide images into editable SVG format, achieving a complete conversion from pixels to structured vectors.
Semantic Feature Purification for Adversarially-Aware RGB-T Tracking
Jiahao Wang (Xidian University), Puhua Chen (Xidian University)
Object TrackingAdversarial AttackConvolutional Neural NetworkTransformerPrompt EngineeringMultimodality
🎯 What it does: This paper proposes a RGB-T tracking defense framework based on semantic feature purification (SFPT), achieving robust tracking under adversarial perturbations through the introduction of text prompts and adaptive perturbation-guided cross-modal fusion;
Semantic Guided Part Relation-aware Network for Point Cloud Completion
Zhensheng Zhou (Shanxi University), Chenghao Fang (Shanxi University)
GenerationTransformerLarge Language ModelVision Language ModelTextPoint Cloud
🎯 What it does: Generate textual descriptions of part relationships for 3D shapes using a multimodal large language model, and design SGPRNet to complete missing point clouds with this semantic guidance.
Semantic Volume: Quantifying and Detecting Both External and Internal Uncertainty in LLMs
Xiaomin Li (Harvard University), Anurag Beniwal (Amazon)
Explainability and InterpretabilityLarge Language ModelTextBenchmark
🎯 What it does: Propose the Semantic Volume method, which quantifies the internal and external uncertainty of LLMs by calculating the determinant of the Gram matrix on perturbed embeddings of queries or answers.
Semantic-Augmented Image Clustering via Adaptive Multi-Modal Collaboration
Xiaohan Zhang (Nanjing University), Huaxiong Li (Nanjing University)
Representation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Combine visual language pre-training models to generate textual descriptions, and perform unsupervised clustering on images through multi-modal collaboration and adaptive weighting.
Semantic-Aware Feature Enhancement for Partial Label Learning
Haowei Mei (Nanjing University), Huaxiong Li (Nanjing University)
ClassificationImageTabular
🎯 What it does: Proposes the Semantic-Aware Feature Enhancement (SAFE) method for addressing the partially labeled learning (PLL) problem, shifting the focus from label denoising to feature recovery and enhancement, and constructing a classifier that jointly integrates complete features with candidate labels.
Semantic-Consistent Bidirectional Contrastive Hashing for Noisy Multi-Label Cross-Modal Retrieval
Likang Peng (Sichuan University), Xu Wang (Sichuan University)
RetrievalConvolutional Neural NetworkSupervised Fine-TuningContrastive LearningMultimodality
🎯 What it does: Proposed a cross-modal hashing framework SCBCH for multi-label noise scenarios, combining CSCC and BSCH modules to achieve robust retrieval.
Semantic-Driven Visual Progressive Refinement for Aerial-Ground Person ReID: A Challenging Large-Scale Benchmark
Aihua Zheng (Anhui University), Bin Luo (Anhui University)
RecognitionRetrievalTransformerPrompt EngineeringMixture of ExpertsVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose a semantic-driven visual evolutionary refinement framework, SVPR-ReID, for aerial-ground person re-identification (AGPReID), and contribute a large-scale real-world scenario dataset, CP2108.
SemanticNN: Compressive and Error-Resilient Semantic Offloading for Extremely Weak Devices
Jiaming Huang (Zhejiang University), Wei Dong (Zhejiang University)
ClassificationObject DetectionCompressionExplainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: Designed and implemented a semantic encoder-decoder framework called SemanticNN for extremely weak embedded devices, capable of withstanding bit errors in wireless transmission, achieving feature compression, and enabling device-edge collaborative inference.
Semantics and Content Matter: Towards Multi-Prior Hierarchical Mamba for Image Deraining
Zhaocheng Yu (Harbin Institute of Technology), Yi Xiao (Harbin Institute of Technology)
RestorationVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Designed a multi-priority hierarchical Mamba network that uses CLIP text prior and DINOv2 visual prior to guide image de-raining.
SemanticVLA: Semantic-Aligned Sparsification and Enhancement for Efficient Robotic Manipulation
Wei Li (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
Robotic IntelligenceTransformerVision-Language-Action ModelMultimodality
🎯 What it does: SemanticVLA proposes a semantic alignment sparsification and enhancement framework for efficiently performing robotic grasping and manipulation tasks.
SEMC: Structure-Enhanced Mixture-of-Experts Contrastive Learning for Ultrasound Standard Plane Recognition
Qing Cai (Ocean University of China), Zhi Liu (Shandong University)
ClassificationRecognitionConvolutional Neural NetworkMixture of ExpertsContrastive LearningImageBiomedical DataUltrasound
🎯 What it does: For the task of ultrasound standard plane recognition, the SEMC framework is proposed, combining structure-aware feature fusion (SSFM) and expert-driven hierarchical contrastive learning with classification (MCRM)
Semi-Supervised High Dynamic Range Image Reconstructing via Bi-Level Uncertain Area Masking
Wei Jiang (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
RestorationImage
🎯 What it does: This paper proposes a semi-supervised HDR image reconstruction framework that utilizes a teacher-student model combined with an uncertainty mask to achieve high-quality HDR restoration with only a small amount of HDR labeled data.
Semi-supervised Latent Disentangled Diffusion Model for Textile Pattern Generation
Chenggong Hu (Zhejiang University), Li Sun (Hangzhou City University)
GenerationData SynthesisDiffusion modelContrastive LearningImage
🎯 What it does: Propose a two-stage model SLDDM-TPG to generate high-fidelity texture pattern images from real clothing images;
Semi-Supervised Regression by Preserving Ranking Relationships Between Close Unlabeled Samples
Ximing Li (Jilin University), Renchu Guan (Shenzhen University of Advanced Technology)
ImageTextAudio
🎯 What it does: Propose a new semi-supervised regression method SSR-RCUS, which combines samples generated by closed Mixup with an auxiliary ranking task to enhance regression performance.
Semi-Supervised Semantic Segmentation via Derivative Label Propagation
Yuanbin Fu (Tianjin University), Xiaojie Guo (Tianjin University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a semi-supervised semantic segmentation framework called DerProp, which corrects pseudo-labels by performing discrete derivative operations on pixel features and incorporating similarity regularization.
Semi-Supervised Synthetic Data Generation with Fine-Grained Relevance Control for Short Video Search Relevance Modeling
Haoran Li (Peking University), Jiao Ran (ByteDance Douyin Content Group)
Data SynthesisRetrievalTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningVideoText
🎯 What it does: Constructed a four-level relevance dataset for Chinese short video retrieval and proposed a semi-supervised controllable relevance synthesis (SSRA) pipeline to generate synthetic query-document pairs with domain-distributed data and controllable relevance labels, enhancing embedding model performance.
Semore: VLM-guided Enhanced Semantic Motion Representations for Visual Reinforcement Learning
Wentao Wang, Yan Wang (Tsinghua University)
Large Language ModelReinforcement LearningVision Language ModelImage
🎯 What it does: Propose the Semore framework, which employs a VLM-guided semantic and motion dual-stream encoder to explicitly supervise and interact features in visual reinforcement learning, thereby enhancing environmental representation capabilities.
Sentient: Detecting APTs via Capturing Indirect Dependencies and Behavioral Logic
Wenhao Yan (Institute of Information Engineering, Chinese Academy of Sciences), Junrong Liu (Institute of Information Engineering, Chinese Academy of Sciences)
Anomaly DetectionGraph Neural NetworkTransformerAuto EncoderGraphSequential
🎯 What it does: Leverage pre-trained graph Transformer and intent analysis module (Bi-Mamba2) to perform global attention learning on the original graph constructed from system audit logs, capturing indirect dependencies and behavioral logic, and detecting APT attacks using only normal logs.
SepPrune: Structured Pruning for Efficient Deep Speech Separation
Yuqi Li (City College of New York), Yao Lu (University of Southern California)
CompressionComputational EfficiencyAudio
🎯 What it does: Developed a structured pruning framework called SepPrune for compressing deep speech separation models and reducing computational costs.
SEQRET: Mining Rule Sets from Event Sequences
Aleena Siji (Helmholtz AI), Jilles Vreeken (CISPA Helmholtz Center for Information Security)
OptimizationExplainability and InterpretabilitySequential
🎯 What it does: This paper proposes a sequence rule set mining method called SEQRET based on the MDL principle, which can discover sequence rules X→Y containing both conditional and unconditional dependencies from long event sequences. The method supports gaps in the rules and ensures the generated rule set is concise and non-redundant.
Sequence-Free for Compound Protein Interaction Prediction
Hongzhi Zhang (Wuhan University), Jia Wu (Macquarie University)
Drug DiscoveryGraph Neural NetworkTransformerLarge Language ModelContrastive LearningMultimodalityBiomedical Data
🎯 What it does: This paper proposes BioText-CPI, a framework for predicting compound-protein interactions based on protein biomedical text descriptions without requiring protein sequences, supporting multi-modal input.
Sequential Selling with Sunk Cost Bias
Yasushi Kawase (University of Tokyo), Tomohiro Nakayoshi (University of Tokyo)
🎯 What it does: Introduce a sunk cost bias model for the sequential selling problem, distinguishing three types of agents: optimistic, ignorant, and sophisticated, and analyzing their optimal strategies and profit losses.
SeqWalker: Sequential-Horizon Vision-and-Language Navigation with Hierarchical Planning
Zebin Han (North University of China), Zhi Han (Chinese Academy Of Sciences)
Robotic IntelligenceRecurrent Neural NetworkLarge Language ModelReinforcement LearningVision-Language-Action ModelImageTextMultimodalitySequentialBenchmark
🎯 What it does: This paper proposes a novel sequential long audio-visual navigation (SH-VLN) task and designs the SeqWalker model to achieve multi-task long-sequence navigation.
SERL: Self-Examining Reinforcement Learning on Open-Domain
Weixuan Ou (Zhejiang University), Yifan Qiao (Alibaba Group)
Large Language ModelReinforcement LearningText
🎯 What it does: Proposed an unsupervised self-checking reinforcement learning framework called SERL, where large language models simultaneously act as both generator (Actor) and evaluator (Judge), achieving self-improvement through internal comparative evaluation.
SEVADE: Self-Evolving Multi-Agent Analysis with Decoupled Evaluation for Hallucination-Resistant Sarcasm Detection
Ziqi Liu (Xi'an Jiaotong-Liverpool University), Yangbin Chen (Xi'an Jiaotong-Liverpool University)
ClassificationLarge Language ModelAgentic AITextRetrieval-Augmented Generation
🎯 What it does: Proposed a self-evolving framework named SEVADE based on dynamic multi-agent reasoning and decoupled evaluation for detecting sarcastic text.
SeViL: Semi-supervised Vision-Language Learning with Text Prompt Guiding for Moving Infrared Small Target Detection
Weiwei Duan (University of Electronic Science and Technology of China), Sicheng Zhu (University of Electronic Science and Technology of China)
Object DetectionPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Proposes a semi-supervised vision-language learning framework SeViL, which utilizes text prompts to guide the detection of small moving targets in infrared images.
SFedHIFI: Fire Rate-Based Heterogeneous Information Fusion for Spiking Federated Learning
Ran Tao (Southwestern University of Finance and Economics), Guisong Liu (Southwestern University of Finance and Economics)
Federated LearningSpiking Neural NetworkImage
🎯 What it does: Proposes the SFedHIFI framework, enabling edge devices to adaptively deploy spiking neural networks (SNNs) with varying widths based on local resources, and aggregates heterogeneous models through channel-level matrix decomposition;
SFGA: Similarity-Constrained Fusion Learning for Unsupervised Anomaly Detection in Multiplex Graphs
Huiliang Zhai (Xidian University), Jing Liu (Xidian University)
Anomaly DetectionGraph Neural NetworkAuto EncoderGraph
🎯 What it does: Designed an unsupervised multi-modal graph anomaly detection framework SFGA, which identifies abnormal nodes by combining multi-view graph autoencoders with cross-modal attention fusion and similarity constraints.
SGAT: Learning Feature Matching with Singularity-enhanced Graph Attention Network
Yizhuo Zhang (China University of Geosciences(Wuhan)), Xin Li (University at Albany)
Pose EstimationRetrievalGraph Neural NetworkTransformerImage
🎯 What it does: Propose a Singularity-enhanced Graph Attention Network (SGAT), which achieves high-precision image feature matching in low-texture and heavily occluded scenarios through a graph attention mechanism guided by single-point singularity and co-potential priors.
SGMHand: Structure-Guided Modulation for Structure-Aware Hand Inpainting
Chuancheng Shi (The University of Sydney), Fei Shen (Northern Arizona University)
RestorationGenerationTransformerDiffusion modelImageMeshGraph
🎯 What it does: Designed a structure-guided hand image inpainting framework called SGMHand, which enhances the detail and realism of hand generation in diffusion models by leveraging structural priors.
SGMT: Social Generating with Multiview-Guided Tuning In Recommender Systems
Jianghong Ma (Harbin Institute of Technology(Shenzhen)), Xiaofeng Zhang (Harbin Institute of Technology(Shenzhen))
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose a recommendation framework named SGMT, addressing the challenge of graph neural networks failing to capture high-order associations under sparse data, through interest-aware social edge generation and dual-view contrastive learning.
SGoT-R1: Social Graph of Thought Reasoning-Enhanced Multimodal Large Language Model for Harmful Meme Detection
Xiuxian Wang (Tianjin University), Anan Liu (Tianjin University)
ClassificationKnowledge DistillationGraph Neural NetworkTransformerSupervised Fine-TuningMultimodalityGraphChain-of-Thought
🎯 What it does: Proposed a multi-chain reasoning framework SGoTRE based on community consensus, and trained SGoT-R1 for harmful meme detection.
SGP4SR: Seperated-Modality Guided User Perference Learning for Multimodal Sequential Reconmmendation
Changhong Li, Chuhang Hong (Huazhong University Of Science And Technology)
Recommendation SystemGraph Neural NetworkTransformerMultimodality
🎯 What it does: Propose the SGP4SR model, which adopts a separated modal framework to learn user preferences, significantly reducing noise impact in multi-modal recommendations.
SGPFeat: Semantic and Geometric Priors for Multi-modal Image Matching
Yuxin Deng (Wuhan University), Jiayi Ma (Wuhan University)
Pose EstimationDepth EstimationKnowledge DistillationConvolutional Neural NetworkTransformerMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose a multi-modal image matching method called SGPFeat, which leverages semantic and geometric priors to enhance feature extraction and keypoint detection;
SGS-3D: High-Fidelity 3D Instance Segmentation via Reliable Semantic Mask Splitting and Growing
Chaolei Wang (Sun Yat-sen University), Ting Han (Sun Yat-sen University)
SegmentationAutonomous DrivingImagePoint Cloud
🎯 What it does: Propose SGS-3D, a training-agnostic framework based on the 'split-then-grow' idea, which first uses geometric primitives to separate and clean 2D-uplifted semantic masks, then gradually grows them into complete instances in 3D voxels based on spatial connectivity and feature similarity, achieving high-fidelity 3D instance segmentation.
ShadeEdit: A Utility-Preserving and Defense-Evasive Knowledge Manipulation Attack in Federated LLMs
Xu Zhang, Tao Xiang (Chongqing University)
Federated LearningAdversarial AttackLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose the ShadeEdit attack under the federated instruction tuning (FedIT) framework, which injects malicious knowledge into the global model by editing models with control over only a few clients, achieving controllable manipulation of the global model.
SHADOW: Dynamic-Aware Credit Assignment Against Long-Horizon Tasks
Yuze Liu (Zhejiang University), Chao Yang (Shanghai AI Laboratory)
Large Language ModelReinforcement LearningAgentic AITextSequential
🎯 What it does: Propose the SHADOW framework to improve credit assignment for LLM agents in long-sequence tasks.
ShaLa: Multimodal Shared Latent Generative Modelling
Jiali Cui (Stevens Institute of Technology), Matthew Klenk (Toyota Research Institute)
GenerationData SynthesisRepresentation LearningConvolutional Neural NetworkMixture of ExpertsDiffusion modelAuto EncoderMultimodality
🎯 What it does: Proposed a new multi-modal shared latent space generation framework ShaLa, combining structured reasoning models and a second-stage diffusion prior to achieve shared latent representation learning for multi-modal data and high-quality cross-modal generation.
ShapBPT: Image Feature Attributions Using Data-Aware Binary Partition Trees
Muhammad Rashid (University of Torino), Damiano Verda (Rulex Innovation Labs)
ClassificationObject DetectionAnomaly DetectionExplainability and InterpretabilityImage
🎯 What it does: Proposes ShapBPT, a data-aware hierarchical Shapley attribution method based on binary partition trees (BPT), for pixel-level image explanations.
Shaping Parameter Contribution Patterns for Out-of-Distribution Detection
Haonan Xu (Nanjing University of Science and Technology), Yang Yang (Nanjing University of Science and Technology)
ClassificationAnomaly DetectionImage
🎯 What it does: This paper proposes a regularization method called SPCP, which truncates the contribution of classifier parameters during training. By limiting the upper bound of parameter contributions, it forms a dense and bounded contribution pattern, thereby enhancing the model's robustness to abnormal inputs and reducing over-reliance on a few dominant parameters.
Shaping Without Tearing: Controllable Diffeomorphic Deformations for Topology-Preserving 3D Point Cloud Augmentation
Jian Bi (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
SegmentationData SynthesisFlow-based ModelPoint Cloud
🎯 What it does: Propose a point cloud augmentation framework called ManiPoint based on diffeomorphism, which can perform controllable geometric deformation without damaging the manifold structure.
Shapley Value Approximation Based on k-Additive Games
Guilherme Dean Pelegrina (Mackenzie Presbyterian University), Eyke Hüllermeier (LMU Munich)
Explainability and InterpretabilityComputational EfficiencyImageTabular
🎯 What it does: Proposes an approximation method for Shapley values called SVA k ADD based on k-additive games, aiming to address the complexity issues in Shapley value calculations.
Share Your Attention: Transformer Weight Sharing via Matrix-based Dictionary Learning
Magauiya Zhussip, Stamatios Lefkimmiatis (ITMO University)
CompressionComputational EfficiencyTransformerImageText
🎯 What it does: Decompose the Transformer attention projection matrix into shared matrix atoms to achieve cross-layer weight sharing, thereby compressing the attention module by approximately 66.7% of parameters while maintaining performance.
Shared & Domain Self-Adaptive Experts with Frequency-Aware Discrimination for Continual Test-Time Adaptation
Jianchao Zhao (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)
Domain AdaptationMixture of ExpertsImage
🎯 What it does: Propose a frequency-aware shared and adaptive expert framework to achieve continual domain adaptation and knowledge preservation for continual test-time adaptation (CTTA) tasks.
Sharp Eyes and Memory for VideoLLMs: Information-Aware Visual Token Pruning for Efficient and Reliable VideoLLM Reasoning
Jialong Qin (Hong Kong University of Science and Technology (Guangzhou)), Xuming Hu (Hong Kong University of Science and Technology (Guangzhou))
Computational EfficiencyTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: Propose a two-stage no-training adaptation pruning framework called SharpV, which dynamically prunes visual tokens using spatial-temporal importance and trims the KV cache during LLM decoding through an information degradation threshold, achieving efficient and reliable video LLM inference.
Sheaf Graph Neural Networks via PAC-Bayes Spectral Optimization
Yoonhyuk Choi (Sookmyung Women's University), Chong-Kwon Kim (Korea Institute of Energy Technology)
ClassificationGraph Neural NetworkGraph
🎯 What it does: Designed and implemented the SGPC (Sheaf Graph Neural Networks with PAC-Bayes Calibration) model, integrating cellular sheaf propagation, Wasserstein-Entropic lifting, Stochastic Variance-Reduced Diffusion (SVR), and Adaptive Frequency Mixing (AFM), and achieving end-to-end training through β-Dirichlet posterior calibration.
Shedding the Facades, Connecting the Domains: Detecting Shifting Multimodal Hate Video with Test-Time Adaptation
Jiao Li (University of Electronic Science and Technology of China), Fan Zhou (University of Electronic Science and Technology of China)
Domain AdaptationAnomaly DetectionVideoMultimodality
🎯 What it does: Proposed a test-time adaptation framework called SCANNER specifically designed for detecting gradually evolving hate videos;
SheetBrain: A Neuro-Symbolic Agent for Accurate Reasoning over Complex and Large Spreadsheets
Ziwei Wang (Carnegie Mellon University), Dongmei Zhang (Microsoft Research)
TransformerLarge Language ModelAgentic AITabularBenchmark
🎯 What it does: Proposes SheetBrain, a multi-stage (understanding–execution–verification) framework based on neuro-symbolic reasoning, to accurately answer and operate complex large-scale spreadsheets.
ShieldRAG: Safeguarding Retrieval-Augmented Generation from Untrusted Knowledge Bases
Peiru Yang (Tsinghua University), Tao Qi (Tsinghua University)
OptimizationSafty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: Design the ShieldRAG framework, enhancing the robustness of Retrieval-Augmented Generation (RAG) against untrustworthy or attacked knowledge bases through joint optimization of safety and relevance for the retrieval model.
ShoppingBench: A Real-World Intent-Grounded Shopping Benchmark for LLM-based Agents
Jiangyuan Wang (Alibaba International Digital Commercial Group), Xiaoyi Zeng (Alibaba International Digital Commercial Group)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AITextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed ShoppingBench, a real-intent-driven end-to-end benchmark for shopping scenarios, containing 3,310 user instructions and a sandbox environment with 25 million real products;
Shrinking the Teacher: An Adaptive Teaching Paradigm for Asymmetric EEG-Vision Alignment
Lukun Wu (Xidian University), Xinbo Gao (Xidian University)
RetrievalRepresentation LearningVision Language ModelContrastive LearningImageBiomedical Data
🎯 What it does: Propose the Adaptive Teaching Paradigm, utilizing a visual model as a teacher to dynamically shrink and adapt its knowledge to fit EEG students, achieving zero-shot EEG-to-Image retrieval;
SIAM: Towards Generalizable Articulated Object Modeling via Single Robot-Object Interaction
Yuyan Liu (Hefei University of Technology), Dan Guo (Hefei University of Technology)
SegmentationPose EstimationDepth EstimationOptimizationRobotic IntelligenceGaussian SplattingOptical FlowImagePoint CloudMesh
🎯 What it does: Propose the SIAM framework, which completes part segmentation, joint parameter estimation, and watertight mesh reconstruction through a single robot-object interaction, generating a URDF model directly applicable for simulation.
SIDE: Surrogate Conditional Data Extraction from Diffusion Models
Yunhao Chen (Fudan University), Xingjun Ma (Fudan University)
GenerationConvolutional Neural NetworkDiffusion modelImageStochastic Differential Equation
🎯 What it does: Propose the SIDE (Surrogate Conditional Data Extraction) framework, which constructs surrogate conditions using clustering information from generated images to enable training data extraction from any diffusion model (whether conditional or unconditional).
SIFThinker: Spatially-Aware Image Focus for Visual Reasoning
Zhangquan Chen (Tsinghua University), Ruqi Huang (Tsinghua University)
Depth EstimationLarge Language ModelReinforcement LearningVision Language ModelImageTextChain-of-Thought
🎯 What it does: Proposed and implemented the SIFThinker framework, which can dynamically focus on and correct visual attention regions during the reasoning process of multimodal large models, and integrate depth information into the reasoning chain.