AAAI 2024 Papers — Page 20
AAAI Conference on Artificial Intelligence · 2331 papers
Shrinking Your TimeStep: Towards Low-Latency Neuromorphic Object Recognition with Spiking Neural Networks
Yongqi Ding (University of Electronic Science and Technology of China), Yongjun Xiao (University of Electronic Science and Technology of China)
RecognitionObject DetectionComputational EfficiencySpiking Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: A low-latency, high-performance neuromorphic object recognition framework SSNN is proposed, which employs multi-stage temporal reduction and early classifiers to enhance the recognition performance of SNNs at short time steps.
Shuffled Deep Regression
Masahiro Kohjima (NTT Corporation)
Tabular
🎯 What it does: Proposes Shuffled Deep Regression (SDR), which learns deep regression models on shuffled data using the Sparse Random EM (SSEM) algorithm;
SIG: Speaker Identification in Literature via Prompt-Based Generation
Zhenlin Su (South China University of Technology), Mingdu Huangfu (Institute of Information Engineering Chinese Academy of Sciences)
RecognitionGenerationTransformerPrompt EngineeringText
🎯 What it does: This paper proposes a prompt-based generative model, SIG, to identify the speakers of quotes in literary texts, supporting open-domain (unknown speakers) and implicit (without direct annotation) situations.
Signed Graph Neural Ordinary Differential Equation for Modeling Continuous-Time Dynamics
Lanlan Chen (Xidian University), Jing Liu (Zhejiang University)
Graph Neural NetworkGraphTime SeriesOrdinary Differential Equation
🎯 What it does: A Signed Graph Neural Ordinary Differential Equation (SGODE) model is proposed for continuous-time dynamic modeling.
SiMA-Hand: Boosting 3D Hand-Mesh Reconstruction by Single-to-Multi-View Adaptation
Yinqiao Wang (Chinese University of Hong Kong), Chi-Wing Fu (Chinese University of Hong Kong)
Pose EstimationKnowledge DistillationImageMesh
🎯 What it does: The SiMA-Hand framework is proposed, which enhances the accuracy of 3D hand mesh reconstruction from single-view RGB images by utilizing multi-view information during the training phase.
SimCalib: Graph Neural Network Calibration Based on Similarity between Nodes
Boshi Tang (Tsinghua University), Helen Meng (Chinese University of Hong Kong)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This study addresses the calibration issue of graph neural networks in semi-supervised node classification and proposes a calibration framework based on node similarity, called SimCalib, aimed at significantly reducing the expected calibration error (ECE) of the model.
SimCS: Simulation for Domain Incremental Online Continual Segmentation
Motasem Alfarra (King Abdullah University of Science and Technology), Matthias Müller (Intel Labs)
SegmentationAutonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This study addresses the problem of online domain incremental continual segmentation and proposes the SimCS method, which utilizes real-time simulated data as regularization to alleviate catastrophic forgetting.
SimDistill: Simulated Multi-Modal Distillation for BEV 3D Object Detection
Haimei Zhao (University of Sydney), Dacheng Tao (University of Sydney)
Object DetectionAutonomous DrivingKnowledge DistillationMultimodalityPoint Cloud
🎯 What it does: The SimDistill method is proposed, which achieves knowledge distillation for BEV 3D object detection through a simulated multimodal teacher-student architecture.
Simple Image-Level Classification Improves Open-Vocabulary Object Detection
Ruohuan Fang (Beihang University), Xiao Bai (Beihang University)
ClassificationObject DetectionTransformerVision Language ModelContrastive LearningImage
🎯 What it does: A context-aware detection framework SIC-CADS based on CLIP image-level multi-label recognition is proposed, which enhances the detection accuracy of open vocabulary object detection by leveraging image-level global knowledge.
Simple Weak Coresets for Non-decomposable Classification Measures
Jayesh Malaviya (Indian Institute of Technology), Rachit Chhaya (DA-IICT)
ClassificationTabular
🎯 What it does: This study constructs a core subset (coreset) for non-decomposable classification metrics (F1, MCC), proves a lower bound for strong core subsets, and provides a theoretical upper bound on the error of weak core subsets obtained through stratified uniform sampling, along with experimental validation.
Simplicity Bias in Overparameterized Machine Learning
Yakir Berchenko (Ben Gurion University of the Negev)
🎯 What it does: This paper theoretically proves the existence of simplicity bias in over-parameterized machine learning, providing examples of wide networks, deep networks, and Boolean functions to illustrate that this bias is independent of the training process and optimizer, which can explain the good generalization of models.
Simplifying Complex Observation Models in Continuous POMDP Planning with Probabilistic Guarantees and Practice
Idan Lev-Yehudi (Technion - Israel Institute of Technology), Vadim Indelman (Technion - Israel Institute of Technology)
OptimizationReinforcement Learning from Human FeedbackMultimodality
🎯 What it does: For POMDP planning in continuous observation spaces, a simplified observation model is proposed for the planning phase, along with a probabilistic value error upper bound based on state-dependent total variation distance.
SimPSI: A Simple Strategy to Preserve Spectral Information in Time Series Data Augmentation
Hyun Ryu (Korea Advanced Institute of Science and Technology), Chang D. Yoo (Korea Advanced Institute of Science and Technology)
Data SynthesisAnomaly DetectionTransformerContrastive LearningTime Series
🎯 What it does: The SimPSI framework is proposed, which retains core spectral information during the time series data augmentation process through spectral mixing and retention mapping.
Simultaneous Optimization of Bid Shading and Internal Auction for Demand-Side Platforms
Yadong Xu (Tsinghua University), Pingzhong Tang (ByteDance)
OptimizationTabular
🎯 What it does: This paper proposes a joint framework for optimizing bid shading and the internal auction mechanism (ex-post IC) simultaneously in a first-price auction environment for demand-side platforms (DSPs), reducing the problem to learning monotonic bidding functions.
Situation-Dependent Causal Influence-Based Cooperative Multi-Agent Reinforcement Learning
Xiao Du (East China Normal University), Ting Wang (East China Normal University)
Reinforcement Learning
🎯 What it does: This paper proposes a multi-agent reinforcement learning algorithm based on context-related causal influence, SCIC, which provides internal rewards to agents using causal interventions and conditional mutual information estimation, thereby enhancing cooperation and exploration capabilities.
SkeletonGait: Gait Recognition Using Skeleton Maps
Chao Fan (Southern University of Science and Technology), Shiqi Yu (Southern University of Science and Technology)
RecognitionPose EstimationConvolutional Neural NetworkGaussian SplattingImageVideo
🎯 What it does: This paper proposes a walking recognition framework based on Skeleton Map, named SkeletonGait and SkeletonGait++, and validates the importance of skeletal structural information in walking recognition through alignment experiments with the traditional contour method DeepGaitV2.
Sketch and Refine: Towards Fast and Accurate Lane Detection
Chao Chen (Nanjing University), Gangshan Wu (Nanjing University)
Object DetectionAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: A Sketch-and-Refine framework is proposed, which first uses a local direction map to quickly draw rough lane candidates, and then refines the candidates through the Lane Segment Association Module (LSAM) to achieve real-time and high-precision lane detection.
Sketched Newton Value Iteration for Large-Scale Markov Decision Processes
Jinsong Liu (Shanghai University of Finance and Economics), Yinyu Ye (Stanford University)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes two value iteration algorithms based on the Newton method: Newton Value Iteration (NVI) and Sketched Newton Value Iteration (SNVI), aimed at solving Markov Decision Processes with large state and action spaces.
SkipDiff: Adaptive Skip Diffusion Model for High-Fidelity Perceptual Image Super-resolution
Xiaotong Luo (Xiamen University), Yun Fu (Northeastern University)
RestorationSuper ResolutionReinforcement LearningDiffusion modelImage
🎯 What it does: SkipDiff is proposed, an adaptive skip diffusion model that initializes with coarse skip steps and then controls fine-grained skip steps through reinforcement learning, achieving a high-fidelity perceptual balance in single image super-resolution.
SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing
Zhecheng Wang (Stanford University), Ram Rajagopal (Stanford University)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: SkyScript has been constructed—a large-scale diversified visual-language dataset containing 2.6 million remote sensing image-text pairs, covering 29,000 semantic labels. Continuous pre-training on this dataset resulted in SkyCLIP, enabling zero-shot scene classification, fine-grained attribute classification, and cross-modal retrieval in the remote sensing field.
SlowTrack: Increasing the Latency of Camera-Based Perception in Autonomous Driving Using Adversarial Examples
Chen Ma (Xi'an Jiaotong University), Chao Shen (Xi'an Jiaotong University)
Object DetectionObject TrackingAutonomous DrivingAdversarial AttackVideo
🎯 What it does: Proposes the SlowTrack framework, which generates low-intrusion delayed adversarial samples for the entire camera perception system (object detection + multi-object tracking).
Small Language Model Can Self-Correct
Haixia Han (East China Normal University), Yanghua Xiao (Fudan University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: An Intrinsic Self-Correction (ISC) mechanism is proposed, enabling small language models (6-13 billion parameters) to automatically verify and correct errors after generating answers.
SMILEtrack: SiMIlarity LEarning for Occlusion-Aware Multiple Object Tracking
Yu-Hsiang Wang (National Yang Ming Chiao Tung University), Xin Li (University at Albany - SUNY)
Object TrackingVideo
🎯 What it does: Proposes the SMILEtrack framework to address occlusion, similar objects, and the efficiency trade-off in multi-object tracking.
SNN-PDE: Learning Dynamic PDEs from Data with Simplicial Neural Networks
Jae Choi (University of Texas at Dallas), Yulia R. Gel (University of Texas at Dallas)
Graph Neural NetworkPoint CloudPhysics RelatedOrdinary Differential Equation
🎯 What it does: This paper proposes a Hodge theory-based degenerate neural network (SNN-PDE) for learning and solving time-varying partial differential equations on irregular grids and time intervals, particularly suitable for physical systems with sparse data and irregular spatial distributions, such as meteorology and fire smoke.
Social Physics Informed Diffusion Model for Crowd Simulation
Hongyi Chen (Tsinghua University), Xiao-Ping Zhang (Tsinghua University)
GenerationData SynthesisRecurrent Neural NetworkGraph Neural NetworkDiffusion modelMultimodalityGraph
🎯 What it does: A social physical information diffusion model, SPDiff, is proposed for generating multimodal crowd movement trajectories.
Social-Aware Group Display Configuration in VR Conference
Bay-Yuan Hsu (National Tsing Hua University), De-Nian Yang (Academia Sinica)
Recommendation SystemOptimizationGraph Neural NetworkGraph
🎯 What it does: In the context of large VR conference scenarios, this paper studies how to allocate limited display slots for each participant, balancing personal preferences and social utility, and proposes the Social-Aware VR Conference Group Display Configuration (SVGD) problem along with a solution.
SocialCVAE: Predicting Pedestrian Trajectory via Interaction Conditioned Latents
Wei Xiang (Zhejiang University), Xiaogang Jin (Zhejiang University)
GenerationData SynthesisAutonomous DrivingRecurrent Neural NetworkAuto EncoderTime SeriesSequential
🎯 What it does: A hybrid framework called SocialCVAE is proposed, which integrates an energy model with a Conditional Variational Autoencoder (CVAE) for predicting pedestrian trajectories and modeling behavioral uncertainty.
SoftCLIP: Softer Cross-Modal Alignment Makes CLIP Stronger
Yuting Gao (Tencent Youtu Lab), Xing Sun (Tencent Youtu Lab)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes SoftCLIP, which relaxes the one-to-one alignment constraint of CLIP, utilizing the fine-grained same-modal similarity between image ROIs and text labels as soft targets, and decoupling negative samples to achieve more flexible and accurate cross-modal alignment.
SOGDet: Semantic-Occupancy Guided Multi-View 3D Object Detection
Qiu Zhou (Independent Researcher), Roger Zimmermann (National University of Singapore)
Object DetectionAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: For multi-view 3D object detection, a semantic occupancy branch (3D Semantic Occupancy) is added in the BEV space to simultaneously predict object detection and environmental semantic occupancy, enhancing the perception of physical context.
Solving Satisfiability Modulo Counting for Symbolic and Statistical AI Integration with Provable Guarantees
Jinzhao Li (Purdue University), Yexiang Xue (Purdue University)
OptimizationGraphTabularAgriculture Related
🎯 What it does: A XOR SMC algorithm is proposed to transform countable problems into SAT, utilizing random XOR constraints to achieve approximate model counting with constant approximation guarantees;
Solving Spectrum Unmixing as a Multi-Task Bayesian Inverse Problem with Latent Factors for Endmember Variability
Dong Wu (Fudan University), Bo Peng (Shanghai Ocean University)
ClassificationRecognitionRestorationAuto EncoderImage
🎯 What it does: A multi-task model based on Bayesian inverse problems (BIPU) is proposed, which simultaneously completes spectral reconstruction, abundance regression, and multi-label classification to address the endmember variability issue in spectral mixing.
SoundCount: Sound Counting from Raw Audio with Dyadic Decomposition Neural Network
Yuhang He (University of Oxford), Andrew Markham (University of Oxford)
RecognitionConvolutional Neural NetworkAudio
🎯 What it does: This paper studies the problem of directly counting different sound sources from raw audio and proposes an end-to-end learnable DyDecNet network;
SpaceGTN: A Time-Agnostic Graph Transformer Network for Handwritten Diagram Recognition and Segmentation
Haoxiang Hu (Institute of Software, Chinese Academy of Sciences), Hongan Wang (Tsinghua University)
RecognitionSegmentationGraph Neural NetworkTransformerImage
🎯 What it does: A time-series-independent graph Transformer network called SpaceGTN is proposed for the recognition and segmentation of handwritten diagrams, along with the construction of a dynamic graph modeling and stroke recovery process; simultaneously, the first large-scale dataset of freely drawn online handwritten diagrams, OHSD, is released.
Span Graph Transformer for Document-Level Named Entity Recognition
Hongli Mao (Beijing Institute of Technology), Heyan Huang (Beijing Institute of Technology)
RecognitionTransformerLarge Language ModelTextBiomedical Data
🎯 What it does: This paper proposes the Span Graph Transformer (SGT), which captures long-distance dependencies at the word and span levels in a document by constructing sentence-level context and span graphs, thereby achieving more accurate entity recognition.
Spanning the Spectrum of Hatred Detection: A Persian Multi-Label Hate Speech Dataset with Annotator Rationales
Zahra Delbari, Mohammad Taher Pilehvar (Cardiff University)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper constructs a new multi-label, annotated rationale Persian hate speech dataset called PHATE, which contains approximately 7,000 manually annotated tweets.
Sparse Bayesian Deep Learning for Cross Domain Medical Image Reconstruction
Jiaxin Huang (University of Electronic Science and Technology of China), Xiaorong Pu (University of Electronic Science and Technology of China)
RestorationDomain AdaptationImageBiomedical DataMagnetic Resonance ImagingComputed TomographyStochastic Differential Equation
🎯 What it does: A lightweight sparse Bayesian deep learning framework is proposed for cross-domain medical image reconstruction, which maintains good generalization to unseen target domains after training in the source domain.
Sparse Enhanced Network: An Adversarial Generation Method for Robust Augmentation in Sequential Recommendation
Junyang Chen (Shenzhen University), Zhiguo Gong (University of Macau)
Recommendation SystemTransformerGenerative Adversarial NetworkContrastive LearningSequential
🎯 What it does: A SparseEnNet based on adversarial generation is proposed for data augmentation in sequential recommendation, generating more robust augmented sequences.
Sparse Variational Student-t Processes
Jian Xu (South China University of Technology), Delu Zeng (South China University of Technology)
Tabular
🎯 What it does: A sparse variational Student-t process (SVTP) is proposed, extending sparse approximation from Gaussian processes to Student-t processes.
Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views
Zixin Zou (Tsinghua University), Song-Hai Zhang
GenerationKnowledge DistillationDiffusion modelNeural Radiance FieldImage
🎯 What it does: Using sparse views to reconstruct 3D objects, a multi-view consistent diffusion model and category score distillation sampling are employed to distill the Stable Diffusion prior into NeRF, achieving high-quality and detail-rich 3D reconstruction and novel view synthesis.
SparseGNV: Generating Novel Views of Indoor Scenes with Sparse RGB-D Images
Weihao Cheng (Tencent), Ying Shan (Tencent)
GenerationData SynthesisDepth EstimationTransformerNeural Radiance FieldGenerative Adversarial NetworkImagePoint Cloud
🎯 What it does: This paper proposes the SparseGNV framework, which generates novel view images in indoor scenes using sparse RGB-D input.
Spatial Transform Decoupling for Oriented Object Detection
Hongtian Yu (University of Chinese Academy of Sciences), Yunfan Liu (University of Chinese Academy of Sciences)
Object DetectionTransformerImage
🎯 What it does: A Spatial Transform Decoupling (STD) framework is designed, utilizing the multi-branch structure of Vision Transformer to predict position, size, and angle separately, and enhancing foreground features layer by layer through cascading activation masks, thereby achieving directed object detection.
Spatial Voting with Incomplete Voter Information
Aviram Imber (Technion Israel Institute of Technology), Benny Kimelfeld (Technion Israel Institute of Technology)
🎯 What it does: The study examines the computational complexity of position-based voting rules (mainly positional scoring rules and approval voting) when the ideal points of voters are given with interval uncertainty in a two-dimensional spatial voting model.
Spatial-Contextual Discrepancy Information Compensation for GAN Inversion
Ziqiang Zhang (Xiamen University), Hanzi Wang (Xiamen University)
RestorationGenerationGenerative Adversarial NetworkImage
🎯 What it does: A GAN inversion method named SDIC is proposed, which first generates a difference map through a Spatial-Contextual Information Prediction Network (DIPN), and then uses this difference map in a Difference Information Compensation Network (DICN) to compensate both the latent code and the early features of the generator, ultimately achieving high-quality reconstruction and editable image output.
Spatial-Related Sensors Matters: 3D Human Motion Reconstruction Assisted with Textual Semantics
Xueyuan Yang (University of Science and Technology Beijing), Xiaojuan Ban (University of Science and Technology Beijing)
Pose EstimationTransformerContrastive LearningMultimodality
🎯 What it does: The research utilizes sparse IMU combined with text semantics to reconstruct 3D human actions.
Spatial-Semantic Collaborative Cropping for User Generated Content
Yukun Su (South China University of Technology), Qingyao Wu (South China University of Technology)
Object DetectionSegmentationGraph Neural NetworkTransformerImageBenchmarkAgriculture Related
🎯 What it does: A spatial semantic collaborative cropping network for user-generated content (UGC), named S²CNet 2, is proposed. It achieves high-quality cropping by constructing a visual object graph and utilizing an adaptive attention mechanism, balancing aesthetics and content integrity.
Spatial-Temporal Interplay in Human Mobility: A Hierarchical Reinforcement Learning Approach with Hypergraph Representation
Zhaofan Zhang (University of Macau), Pengyang Wang (University of Macau)
Recommendation SystemReinforcement LearningTime SeriesSequential
🎯 What it does: A space-time decoupling fusion framework based on hierarchical reinforcement learning (STI-HRL) is proposed, which represents historical records by constructing a mobility hypergraph to predict the user's next visit location.
Spatio-Temporal Fusion for Human Action Recognition via Joint Trajectory Graph
Yaolin Zheng (Beijing Information Science and Technology University), Longfei Xu (Beijing Information Science and Technology University)
RecognitionPose EstimationGraph Neural NetworkTransformerVideo
🎯 What it does: Proposes the Joint Trajectory Graph (JTG) structure and the JT-GraphFormer model for skeleton action recognition;
Spatio-Temporal Pivotal Graph Neural Networks for Traffic Flow Forecasting
Weiyang Kong (Sun Yat-Sen University), Yubao Liu (Sun Yat-Sen University)
Anomaly DetectionOptimizationGraph Neural NetworkTime Series
🎯 What it does: This paper proposes a new traffic flow prediction model STPGNN, focusing on identifying pivotal nodes in the traffic network, and achieving joint modeling of spatiotemporal features for pivotal and non-pivotal nodes through a pivotal graph convolution module and a parallel framework.
SPD-DDPM: Denoising Diffusion Probabilistic Models in the Symmetric Positive Definite Space
Yunchen Li (East China Normal University), Shaohui Lin (Tencent)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelGraphTime Series
🎯 What it does: This paper proposes a denoising diffusion probabilistic model (SPD-DDPM) that operates in the space of symmetric positive definite matrices (SPD) for unconditional and conditional generation of SPD matrices, and introduces a deeper SPD-U-Net to enhance the network's fitting capability.
SPEAL: Skeletal Prior Embedded Attention Learning for Cross-Source Point Cloud Registration
Kezheng Xiong (Xiamen University), Cheng Wang (Xiamen University)
Autonomous DrivingOptimizationTransformerPoint Cloud
🎯 What it does: A cross-source point cloud registration method based on skeleton priors, SPEAL, is proposed, which achieves high-precision registration of both cross-source and same-source point clouds using unsupervised skeleton extraction, skeleton-aware GeoTransformer, and correspondence dual sampler.
Spear and Shield: Adversarial Attacks and Defense Methods for Model-Based Link Prediction on Continuous-Time Dynamic Graphs
Dongjin Lee (Korea Advanced Institute of Science and Technology), Kijung Shin (Korea Advanced Institute of Science and Technology)
Recommendation SystemAdversarial AttackGraph Neural NetworkGraphTime Series
🎯 What it does: Designed and implemented an adversarial attack method T-SPEAR for continuous-time dynamic graph link prediction and the corresponding robust training defense method T-SHIELD.
Spectral Prompt Tuning: Unveiling Unseen Classes for Zero-Shot Semantic Segmentation
Wenhao Xu (Beijing University of Posts and Telecommunications), Xiaopeng Zhang
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: A zero-shot semantic segmentation method based on CLIP, called SPT-SEG, is proposed, achieving direct pixel-level segmentation in a single stage.
Spectral-Based Graph Neural Networks for Complementary Item Recommendation
Haitong Luo (Institute of Computing Technology), Yujun Zhang (Institute of Computing Technology)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A spectral domain-based graph neural network SComGNN is proposed, which uses low-pass and band-pass filters to extract the correlation and diversity features of complementary goods, respectively, and adaptively fuses them through a dual-stage attention mechanism to address the balance between correlation and diversity in complementary recommendations.
SpectralNeRF: Physically Based Spectral Rendering with Neural Radiance Field
Ru Li (Harbin Institute of Technology), Shuaicheng Liu (University of Electronic Science and Technology of China)
GenerationData SynthesisOptimizationNeural Radiance FieldImage
🎯 What it does: A spectral rendering framework called SpectralNeRF based on NeRF is proposed, which can generate high-quality white light RGB images from multi-wavelength perspectives.
Spectrum Translation for Refinement of Image Generation (STIG) Based on Contrastive Learning and Spectral Filter Profile
Seokjun Lee (Korea Institute of Science and Technology), Hyunseok Seo (Korea University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A spectrum translation framework STIG based on contrastive learning is proposed, which directly corrects the amplitude spectrum of generated images in the frequency domain to reduce frequency domain distortion.
SpFormer: Spatio-Temporal Modeling for Scanpaths with Transformer
Wenqi Zhong (Northwestern Polytechnical University), Dingwen Zhang (Northwestern Polytechnical University)
ClassificationRecognitionTransformerImage
🎯 What it does: This paper proposes SpFormer, a Transformer-based scanpath modeling framework that can simultaneously capture location, time, and dwell time information.
SPGroup3D: Superpoint Grouping Network for Indoor 3D Object Detection
Yun Zhu (Nanjing University of Science and Technology), Jin Xie
Object DetectionConvolutional Neural NetworkPoint Cloud
🎯 What it does: A 3D indoor object detection network based on superpoint grouping, SPGroup3D, is proposed.
SphereDiffusion: Spherical Geometry-Aware Distortion Resilient Diffusion Model
Tao Wu (Zhejiang University), Xi Li (Renmin University of China)
SegmentationGenerationData SynthesisDiffusion modelContrastive LearningImage
🎯 What it does: The SphereDiffusion framework is proposed, achieving controllable generation of high-quality spherical panoramic images from a single NFOV semantic segmentation map and text prompts.
Spherical Pseudo-Cylindrical Representation for Omnidirectional Image Super-resolution
Qing Cai (Ocean University of China), Yee-Hong Yang (University of Alberta)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerImage
🎯 What it does: A panorama image super-resolution framework based on Latitude Adaptive Pseudo-Cylindrical Projection (LAPR) is proposed, along with a viewport-based loss function and a recursive progressive network to achieve high-quality super-resolution.
Spiking NeRF: Representing the Real-World Geometry by a Discontinuous Representation
Zhanfeng Liao (Zhejiang University), Gang Pan (Zhejiang University)
GenerationData SynthesisSpiking Neural NetworkNeural Radiance FieldPoint CloudMesh
🎯 What it does: This paper proposes a Spiking NeRF using a hybrid ANN-SNN framework, which implements a discontinuous density field with discrete spiking neurons, thereby enabling more accurate reconstruction of 3D geometric structures.
SpikingBERT: Distilling BERT to Train Spiking Language Models Using Implicit Differentiation
Malyaban Bal, Abhronil Sengupta (Pennsylvania State University)
Computational EfficiencyKnowledge DistillationSpiking Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A BERT model based on spiking neural networks, SpikingBERT, is proposed and trained through implicit differentiation and knowledge distillation.
Spot the Error: Non-autoregressive Graphic Layout Generation with Wireframe Locator
Jieru Lin (Microsoft Research), Chin-Yew Lin (Microsoft Research)
Object DetectionGenerationTransformerImage
🎯 What it does: This paper studies non-autoregressive (NAR) graphic layout generation and proposes a learning-based locator that corrects generation errors through iterative mask prediction using rendered wireframe images.
Spotting the Unseen: Reciprocal Consensus Network Guided by Visual Archetypes
Wenbo Hu (East China Normal University), Ching Y. Suen (Concordia University)
Object DetectionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: The CoNet framework is proposed, which achieves one-shot open-set object/text detection through visual prototypes, utilizing self-correlation and cross-correlation modules for dense correspondence and spatial alignment.
SQLdepth: Generalizable Self-Supervised Fine-Structured Monocular Depth Estimation
Youhong Wang (Northwestern Polytechnical University), Hongkai Yu (Bytedance Inc)
Depth EstimationAutonomous DrivingTransformerImage
🎯 What it does: A self-supervised monocular depth estimation method based on Self Query Layer (SQL) called SQLdepth is proposed, which can learn fine-grained scene geometry through self-cost volume and directly predict depth maps from single-frame images.
SRFormer: Text Detection Transformer with Incorporated Segmentation and Regression
Qingwen Bu (Shanghai Jiao Tong University), Yichuan Cheng (City University of Hong Kong)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes SRFormer, a text detection model that integrates segmentation and regression within the DETR framework.
SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation
Chengyou Jia (Xi'an Jiaotong University), Jingdong Wang (Baidu Inc)
Object DetectionGenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes a diffusion model based on spatial-semantic mapping (SSMG) for generating realistic multi-object images from user-defined layouts (which can be boxes, masks, or keypoints) and free-text descriptions for each instance.
Stability in Online Coalition Formation
Martin Bullinger (University of Oxford), René Romen (Technical University of Munich)
Optimization
🎯 What it does: A framework for online coalition formation is proposed, studying how to achieve various stable structures (Nash, Individual, Contractual Nash, Contractual Individual) and Pareto optimal allocations under the conditions where agents arrive over time and must be assigned to coalitions immediately and irreversibly.
Stability of Multi-Agent Learning in Competitive Networks: Delaying the Onset of Chaos
Aamal Hussain (Imperial College London), Francesco Belardinelli (Imperial College London)
Reinforcement LearningGraph
🎯 What it does: This paper studies the convergence stability of Q-Learning dynamics in competitive network games, using statistical physics methods to perform average analysis on random game families, obtaining stability boundaries that are proven to depend only on the number of neighbors rather than the total number of players.
Stable Model Semantics for Description Logic Terminologies
Federica Di Stefano (Institute of Logic and Computation TU Wien), Mantas Šimkus (Umeå University)
🎯 What it does: A description logic knowledge base and stable model semantics for cyclic terms based on Quantified Equilibrium Logic is proposed, and its definition is provided.
Stable Unlearnable Example: Enhancing the Robustness of Unlearnable Examples via Stable Error-Minimizing Noise
Yixin Liu (Lehigh University), Lichao Sun (Drexel University)
ClassificationSafty and PrivacyAdversarial AttackImage
🎯 What it does: A new robust unlearnable sample generation method called Stable Error-Minimizing Noise (SEM) is proposed to protect data privacy in adversarial training environments.
STAIR: Spatial-Temporal Reasoning with Auditable Intermediate Results for Video Question Answering
Yueqian Wang (Peking University), Dongyan Zhao (Peking University)
RecognitionExplainability and InterpretabilityRecurrent Neural NetworkLarge Language ModelPrompt EngineeringVideoMultimodality
🎯 What it does: This paper proposes STAIR, a video question-answering framework based on neural module networks that can decompose questions into multi-level sub-tasks and reason step by step, while also returning auditable intermediate results, significantly enhancing the ability to handle complex spatiotemporal reasoning in long videos.
STAR: Boosting Low-Resource Information Extraction by Structure-to-Text Data Generation with Large Language Models
Mingyu Derek Ma (University of California), Wei Wang (University of California)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: STAR proposes a reverse structure-to-text data generation pipeline that utilizes large models to first generate event/relationship structures from unlabeled data, and then generate corresponding text. It achieves unsupervised self-correction through self-reflection and template feedback.
STAS: Spatial-Temporal Return Decomposition for Solving Sparse Rewards Problems in Multi-agent Reinforcement Learning
Sirui Chen (Renmin University of China), Yali Du (King's College London)
TransformerReinforcement LearningSequential
🎯 What it does: This paper proposes STAS, a multi-agent reward splitting method based on spatial-temporal attention and Shapley values, addressing the credit assignment problem under sparse rewards.
Statistical Spatially Inhomogeneous Diffusion Inference
Yinuo Ren (Stanford University), Grant M. Rotskoff (Stanford University)
Time SeriesFinance RelatedPhysics RelatedStochastic Differential Equation
🎯 What it does: A neural network-based estimation method is proposed to infer the drift and diffusion tensor of a multidimensional spatial non-homogeneous diffusion process from discrete observational data.
STDiff: Spatio-Temporal Diffusion for Continuous Stochastic Video Prediction
Xi Ye (Polytechnic Montreal), Guillaume-Alexandre Bilodeau (Polytechnic Montreal)
GenerationData SynthesisAutonomous DrivingRecurrent Neural NetworkDiffusion modelScore-based ModelVideoStochastic Differential Equation
🎯 What it does: A spatiotemporal diffusion model named STDiff is proposed for continuous random video prediction.
Stealthy Adversarial Attacks on Stochastic Multi-Armed Bandits
Zhiwei Wang (Tsinghua University), Hongning Wang (Oregon State University)
Adversarial AttackReinforcement LearningTabular
🎯 What it does: This study focuses on reward poisoning attacks on stochastic multi-armed bandit (MAB) algorithms, proposing a detection method based on homogeneity testing, and exploring the concept and feasibility of covert attacks.
StegaStyleGAN: Towards Generic and Practical Generative Image Steganography
Wenkang Su (Guangzhou University), Yiyan Sun (Sun Yat-Sen University)
GenerationData SynthesisSafty and PrivacyGenerative Adversarial NetworkImage
🎯 What it does: A general generative image steganography framework, StegaStyleGAN, is proposed, which embeds secret information into the noise injection of StyleGAN2 using a Distribution Preserving Secret Data Modulator (DP-SDM) and is equipped with a densely connected Secret Data Extractor (SDE). By incorporating an Image Attack Simulator (IAS), two models, StegaStyleGAN-Ls and StegaStyleGAN-Ly, can be obtained for lossless and lossy channels, respectively, and it supports the combination of GAN inversion for conditional steganography.
StegFormer: Rebuilding the Glory of Autoencoder-Based Steganography
Xiao Ke, Wenzhong Guo (Fuzhou University)
Data SynthesisSafty and PrivacyTransformerAuto EncoderImage
🎯 What it does: A StegFormer model based on autoencoders is proposed, capable of embedding one or more secret images into carrier images of the same resolution, enhancing reliability in real-world scenarios through a normalization training strategy and constrained loss.
STEM: Unleashing the Power of Embeddings for Multi-Task Recommendation
Liangcai Su (Tsinghua University), Jie Jiang (Tsinghua University)
Recommendation SystemMixture of ExpertsVideo
🎯 What it does: This paper proposes the Shared and Task-Specific Embedding (STEM) paradigm and implements the STEm-Net model, specifically addressing negative transfer in multi-task recommendation, achieving positive transfer especially on comparable samples with a balanced positive-negative sample ratio; it conducts fine-grained partitioning of comparable samples, revealing that traditional shared embedding methods are prone to negative transfer on these samples.
Step Vulnerability Guided Mean Fluctuation Adversarial Attack against Conditional Diffusion Models
Hongwei Yu (University of Science and Technology Beijing), Huimin Ma (University of Science and Technology Beijing)
RestorationSuper ResolutionAdversarial AttackDiffusion modelImage
🎯 What it does: This paper proposes an adversarial attack method for conditional diffusion models called Mean Fluctuation Attack (MFA). It exploits the sensitivity of diffusion models to the mean of noise by inducing a mean shift during the reverse sampling process, thereby degrading the generation quality. Furthermore, it investigates the vulnerabilities of different reverse steps and introduces two attack variants guided by vulnerability: MFA-VT and MFA-MVS.
Stereo Vision Conversion from Planar Videos Based on Temporal Multiplane Images
Shanding Diao (Hefei University of Technology), Ronggang Wang (Peking University)
Image TranslationData SynthesisDepth EstimationOptical FlowVideo
🎯 What it does: This paper proposes a Time-domain Multi-Plane Image (TMPI) model that generates high-quality stereo or light field videos from planar videos.
Sterling: Synergistic Representation Learning on Bipartite Graphs
Baoyu Jing (University of Illinois), Hanghang Tong (University of Illinois)
Recommendation SystemRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A no-negative-sample non-contrastive learning framework named STERLING is proposed for learning node embeddings on bipartite graphs while simultaneously preserving local (cross-type and same-type neighbors) and global (co-clustering) collaborative characteristics.
Stitching Segments and Sentences towards Generalization in Video-Text Pre-training
Fan Ma (Zhejiang University), Yi Yang (Zhejiang University)
GenerationRetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: A pre-training task based on the splicing and matching of video segments and sentences is proposed, utilizing boundary prediction in the spliced long sequence to achieve fine-grained video-text alignment.
Stitching Sub-trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL
Sungyoon Kim (KAIST), Kee-Eung Kim (KAIST)
TransformerReinforcement LearningDiffusion modelSequential
🎯 What it does: Proposed the SSD method, which uses conditional diffusion models and multi-step goal chain technology to achieve offline goal-oriented reinforcement learning sub-trajectory stitching.
Stochastic Bayesian Optimization with Unknown Continuous Context Distribution via Kernel Density Estimation
Xiaobin Huang (Nanjing University), Chao Qian (Nanjing University)
OptimizationTabularFinance Related
🎯 What it does: The research focuses on Bayesian optimization under unknown continuous contextual distributions and proposes two algorithms: SBO-KDE and DRBO-KDE.
StockMixer: A Simple Yet Strong MLP-Based Architecture for Stock Price Forecasting
Jinyong Fan (Shanghai Jiao Tong University), Yanyan Shen (Shanghai Jiao Tong University)
Recommendation SystemOptimizationTabularTime SeriesFinance Related
🎯 What it does: A lightweight architecture based on MLP called StockMixer is proposed for stock price prediction.
Stop! Planner Time: Metareasoning for Probabilistic Planning Using Learned Performance Profiles
Matthew Budd (Oxford Robotics Institute), Nick Hawes (Oxford Robotics Institute)
OptimizationMeta LearningReinforcement Learning
🎯 What it does: A non-greedy meta-reasoning framework based on deep reinforcement learning has been developed, which can decide when to stop planning, when to execute, and how to dynamically adjust the planner's hyperparameters in probabilistic planning (such as MDP).
Strategyproof Mechanisms for Group-Fair Obnoxious Facility Location Problems
Jiaqian Li (City University of Hong Kong), Hau Chan (University of Nebraska Lincoln)
Optimization
🎯 What it does: This paper studies the aversion facility location problem concerning group fairness, designing strategyproof mechanisms and providing both deterministic and randomized mechanisms that approximate the optimal solutions for maximum total/average group costs as well as inter-group and intra-group fairness (IIF).
Stratified GNN Explanations through Sufficient Expansion
Yuwen Ji (Beihang University), Ge Wang (University of Science and Technology Beijing)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: This paper proposes a hierarchical GNN explainer called STFExplainer based on sufficient expansion, which can generate explanation subgraphs at multiple levels (such as node clustering level, subgraph level, etc.) and directly extract multi-level interpretability information from existing GNN models.
Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning
Samyadeep Basu (University of Maryland), Soheil Feizi (Microsoft Research)
ClassificationRepresentation LearningTransformerSupervised Fine-TuningImageBenchmark
🎯 What it does: Conducted large-scale consistency experiments on parameter-efficient fine-tuning (PEFT) of Vision Transformers in few-shot classification tasks, and proposed two minimal yet powerful baseline methods, LN-TUNE and ATTNSCALE.
Structural Entropy Based Graph Structure Learning for Node Classification
Liang Duan (Yunnan University), Angsheng Li (Beihang University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: A graph structure learning framework based on structural entropy is proposed, which enhances and integrates multiple views by constructing encoding trees to extract hierarchical community information, thereby improving node classification performance.
Structural Information Enhanced Graph Representation for Link Prediction
Lei Shi (Ant Group), Jun Zhou (Ant Group)
Graph Neural NetworkTransformerGraph
🎯 What it does: In the task of link prediction in graphs, a structure information enhanced graph representation framework (SIEG) is proposed, which improves prediction performance by removing neighbor node features, utilizing GNN to encode neighborhood structures, and introducing a Binary Structural Transformer (BST) to encode the structural relationships of target node pairs.
Structural Information Guided Multimodal Pre-training for Vehicle-Centric Perception
Xiao Wang (Anhui University), Jin Tang (La Trobe University)
RecognitionObject DetectionSegmentationAutonomous DrivingKnowledge DistillationTransformerAuto EncoderContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes VehicleMAE, a multimodal pre-training framework for vehicle perception, which utilizes structural priors (vehicle contours) and semantic priors (natural language descriptions) to guide MAE in vehicle image reconstruction.
Structure-Aware Multimodal Sequential Learning for Visual Dialog
Young-Jin Kim (Hanyang University), Eun-Sol Kim (KT Corporation)
GenerationRetrievalTransformerVision Language ModelMultimodalitySequential
🎯 What it does: A structure-aware cross-modal alignment module utilizing pre-trained visual models and language models has been developed for visual dialogue generation.
Structure-CLIP: Towards Scene Graph Knowledge to Enhance Multi-Modal Structured Representations
Yufeng Huang (Zhejiang University), Wen Zhang (Zhejiang University)
RetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A new multimodal pre-training framework called Structure-CLIP is proposed, which utilizes scene graph knowledge to enhance the representation of fine-grained structured semantics of images and texts through semantic negative sampling and a knowledge-enhanced encoder.
Structured Probabilistic Coding
Dou Hu (Institute of Information Engineering Chinese Academy of Sciences), Songlin Hu (Institute of Information Engineering Chinese Academy of Sciences)
ClassificationRepresentation LearningTransformerSupervised Fine-TuningText
🎯 What it does: A supervised representation learning framework named Structured Probabilistic Coding (SPC) is proposed, which utilizes an encoder to simultaneously perform probabilistic coding and task prediction, extracting compact and information-rich representations from the input.
Style2Talker: High-Resolution Talking Head Generation with Emotion Style and Art Style
Shuai Tan (Shanghai Jiao Tong University), Ye Pan (Shanghai Jiao Tong University)
GenerationDiffusion modelGenerative Adversarial NetworkVideoTextAudio
🎯 What it does: This paper proposes Style 2 Talker, a high-resolution audio-driven speaker avatar generation method based on emotional text and artistic images.
StyleSinger: Style Transfer for Out-of-Domain Singing Voice Synthesis
Yu Zhang (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationData SynthesisDiffusion modelAudio
🎯 What it does: Designed and implemented the StyleSinger model, achieving zero-shot style transfer for singing voice synthesis.
Submodel Enumeration for CTL Is Hard
Nicolas Fröhlich, Arne Meier (Leibniz University Hannover)
🎯 What it does: This paper studies the complexity of the submodel enumeration problem in Computation Tree Logic (CTL) and proves that this problem is DelNP-complete in all sub-languages that include Boolean connectives. It also presents a polynomial delay enumeration algorithm when only using non-negative Boolean operations and restricting certain CTL operators.
Successive POI Recommendation via Brain-Inspired Spatiotemporal Aware Representation
Gehua Ma (Zhejiang University), Huajin Tang (Zhejiang University)
Recommendation SystemRecurrent Neural NetworkGraph Neural NetworkTime SeriesSequential
🎯 What it does: By constructing a graph context based on the brain's mesh-hippocampal system, the STE framework is proposed and applied to continuous recommendation of POIs (STEP).