AAAI 2026 Papers — Page 36
AAAI Conference on Artificial Intelligence · 4149 papers
SpikCommander: A High-performance Spiking Transformer with Multi-view Learning for Efficient Speech Command Recognition
Jiaqi Wang (Harbin Institute of Technology), Zhengyu Ma (Pengcheng Laboratory)
RecognitionSpiking Neural NetworkTransformerAudio
🎯 What it does: Proposed a novel speech command recognition model called SpikCommander based on Spiking Transformer
Spike Imaging Velocimetry: Dense Motion Estimation of Fluids Using Spike Streams
Yunzhong Zhang (Nanjing University), Xun Cao (Peking University)
Graph Neural NetworkSpiking Neural NetworkOptical FlowTime SeriesPhysics Related
🎯 What it does: This paper utilizes a high temporal resolution and high dynamic range spike camera to propose a Spike Imaging Velocimetry (SIV) network, achieving end-to-end estimation from spike flow to velocity fields.
Spike Stream Memory Transfer for Dynamic Scene Reconstruction
Yanchen Dong (Peking University), Tiejun Huang (University of Chinese Academy of Sciences)
RestorationConvolutional Neural NetworkOptical FlowVideoSequential
🎯 What it does: Proposed and implemented the Spike Stream Memory Transfer (SSMT) framework to reconstruct high-frame-rate, clear dynamic scene images from discrete pulse streams of neuromorphic spike cameras.
Spiking Heterogeneous Graph Attention Networks
Buqing Cao, Jianxun Liu (Hunan University of Science and Technology)
ClassificationComputational EfficiencyRepresentation LearningGraph Neural NetworkSpiking Neural NetworkGraph
🎯 What it does: Proposed SpikingHAN, which integrates spiking neural networks (SNN) with heterogeneous graph attention networks, achieving node classification through single-layer shared convolution, semantic-level attention, and SNN to generate 1-bit representations;
Spiking-Aided Neural Architecture for Efficient and Robust WiFi Sensing
Yisha Lu (Pengcheng Laboratory), Bowen Zhang (Shenzhen X-institute)
RecognitionComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkSpiking Neural NetworkTime SeriesBenchmark
🎯 What it does: Propose a hybrid ANN-SNN network called SWS-Net, combining the noise suppression and time integration capabilities of Spiking Neural Networks (SNNs) with the feature extraction advantages of traditional Artificial Neural Networks (ANNs) to achieve efficient and robust WiFi sensing;
Spikingformer: A Key Foundation Model for Spiking Neural Networks
Chenlin Zhou (Peking University), Yonghong Tian (Peking University)
ClassificationRecognitionComputational EfficiencySpiking Neural NetworkTransformerImageText
🎯 What it does: Proposed Spikingformer, a full-spike Transformer architecture integrating MS Residual and self-attention, addressing the non-spike computation issues in models like Spikformer;
SpikingIR: A Novel Converted Spiking Neural Network for Efficient Image Restoration
Yang Ouyang (Xiamen University), Yanyun Qu (Xiamen University)
RestorationSuper ResolutionSpiking Neural NetworkImage
🎯 What it does: Proposed a framework called SpikingIR based on ANN-to-SNN conversion for image restoration tasks.
SPIRAL: Symbolic LLM Planning via Grounded and Reflective Search
Yifan Zhang (Vanderbilt University), Achille Fokoue (IBM T.J. Watson Research Center)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose a framework named SPIRAL that integrates three types of LLM agents (Planner, Simulator, Critic) into the MCTS loop, achieving symbolic planning through 'rooted reflective search';
SPJFNet: Self-Mining Prior-Guided Joint Frequency Enhancement for Ultra-Efficient Dark Image Restoration
Tongshun Zhang (Jilin University), Qiuzhan Zhou (Jilin University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: Proposed an efficient network called SPJFNet for dark image recovery, which significantly reduces the number of parameters and computational cost while maintaining or improving visual quality.
Splat-SAP: Feed-Forward Gaussian Splatting for Human-Centered Scene with Scale-Aware Point Map Reconstruction
Boyao Zhou (Ant Group), Yebin Liu (Tsinghua University)
GenerationData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkTransformerGaussian SplattingImageVideoPoint Cloud
🎯 What it does: Proposed an end-to-end Splat-SAP framework that can first self-supervisedly recover scale-aware point maps (metric geometry) from sparse dual-camera inputs with only two views, and then achieve high-quality free-viewpoint video synthesis for human-centric scenes through refinement and Gaussian-plane-based rendering.
Splats in Splats: Robust and Effective 3D Steganography Towards Gaussian Splatting
Yijia Guo (Peking University), Lei Ma (Peking University)
Safty and PrivacyConvolutional Neural NetworkAuto EncoderGaussian SplattingPoint Cloud
🎯 What it does: Proposes 'Splats in Splats,' a steganography framework that can embed 3D content into 3D Gaussian Splatting (3DGS) without altering the original attributes.
SplatSSC: Decoupled Depth-Guided Gaussian Splatting for Semantic Scene Completion
Rui Qian (Nanyang Technological University), Lihua Xie (Nanyang Technological University)
SegmentationDepth EstimationTransformerGaussian SplattingImage
🎯 What it does: This study proposes the SplatSSC framework for 3D semantic scene completion from monocular images, addressing the issues of redundancy and 'floating bodies' caused by traditional random initialization of Gaussian primitives;
Split-Layer: Enhancing Implicit Neural Representation by Maximizing the Dimensionality of Feature Space
Zhicheng Cai (Nanjing University), Xun Cao (Nanjing University)
Representation LearningNeural Radiance FieldImageMeshComputed Tomography
🎯 What it does: Proposes the Split-Layer mechanism, which splits each fully connected layer into multiple parallel branches and aggregates them through Hadamard product to construct a high-order polynomial feature space, thereby significantly enhancing the representational power of implicit neural representations (INR).
Spontaneous Yet Predictable: Shapelet-Driven, Channel-Aware Intention Decoding from Multi-Region ECoG
Keren Cao (Xi'an Jiaotong University), Liangjun Chen (Xi'an Jiaotong University)
ClassificationExplainability and InterpretabilityTransformerContrastive LearningBiomedical Data
🎯 What it does: This study proposes a shapelet-driven, channel-aware dual-region ECoG intent decoding framework that can accurately predict spontaneous vocalizations of marmosets 200 ms in advance.
SPP-SCL: Semi-Push-Pull Supervised Contrastive Learning for Image-Text Sentiment Analysis and Beyond
Jiesheng Wu, Shengrong Li (Anhui Normal University)
ClassificationConvolutional Neural NetworkTransformerContrastive LearningMultimodality
🎯 What it does: This paper proposes a semi-pull supervised contrastive learning (SPP-SCL) framework, which first performs contrastive alignment of same-sentiment category samples within the image and text modalities individually, then conditionally pulls cross-modal sentiment representations closer based on similarity threshold conditions, thereby achieving intra-modal and cross-modal consistency in the sentiment embedding space before fusion; meanwhile, hierarchical attention (HA) and cross-modal fusion (CMF) modules are combined to enhance feature expression and fusion effects.
SPSC: Sparse and Scalable Multi-Modal 3D Occupancy Prediction for Autonomous Driving
Qingju Guo (Beijing Institute Of Technology), Wei Li (Nanjing University)
Autonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: Propose the SPSC framework, achieving multi-modal high-resolution 3D semantic occupancy prediction by leveraging sparse occupancy queries and query serialization;
SR-KI: Scalable and Real-Time Knowledge Integration into LLMs via Supervised Attention
Bohan Yu (University of Chinese Academy of Sciences), Kang Liu (University of Chinese Academy of Sciences)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelText
🎯 What it does: Propose SR-KI, a method for scalable, real-time knowledge injection into large language models through supervised attention;
SRACG: A Code Generation Framework with Selective Retrieval Augmentation
Mengzhen Wang (South China University of Technology), Yi Cai (South China University of Technology)
GenerationRetrievalAI Code AssistantTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Propose the SRACG framework to address the retrieval enhancement issue in LLM code generation, adopting necessity-aware retrieval selection and multi-dimensional enhancement strategies;
SRD: Reinforcement-Learned Semantic Perturbation for Backdoor Defense in VLMs
Shuhan Xu (Wuhan University), Dacheng Tao (AGH University of Krakow)
Adversarial AttackTransformerReinforcement LearningVision Language ModelMultimodality
🎯 What it does: To defend against backdoor attacks on vision-language models, the authors propose a semantic reward defense framework based on reinforcement learning, which generates red mask perturbations on input images to disrupt the model's attention distribution, thereby suppressing trigger activation and reducing attack success rate.
SRSplat: Feed-Forward Super-Resolution Gaussian Splatting from Sparse Multi-View Images
Xinyuan Hu (Hangzhou Dianzi University), Min Tan (Li Auto Inc.)
Super ResolutionConvolutional Neural NetworkTransformerLarge Language ModelDiffusion modelGaussian SplattingImageMultimodality
🎯 What it does: Proposes SRSplat, a forward network capable of performing real-time high-resolution 3D scene reconstruction using only a few low-resolution view images.
SSCL: Adversarially Guided Image Compression via Semantic and Spectral Consistency Learning
Wei Jiang (Peking University), Ronggang Wang (Alibaba Cloud Computing)
CompressionTransformerGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the SSCL (Semantic and Spectral Consistency Learning) strategy, which enhances perceptual quality in learned image compression by incorporating semantic consistency and spectral consistency discriminators.
SSHPool: The Separated Subgraph-based Hierarchical Pooling
Zhuo Xu, Edwin R. Hancock (University Of York)
ClassificationGraph Neural NetworkGraph
🎯 What it does: Proposed a hierarchical pooling method called SSHPool based on separated subgraphs for graph classification, and constructed a complete end-to-end GNN framework.
SSL-CST: Cell Segmentation for Single-Cell Spatial Transcriptome Based on Self-Supervised Learning
Weiliang Huo (Hainan University), Qingchen Zhang (Hainan University)
SegmentationTransformerMultimodalityBiomedical Data
🎯 What it does: Proposed a self-supervised learning-based method for single-cell spatial transcriptomics cell segmentation called SSL-CST, which accurately segments cells by leveraging nuclear staining images and gene expression information;
SSR-SAM: Retrieval-Style Segment Anything Model for Semi-Supervised Ultra-High-Resolution Image Segmentation
Shijie Li (Fudan University), Xieping Gao (Hunan Normal University)
SegmentationRetrievalTransformerPrompt EngineeringImageRetrieval-Augmented Generation
🎯 What it does: Develop a retrieval-based semi-supervised Ultra-High-Resolution (UHR) image segmentation framework called SSR-SAM based on the Segment Anything Model (SAM), which generates visual semantic prompts using locally annotated regions and retrieves similar pixels across the entire image, further achieving consistency regularization through prompt layer perturbation;
SSR: Semantic and Spatial Rectification for CLIP-based Weakly Supervised Segmentation
Xiuli Bi (Chongqing University of Posts and Telecommunications), Bin Xiao (Jinan Inspur Data Technology Co Ltd)
SegmentationTransformerContrastive LearningImageMultimodality
🎯 What it does: Propose an SSR method combining semantic and spatial correction for weakly supervised semantic segmentation based on CLIP, addressing issues of over-activation in non-target foreground and background regions.
ST-LLM: Spatial Transcriptomics Embedding with Large Language Models
Zhetao Xu (Beijing Institute of Technology), Bin Hu (Beijing Institute of Technology)
Representation LearningGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningPrompt EngineeringGraphBiomedical Data
🎯 What it does: The ST-LLM framework converts spatial transcriptomics graph structures into structured text, utilizing large language models to generate spatially aware embeddings for clustering and region detection.
ST-SAM: Multimodal Scene Text Segmentation with Dense Visual and Sparse Textual Prompts via SAM
Jin Wei, QianYing Wang (Xi'an Jiaotong University)
SegmentationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelMultimodality
🎯 What it does: Propose ST-SAM, a multimodal method that integrates dense visual prompts and sparse text prompts, adapting SAM into a scene text segmentation model.
ST-TPP: Learning Semi-Transductive Temporal Point Processes with Gromov-Wasserstein Barycentric Regularization
Qingmei Wang (Renmin University of China), Hongteng Xu (Renmin University of China)
Representation LearningTime SeriesSequential
🎯 What it does: Proposed a semi-supervised temporal point process model called ST-TPP, which improves event prediction by leveraging sequence clustering information.
ST-VLM: A Spatial-to-Image Multimodal Spatial-Temporal Prediction Framework with Vision-Language Model
Tong Zhao (Beijing University of Posts and Telecommunications), Dandan Liu (Beijing University of Posts and Telecommunications)
TransformerSupervised Fine-TuningVision Language ModelImageTime Series
🎯 What it does: The ST-VLM framework utilizes multimodal vision-language models for spatiotemporal prediction;
Stability-Aware Reinforcement Learning for Robust Class Integration Test Order Generation
Yanru Ding (China University of Mining and Technology), Luciano Baresi (China University of Mining and Technology)
Reinforcement LearningText
🎯 What it does: Propose the LM-CITO framework, using Lyapunov-guided reward shaping and modified testing (MT) to generate robust class integration test sequences
Stabilizing Cross-Modal Bidirectional Attribution: Few-Shot Adversarial Prompt Tuning for Robust Vision-Language Models
Jun Feng (Huazhong University of Science and Technology), Shunli Zhang (Anhui University of Science and Technology)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningAdversarial AttackTransformerPrompt EngineeringGenerative Adversarial NetworkContrastive LearningMultimodality
🎯 What it does: This paper proposes a few-shot adversarial prompt tuning framework called CBA-FAPT based on cross-modal bidirectional attribution graphs, aiming to enhance the robustness of vision-language models by stabilizing the internal reasoning process of the model.
Stabilizing Policy Gradient Methods via Reward Profiling
Shihab Ahmed (University of Central Florida), Aritra Dutta (University of Central Florida)
Robotic IntelligenceReinforcement Learning
🎯 What it does: Proposed and implemented a general reward profiling framework to selectively update policies in policy gradient (PG) methods based on high-confidence performance evaluation, thereby reducing gradient estimation variance, improving convergence speed, and enhancing learning stability.
Stabilizing Self-Consuming Diffusion Models with Latent Space Filtering
Zhongteng Cai (Ohio State University), Xueru Zhang (Ohio State University)
GenerationConvolutional Neural NetworkSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Proposes a self-supervised diffusion model training method based on Latent Space Filtering (LSF), which utilizes the low-dimensional structural degradation in the latent space to filter out unrealistic synthetic samples, thereby suppressing model collapse and maintaining generation quality.
Stabilizing Spiking Neurons Through Biologically Inspired Polarization
Matthew Lai (University of Technology Sydney), Longbing Cao (Macquarie University)
ClassificationSpiking Neural NetworkImage
🎯 What it does: Proposed the POLARA neuron model, achieving gradient propagation stability through a biology-inspired polarization phase;
Stable and Adaptive Fusion for Multi-domain Multi-task Recommendation
Ke Fei (Tencent), Jingjing Li (Tencent)
Recommendation SystemMixture of ExpertsTabular
🎯 What it does: Designed and implemented the Stable and Adaptive Fusion (SAF) framework to address the negative transfer problem in multi-domain multi-task recommendation.
Stable Voting and the Splitting of Cycles
Wesley H. Holliday (University of California, Berkeley), Cynthia Wang (Carnegie Mellon University)
🎯 What it does: Investigated the relationship between Simple Stable Voting (SSV) and Splitting Cycle (SC) methods, proving that when the number of candidates does not exceed 6, the winner of SSV is necessarily the winner of SC, and provided a minimal counterexample for 7 or more candidates.
Stage-Aware Graph Contrastive Learning with Node-oriented Mixture of Experts
Xiangkai Zhu, Longsheng Su (Shandong University Of Science And Technology)
Representation LearningGraph Neural NetworkLarge Language ModelMixture of ExpertsContrastive LearningGraph
🎯 What it does: Propose Stage-Aware Graph Contrastive Learning (SAGCL), which combines multilingual models through Node-oriented Mixture of Experts (NodeMoE), and uses self-supervised contrastive learning to align LLM embeddings with graph structures during the feature transformation and propagation stages of GNN.
STaR: Sensitive Trajectory Regulation for Unlearning in Large Reasoning Models
Jingjing Zhou (University of Chinese Academy of Sciences), Liang Li (University of Chinese Academy of Sciences)
Safty and PrivacyLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Proposed a parameter-agnostic framework STaR for privacy-delearning in large reasoning models during inference, capable of dynamically suppressing sensitive information throughout the entire chain-of-thought reasoning process.
Start Small, Think Big: Curriculum-based Relative Policy Optimization for Visual Grounding
Qingyang Yan (Huazhong University of Science and Technology), Yixiong Zou (Huazhong University of Science and Technology)
Object DetectionReinforcement LearningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: For the visual localization task, the impact of Chain-of-Thought (CoT) generation on model performance is studied, and a curriculum learning strategy named CuRPO based on CoT length and reward is proposed. Reinforcement learning (GRPO) is used to gradually introduce data with increasing difficulty, improving localization accuracy.
State Mamba: Spatiotemporal EEG State-Space Model with Dynamic Brain Alignment for Cross-Subject Representation
Weining Weng (Institute of Computing Technology, Chinese Academy of Sciences), Yiqiang Chen (Institute of Computing Technology, Chinese Academy of Sciences)
ClassificationRepresentation LearningBiomedical Data
🎯 What it does: Proposed a new spatiotemporal EEG state space model called State Mamba, aiming to learn cross-subject consistent EEG representations by modeling neural responses and their spatiotemporal state transitions.
State Proficiency-Based Adaptive Fine-Tuning for Offline-to-Online Reinforcement Learning
Songlin Li (Jilin University), Shuai Lü
Supervised Fine-TuningReinforcement LearningBenchmark
🎯 What it does: This paper proposes an adaptive fine-tuning method called SPA based on state proficiency, aimed at performing online fine-tuning from offline-trained policies, with the goal of significantly improving sample efficiency and final performance while maintaining training stability.
State-Derivative-Aware Neural Controlled Differential Equations for Multivariate Time Series Anomaly Detection and Diagnosis
Xin Sun (Zhejiang University), Chao Li (Zhejiang University)
Anomaly DetectionComputational EfficiencyTransformerTime SeriesStochastic Differential Equation
🎯 What it does: This paper proposes a multivariate time series anomaly detection framework named SDA-D that simultaneously leverages time-point reconstruction error and system state derivatives.
State-Space Hierarchical Compression with Gated Attention and Learnable Sampling for Hour-Long Video Understanding in Large Multimodal Models
Geewook Kim (NAVER Cloud AI), Minjoon Seo (KAIST AI)
CompressionLarge Language ModelVideoMultimodalityBenchmark
🎯 What it does: Propose a multi-level compression framework named MambaMia, which compresses a large number of frame features from hour-level long videos before inputting them into large-scale multimodal models, thereby alleviating the token explosion problem.
Stationary and Clustering Transformer Hashing for Cross-modal Retrieval
Zhan Yang (Central South University), Yinan Li (Central South University)
RetrievalTransformerContrastive LearningMultimodality
🎯 What it does: Propose an unsupervised cross-modal hashing method called SCTH, combining Transformer fusion, soft clustering, and Markov steady-state distribution to enhance cross-modal retrieval performance.
Statistical Learning Theory for Distributional Classification
Christian Fiedler (Technical University of Munich)
Classification
🎯 What it does: Analyzed the support vector machine classification method for distributed inputs under two-stage sampling, and provided new generalization error upper bounds, convergence, and learning rates
Statistically Robust Sparse High-order Interaction Model
Diptesh Das (University of Tokyo), Koji Tsuda (University of Tokyo)
OptimizationTabularTime SeriesBiomedical Data
🎯 What it does: Developed Huberized-SHIM, which integrates Huber loss robust regression, ℓ1+ℓ2 regularization, and an exact homotopy regularization path algorithm into the sparse high-order interaction model SHIM, combined with tree-based pruning and synthetic prediction framework;
Steering One-Step Diffusion Model with Fidelity-Rich Decoder for Fast Image Compression
Zheng Chen (Shanghai Jiao Tong University), Yulun Zhang (Shanghai Jiao Tong University)
CompressionDiffusion modelAuto EncoderImage
🎯 What it does: Designed and implemented SODEC, a single-step diffusion image compression model that combines high-information potential codes generated by VAE with one-time diffusion decoding, along with a fidelity guidance module and rate annealing training.
Steering Pretrained Drafters During Speculative Decoding
Frédéric Berdoz (ETH Zurich), Roger Wattenhofer (ETH Zurich)
GenerationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This study proposes a lightweight dynamic steering mechanism that injects a steering vector into the pre-trained drafter by leveraging the hidden representations of the verifier, significantly improving the token acceptance rate and overall inference throughput during speculative decoding.
Steering Representations, Safeguarding Privacy: A Cross-Modal Privacy Protection Method for Generative AI
Jie Zhang, Jinta Weng (HiThink Research)
Safty and PrivacyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelTextMultimodality
🎯 What it does: Propose a 'Know-Then-Do' framework that constructs privacy-enhancing Steering Vectors in the representation space, injecting these vectors during inference to enhance privacy awareness in large language models and multimodal models without requiring additional training.
Steering Visuomotor Policy in Open Worlds via Cross-View Goal Alignment
Shaofei Cai (Peking University), Yitao Liang (Peking University)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerReinforcement LearningVision-Language-Action ModelImageVideoBenchmark
🎯 What it does: Propose a cross-view goal alignment framework that allows humans to specify goals using segmentation masks from their own camera views, training visual-motor policies that can complete tasks based on their own observations, significantly improving human-robot interaction efficiency.
SteerMusic: Enhanced Musical Consistency for Zero-shot Text-Guided and Personalized Music Editing
Xinlei Niu (Australian National University), Yuki Mitsufuji (Sony AI)
GenerationDiffusion modelScore-based ModelContrastive LearningBenchmarkAudio
🎯 What it does: Proposes two zero-shot music editing methods, SteerMusic and SteerMusic+, which directly edit raw music in the data space using score distillation and delta denoising score.
StegaVAR: Privacy-Preserving Video Action Recognition via Steganographic Domain Analysis
Lixin Chen (Great Bay University), Xun Lin (Westlake University)
RecognitionSafty and PrivacyVideo
🎯 What it does: Propose the StegaVAR framework, which embeds action videos into regular cover videos and performs action recognition directly in the steganography domain;
Step Back to Leap Forward: Self-Backtracking for Symbolic Reasoning and Planning in Language Models
Xiao-Wen Yang (Nanjing University), Yu-Feng Li (Nanjing University)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningTextSequential
🎯 What it does: Propose a self-backtracking mechanism that enables large language models to automatically determine when to backtrack and improve search paths in reasoning and planning tasks.
Step-by-step Layered Design Generation
Faizan Farooq Khan, Balaji Vasan Srinivasan (King Abdullah University of Science and Technology)
SegmentationGenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelMultimodalityBenchmark
🎯 What it does: This study proposes a system called SLEDGE that can update graphic designs based on step-by-step instructions, and constructs the IDeation dataset and benchmark for training and evaluation.
Step-GRPO: Enhancing Reasoning Quality and Efficiency via Structured PRM-Based Reinforcement Learning
Weijie Li (Yunnan University), Xuejie Zhang (Yunnan University)
Computational EfficiencyReinforcement LearningText
🎯 What it does: Proposed a method called Step-GRPO, which improves inference quality and efficiency by integrating step-level process reward model (PRM) signals into sparse trajectory-level feedback.
STEP-Nav: Spatial-Temporal Efficient Visual Token Pruning for Vision-and-Language Navigation with Large Language Models
Yantao Lu, Chenglie Du (Didi Chuxing)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelImageVideoTextMultimodalityBenchmark
🎯 What it does: Propose the STEP-Nav framework, which simultaneously prunes visual tokens at the spatial and temporal levels to reduce computational costs in LLMs for visual navigation tasks;
StepFun-Formalizer: Unlocking the Autoformalization Potential of LLMs Through Knowledge-Reasoning Fusion
Yutong Wu (Institute of Computing Technology, Chinese Academy of Sciences), Xing Hu (Institute of Computing Technology, Chinese Academy of Sciences)
Knowledge DistillationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
🎯 What it does: This paper proposes a data synthesis and training pipeline named ThinkingF to improve the accuracy of large models in automatic formalization of natural language mathematical expressions (autoformalization).
Stepwise Contrastive Reasoning for Retrieval-Augmented Generation over Knowledge Graphs
Chenxiao Lin, Qingqiang Wu (Xiamen University)
Explainability and InterpretabilityComputational EfficiencyGraph Neural NetworkTransformerContrastive LearningGraphRetrieval-Augmented Generation
🎯 What it does: This paper proposes a lightweight retrieval-augmented generation framework called Stepwise Contrastive Reasoning (SCR), which achieves interpretable reasoning and knowledge retrieval on knowledge graphs by progressively aligning the semantic embeddings of questions and graph entities;
STMI: Segmentation-Guided Token Modulation with Cross-Modal Hypergraph Interaction for Multi-Modal Object Re-Identification
Xingguo Xu (Dalian University of Technology), Dell Zhang (China Telecom)
RecognitionGraph Neural NetworkTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: Proposed the STMI framework, combining segmentation-guided feature modulation, semantic token reallocation, and cross-modal hypergraph interaction to achieve multimodal person re-identification.
Stochastic Decentralized Optimization of Non-Smooth Convex and Convex-Concave Problems over Time-Varying Networks
Maxim Divilkovskiy (Moscow Institute of Physics and Technology), Alexander Gasnikov (Innopolis University)
OptimizationGraph
🎯 What it does: Proposed a nonsmooth stochastic distributed optimization algorithm applicable to time-varying networks, covering convex minimization and convex-concave saddle point problems.
Stochastic Universal Adversarial Perturbations with Fixed Optimization Constraint and Ensured High-probability Transferability
Yulin Jin (Hong Kong Polytechnic University), Xiaofeng Chen (Xidian University)
Adversarial AttackImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper theoretically explains the transferability of universal adversarial perturbations (UAP), proposes and analyzes the relationship between transferability gap and algorithm stability, and designs an Expected Constraint and Noisy Stochastic Universal Adversarial Perturbation (SUAP) algorithm based on this.
STOLA: Self-Adaptive Touch-Language Framework for Tactile Commonsense Reasoning in Open-Ended Scenarios
Ning Cheng (Beijing Jiaotong University), Wenjuan Han (Beijing University of Posts and Telecommunications)
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsMultimodalityPhysics Related
🎯 What it does: Proposes STOLA—an adaptive tactile-language framework that utilizes Mixture of Experts (MoE) and two-stage training to achieve open-ended tactile common-sense reasoning.
Stop Mixing Things Up! BISCUIT Teaches Vision-Language Models to Learn New Concepts from Images on the Spot
Jiahua Bao (Harbin Institute of Technology), Jie Liu (Harbin Institute of Technology)
TransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: BISCUIT proposes a two-step training method that enables vision-language models to instantly learn and utilize new concepts appearing in images during inference, without relying on text injection or vocabulary expansion.
StoryBox: Collaborative Multi-Agent Simulation for Hybrid Bottom-Up Long-Form Story Generation Using Large Language Models
Zehao Chen (Sun Yat-sen University), Haoran Li (Sun Yat-sen University)
GenerationTransformerLarge Language ModelAgentic AITextRetrieval-Augmented Generation
🎯 What it does: This paper proposes the StoryBox framework, which generates events in dynamic sandbox environments through multi-agent simulation and transforms these events into coherent, long-form stories via the Storyteller Agent, achieving narrative lengths exceeding 10,000 words.
STrans: Spontaneous Architecture Evolution for Adaptive Time Series Forecasting
Haoyi Jia (University of New South Wales)
Neural Architecture SearchTransformerTime Series
🎯 What it does: Propose the STrans framework, which automatically discovers the optimal attention, normalization, encoding, and activation combination for time series Transformers through differentiable architecture search, achieving an end-to-end adaptive prediction model.
Strategic Reasoning over Golog Programs in the Nondeterministic Situation Calculus
Giuseppe De Giacomo (University of Oxford), Matteo Mancanelli (York University)
🎯 What it does: This paper proposes a strategy synthesis framework for non-deterministic Golog programs, which constructs symbolic program graphs and cross-products with domain models, employs game theory for strategy reasoning, and provides a reduction-based minimal closure iteration algorithm.
Stratified Knowledge-Density Super-Network for Scalable Vision Transformers
Longhua Li (Southeast University), Xin Geng (Southeast University)
ClassificationComputational EfficiencyTransformerImageBenchmark
🎯 What it does: Convert a pre-trained ViT into a hierarchical knowledge density (SKD) super network, enabling zero-cost extraction of subnetworks of arbitrary scale;
Streaming Generated Gaussian Process Experts for Online Learning and Control
Zewen Yang (Technical University Of Munich), Sami Haddadin (Technical University Of Munich)
Computational EfficiencyMixture of ExpertsTabular
🎯 What it does: Proposed the SkyGP framework, which can online generate and maintain a finite number of Gaussian Process experts in a streaming data environment, achieving efficient online learning and control;
Streaming Generation of Co-Speech Gestures via Accelerated Rolling Diffusion
Evgeniia Vu (Constructor University), Dmitry Vetrov (Constructor University)
GenerationDiffusion modelMultimodality
🎯 What it does: This study proposes a framework for real-time generation of co-speech gestures, achieving continuous long-sequence synthesis using accelerated rolling diffusion technology.
StreamingTalker: Audio-driven 3D Facial Animation with Autoregressive Diffusion Model
Yifan Yang (Zhejiang University), Hujun Bao (Ant Group)
GenerationTransformerDiffusion modelAuto EncoderMeshAudio
🎯 What it does: Designed and implemented a real-time voice-driven 3D facial animation method based on autoregressive diffusion models, capable of streaming high-quality facial mesh generation for audio of arbitrary length.
StreamKV: Streaming Video Question-Answering with Segment-based KV Cache Retrieval and Compression
Yilong Chen (Peking University), Ming Lu (Peking University)
RetrievalCompressionTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideo
🎯 What it does: This paper proposes StreamKV, a no-training framework that provides dynamic semantic segmentation, KV cache retrieval, and compression for video large language models, thereby achieving efficient streaming video question answering.
StreamSTGS: Streaming Spatial and Temporal Gaussian Grids for Real-Time Free-Viewpoint Video
Zhihui Ke (Tianjin University), Tie Qiu (Tianjin University)
CompressionTransformerGaussian SplattingVideo
🎯 What it does: A dynamic 3D Gaussian representation based on StreamSTGS (Stream-based Spatiotemporal Gaussian Grids) is proposed for real-time free-viewpoint video streams.
STRIDE-QA: Visual Question Answering Dataset for Spatiotemporal Reasoning in Urban Driving Scenes
Keishi Ishihara (Turing Inc), Yu Yamaguchi (Turing Inc)
Autonomous DrivingSupervised Fine-TuningPrompt EngineeringVision Language ModelImageMultimodalityPoint CloudTime SeriesBenchmark
🎯 What it does: Explored the spatiotemporal reasoning visual question answering dataset STRIDE-QA for urban driving scenarios.
Strip R-CNN: Large Strip Convolution for Remote Sensing Object Detection
Xinbin Yuan (Nankai University), Ming-Ming Cheng (Nankai University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Propose Strip R-CNN, a framework that utilizes large-scale strip convolution to enhance high aspect ratio object detection in remote sensing images.
StrokeFusion: Vector Sketch Generation via Joint Stroke-UDF Encoding and Latent Sequence Diffusion
Jin Zhou (Shenzhen University), Hui Huang (Shenzhen University)
GenerationConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: Proposed a two-stage vector sketch generation framework called StrokeFusion, which first uses dual-modal stroke-UDF encoding to jointly represent stroke geometry and unsigned distance fields, then generates strokes through an unordered, variable-length latent space diffusion model, enabling synchronized learning of position, scale, and trajectory while achieving stroke interpolation and editing.
Structural Approach to Guiding a Present-Biased Agent
Tatiana Belova (ITMO University), Danil Sagunov (ITMO University)
OptimizationGraph
🎯 What it does: Studied and addressed the T-PATH-EDITING problem, providing multiple parameterized algorithms along with corresponding hard complexity bounds.
Structural Entropy Guided Incremental Learning for Open-World Multimodal Social Event Detection
Zhiwei Yang (Chinese Academy of Sciences), Zhiqin Yang (Beijing Institute of Technology)
RecognitionKnowledge DistillationTransformerLarge Language ModelContrastive LearningMultimodality
🎯 What it does: This paper proposes the SeInEvent framework, achieving unsupervised, incremental multi-modal social event detection, capable of continuous learning and automatic denoising in open-world scenarios.
Structure Detection for Contextual Reinforcement Learning
Tianyue Zhou (Massachusetts Institute of Technology), Cathy Wu (Massachusetts Institute of Technology)
Reinforcement LearningBenchmark
🎯 What it does: Studied the SD-MBTL framework for multi-strategy transfer learning based on structural detection in multi-dimensional context reinforcement learning, and proposed M/GP-MBTL to automatically identify and leverage the MOUNTAIN structure to select source tasks, significantly improving transfer efficiency.
Structure-Aware Encodings of Argumentation Properties for Clique-width
Yasir Mahmood (Paderborn University), Johannes K. Fichte (Linkoping University)
Computational EfficiencyGraph
🎯 What it does: This paper designs a structure-aware encoding based on k-expressions (DDG reduction), compressing various semantics in the abstract argumentation framework (such as stable, acceptable, complete, preferred, semi-stable, phased, etc.) into (Q)SAT problems, and proves that the encoding is linearly effective in preserving clique-width.
Structure-based RNA Design by Step-wise Optimization of Latent Diffusion Model
Qi Si (Shanghai Academy of Artificial Intelligence for Science), Yuan Cheng (Shanghai Academy of Artificial Intelligence for Science)
GenerationOptimizationGraph Neural NetworkTransformerSupervised Fine-TuningReinforcement LearningDiffusion modelGraphBiomedical Data
🎯 What it does: Proposed a framework named SOLD that combines implicit diffusion models with reinforcement learning for RNA inverse folding, directly optimizing 2D and 3D structural metrics through one-step sampling and segmented rewards, thereby improving the accuracy of RNA sequence design.
Structure-Enhanced Adapter for Self-Supervised Heterogeneous Graph Learning
Fengyu Yan (Tianjin University), Dongxiao He (Tianjin University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose a plug-and-play structure-enhancing adapter named HSADP for self-supervised heterogeneous graph learning, which utilizes a dual-path mechanism (homogeneous subgraph enhancement and heterogeneous edge discriminator) to simultaneously address edge missing and redundancy issues in graphs;
Structures Meet Semantics: Multimodal Fusion via Graph Contrastive Learning
Jiangfeng Sun (Beijing University of Posts and Telecommunications), Meina Song (Beijing University of Posts and Telecommunications)
ClassificationGraph Neural NetworkContrastive LearningMultimodality
🎯 What it does: Propose a structure-semantic unified multimodal sentiment analysis framework (SSU), achieving cross-modal feature fusion by constructing modality-specific graphs, introducing semantic anchors, and employing multi-perspective contrastive learning.
Studying Classifier(-Free) Guidance from a Classifier-Centric Perspective
Xiaoming Zhao (University of Illinois Urbana Champaign), Alex Schwing (University of Illinois Urbana Champaign)
GenerationData SynthesisExplainability and InterpretabilityDiffusion modelScore-based ModelFlow-based ModelRectified FlowImage
🎯 What it does: Systematically studied and explained the behavior of classifier-free guidance in diffusion models from the perspective of classifiers, and proposed a post-processing step based on flow matching to verify this theory.
Style4D-Bench: A Benchmark Suite for 4D Stylization
Beiqi Chen (Harbin Institute of Technology), Guangcong Wang (Great Bay University)
GenerationContrastive LearningGaussian SplattingVideoBenchmark
🎯 What it does: Established Style4D-Bench, providing a unified dataset, evaluation protocol, and benchmark model for 4D stylization, and proposed the Style4D framework to achieve high-quality, multi-view, temporally consistent 4D stylization.
StyleDrive: Towards Driving-Style Aware Benchmarking of End-To-End Autonomous Driving
Ruiyang Hao (Tsinghua University), Zaiqing Nie (Tsinghua University)
Autonomous DrivingSupervised Fine-TuningDiffusion modelVideoMultimodalityBenchmark
🎯 What it does: This paper introduces the StyleDrive dataset and the corresponding benchmark for studying and evaluating personalized end-to-end autonomous driving models.
StyleFM: Frequency Manipulation Empowered by Recursive Attention on Diffusion Models for Arbitrary Style Transfer
Yingnan Ma (University of Alberta), Anup Basu (University of Alberta)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: Proposed a training-free diffusion-based arbitrary style transfer method called StyleFM, combining three-frequency operations in the frequency domain and recursive attention to achieve high-quality content preservation and style embedding.
StyleProto: Style-Augmented Prototype Learning for Cross-Domain Few-Shot Object Detection
Xi Yang (Xidian University), Quantao Xie (Xidian University)
Object DetectionDomain AdaptationTransformerImage
🎯 What it does: Propose the StyleProto framework, which enhances cross-domain few-shot object detection performance by generating diverse styles on the support set, constructing semantically focused prototypes, and performing hierarchical fusion.
StyleSentinel: Reliable Artistic Copyright Verification via Stylistic Fingerprints
Lingxiao Chen (Sun Yat-sen University), Wei Lu (Sun Yat-sen University)
RecognitionAnomaly DetectionConvolutional Neural NetworkVision Language ModelDiffusion modelImageText
🎯 What it does: Propose StyleSentinel, an art copyright verification method based on inherent style fingerprints of works, capable of identifying unauthorized copies without preprocessing;
StyleTailor: Towards Personalized Fashion Styling via Hierarchical Negative Feedback
Hongbo Ma (Tsinghua University), Ming Li (Hangzhou Dianzi University)
GenerationRetrievalRecommendation SystemVision Language ModelImageTextMultimodality
🎯 What it does: Propose the StyleTailor framework to realize a closed-loop system for personalized clothing design, shopping recommendations, virtual try-on, and evaluation.
Sub-MoE: Efficient Mixture-of-Expert LLMs Compression via Subspace Expert Merging
Lujun Li (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
Computational EfficiencyLarge Language ModelMixture of ExpertsText
🎯 What it does: Propose the Sub-MoE framework, achieving compression of MoE LLMs through subspace expert merging;
SubGCache: Accelerating Graph-based RAG with Subgraph-level KV Cache
Qiuyu Zhu (Nanyang Technological University), Jie Zhang (Nanyang Technological University)
RetrievalComputational EfficiencyGraph Neural NetworkLarge Language ModelGraphRetrieval-Augmented Generation
🎯 What it does: Designed and implemented SubGCache, a graph structure retrieval-augmented generation framework based on subgraph-level KV caching, significantly reducing inference latency in batch query scenarios.
Subgraph Encoding with Bicentric Sphere Node Labeling and Pooling for Link Prediction
Zhihong Fang, Qing Gao (Hunan University)
Graph Neural NetworkGraph
🎯 What it does: Proposed a new subgraph encoding architecture called BS-SubGNN based on double-heart spheres for link prediction tasks, aiming to effectively distinguish and aggregate the contributions of nodes in subgraphs.
Subspace-Aware Graph Construction and Contrastive Alignment for Multimodal Recommendation with Large Language Models
Haodong Li (China University of Petroleum (East China)), Xiaokang Zhou (Peking University)
Recommendation SystemGraph Neural NetworkLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a multi-modal recommendation framework named SCALE, which generates user and item descriptions using a large language model, constructs a subspace-aware semantic graph, and fuses collaborative and semantic information through graph convolution and contrastive loss to achieve high-quality personalized recommendations.
SUGAR: Learning Skeleton Representation with Visual-Motion Knowledge for Action Recognition
Qilang Ye (Nankai University), Zitong Yu (Shenzhen University)
RecognitionPose EstimationGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningVideoTextMultimodalityGraph
🎯 What it does: Propose the SUGAR framework, which supervises skeleton representation learning through visual and motion knowledge, and utilizes large language models (LLMs) for action recognition and description.
Suit the Remedy to the Retriever: Interpretable Query Optimization with Retriever Preference Alignment for Vision-Language Retrieval
GuangHao Meng (Tsinghua University), Qing Li (Pengcheng Laboratory)
RetrievalOptimizationExplainability and InterpretabilityLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper proposes Retriever-Adaptive Query Optimization (RAQO), an interpretable visual-language retrieval query optimization framework that automatically rewrites ambiguous queries based on the retriever's preferences to enhance retrieval performance.
Superior Runtime Guarantees for the MOEA/D Multi-Objective Optimizer via Weighted-Sum Decomposition
Danyang Zhang (Harbin Institute of Technology), Benjamin Doerr (Ecole Polytechnique)
OptimizationBenchmark
🎯 What it does: This paper completes a full runtime analysis of MOEA/D using the original weighted-sum decomposition, proving that it can achieve significant acceleration on the classic OMM and multi-modal OJZJ benchmarks;
Supervised Dynamic Dimension Reduction with Deep Neural Network
Zhanye Luo (University of Chicago), Xiufan Yu (University of Notre Dame)
Representation LearningConvolutional Neural NetworkRecurrent Neural NetworkTime Series
🎯 What it does: Propose a supervised deep dynamic principal component analysis (SDDP) framework that integrates the target variable and lagged observations into factor extraction, generating target-aware low-dimensional features for nonlinear dynamic prediction.
Surgical AI Copilot: Energy-Based Fourier Gradient Low-Rank Adaptation for Surgical LLM Agent Reasoning and Planning
Jiayuan Huang (University College London), Mobarak I. Hoque (University of Manchester)
TransformerLarge Language ModelSupervised Fine-TuningAgentic AIPrompt EngineeringTextMultimodalityBiomedical Data
🎯 What it does: Propose Surgical AI Copilot—a Planner-Worker architecture-based LLM robot designed to provide real-time task planning, dialogue, and multimodal execution for endonasal pituitary surgery;
SurgPub-Video: A Comprehensive Surgical Video Framework for Enhanced Surgical Intelligence in Vision-Language Model
Yaoqian Li (Chinese University of Hong Kong), Pheng-Ann Heng (Chinese University of Hong Kong)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextBiomedical DataBenchmark
🎯 What it does: Constructed the SurgPub-Video dataset, a large-scale collection of peer-reviewed journal videos, and designed the SurgLLaVA-Video model based on TinyLLaVA-Video to support video-level inputs, aiming to enhance the performance of surgical vision-language models in tasks such as video scene understanding, VQA, action recognition, and technical evaluation.
Surrogate as Teacher: Distillation-Guided Graph Poisoning Attack
Xingyu Peng (State Key Laboratory of Complex and Critical Software Environment, Beihang University), Ke Xu (State Key Laboratory of Complex and Critical Software Environment, Beihang University)
Knowledge DistillationAdversarial AttackGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose MetaDist, a framework that reinterprets graph structure poisoning attacks as self-supervised knowledge distillation, inducing misclassification by maximizing prediction distribution differences through teacher-student model interactions.