AAAI 2023 Papers — Page 14
AAAI Conference on Artificial Intelligence · 1578 papers
Semi-supervised Learning with Support Isolation by Small-Paced Self-Training
Zheng Xie (Nanjing University), Ming Li (Nanjing University)
ClassificationSupervised Fine-TuningImage
🎯 What it does: This paper proposes a semi-supervised learning framework for supporting spatial isolation (label missing determined by a selection mechanism) — Small-Paced Self-Training, which uses Wasserstein distance to control the distribution difference between the training and pseudo-label subsets, gradually generating reliable pseudo-labels.
Semi-transductive Learning for Generalized Zero-Shot Sketch-Based Image Retrieval
Ce Ge (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
RetrievalConvolutional Neural NetworkImage
🎯 What it does: A semi-transfer learning paradigm is proposed, utilizing unlabeled unseen category images for distribution correction, and achieving cross-modal embedding and ranking learning through a semi-heterogeneous network, addressing the challenge of General Zero-Shot Sketch-Based Image Retrieval (GZS-SBIR).
Semidefinite Programming versus Burer-Monteiro Factorization for Matrix Sensing
Baturalp Yalçın (University of California Berkeley), Somayeh Sojoudi (University of California Berkeley)
OptimizationGraph
🎯 What it does: This paper compares two mainstream solving strategies in matrix sensing/completion problems—semidefinite programming (SDP) and Burer-Monteiro (B-M) factorization—through theoretical analysis and simulations, clarifying their successes and failures under different instances.
SEnsor Alignment for Multivariate Time-Series Unsupervised Domain Adaptation
Yucheng Wang (Nanyang Technological University), Lihua Xie (Nanyang Technological University)
Domain AdaptationRecurrent Neural NetworkGraph Neural NetworkContrastive LearningTime Series
🎯 What it does: This paper studies unsupervised domain adaptation for multivariate time series data and proposes the SEnsor Alignment (SEA) framework, which utilizes local (sensor feature and correlation) and global (global feature) alignment to achieve cross-domain knowledge transfer.
Separate but Equal: Equality in Belief Propagation for Single Cycle Graphs
Erel Cohen (Ben Gurion University of the Negev), Roie Zivan (Ben Gurion University of the Negev)
Graph Neural NetworkGraph
🎯 What it does: This paper studies the phenomenon of belief equality and assignment equality in single-ring graphs under Min-sum belief propagation, proving that even with unary constraints breaking the equality, these equalities cannot be avoided, and provides necessary and sufficient conditions for the occurrence of equality.
SEPT: Towards Scalable and Efficient Visual Pre-training
Yiqi Lin (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
ClassificationObject DetectionRetrievalComputational EfficiencyTransformerContrastive LearningImage
🎯 What it does: The SEPT framework is proposed, which addresses the efficiency and scalability issues of large-scale unlabeled data by retrieving task-related subsets and performing self-supervised pre-training on these subsets.
Sequence Generation with Label Augmentation for Relation Extraction
Bo Li (Peking University), Shikun Zhang (Peking University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study explores the application of sequence generation (Seq2Seq) models to the relation extraction (RE) task, finding that directly generating relation names can leverage the semantic and associative information within those names. Subsequently, the RELA method is proposed, which enhances relation names through three automatic label augmentation strategies: paraphrasing, inquiry, and synonym retrieval using GPT-2, thereby improving model performance. Finally, an in-depth analysis of BART's attention and hidden states in the RE task is conducted.
Set-to-Sequence Ranking-Based Concept-Aware Learning Path Recommendation
Xianyu Chen (Shanghai Jiao Tong University), Yong Yu (Shanghai Jiao Tong University)
Recommendation SystemRecurrent Neural NetworkTransformerReinforcement LearningSequential
🎯 What it does: Proposes modeling the learning path recommendation as a set-to-sequence ranking task and designs the SRC framework to generate personalized learning paths.
ShadowFormer: Global Context Helps Shadow Removal
Lanqing Guo (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
RestorationTransformerImage
🎯 What it does: This paper proposes a single-stage shadow removal network called ShadowFormer based on Transformer, which utilizes global contextual information to restore shadowed areas.
Sharing Pattern Submodels for Prediction with Missing Values
Lena Stempfle (Chalmers University of Technology), Fredrik D. Johansson (Chalmers University of Technology)
Supervised Fine-TuningTabularBiomedical DataAlzheimer's Disease
🎯 What it does: Proposes a Shared Pattern Submodel (SPSM) that achieves predictions in the presence of missing values during testing;
SharpSSAT: A Witness-Generating Stochastic Boolean Satisfiability Solver
Yu-Wei Fan (National Taiwan University), Jie-Hong R. Jiang (National Taiwan University)
TabularBenchmark
🎯 What it does: This paper presents a new SSAT solver called SharpSSAT, which includes the capability to generate Skolem function proofs; it also implements a proof generation scheme for the existing ClauSSat solver.
SheetPT: Spreadsheet Pre-training Based on Hierarchical Attention Network
Ran Jia (Microsoft Research Asia), Dongmei Zhang (Microsoft Research Asia)
TransformerLarge Language ModelTabular
🎯 What it does: Developed the SHEETPT pre-training model, which learns the semantic information of complete spreadsheets through a hierarchical attention network.
ShiftDDPMs: Exploring Conditional Diffusion Models by Shifting Diffusion Trajectories
Zijian Zhang (Zhejiang University), Qi Tian (Huawei Cloud and AI)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes ShiftDDPMs, which introduces a condition-related mean shift during the forward diffusion process to achieve a 'drift' of the conditional diffusion trajectory, thereby utilizing conditional information at all time steps.
Show Me the Way! Bilevel Search for Synthesizing Programmatic Strategies
David S. Aleixo (Universidade Federal de Viçosa), Levi H.S. Lelis
OptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: A dual-layer search algorithm Bi-S is proposed for synthesizing interpretable procedural strategies.
Show, Interpret and Tell: Entity-Aware Contextualised Image Captioning in Wikipedia
Khanh Nguyen (Computer Vision Center Universitat Autonoma de Barcelona), Dimosthenis Karatzas (Computer Vision Center Universitat Autonoma de Barcelona)
GenerationDomain AdaptationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: The study focuses on the task of context-aware image captioning for entities on Wikipedia images, utilizing images, paragraphs, and descriptions to generate captions that are consistent with specific contexts.
SHUNIT: Style Harmonization for Unpaired Image-to-Image Translation
Seokbeom Song (Yonsei University), Euntai Kim (Korea Electronics Technology Institute)
Image TranslationImage HarmonizationObject DetectionSegmentationGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes SHUNIT, a method for unpaired image-to-image translation that achieves 'style harmonization' to convert source domain images into target domain styles while maintaining semantic consistency.
Siamese-Discriminant Deep Reinforcement Learning for Solving Jigsaw Puzzles with Large Eroded Gaps
Xingke Song (University of Nottingham), Ruibin Bai (University of Nottingham)
Convolutional Neural NetworkReinforcement LearningImage
🎯 What it does: Proposes Siamese-Discriminant Deep Reinforcement Learning (SD-RL) to solve the Jigsaw Puzzle with Large Erasure Gaps (JPwLEG) problem.
SigMaNet: One Laplacian to Rule Them All
Stefano Fiorini (University of Milano-Bicocca), Enza Messina (University of Milano-Bicocca)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes SigMaNet, a spectral convolutional neural network capable of handling both directed and undirected graphs with edge weights that can be positive or negative.
Signed Laplacian Graph Neural Networks
Yu Li (Jilin University), Yi Chang (Jilin University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a new Signed Laplacian Graph Neural Network (SLGNN) framework for learning node representations in signed graphs with both positive and negative edges.
Simple and Effective Synthesis of Indoor 3D Scenes
Jing Yu Koh (Google Research), Peter Anderson (Google Research)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImageVideoPoint Cloud
🎯 What it does: Synthesize high-resolution 3D indoor scenes from a small number of indoor images (RGB-D or single RGB), and generate consistent images and videos under large viewpoint changes;
Simple and Efficient Heterogeneous Graph Neural Network
Xiaocheng Yang (Chinese Academy of Sciences), Dongrui Fan (Griffith University)
Computational EfficiencyRepresentation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: This paper proposes a simple and efficient heterogeneous graph neural network, SeHGNN, which utilizes mean aggregation to precompute neighbor information and integrates the semantic features of various long meta-paths through a single-layer structure using Transformer, to achieve node representation learning.
Simulating Network Paths with Recurrent Buffering Units
Divyam Anshumaan (Microsoft Research India), Venkat N. Padmanabhan
Recurrent Neural NetworkTime SeriesSequential
🎯 What it does: This paper proposes a new gray-box model-based network path simulation method called Recurrent Buffering Unit (RBU), aimed at generating realistic end-to-end delay and packet loss sequences under different sending protocols.
Simultaneously Updating All Persistence Values in Reinforcement Learning
Luca Sabbioni (Politecnico di Milano), Marcello Restelli (Politecnico di Milano)
Reinforcement LearningTabular
🎯 What it does: This paper proposes a new All-Persistence Bellman Operator that can simultaneously update action value functions across all time scales in a single experience transfer, and based on this, extends classical Q-learning and DQN to obtain Persistent Q-learning and Persistent DQN.
Skating-Mixer: Long-Term Sport Audio-Visual Modeling with MLPs
Jingfei Xia (Southern University of Science and Technology), Feng Zheng (Harbin Institute of Technology)
TransformerVideoMultimodalityAudio
🎯 What it does: Proposes Skating-Mixer, a multimodal model based on MLP-Mixer, to score long-duration figure skating videos using audio and video information.
SKDBERT: Compressing BERT via Stochastic Knowledge Distillation
Zixiang Ding (Meituan), Wei Lin
CompressionKnowledge DistillationTransformerLarge Language ModelText
🎯 What it does: This paper proposes a Stochastic Knowledge Distillation (SKD) method for compressing the BERT language model, resulting in a smaller and faster SKDBERT.
SKIER: A Symbolic Knowledge Integrated Model for Conversational Emotion Recognition
Wei Li (Nanyang Technological University), Erik Cambria (Nanyang Technological University)
ClassificationRecognitionGraph Neural NetworkSupervised Fine-TuningText
🎯 What it does: A neural symbolic model named SKIER is proposed to recognize emotions in multi-party dialogues, explicitly modeling discourse relations and integrating symbolic knowledge.
SlideVQA: A Dataset for Document Visual Question Answering on Multiple Images
Ryota Tanaka (NTT Corporation), Kuniko Saito (NTT Corporation)
RecognitionObject DetectionExplainability and InterpretabilityTransformerVision Language ModelImageMultimodality
🎯 What it does: A new task of multi-image visual question answering (SlideVQA) is proposed, along with a novel dataset that includes over 2.6k slide groups, over 52k slide images, 14.5k question-answer pairs, approximately 890k bounding boxes, and arithmetic expressions. A unified end-to-end model, M3D, is designed to simultaneously perform evidence selection and answer generation, including the generation of arithmetic expressions in the answers.
SLIQ: Quantum Image Similarity Networks on Noisy Quantum Computers
Daniel Silver (Northeastern University), Devesh Tiwari (Northeastern University)
RecognitionRetrievalOptimizationImageBiomedical Data
🎯 What it does: Designed and implemented SLIQ, a resource-efficient unsupervised image similarity detection network for NISQ-era quantum computers.
SLOTH: Structured Learning and Task-Based Optimization for Time Series Forecasting on Hierarchies
Fan Zhou (Ant Group), Hu Yun (Ant Group)
OptimizationConvolutional Neural NetworkRecurrent Neural NetworkTransformerReinforcement LearningTime Series
🎯 What it does: Proposes the SLOTH framework, which utilizes a hierarchical structure with top-down convolution and bottom-up attention for multi-level time series forecasting, and incorporates a differentiable OptNet optimization layer in the reconciliation step to achieve end-to-end learning and task constraint satisfaction.
Smoothed Online Combinatorial Optimization Using Imperfect Predictions
Kai Wang (Harvard University), Sridhar Mahadevan (Adobe Research)
OptimizationTime Series
🎯 What it does: This paper studies the smooth online combinatorial optimization problem with switching costs and proposes an algorithm that utilizes uncertain prediction dynamic programming windows.
SMT Safety Verification of Ontology-Based Processes
Diego Calvanese (Free University of Bozen-Bolzano), Marco Montali (Free University of Bozen-Bolzano)
Safty and Privacy
🎯 What it does: The research focuses on the security verification of ontology-based processes under RDFS+ ontology, proposing a backward reachability algorithm based on SMT and proving it to be decidable in PSPACE.
Social Bias Meets Data Bias: The Impacts of Labeling and Measurement Errors on Fairness Criteria
Yiqiao Liao (Ohio State University), Parinaz Naghizadeh (Ohio State University)
OptimizationData-Centric LearningTabularFinance Related
🎯 What it does: This paper studies the robustness of different group fairness constraints (DP, TPR, FPR, EO) on model decision thresholds and corporate profits when there are label errors or feature measurement errors in the training data.
Social Relation Reasoning Based on Triangular Constraints
Yunfei Guo (National Laboratory of Pattern Recognition), Cheng-Lin Liu (National Laboratory of Pattern Recognition)
RecognitionObject DetectionRepresentation LearningGraph Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a triangle constraint-based graph attention network (TRGAT) that combines node contrastive learning to identify and infer social relationships between individuals from images.
Socially Optimal Non-discriminatory Restrictions for Continuous-Action Games
Michael Oesterle (University of Mannheim), Guni Sharon (Texas A&M University)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes to enhance the social utility of the minimum Nash equilibrium in continuous action multi-player games through a unified action space constraint (non-discriminatory constraint) and presents the SOAR algorithm to achieve this goal.
Soft Action Priors: Towards Robust Policy Transfer
Matheus Centa (University Lille), Philippe Preux (University Lille)
Knowledge DistillationReinforcement LearningTabular
🎯 What it does: Introducing soft action priors in reinforcement learning and achieving robust policy distillation through an adaptive reward weighting mechanism;
Soft Target-Enhanced Matching Framework for Deep Entity Matching
Wenzhou Dou (Northeastern University), Ge Yu (RMIT University)
Knowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningTabular
🎯 What it does: A STEAM framework is proposed, which enhances the generalization of deep entity matching through soft target augmentation.
SoftCorrect: Error Correction with Soft Detection for Automatic Speech Recognition
Yichong Leng (University of Science and Technology of China), Tie-Yan Liu (Microsoft Research Asia)
RecognitionTransformerSupervised Fine-TuningAudio
🎯 What it does: This paper studies error correction of automatic speech recognition (ASR) outputs and proposes a SoftCorrect system.
Solving Explainability Queries with Quantification: The Case of Feature Relevancy
Xuanxiang Huang (University of Toulouse), Joao Marques-Silva (National University of Singapore)
Explainability and InterpretabilityComputational EfficiencyTabularBenchmark
🎯 What it does: This study investigates the quantification problem of Feature Relevance Queries (FRP) and proposes an algorithm applicable to any machine learning classifier, with experimental validation conducted on random forests.
Solving Large-Scale Pursuit-Evasion Games Using Pre-trained Strategies
Shuxin Li (Nanyang Technological University), Bo An (Nanyang Technological University)
OptimizationMeta LearningGraph Neural NetworkSupervised Fine-TuningReinforcement LearningGraph
🎯 What it does: A two-stage method is proposed, embedding the pre-training/fine-tuning strategy into the PSRO framework to efficiently solve large-scale pursuit-evasion games.
Sparse Coding in a Dual Memory System for Lifelong Learning
Fahad Sarfraz (NavInfo Europe), Bahram Zonooz (Eindhoven University of Technology)
ClassificationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: The SCoMMER method is proposed, which combines sparse activation and multiple memory replay in continual learning, mimicking the brain's sparse coding and multi-memory system to alleviate catastrophic forgetting.
Sparse Maximum Margin Learning from Multimodal Human Behavioral Patterns
Ervine Zheng (Rochester Institute of Technology), Zhi Zheng (Rochester Institute of Technology)
ClassificationExplainability and InterpretabilityRecurrent Neural NetworkMultimodalityBiomedical Data
🎯 What it does: A dynamic multimodal data fusion framework based on a two-layer probabilistic mixture model (SM2-MRS) is proposed, which simultaneously conducts pattern mining and maximum margin learning, utilizing group sparse priors to filter patterns that contribute to classification.
Spatial-Spectral Transformer for Hyperspectral Image Denoising
Miaoyu Li (Beijing Institute of Technology), Yulun Zhang (ETH Zurich)
RestorationTransformerImage
🎯 What it does: A spatial-spectral Transformer (SST) is proposed for hyperspectral image (HSI) denoising.
SpatialFormer: Semantic and Target Aware Attentions for Few-Shot Learning
Jinxiang Lai (Tencent Youtu Lab), Chengjie Wang (Tencent Youtu Lab)
ClassificationMeta LearningTransformerImage
🎯 What it does: This paper proposes a novel attention mechanism based on SpatialFormer, specifically designed for few-shot learning with semantically similar support and query images. It constructs the Semantic and Target Attentions (STA) module (including SFSA and SFTA) and the Novel Task Attention (NTA) module, significantly enhancing the focus on target objects while suppressing background interference.
Spatio-Temporal Meta-Graph Learning for Traffic Forecasting
Renhe Jiang (Toyota Motor Corporation), Toyotaro Suzumura (Toyota Motor Corporation)
Recurrent Neural NetworkGraph Neural NetworkContrastive LearningGraphTime Series
🎯 What it does: A spatiotemporal graph convolutional recurrent network named MegaCRN is proposed for traffic prediction through meta-graph learning.
Spatio-Temporal Neural Structural Causal Models for Bike Flow Prediction
Pan Deng (Beihang University), Mulan Wang (Beihang University)
Graph Neural NetworkTime Series
🎯 What it does: This paper proposes a spatiotemporal neural structural causal model (STNSCM) based on structural causal models, which eliminates confounding interference through the frontdoor criterion and combines dynamic causal graphs with a counterfactual representation reasoning module to predict the flow of bike-sharing systems.
Spatio-Temporal Self-Supervised Learning for Traffic Flow Prediction
Jiahao Ji (Beihang University), Yu Zheng (JD Technology)
Graph Neural NetworkContrastive LearningTime Series
🎯 What it does: A spatial-temporal model ST-SSL based on self-supervised learning is proposed for urban traffic flow prediction.
Spatiotemporal Deformation Perception for Fisheye Video Rectification
Shangrong Yang (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
Image TranslationRestorationConvolutional Neural NetworkGenerative Adversarial NetworkOptical FlowVideo
🎯 What it does: A fisheye video correction framework based on spatiotemporal deformation perception is proposed, achieving stable and accurate correction through optical flow estimation and enhancement.
Spearman Rank Correlation Screening for Ultrahigh-Dimensional Censored Data
Hongni Wang (Shandong University of Finance and Economics), Xiaodong Yan (Shandong University)
Biomedical Data
🎯 What it does: This paper proposes a model-independent feature selection method based on Spearman rank correlation (SRCS cen) for high-dimensional right-censored data.
Spectral Feature Augmentation for Graph Contrastive Learning and Beyond
Yifei Zhang (Chinese University of Hong Kong), Irwin King (Chinese University of Hong Kong)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningImageGraph
🎯 What it does: This paper proposes a Spectral Feature Augmentation (SFA) method based on incomplete power iteration, aimed at rebalancing the singular values of the feature matrix and injecting noise in graph contrastive learning (GCL) and image contrastive learning, thereby enhancing the alignment and generalization capabilities of representations.
SplitNet: A Reinforcement Learning Based Sequence Splitting Method for the MinMax Multiple Travelling Salesman Problem
Hebin Liang (Tianjin University), Jianye Hao (Tianjin University)
OptimizationTransformerReinforcement LearningSequential
🎯 What it does: SplitNet is proposed, which obtains a MinMax multi-vehicle TSP solution by splitting a single TSP solution into multiple subsequences and connecting back to the starting point.
Splitting Answer Set Programs with Respect to Intensionality Statements
Jorge Fandinno (University of Nebraska at Omaha), Yuliya Lierler (University of Nebraska at Omaha)
🎯 What it does: This paper proposes a method for answer set programming (ASP) program partitioning based on 'intensionality statements', which can decompose an entire program into several subprograms while considering the parameters of predicates and their context, and maintain the stability of the models.
SPRING: Situated Conversation Agent Pretrained with Multimodal Questions from Incremental Layout Graph
Yuxing Long (Beijing University of Posts and Telecommunications), Xiaojie Wang (Beijing University of Posts and Telecommunications)
TransformerMultimodality
🎯 What it does: This paper proposes a multimodal question-answering pre-training framework called SPRING, which is automatically generated based on Incremental Layout Graphs (ILG), significantly enhancing the understanding and generation of visual attributes and spatial relationships in virtual scene dialogues.
SRoUDA: Meta Self-Training for Robust Unsupervised Domain Adaptation
Wanqing Zhu (Fuzhou University), Ximeng Liu (Yuan Ze University)
Domain AdaptationAdversarial AttackMeta LearningImage
🎯 What it does: A self-supervised learning-based meta-training framework SRoUDA is proposed to enhance the adversarial robustness of unsupervised domain adaptation models.
SSDA3D: Semi-supervised Domain Adaptation for 3D Object Detection from Point Cloud
Yan Wang (Beijing Institute of Technology), Jianbing Shen (University of Macau)
Object DetectionDomain AdaptationAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes a framework for semi-supervised domain adaptation (SSDA) in 3D object detection, named SSDA3D, which addresses the significant domain gap between the source and target domains and enhances performance using a small amount of target annotations.
SSMI: Semantic Similarity and Mutual Information Maximization Based Enhancement for Chinese NER
Pengnian Qi (Renmin University of China), Biao Qin (Renmin University of China)
RecognitionRecurrent Neural NetworkTransformerContrastive LearningText
🎯 What it does: This paper proposes a Chinese named entity recognition model based on semantic similarity and mutual information maximization, called SSMI. It first filters the word boundaries most relevant to character semantics from a pre-trained word list, then concatenates these words with character vectors and enhances them through local/global mutual information before feeding them into a multi-layer network for entity recognition.
SSPAttack: A Simple and Sweet Paradigm for Black-Box Hard-Label Textual Adversarial Attack
Han Liu (Dalian University of Technology), Xianchao Zhang (Dalian University of Technology)
Adversarial AttackText
🎯 What it does: This study proposes a black-box hard-label text adversarial attack method based on simple word replacement—SSPAttack.
Stability-Based Generalization Analysis for Mixtures of Pointwise and Pairwise Learning
Jiahuan Wang (Huazhong Agricultural University), Xin Tang (Ping An Property and Casualty Insurance Company)
Recommendation System
🎯 What it does: This paper systematically derives the generalization error upper bound of the point-to-point learning and pairwise learning hybrid framework (PPL) based on algorithm stability theory, and for the first time provides a high-probability generalization and convergence rate analysis of PPL under two types of algorithms: SGD and RRM.
Stability-Based Generalization Analysis of the Asynchronous Decentralized SGD
Xiaoge Deng (National University of Defense Technology), Dongsheng Li (National University of Defense Technology)
OptimizationImage
🎯 What it does: This paper, based on the theory of algorithm stability, conducts a quantitative analysis of the generalization error and the super generalization error of Asynchronous Decentralized Stochastic Gradient Descent (AD-SGD) for strongly convex, convex, and non-convex cases for the first time, and provides theoretical upper bounds.
Stable Learning via Sparse Variable Independence
Han Yu (Tsinghua University), Xingxuan Zhang (Tsinghua University)
ClassificationOptimizationTabularStochastic Differential Equation
🎯 What it does: The Sparse Variable Independence (SVI) algorithm is proposed, which addresses the stability prediction problem caused by covariate shift under limited samples by incorporating sparse constraints and iterative feedback based on sample weighting.
STAGE: Span Tagging and Greedy Inference Scheme for Aspect Sentiment Triplet Extraction
Shuo Liang (Huazhong University of Science and Technology), Dangyang Chen (Ping An Property and Casualty Insurance Company of China)
TransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes the STAGE framework, which employs span-level labeling and greedy inference for the Aspect Sentiment Triplet Extraction (ASTE) task, constructing an end-to-end BERT-based model capable of simultaneously identifying aspect terms, opinion terms, and their sentiment polarity.
STARS: Spatial-Temporal Active Re-sampling for Label-Efficient Learning from Noisy Annotations
Dayou Yu (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
ClassificationData-Centric LearningConvolutional Neural NetworkReinforcement LearningTabularBiomedical Data
🎯 What it does: Proposes the use of Space-Time Active Resampling (STARS) in active learning with noisy labels to enhance labeling efficiency and generalization performance.
State-Conditioned Adversarial Subgoal Generation
Vivienne Huiling Wang (Tampere University), Joni-Kristian Kämäräinen (Aalto University)
Robotic IntelligenceReinforcement LearningGenerative Adversarial NetworkSequential
🎯 What it does: In goal-conditioned hierarchical reinforcement learning, a generative adversarial subgoal generation method (SAGA) is proposed, which alleviates the non-stationarity problem in offline training by allowing the high-level policy to generate subgoals compatible with the current low-level policy.
Steganography of Steganographic Networks
Guobiao Li (Fudan University), Zhenxing Qian (Fudan University)
Convolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This study investigates a method to disguise a secret deep neural network model as a model performing ordinary tasks, achieving covert transmission and recovery of the secret model.
Stepdown SLOPE for Controlled Feature Selection
Jingxuan Liang (Huazhong Agricultural University), Xin Tang (Ping An Property and Casualty Insurance Company)
🎯 What it does: This paper proposes two step-down methods based on SLOPE, k-SLOPE and F-SLOPE, to control k-FWER and FDP in high-dimensional feature selection.
StereoDistill: Pick the Cream from LiDAR for Distilling Stereo-Based 3D Object Detection
Zhe Liu (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
Object DetectionAutonomous DrivingKnowledge DistillationPoint Cloud
🎯 What it does: This paper proposes a cross-modal distillation framework called StereoDistill to enhance the performance of 3D object detection based on stereo cameras.
STOA-VLP: Spatial-Temporal Modeling of Object and Action for Video-Language Pre-training
Weihong Zhong (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
Object DetectionRetrievalTransformerVision Language ModelVideoTextMultimodality
🎯 What it does: The STOA-VLP framework is designed to jointly model object trajectories and multiple action information in videos, guiding the model to learn more accurate cross-modal representations through two fine-grained auxiliary tasks (Object-Text Alignment OTA and Action Set Prediction ASP).
Stochastic Contextual Bandits with Long Horizon Rewards
Yuzhen Qin (University of California), Samet Oymak (University of Michigan)
Reinforcement LearningTime Series
🎯 What it does: This paper proposes a new context-linear multi-armed bandit model, where the reward varies according to the weighted sum of at most s previous contexts in the time series, and the time window h can be very large. Two sample-efficient algorithms utilizing sparse structures (Doubling Lasso and Adaptive Doubling Lasso) are provided for both data-scarce and data-rich scenarios, and asymptotic upper bounds on the reward are proven to have no polynomial relationship with h.
Stop-Gradient Softmax Loss for Deep Metric Learning
Lu Yang (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
RetrievalConvolutional Neural NetworkImage
🎯 What it does: This paper proposes two techniques: Stop-Gradient Softmax Loss (SGSL) and Remove the last BN-ReLU (RBR), aimed at deep metric learning under L2-normalization conditions, addressing the convergence difficulties of softmax training and enhancing feature clustering effects.
Store and Fetch Immediately: Everything Is All You Need for Space-Time Video Super-resolution
Mengshun Hu (Wuhan University), Zheng Wang (Peking University)
RestorationSuper ResolutionConvolutional Neural NetworkVideo
🎯 What it does: A spatiotemporal video super-resolution network based on a store-and-fetch mechanism is proposed, which captures super-resolution information from the past, present, and future in video sequences simultaneously through forward and backward recursive modules (FRM, BRM), achieving a thorough exploration of long-distance spatiotemporal correlations.
Strategic Facility Location with Clients That Minimize Total Waiting Time
Simon Krogmann (Hasso Plattner Institute), Alexander Skopalik (University of Twente)
Optimization
🎯 What it does: A bilateral facility location game is proposed, studying the strategic allocation of customers under the minimization of total waiting time, analyzing the mutual influence between facilities and customers, proving the uniqueness and polynomial solvability of customer equilibrium, demonstrating that subgame perfect equilibrium does not necessarily exist and its determination is NP-hard, but providing a polynomial-time solvable 3-approximation of subgame perfect equilibrium.
Strategyproofness and Proportionality in Party-Approval Multiwinner Elections
Théo Delemazure (Paris Dauphine University), Patrick Lederer (Technical University of Munich)
🎯 What it does: This paper studies the incompatibility of strategic untrustworthiness and proportional representation in party-approved multi-winner elections; it proposes weak strategic untrustworthiness and proves that CCAV is the only Thiele rule that satisfies this property.
Stroke Extraction of Chinese Character Based on Deep Structure Deformable Image Registration
Meng Li (Institute of Automation, Chinese Academy of Sciences), Jian Wang (Institute of Automation, Chinese Academy of Sciences)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A method for extracting Chinese character strokes based on deep learning is proposed, utilizing structure-variable image registration, semantic segmentation, and single-stroke extraction networks to achieve precise separation and matching of strokes.
Structurally Restricted Fragments of Numeric Planning – a Complexity Analysis
Alexander Shleyfman (Bar Ilan University), Peter Jonsson (Linkoping University)
Optimization
🎯 What it does: This paper studies the complexity of Simple Numerical Planning (SNP) under constrained structures and proposes achieving decidability and PSPACE solvability by restricting the positions of numerical variables in the Causal Graph (CG).
Structure Aware Incremental Learning with Personalized Imitation Weights for Recommender Systems
Yuening Wang (Huawei Noah's Ark Lab), Mark Coates (McGill University)
Recommendation SystemKnowledge DistillationGraph Neural NetworkTabular
🎯 What it does: This paper addresses the incremental learning problem in graph neural network recommendation systems and proposes an adaptive personalized imitation weight learning strategy that can dynamically adjust the strength of knowledge distillation based on changes in user interest distribution, balancing the contributions of new and old data.
Structure Flow-Guided Network for Real Depth Super-resolution
Jiayi Yuan (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
RestorationDepth EstimationSuper ResolutionOptical FlowImageMultimodality
🎯 What it does: A structure flow-guided deep depth map super-resolution framework (SFG) is proposed to recover structural distortions and edge noise in real scenes.
Structured BFGS Method for Optimal Doubly Stochastic Matrix Approximation
Dejun Chu (Hefei University of Technology), Qing Tao (Tsinghua University)
OptimizationTabular
🎯 What it does: A structured BFGS algorithm is proposed to find the optimal approximation of a given matrix on the Birkhoff polytope (i.e., the set of doubly stochastic matrices), transforming the problem into unconstrained smooth optimization using the dual-dual form;
Structured Case-Based Reasoning for Inference-Time Adaptation of Text-to-SQL Parsers
Abhijeet Awasthi (Indian Institute of Technology Bombay), Sunita Sarawagi (Indian Institute of Technology Bombay)
Domain AdaptationOptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a structured case-based reasoning (StructCBR) method that does not require retraining parameters during inference, aimed at quickly adapting Text-to-SQL models to new database schemas.
Style-Content Metric Learning for Multidomain Remote Sensing Object Recognition
Wenda Zhao (Dalian University of Technology), You He (Tsinghua University)
RecognitionObject DetectionContrastive LearningImage
🎯 What it does: A style-content metric learning framework is proposed to address the generalization problem in multi-domain remote sensing object recognition.
StyleTalk: One-Shot Talking Head Generation with Controllable Speaking Styles
Yifeng Ma (Tsinghua University), Xin Yu (University of Technology Sydney)
GenerationTransformerGenerative Adversarial NetworkImageVideoAudio
🎯 What it does: A controllable talking head generation framework called StyleTalk is proposed, which can synthesize realistic talking videos with any reference video speaking style from a single target image and a segment of audio.
Submodular Maximization under the Intersection of Matroid and Knapsack Constraints
Yu-Ran Gu (Nanjing University), Chao Qian (Nanjing University)
Recommendation SystemOptimizationSimultaneous Localization and MappingGraph
🎯 What it does: Two algorithms, SPROUT and SPROUT++, are proposed to solve the non-monotonic submodular maximization problem under k-matroid and m-knapsack constraints;
Subspace-Aware Exploration for Sparse-Reward Multi-Agent Tasks
Pei Xu (University of Chinese Academy of Sciences), Kaiqi Huang (University of Chinese Academy of Sciences)
Reinforcement Learning
🎯 What it does: This study explores the exploration problem in multi-agent sparse reward environments, proposing a low-cost algorithm called SAME that utilizes a reward function relying solely on the structural prior of the state subspace to design a subspace entropy exploration objective.
Substructure Aware Graph Neural Networks
DingYi Zeng, Hong Qu (University of Electronic Science and Technology of China)
Graph Neural NetworkGraph
🎯 What it does: A substructure-based graph neural network framework (SAGNN) is proposed, which enhances the expressive power of traditional GNNs by encoding subgraphs obtained from the original graph through continuous and selective edge deletion into Cut subgraphs and Ego subgraphs, using the return probability of random walks to inject the encoded results into node features.
SumREN: Summarizing Reported Speech about Events in News
Revanth Gangi Reddy (University of Illinois Urbana-Champaign), Heng Ji (University of Macau)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes the task of summarizing reporting statements from different speakers about events in news texts and constructs the SUMREN benchmark dataset.
Super-efficient Echocardiography Video Segmentation via Proxy- and Kernel-Based Semi-supervised Learning
Huisi Wu (Shenzhen University), Jing Qin (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkVideoUltrasound
🎯 What it does: A semi-supervised cardiac ultrasound video segmentation network based on agents and convolutional kernels, PKEcho-Net, is proposed to achieve real-time segmentation of the left ventricular endocardium.
Superpoint Transformer for 3D Scene Instance Segmentation
Jiahao Sun (South China University of Technology), Xiangmin Xu (South China University of Technology)
Object DetectionSegmentationTransformerPoint Cloud
🎯 What it does: Proposes an end-to-end 3D instance segmentation method SPFormer based on Superpoint Transformer.
Supervised Contrastive Few-Shot Learning for High-Frequency Time Series
Xi Chen (Alibaba Group), Jin Wang (Alibaba Group)
ClassificationAnomaly DetectionRepresentation LearningConvolutional Neural NetworkContrastive LearningTime Series
🎯 What it does: A supervised contrastive learning framework SCFSL has been developed for few-shot representation learning and classification of high-frequency vibration time series.
Sustaining Fairness via Incremental Learning
Somnath Basu Roy Chowdhury (University of North Carolina at Chapel Hill), Snigdha Chaturvedi (University of North Carolina at Chapel Hill)
Adversarial AttackTabular
🎯 What it does: Proposes the FaIRL framework, which simultaneously learns fair representations and maintains performance on new tasks in an incremental learning environment.
SVFI: Spiking-Based Video Frame Interpolation for High-Speed Motion
Lujie Xia (Peking University), Tiejun Huang (Peking University)
RestorationData SynthesisSpiking Neural NetworkOptical FlowVideo
🎯 What it does: A dual-modal video frame interpolation framework SVFI is proposed, which generates intermediate frames using the high temporal resolution binary pulse stream produced by a pulsed camera in conjunction with traditional RGB frames.
SVP-T: A Shape-Level Variable-Position Transformer for Multivariate Time Series Classification
Rundong Zuo (Hong Kong Baptist University), Grace L.H. Wong (Chinese University of Hong Kong)
ClassificationTransformerTime Series
🎯 What it does: A shape-level variable-position transformer (SVP-T) is proposed for multivariate time series classification, using subsequences (shapes) as input tokens and modeling the interdependencies between variables and time intervals through a transformer.
SWBNet: A Stable White Balance Network for sRGB Images
Chunxiao Li (Beijing University of Posts and Telecommunications), Anlong Ming (Beijing University of Posts and Telecommunications)
TransformerContrastive LearningImage
🎯 What it does: Proposes a stable white balance network SWBNet to address the instability of white balance in sRGB images.
SwiftAvatar: Efficient Auto-Creation of Parameterized Stylized Character on Arbitrary Avatar Engines
Shizun Wang (Beijing University of Posts and Telecommunications), Jing Liu (Douyin Vision)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an unsupervised SwiftAvatar framework that can automatically generate avatar parameters that meet user needs based on user selfies in any stylized avatar engine.
SwinRDM: Integrate SwinRNN with Diffusion Model towards High-Resolution and High-Quality Weather Forecasting
Lei Chen (Alibaba Group), Fan Wang (Alibaba Group)
Recurrent Neural NetworkTransformerDiffusion modelTime Series
🎯 What it does: A framework called SwinRDM is proposed, which integrates an improved SwinRNN (SwinRNN+) with a conditional diffusion model for medium-term weather forecasting at a resolution of 0.25°.
SWL-Adapt: An Unsupervised Domain Adaptation Model with Sample Weight Learning for Cross-User Wearable Human Activity Recognition
Rong Hu (Zhejiang University), Xing Tang (Zhejiang University)
RecognitionDomain AdaptationMeta LearningTime Series
🎯 What it does: This paper proposes an unsupervised domain adaptation model SWL-Adapt, which achieves adaptive cross-user wearable human activity recognition through sample weight learning.
Symbolic Metamodels for Interpreting Black-Boxes Using Primitive Functions
Mahed Abroshan (Alan Turing Institute), Mohammad Mahdi Khalili (Yahoo Research)
OptimizationExplainability and InterpretabilityTabular
🎯 What it does: A symbolic meta-modeling method based on the Kolmogorov Superposition Theorem (SMPF) is proposed, which uses genetic programming to search tree structures and gradient descent to train parameters, generating interpretable black-box approximation models.
Symbolic Replay: Scene Graph as Prompt for Continual Learning on VQA Task
Stan Weixian Lei (National University of Singapore), Mike Zheng Shou (Agency for Science Technology and Research)
RecognitionData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextMultimodalityBenchmark
🎯 What it does: A benchmark called CLOVE is proposed, specifically designed to evaluate the continual learning capability of Visual Question Answering (VQA), covering two learning settings: scene increment and function increment.
Symmetry-Aware Transformer-Based Mirror Detection
Tianyu Huang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
Object DetectionSegmentationTransformerImage
🎯 What it does: A dual-path symmetric perception Transformer network (SATNet) is proposed for mirror detection.
Symphony in the Latent Space: Provably Integrating High-Dimensional Techniques with Non-linear Machine Learning Models
Qiong Wu (College of William and Mary), Mihai Cucuringu (University of Oxford)
Recommendation SystemAnomaly DetectionOptimizationTabularTime SeriesFinance Related
🎯 What it does: An additive impact model that separates entity interaction and feature nonlinearity learning is proposed, with theoretical guarantees provided.
Synchronization and Diversity of Solutions
Emmanuel Arrighi (University of Bergen), Petra Wolf (University of Trier)
🎯 What it does: This paper studies the diversity problem of the set of minimal synchronizing words for a given deterministic finite automaton, proposing a definition of diversity based on edit distance and subsequence minimality. It also provides corresponding parameterized algorithms and complexity analysis, while extending this framework to the field of compatible planning.
Synthetic Data Can Also Teach: Synthesizing Effective Data for Unsupervised Visual Representation Learning
Yawen Wu (University of Pittsburgh), Jingtong Hu (University of Pittsburgh)
GenerationData SynthesisRepresentation LearningGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A joint training framework for synthetic data generation is proposed, utilizing a generator to dynamically produce hard samples and hard positive samples to enhance the representation quality of unsupervised contrastive learning.
T-distributed Spherical Feature Representation for Imbalanced Classification
Xiaoyu Yang (Tongji University), Chao Ma (Changhai Hospital of Shanghai)
ClassificationRecognitionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes the t-distribution spherical metric (tSP) and the corresponding tSP classifier to address the issue of feature space congestion in extremely imbalanced classification, achieving feature space balance and performance improvement in visual classification tasks.