AAAI 2024 Papers — Page 19
AAAI Conference on Artificial Intelligence · 2331 papers
S3A: Towards Realistic Zero-Shot Classification via Self Structural Semantic Alignment
Sheng Zhang (Mohamed bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Linköping University)
ClassificationRecognitionTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: The Self Structural Semantic Alignment (S3A) framework is proposed, utilizing unsupervised image clustering, voting, LLM prompting, and realignment self-learning to address the real zero-shot classification problem under large vocabularies.
SA²VP: Spatially Aligned-and-Adapted Visual Prompt
Wenjie Pei (Harbin Institute of Technology), Guangming Lu (Harbin Institute of Technology)
ClassificationTransformerPrompt EngineeringImage
🎯 What it does: A spatially aligned and adaptive visual prompt model SA VP 2 is proposed, which aligns a two-dimensional prompt map with the image feature map to improve visual prompt tuning.
Safe Abductive Learning in the Presence of Inaccurate Rules
Xiao-Wen Yang (Nanjing University), Zhi-Hua Zhou
ClassificationOptimizationExplainability and InterpretabilityGraph Neural NetworkTabular
🎯 What it does: In the context of combining raw data with logical rules, a Safe-ABL safe contradiction learning framework is proposed, which can maintain or improve model performance even when the knowledge base contains inaccurate rules.
SafeAR: Safe Algorithmic Recourse by Risk-Aware Policies
Haochen Wu (University of Michigan), Sriram Gopalakrishnan (J.P. Morgan)
Safty and PrivacyReinforcement LearningTabularFinance Related
🎯 What it does: Proposes the SafeAR (Safe Algorithmic Retrospection) framework, which provides feasible action paths for individuals adversely affected by machine learning decisions through risk-aware strategies.
SALSA: Semantically-Aware Latent Space Autoencoder
Kathryn E. Kirchoff (University of North Carolina), Shawn M. Gomez (University of North Carolina)
Drug DiscoveryTransformerAuto EncoderContrastive LearningGraph
🎯 What it does: By incorporating supervised contrastive loss into the SMILES transformer autoencoder, a semantically aware molecular latent space is learned.
SAM-PARSER: Fine-Tuning SAM Efficiently by Parameter Space Reconstruction
Zelin Peng (Shanghai Jiao Tong University), Wei Shen (Shanghai Jiao Tong University)
Object DetectionSegmentationSupervised Fine-TuningImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a method to efficiently fine-tune different scenes by reconstructing the parameter space through matrix decomposition on the Segment Anything Model (SAM), allowing for fine-tuning with only coefficient adjustments.
SAME: Sample Reconstruction against Model Extraction Attacks
Yi Xie (Xidian University), Xiaofeng Chen (Nanyang Technological University)
Anomaly DetectionAdversarial AttackAuto EncoderImage
🎯 What it does: A sample reconstruction-based model extraction attack detection method (SAME) is proposed, which can identify malicious queries without relying on additional OOD data, user query records, or access to white-box models, and can be used in conjunction with active defense strategies.
SAMFlow: Eliminating Any Fragmentation in Optical Flow with Segment Anything Model
Shili Zhou (Fudan University), Bo Yan (Fudan University)
Autonomous DrivingTransformerOptical FlowImageVideo
🎯 What it does: Using the frozen Segment Anything Model (SAM) image encoder combined with FlowFormer, we propose SAMFlow to eliminate the fragmentation problem in optical flow estimation.
Sample Efficient Reinforcement Learning with Partial Dynamics Knowledge
Meshal Alharbi (Massachusetts Institute of Technology), Munther Dahleh (Massachusetts Institute of Technology)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper studies the sample complexity problem in online reinforcement learning with known or efficiently learnable partial dynamics knowledge (in the form of additive disturbance). It proposes an optimization Q-learning algorithm based on UCB (UCB-f) that achieves a √T regret independent of S and A when the dynamics are fully known; when only noise estimates are available, the sample complexity remains independent of the state and action dimensions and is related to the approximation error of f and the Lipschitz constant of the value function.
Sample-and-Bound for Non-convex Optimization
Yaoguang Zhai (University of California), Sicun Gao
Optimization
🎯 What it does: A non-convex optimization method that combines sampling and tree search, called MCIR, is proposed.
Sample-Constrained Black Box Optimization for Audio Personalization
Rajalaxmi Rajagopalan (University of Illinois), Romit Roy Choudhury (University of Illinois)
Recommendation SystemOptimizationAudio
🎯 What it does: A hybrid strategy ORACLEBO is proposed for audio personalization, which simultaneously uses full filter queries and dimension queries in black-box optimization.
Sample-Level Cross-View Similarity Learning for Incomplete Multi-View Clustering
Suyuan Liu (National University of Defense Technology), Xinwang Liu (National University of Defense Technology)
OptimizationRepresentation LearningMultimodality
🎯 What it does: A method for incomplete multi-view clustering based on sample-level cross-view similarity learning (SCSL) is proposed, which can construct a complete similarity matrix between all sample pairs and perform spectral clustering.
Sampling for Beyond-Worst-Case Online Ranking
Qingyun Chen (University of California), Ruilong Zhang (University at Buffalo)
Time Series
🎯 What it does: An online model for the feedback arc set (FAS) problem is proposed and analyzed, particularly focusing on semi-online and additional sampling models that include prior samples;
Sampling-Resilient Multi-Object Tracking
Zepeng Li (Zhejiang University), Gang Chen (Zhejiang University)
Object TrackingRecurrent Neural NetworkReinforcement LearningVideo
🎯 What it does: A robust multi-object tracking framework for downsampled videos, SR-Track, is proposed to address the accuracy degradation issue under high frame rate compression.
SasWOT: Real-Time Semantic Segmentation Architecture Search WithOut Training
Chendi Zhu (Nanjing University), Zhengxing Sun (Nanjing University)
SegmentationNeural Architecture SearchImage
🎯 What it does: A completely training-free semantic segmentation architecture search framework, SasWOT, has been designed and implemented, which first automatically searches for zero-cost proxies and then uses them for training-free network search.
SAT-Based Algorithms for Regular Graph Pattern Matching
Miguel Terra-Neves (OutSystems), Antonio Alegria (Zharta)
Graph Neural NetworkGraph
🎯 What it does: A SAT-based Regular Graph Pattern Matching (ReGaP) algorithm is proposed, which can check complex structural properties in graphs through declarative specifications, extending the traditional graph isomorphism problem.
SAT-Based Techniques for Lexicographically Smallest Finite Models
Mikoláš Janota (Czech Technical University in Prague), Petr Vojtěchovský (Ben-Gurion University of the Negev)
OptimizationTabular
🎯 What it does: Using a SAT solver to incrementally construct and compute the lexicographically minimal representation (lexmin) of a given finite algebraic structure (single binary operation).
SAT-Based Tree Decomposition with Iterative Cascading Policy Selection
Hai Xia (TU Wien), Stefan Szeider (TU Wien)
OptimizationGraph Neural NetworkGraph
🎯 What it does: The study investigates how to automatically configure the TW-SLIM algorithm to improve the tree decomposition width of graphs.
SAUI: Scale-Aware Unseen Imagineer for Zero-Shot Object Detection
Jiahao Wang (Xi'an Jiaotong University), Qinghua Zheng (Xi'an Jiaotong University)
Object DetectionGenerative Adversarial NetworkImage
🎯 What it does: Designed and implemented the Scale-Aware Unseen Imagineer (SAUI), which generates unseen class visual features of different scales for zero-shot object detection using multi-scale feature extraction and series GANs;
SAVSR: Arbitrary-Scale Video Super-Resolution via a Learned Scale-Adaptive Network
Zekun Li (Xidian University), Wei Feng (Tianjin University)
RestorationSuper ResolutionConvolutional Neural NetworkVideo
🎯 What it does: A SAVSR network is designed to achieve video super-resolution at arbitrary scales (non-integer and non-symmetric ratios) under a single model.
Say Anything with Any Style
Shuai Tan (Shanghai Jiao Tong University), Ye Pan (Shanghai Jiao Tong University)
GenerationData SynthesisPose EstimationRecurrent Neural NetworkTransformerAuto EncoderVideoAudio
🎯 What it does: Generate speaker videos synchronized with audio, customizable styles, and head movements, supporting video-driven style editing.
SayCanPay: Heuristic Planning with Large Language Models Using Learnable Domain Knowledge
Rishi Hazra (Orebro University), Luc De Raedt (KU Leuven)
OptimizationRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes the SayCanPay framework, which combines large language models with heuristic search to generate feasible and cost-effective plans through three steps (Say, Can, Pay);
SC-NeuS: Consistent Neural Surface Reconstruction from Sparse and Noisy Views
Shi-Sheng Huang (Beijing Normal University), Ying Shan (Tencent)
Pose EstimationDepth EstimationNeural Radiance FieldPoint Cloud
🎯 What it does: A framework called SC-NeuS is proposed, which can jointly learn neural implicit surfaces and camera poses from sparse and noisy camera poses.
Scalable Enumeration of Trap Spaces in Boolean Networks via Answer Set Programming
Giang Trinh (Aix-Marseille University), Sylvain Soliman (Inria)
Biomedical Data
🎯 What it does: This paper studies a Boolean network trap space enumeration method based on answer set programming (tsconj), achieving efficient computation of both minimum and maximum trap spaces.
Scalable Geometric Fracture Assembly via Co-creation Space among Assemblers
Ruiyuan Zhang (Zhejiang University), Chao Wu (Zhejiang University)
Pose EstimationOptimizationTransformerPoint CloudMesh
🎯 What it does: A scalable geometric fracture assembly framework is proposed, utilizing a collaborative creation space to allow multiple assemblers to gradually complete the assembly, and introducing geometric conflict loss to avoid local optima and collision issues.
Scalable Motion Style Transfer with Constrained Diffusion Generation
Wenjie Yin (KTH Royal Institute of Technology), Mårten Björkman (University of Copenhagen)
GenerationData SynthesisDiffusion modelVideoOrdinary Differential Equation
🎯 What it does: A scalable motion style transfer framework based on Dual Diffusion Implicit Bridges (DDIBs) and Keyframe Manifold Constraint Gradient (KMCG) is proposed, supporting the transfer between up to ten dance styles while enhancing the quality of style transfer while maintaining content consistency.
Scale Optimization Using Evolutionary Reinforcement Learning for Object Detection on Drone Imagery
Jialu Zhang (University of Nottingham), Jiang Liu (Southern University of Science and Technology)
Object DetectionOptimizationReinforcement LearningImage
🎯 What it does: An evolutionary reinforcement learning agent (EVORL) combined with a coarse-fine detection framework is used to adaptively optimize the target scale in drone aerial images to improve detection accuracy.
Scaling and Masking: A New Paradigm of Data Sampling for Image and Video Quality Assessment
Yongxu Liu (Xidian University), Jinjian Wu (Xidian University)
TransformerImageVideo
🎯 What it does: This paper proposes a sampling method based on scaling and masking, called SAMA, which captures both local details and global semantics of images/videos while maintaining a single-branch model.
Scaling Few-Shot Learning for the Open World
Zhipeng Lin (National University of Defense Technology), Ji Wang (National University of Defense Technology)
Representation LearningMeta LearningTransformerImage
🎯 What it does: This paper proposes a new problem called FSL-MNC (Few-Shot Learning with Many Novel Classes), addressing few-shot learning under thousands of new classes in an open world, and designs the SHA-Pipeline for efficient training and testing under large-scale categories.
Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data
Shuvendu Roy (Queens University), Ali Etemad (Queens University)
ClassificationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes the UnMixMatch semi-supervised learning framework, which can learn effective representations from unconstrained unlabeled data and improve classification performance.
ScanERU: Interactive 3D Visual Grounding Based on Embodied Reference Understanding
Ziyang Lu (University of Electronic Science and Technology of China), Heng Tao Shen (University of Science and Technology of China)
RecognitionObject DetectionTransformerVision Language ModelTextPoint Cloud
🎯 What it does: The task of 'Embodied Reference Understanding (ERU)' is proposed, which utilizes natural language and human gestures to jointly locate target objects in 3D point cloud scenes, and based on this, a complete system from data collection, synthesis to model training is designed.
SCD-Net: Spatiotemporal Clues Disentanglement Network for Self-Supervised Skeleton-Based Action Recognition
Cong Wu (Jiangnan University), Zhenhua Feng (University of Surrey)
RecognitionRetrievalGraph Neural NetworkTransformerContrastive LearningVideoMultimodality
🎯 What it does: A skeleton action recognition framework SCD-Net based on self-supervised contrastive learning is proposed, which utilizes a dual-path decoupled encoder to extract pure spatial and temporal features from skeleton sequences and achieves cross-domain comparison through global anchors.
SciEval: A Multi-Level Large Language Model Evaluation Benchmark for Scientific Research
Liangtai Sun (Shanghai Jiao Tong University), Kai Yu (Shanghai Jiao Tong University)
TransformerLarge Language ModelTextBenchmarkPhysics RelatedChain-of-Thought
🎯 What it does: SciEval is proposed, a multi-disciplinary, hierarchical scientific research assessment benchmark that covers four dimensions (fundamental knowledge, knowledge application, scientific computation, research capability) in chemistry, physics, and biology, while also providing static, dynamic, and experimental question banks.
Scores for Learning Discrete Causal Graphs with Unobserved Confounders
Alexis Bellot (Google DeepMind), Elias Bareinboim (Columbia University)
Graph Neural NetworkGraphTabular
🎯 What it does: This paper proposes a new Bayesian scoring method based on Watanabe's asymptotic expansion, aimed at capturing equality and inequality constraints in the learning of discrete causal graph structures in the presence of unobserved confounders.
SCP: Spherical-Coordinate-Based Learned Point Cloud Compression
Ao Luo (KDDI Research), Jiro Katto (Yokohama National University)
CompressionAutonomous DrivingTransformerAuto EncoderPoint Cloud
🎯 What it does: Convert LiDAR point clouds from Cartesian coordinates to spherical coordinates, and based on this, construct an Octree for learned compression.
Scribble Hides Class: Promoting Scribble-Based Weakly-Supervised Semantic Segmentation with Its Class Label
Xinliang Zhang (Peking University), Yanye Lu (Peking University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Class-Driven Scribble Enhancement Network (CDSP) that utilizes image-level category information in scribbles to generate global pseudo-labels, and enhances weakly supervised semantic segmentation performance through local feature correction and distance entropy loss.
SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation
Zhengze Xu (Huazhong University of Science and Technology), Changxin Gao (Huazhong University of Science and Technology)
SegmentationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes SCTNet, a single-branch CNN combined with a Transformer during training, to achieve real-time semantic segmentation.
SD-MVS: Segmentation-Driven Deformation Multi-View Stereo with Spherical Refinement and EM Optimization
Zhenlong Yuan (Institute of Computing Technology, Chinese Academy of Sciences), Zhaoqi Wang (Institute of Computing Technology, Chinese Academy of Sciences)
SegmentationDepth EstimationOptimizationComputational EfficiencyImageBenchmark
🎯 What it does: This paper proposes a PatchMatch multi-view stereo reconstruction method driven by instance segmentation (SD-MVS), which achieves pixel-level patch deformation through SAM segmentation, a multi-scale parallel consistency architecture, spherical gradient refinement, and EM optimization of hyperparameters, significantly improving the reconstruction integrity and accuracy in texture-missing areas.
SDAC: A Multimodal Synthetic Dataset for Anomaly and Corner Case Detection in Autonomous Driving
Lei Gong (University of Science and Technology of China), Jianmin Ji (University of Science and Technology of China)
Object DetectionSegmentationData SynthesisAnomaly DetectionAutonomous DrivingImageMultimodalityPoint Cloud
🎯 What it does: A multi-modal synthetic dataset SDAC is proposed for anomaly and corner case detection in autonomous driving, and a systematic evaluation of closed-set and open-set detection methods is conducted.
SDGAN: Disentangling Semantic Manipulation for Facial Attribute Editing
Wenmin Huang (Sun Yat-sen University), Xiaochun Cao (Shenzhen University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A semantic decoupling GAN (SDGAN) is proposed, achieving precise control of facial attribute editing through an attribute-specific editing module and semantic masks.
SDGMNet: Statistic-Based Dynamic Gradient Modulation for Local Descriptor Learning
Yuxin Deng (Wuhan University), Jiayi Ma (Wuhan University)
Pose EstimationRetrievalContrastive LearningImage
🎯 What it does: A statistical-based dynamic gradient modulation method called SDGMNet is proposed for local descriptor learning.
SEA-GWNN: Simple and Effective Adaptive Graph Wavelet Neural Network
Swakshar Deb (University of Dhaka), Shafin Rahman (North South University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: The SEA-GWNN model is proposed, which adaptively learns dual-channel wavelet filters on arbitrary graphs using a multi-hop computation tree-based boosting transformation, and achieves node classification by fusing high and low-frequency information.
SEC: More Accurate Clustering Algorithm via Structural Entropy
Junyu Huang (Central South University), Jianxin Wang (Central South University)
OptimizationTabular
🎯 What it does: The SEC clustering algorithm is proposed, which extracts global structural information through a structural entropy encoding tree and achieves better clustering by combining an iterative pre-deletion and reallocation mechanism.
SECap: Speech Emotion Captioning with Large Language Model
Yaoxun Xu (Tsinghua University), Rongzhi Gu (Chinese University of Hong Kong)
GenerationTransformerLarge Language ModelContrastive LearningTextAudio
🎯 What it does: Proposes a speech emotion description (caption) task, utilizing large language models to generate natural language emotion descriptions;
Secure Distributed Sparse Gaussian Process Models Using Multi-Key Homomorphic Encryption
Adil Nawaz (Shenzhen University), Jie Chen (Shenzhen University)
Federated LearningSafty and PrivacyTabular
🎯 What it does: A distributed sparse Gaussian process regression model based on multi-key homomorphic encryption has been designed and implemented to ensure privacy and security during distributed training among IoT devices.
Seed-Guided Fine-Grained Entity Typing in Science and Engineering Domains
Yu Zhang (University of Illinois), Jiawei Han (University of Illinois)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: A seed-guided fine-grained entity type classification framework SETYPE is proposed, which can be trained using only type names and a small number of seed entities, while simultaneously predicting both seen and unseen type entities.
Seeing Dark Videos via Self-Learned Bottleneck Neural Representation
Haofeng Huang (Peking University), Jiaying Liu (Peking University)
RestorationGaussian SplattingVideo
🎯 What it does: A completely self-supervised low-light video enhancement method is proposed, utilizing bottleneck neural representation to restore high-quality videos without relying on external data.
SEER: Backdoor Detection for Vision-Language Models through Searching Target Text and Image Trigger Jointly
Liuwan Zhu (University of Hawaii), Hongyi Wu (University of Arizona)
Anomaly DetectionOptimizationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: The SEER algorithm is proposed for jointly searching image triggers and malicious target texts in pre-trained vision-language models to detect whether a backdoor has been implanted.
SeGA: Preference-Aware Self-Contrastive Learning with Prompts for Anomalous User Detection on Twitter
Ying-Ying Chang (National Yang Ming Chiao Tung University), Wen-Chih Peng (National Yang Ming Chiao Tung University)
Anomaly DetectionGraph Neural NetworkLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: The SeGA framework is proposed for detecting anomalous users (including bots and trolls) on Twitter, achieving multi-class classification through heterogeneous information networks (HIN) and learning user content preferences.
Segment beyond View: Handling Partially Missing Modality for Audio-Visual Semantic Segmentation
Renjie Wu (University of Adelaide), Hsiang-Ting Chen (University of Adelaide)
SegmentationAutonomous DrivingKnowledge DistillationTransformerVideoMultimodalityAudio
🎯 What it does: Proposes the Segment Beyond View (SBV) method to achieve semantic segmentation of super-view vehicles in road scenes under the conditions of only first-person perspective and binaural audio.
SEIT: Structural Enhancement for Unsupervised Image Translation in Frequency Domain
Zhifeng Zhu (Xi'an Jiaotong University), Yuehu Liu (Xi'an Jiaotong University)
Image TranslationDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an unsupervised image translation framework called SEIT that enhances structure in the frequency domain, combining FDA and WSE modules to achieve style transfer and structural detail preservation.
Selective and Orthogonal Feature Activation for Pedestrian Attribute Recognition
Junyi Wu (Xiamen Meiya Pico Information Company Ltd.), Jianqiang Zhao (Xiamen Meiya Pico Information Company Ltd.)
RecognitionTransformerImage
🎯 What it does: This paper proposes a pedestrian attribute recognition framework called SOFAFormer based on Vision Transformer, and enhances the recognition capability for minority attributes through adaptive feature suppression and orthogonal feature activation loss.
Selective Deep Autoencoder for Unsupervised Feature Selection
Wael Hassanieh (University of Michigan), Abdallah Chehade (University of Michigan)
Representation LearningAuto EncoderTabular
🎯 What it does: This paper proposes an unsupervised feature selection framework called Selective Deep AutoEncoder (SDAE), which utilizes deep autoencoders and a custom Selective Layer to automatically learn the minimal feature subset that can reconstruct the original feature space.
Selective Focus: Investigating Semantics Sensitivity in Post-training Quantization for Lane Detection
Yunqian Fan (ShanghaiTech University), Xianglong Liu (Beihang University)
Autonomous DrivingOptimizationImage
🎯 What it does: A 'Selective Focus' framework is proposed for post-training quantization (PTQ) of lane detection models, utilizing semantic sensitivity to guide the quantization process.
Self-Distillation Regularized Connectionist Temporal Classification Loss for Text Recognition: A Simple Yet Effective Approach
Ziyin Zhang (Huawei Technologies), Wei Peng (Huawei Technologies)
RecognitionKnowledge DistillationRecurrent Neural NetworkText
🎯 What it does: A self-distillation based CTC loss (DCTC) is proposed, which generates high-quality implicit alignment through MAP estimation and incorporates frame-level supervision to enhance text recognition accuracy.
Self-Interpretable Graph Learning with Sufficient and Necessary Explanations
Jiale Deng (Shanghai Jiao Tong University), Yanyan Shen (Shanghai Jiao Tong University)
Explainability and InterpretabilityRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: The SUNNY-GNN framework is proposed, combining self-explanatory graph learning and contrastive learning to generate both sufficient and necessary explanatory subgraphs to enhance GNN prediction performance.
Self-Paced Unified Representation Learning for Hierarchical Multi-Label Classification
Zixuan Yuan (Hong Kong University of Science and Technology), Hui Xiong
ClassificationRepresentation LearningGraph Neural NetworkTextBiomedical Data
🎯 What it does: Proposes an adaptive unified representation learning framework (SPUR) for hierarchical multi-label classification (HMLC).
Self-Prompt Mechanism for Few-Shot Image Recognition
Mingchen Song (Ocean University of China), Guoqiang Zhong (Ocean University of China)
RecognitionMeta LearningTransformerPrompt EngineeringImage
🎯 What it does: Proposes a Self-Prompt Mechanism (SPM) that generates self-prompts through spatial and channel selection on deep features of Vision Transformer, and injects these prompts into self-attention, enhancing few-shot image recognition performance with only 2% of parameters for efficient fine-tuning.
Self-Supervised 3D Human Mesh Recovery from a Single Image with Uncertainty-Aware Learning
Guoli Yan (Wayne State University), Jing Hua (Wayne State University)
Pose EstimationDepth EstimationConvolutional Neural NetworkImageMesh
🎯 What it does: A self-supervised monocular 3D human mesh recovery framework is proposed, trained using the self-consistency of 2D joints and depth maps.
Self-Supervised Bird’s Eye View Motion Prediction with Cross-Modality Signals
Shaoheng Fang (Shanghai Jiao Tong University), Siheng Chen (Shanghai Jiao Tong University)
Autonomous DrivingOptical FlowMultimodalityPoint Cloud
🎯 What it does: This paper proposes a cross-modal self-supervised framework for learning dense bird's-eye view (BEV) motion flow without labeled data;
Self-Supervised Disentangled Representation Learning for Robust Target Speech Extraction
Zhaoxi Mu (Xi'an Jiaotong University), Qing Yang (Xi'an Jiaotong University)
Representation LearningTransformerAuto EncoderContrastive LearningAudio
🎯 What it does: This study proposes a self-supervised separation representation learning framework (SDR-TSE) that enhances the robustness of target speech extraction by separating global, semantic, and speaker identity information from reference speech.
Self-Supervised Multi-Modal Knowledge Graph Contrastive Hashing for Cross-Modal Search
Meiyu Liang (Beijing University of Posts and Telecommunications), Zhe Xue (Beijing University of Posts and Telecommunications)
RetrievalGraph Neural NetworkContrastive LearningImageTextMultimodality
🎯 What it does: This paper studies a self-supervised multi-granularity multi-modal knowledge graph contrastive hashing method for cross-modal retrieval.
Self-Supervised Representation Learning with Meta Comprehensive Regularization
Huijie Guo (Beihang University), Lei Shi (Beihang University)
Object DetectionRepresentation LearningContrastive LearningImage
🎯 What it does: Proposes the CompMod module, which combines Meta Comprehensive Regularization (MCR) to improve self-supervised learning models, enabling them to learn more comprehensive representations and reduce task-related information loss caused by data augmentation.
Self-Training Based Few-Shot Node Classification by Knowledge Distillation
Zongqian Wu (University of Electronic Science and Technology of China), Xiaofeng Zhu (University of Electronic Science and Technology of China)
ClassificationKnowledge DistillationGraph Neural NetworkSupervised Fine-TuningGraph
🎯 What it does: A self-training few-shot node classification method based on knowledge distillation (KD-FSNC) is proposed, which enhances the performance of the student model through representation distillation (local + global) and pseudo-label distillation after pre-training on the teacher model.
SelfPromer: Self-Prompt Dehazing Transformers with Depth-Consistency
Cong Wang (Hong Kong Polytechnic University), Xiao-Ming Wu (Nanjing University of Science and Technology)
RestorationDepth EstimationTransformerPrompt EngineeringImage
🎯 What it does: This study investigates an image dehazing method using a self-prompting transformer called SelfPromer guided by depth consistency.
Semantic Complete Scene Forecasting from a 4D Dynamic Point Cloud Sequence
Zifan Wang (Tsinghua University), Li Yi (Tsinghua University)
SegmentationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: Proposes the Semantic Complete Scene Forecasting (SCSF) task and designs the SCSFNet model.
Semantic Lens: Instance-Centric Semantic Alignment for Video Super-resolution
Qi Tang (Beijing Jiaotong University), Chao Yao (University of Science and Technology Beijing)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerVideo
🎯 What it does: Proposes the Semantic Lens framework, which decomposes videos into instance, event, and scene semantics, and enhances pixel-level features driven by semantic priors to achieve video super-resolution.
Semantic Segmentation in Multiple Adverse Weather Conditions with Domain Knowledge Retention
Xin Yang (National University of Singapore), Robby T. Tan (National University of Singapore)
SegmentationDomain AdaptationImage
🎯 What it does: A semantic segmentation method for continuous multi-target unsupervised domain adaptation is proposed, which can sequentially adapt to various adverse weather conditions without accessing previous target data while retaining previously learned knowledge.
Semantic-Aware Autoregressive Image Modeling for Visual Representation Learning
Kaiyou Song (Megvii Technology), Tong Wang (Megvii Technology)
Object DetectionSegmentationRepresentation LearningTransformerImage
🎯 What it does: A semantic-aware autoregressive image modeling method SemAIM is proposed for visual self-supervised pre-training.
Semantic-Aware Data Augmentation for Text-to-Image Synthesis
Zhaorui Tan (Xi'an Jiaotong-Liverpool University), Kaizhu Huang (Duke Kunshan University)
GenerationData SynthesisTransformerDiffusion modelGenerative Adversarial NetworkImageText
🎯 What it does: A semantic-aware data augmentation framework (SADA) for text-to-image synthesis is proposed, which includes implicit text semantic preservation augmentation (ITA) and generated image semantic preservation (GisC).
Semantic-Aware Transformation-Invariant RoI Align
Guo-Ye Yang (Tsinghua University), Shi-Min Hu (Tsinghua University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: A new RoI feature extractor called Semantic RoI Align (SRA) is proposed, which can extract RoI features that are invariant to various geometric transformations in two-stage detectors.
Semantic-Guided Generative Image Augmentation Method with Diffusion Models for Image Classification
Bohan Li (Harbin Institute of Technology), Wanxiang Che (Harbin Institute of Technology)
ClassificationGenerationTransformerPrompt EngineeringDiffusion modelImage
🎯 What it does: A semantic-guided image enhancement method SGID based on diffusion models is proposed, utilizing image labels and titles generated by BLIP as prompts to enhance the diversity of generated images while maintaining semantic consistency.
Semantic-Guided Novel Category Discovery
Weishuai Wang (Peking University), Yang Liu (Peking University)
ClassificationRecognitionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A new semantic-guided novel category discovery method (Semantic-guided Novel Category Discovery, SNCD) is proposed, which simultaneously completes clustering of unlabeled images and semantic recognition under the name information of existing labeled categories and unknown categories.
Semi-supervised 3D Object Detection with PatchTeacher and PillarMix
Xiaopei Wu (Zhejiang University), Wanli Ouyang (Zhejiang University)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes two key technologies, PatchTeacher and PillarMix, for semi-supervised 3D object detection. It utilizes high-resolution voxelization of partial scenes to generate high-quality pseudo-labels and enhances data diversity through pillar mixing, ultimately achieving better detection performance.
Semi-supervised Active Learning for Video Action Detection
Ayush Singh (Indian Institute of Technology Indian School of Mines Dhanbad), Yogesh Singh Rawat (University of Central Florida)
RecognitionConvolutional Neural NetworkSupervised Fine-TuningVideo
🎯 What it does: A framework for video action detection that combines semi-supervised learning with active learning is proposed, utilizing noise augmentation and FFT high-pass filtering to enhance labeling efficiency.
Semi-Supervised Blind Image Quality Assessment through Knowledge Distillation and Incremental Learning
Wensheng Pan (Xiamen University), Pingyang Dai (Ocean University of China)
Knowledge DistillationTransformerImage
🎯 What it does: A semi-supervised unsupervised BIQA framework SS-IQA is proposed, combining knowledge distillation and incremental learning to address the issue of insufficient labeled data;
Semi-supervised Class-Agnostic Motion Prediction with Pseudo Label Regeneration and BEVMix
Kewei Wang (Huazhong University of Science and Technology), Guosheng Lin (Nanyang Technological University)
Autonomous DrivingSupervised Fine-TuningPoint Cloud
🎯 What it does: This study investigates the application of semi-supervised learning in predicting the motion of unlabeled targets, proposing pseudo-label regeneration and BEVMix enhancement.
Semi-supervised Learning of Dynamical Systems with Neural Ordinary Differential Equations: A Teacher-Student Model Approach
Yu Wang (University of California), Peng Li (Oak Ridge National Laboratory)
Reinforcement LearningTime SeriesOrdinary Differential Equation
🎯 What it does: This paper proposes TS-NODE, a neural ODE method that utilizes a teacher-student model and pseudo rollouts for semi-supervised learning to model dynamic systems when labeled samples are scarce.
Semi-supervised Open-World Object Detection
Sahal Shaji Mullappilly (Mohamed bin Zayed University of Artificial Intelligence), Hisham Cholakkal (Mohamed bin Zayed University of Artificial Intelligence)
Object DetectionTransformerReinforcement LearningImage
🎯 What it does: A semi-supervised open-world object detection framework SS-OWFormer is proposed to address the problem of unknown object detection and incremental learning in the absence of complete labels.
Semi-supervised TEE Segmentation via Interacting with SAM Equipped with Noise-Resilient Prompting
Sen Deng (Hong Kong Polytechnic University), Jing Qin (Chinese University of Hong Kong)
SegmentationPrompt EngineeringBiomedical DataUltrasound
🎯 What it does: This paper proposes a semi-supervised TEE image left atrial appendage segmentation method that combines the large-scale foundation model SAM with a self-reconstruction mechanism.
SemTra: A Semantic Skill Translator for Cross-Domain Zero-Shot Policy Adaptation
Sangwoo Shin (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
Domain AdaptationAutonomous DrivingRobotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelVision Language ModelContrastive LearningVideoMultimodality
🎯 What it does: The SemTra framework is proposed to achieve cross-domain zero-shot strategy adaptation under multimodal task prompts. It decomposes tasks into transferable skill sequences through semantic skill translation and performs parameterized instantiation in the target domain.
SENCR: A Span Enhanced Two-Stage Network with Counterfactual Rethinking for Chinese NER
Hang Zheng (Chongqing University), Li Liu (Chongqing University)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkSupervised Fine-TuningText
🎯 What it does: A dictionary-free Chinese named entity recognition framework called SENCR is proposed, which includes a boundary detector, a convolution-enhanced span classifier, and a counterfactual reasoning strategy during the inference phase.
Separate the Wheat from the Chaff: Model Deficiency Unlearning via Parameter-Efficient Module Operation
Xinshuo Hu (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: By extracting and subtracting the defect capability in the Anti-Expert Parameter Efficient Module (anti-expert PEM), the authenticity and detoxification of large language models are enhanced while maintaining their foundational capabilities.
SeqGPT: An Out-of-the-Box Large Language Model for Open Domain Sequence Understanding
Tianyu Yu (Tsinghua University), Yong Jiang (Zhejiang University)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextSequential
🎯 What it does: SeqGPT achieves zero-shot reasoning across tasks by unifying natural language understanding tasks into two types of atomic tasks, constructing a consistent input-output format.
SeqRank: Sequential Ranking of Salient Objects
Huankang Guan (City University of Hong Kong), Rynson W.H. Lau (City University of Hong Kong)
Object DetectionSegmentationTransformerImage
🎯 What it does: Proposes the SeqRank model, which predicts the attention order of salient objects in an image based on the eye movement mechanism of human vision.
Sequential Fusion Based Multi-Granularity Consistency for Space-Time Transformer Tracking
Kun Hu (National University of Defense Technology), Huibin Tan (National University of Defense Technology)
Object TrackingTransformerVideo
🎯 What it does: This paper proposes a Transformer-based spatial-temporal consistency tracker called STCFormer, which employs a sequential fusion framework and enhances the model's robustness to targets through three types of multi-granularity consistency losses (label consistency, attention consistency, and semantic consistency).
Set Prediction Guided by Semantic Concepts for Diverse Video Captioning
Yifan Lu (Chinese Academy of Sciences), Weiming Hu (Chinese Academy of Sciences)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelVideoText
🎯 What it does: A semantic concept-guided set prediction framework is proposed for generating diverse video captions.
SeTformer Is What You Need for Vision and Language
Pourya Shamsolmoali (East China Normal University), Michael Felsberg (Linkoping University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a Transformer architecture based on Self-optimal Transport (SeT) — SeTformer, to replace traditional dot-product self-attention;
Settling Decentralized Multi-Agent Coordinated Exploration by Novelty Sharing
Haobin Jiang (Peking University), Zongqing Lu (Peking University)
Reinforcement Learning
🎯 What it does: A coordination exploration method named MACE is proposed in distributed multi-agent reinforcement learning, which utilizes limited communication to achieve global novelty estimation and promotes cooperation among agents through hindsight rewards.
SFC: Shared Feature Calibration in Weakly Supervised Semantic Segmentation
Xinqiao Zhao (Xi'an Jiaotong-Liverpool University), Jimin Xiao (Xi'an Jiaotong-Liverpool University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A Shared Feature Calibration (SFC) method is proposed, which improves the pseudo-labels generated by CAM in weakly supervised semantic segmentation through image bank resampling and multi-scale distribution weighted consistency loss, enhancing the final segmentation performance.
SGFormer: Semantic Graph Transformer for Point Cloud-Based 3D Scene Graph Generation
Changsheng Lv (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
Object DetectionGenerationGraph Neural NetworkTransformerLarge Language ModelPoint Cloud
🎯 What it does: This paper proposes a Transformer structure named SGFormer for 3D scene graph generation based on point clouds, which can accurately predict object categories and their relationships through global information transmission.
SGNet: Structure Guided Network via Gradient-Frequency Awareness for Depth Map Super-resolution
Zhengxue Wang (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
RestorationDepth EstimationSuper ResolutionImage
🎯 What it does: A structure-guided network SGNet is proposed, which uses gradient domain and frequency domain information to enhance the super-resolution quality of low-resolution depth maps.
Shadow Generation with Decomposed Mask Prediction and Attentive Shadow Filling
Xinhao Tao (Shanghai Jiao Tong University), Li Niu (Ant Group)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A two-stage DMASNet network is proposed to generate realistic shadows for foreground objects in image synthesis, and a large-scale rendered shadow dataset RdSOBA is constructed.
SHAP@k: Efficient and Probably Approximately Correct (PAC) Identification of Top-K Features
Sanjay Kariyappa (JPMorganChase AI Research), Daniele Magazzeni (JPMorganChase AI Research)
Recommendation SystemAnomaly DetectionOptimizationExplainability and InterpretabilityComputational EfficiencyReinforcement LearningTabularFinance Related
🎯 What it does: This study investigates the identification of Top-k and ordered Top-k feature importance based on SHAP in credit risk models, and proposes KernelSHAP@k, SamplingSHAP@k, and their ordered versions, which offer PAC guarantees with higher sample efficiency.
ShapeBoost: Boosting Human Shape Estimation with Part-Based Parameterization and Clothing-Preserving Augmentation
Siyuan Bian (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
SegmentationPose EstimationConvolutional Neural NetworkImage
🎯 What it does: Proposes the ShapeBoost framework, which utilizes part-based shape parameterization and clothing-preserving data augmentation to achieve high-precision recovery of human shape from a single RGB image.
Shaping Up SHAP: Enhancing Stability through Layer-Wise Neighbor Selection
Gwladys Kelodjou (University of Rennes), Alexandre Termier (University of Rennes)
Explainability and InterpretabilityTabular
🎯 What it does: This paper studies the instability of Kernel SHAP and proposes a ST-SHAP method based on hierarchical neighbor selection to enhance the stability of explanations. It further demonstrates that using only the first layer of neighbors can yield a fast interpretable solution that is highly consistent with SHAP values.
ShareBERT: Embeddings Are Capable of Learning Hidden Layers
Jia Cheng Hu (University of Modena and Reggio Emilia), Alessandro Capotondi (University of Modena and Reggio Emilia)
CompressionOptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes the ShareBERT architecture, which constructs a near-zero parameter Transformer encoder using embedding parameter sharing (EPS and VEPS), thereby compressing the BERT model without using knowledge distillation;
Sharpness-Aware Model-Agnostic Long-Tailed Domain Generalization
Houcheng Su (University of Macau), Zhenghan Chen (Peking University)
ClassificationDomain AdaptationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a combination of smoothness-enhanced adversarial training and maximum squared loss in multi-source domain generalization, aiming to improve the model's performance on unseen domains under long-tailed distributions.
SHaRPose: Sparse High-Resolution Representation for Human Pose Estimation
Xiaoqi An (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Pose EstimationTransformerImage
🎯 What it does: A pose estimation framework based on sparse high-resolution representation, SHaRPose, is proposed.
SHoP: A Deep Learning Framework for Solving High-Order Partial Differential Equations
Tingxiong Xiao (Tsinghua University), Jinli Suo (Tsinghua University)
Explainability and InterpretabilityComputational EfficiencyPhysics RelatedOrdinary Differential Equation
🎯 What it does: This paper proposes the SHoP framework, which utilizes higher-order derivative rules to train MLPs for solving higher-order partial differential equations, and provides explicit solutions through Taylor expansion.