NeurIPS 2023 Papers — Page 28
Conference on Neural Information Processing Systems · 3218 papers
SpatialRank: Urban Event Ranking with NDCG Optimization on Spatiotemporal Data
BANG AN, Tianbao Yang (Texas A&M University)
Recommendation SystemOptimizationRecurrent Neural NetworkGraph Neural NetworkTabularTime Series
🎯 What it does: A deep learning framework called SpatialRank is proposed for ranking risk locations of urban events (such as traffic accidents and crimes).
Spatio-Angular Convolutions for Super-resolution in Diffusion MRI
Matthew Lyon (University of Manchester), Mauricio A Álvarez (University of Manchester)
RestorationSuper ResolutionConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingDiffusion Tensor Imaging
🎯 What it does: In diffusion magnetic resonance imaging (dMRI), a PCCNN network is constructed using parameterized continuous convolution (PCConv) to achieve high angular resolution super-resolution from low angular resolution dMRI, which is then used for subsequent FBA and NODDI analysis.
SpecTr: Fast Speculative Decoding via Optimal Transport
Ziteng Sun (Google Research), Felix Yu (Google Research)
GenerationOptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes SpecTr, a multi-candidate explicit sampling method improved by optimal transport theory, aimed at accelerating autoregressive inference of large language models.
Spectral Co-Distillation for Personalized Federated Learning
Zihan Chen (Singapore University of Technology and Design), Kai Fong Ernest Chong (Singapore University of Technology and Design)
Federated LearningKnowledge DistillationImage
🎯 What it does: A spectral co-distillation framework and a no-wait local training protocol are proposed to jointly train a global model and a personalized model, enhancing the personalization performance of Federated Learning.
Spectral Entry-wise Matrix Estimation for Low-Rank Reinforcement Learning
Stefan Stojanovic (KTH), Alexandre Proutiere (KTH)
OptimizationReinforcement Learning
🎯 What it does: The research focuses on low-rank reinforcement learning and low-rank bandit problems, utilizing spectral decomposition to estimate low-rank matrices (reward matrices or transition matrices), and provides theoretical guarantees for entry-level errors and subspace recovery based on spectral methods; it constructs a regret reduction algorithm SME-AE for low-rank bandits based on these estimates, as well as an optimal policy identification algorithm for reward-free low-rank MDPs, along with corresponding sample complexity and performance upper bounds.
Spectral Evolution and Invariance in Linear-width Neural Networks
Zhichao Wang (University of California San Diego), Tony Chiang (Pacific Northwest National Laboratory)
OptimizationRepresentation LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningImageText
🎯 What it does: Under the setting of linear width ratio (equal sample size, input dimension, and hidden width ratio), the spectral evolution of a fully connected two-layer network during the training process with different optimizers and learning rates is systematically studied. Through experiments and theoretical analysis, the spectral invariance, emergence of outliers, and heavy-tail behaviors of the weight matrix, conjugate kernel (CK), and neural tangent kernel (NTK) are demonstrated and analyzed, along with their associations with feature learning and generalization performance.
Spectral Invariant Learning for Dynamic Graphs under Distribution Shifts
Zeyang Zhang (Tsinghua University), Wenwu Zhu (Tsinghua University)
Domain AdaptationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Proposes to capture and utilize variable and invariant patterns of dynamic spectra in the frequency domain to achieve robust predictions against distribution shifts.
Speculative Decoding with Big Little Decoder
Sehoon Kim (University of California), Kurt Keutzer (University of California)
GenerationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The Big Little Decoder (BiLD) framework is proposed, which generates text collaboratively using a small model (autoregressive) and a large model (non-autoregressive), significantly reducing inference latency through fallback and rollback strategies without altering the training process.
Spike-driven Transformer
Man Yao (Institute of Automation Chinese Academy of Sciences), Guoqi Li (Institute of Automation Chinese Academy of Sciences)
ClassificationRecognitionSpiking Neural NetworkTransformerImage
🎯 What it does: Proposes Spike-Driven Transformer, which integrates the spike-driven principle of SNN with Transformer, allowing the network to achieve attention and MLP computation solely through sparse addition.
Spiking PointNet: Spiking Neural Networks for Point Clouds
Dayong Ren (Intelligent Science and Technology Academy of CASIC), Yufei Guo (Scientific Research Laboratory of Aerospace Intelligent Systems and Technology)
ClassificationSpiking Neural NetworkPoint Cloud
🎯 What it does: This paper studies the use of spiking neural networks in point cloud classification tasks and proposes the Spiking PointNet model.
Spontaneous symmetry breaking in generative diffusion models
Gabriel Raya (Jheronimus Academy of Data Science), Luca Ambrogioni (Radboud University)
GenerationData SynthesisDiffusion modelImageStochastic Differential Equation
🎯 What it does: This paper studies the phenomenon of spontaneous symmetry breaking in the generation process of diffusion models and utilizes this phenomenon to improve fast samplers.
SPQR: Controlling Q-ensemble Independence with Spiked Random Model for Reinforcement Learning
Dohyeok Lee (Seoul National University), Jungwoo Lee (Seoul National University)
Reinforcement LearningSequential
🎯 What it does: A Q-ensemble independence regularization method SPQR based on random matrix theory is proposed, which uses the KL divergence between the eigenvalue distribution and the Wigner semicircle distribution to reduce the correlation of the Q-ensemble and suppress estimation bias.
SPRING: Studying Papers and Reasoning to play Games
Yue Wu (Carnegie Mellon University), Yuanzhi Li (Microsoft Research)
TransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: By reading the LaTeX source code of the Crafter game paper and utilizing LLM for chain reasoning, a zero-training game agent SPRING is achieved.
Spuriosity Didn’t Kill the Classifier: Using Invariant Predictions to Harness Spurious Features
Cian Eastwood (University of Edinburgh), Bernhard Schölkopf (Max Planck Institute for Intelligent Systems)
Domain AdaptationImage
🎯 What it does: The Stable Feature Boosting (SFB) algorithm is proposed, which utilizes stable feature predictions to adjust and leverage unstable (artifact) features in an unsupervised manner to enhance domain generalization performance.
Spuriosity Rankings: Sorting Data to Measure and Mitigate Biases
Mazda Moayeri (University of Maryland), Soheil Feizi (University of Maryland)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a method based on 'spuriosity ranking' that measures and ranks the occurrence of spurious cues in images by utilizing robust neural features from interpretable models. This allows for the identification and quantification of the model's bias towards spurious cues on existing data, and fine-tuning the classification head on low-spurious cue samples to reduce bias.
SQ Lower Bounds for Learning Mixtures of Linear Classifiers
Ilias Diakonikolas (University of Wisconsin Madison), Yuxin Sun (University of Wisconsin Madison)
ClassificationOptimization
🎯 What it does: Under high-dimensional Gaussian covariates, a lower bound for the learning problem of uniform mixture linear classifiers is provided under statistical queries (SQ), proving that the known algorithms are nearly optimal.
SQ Lower Bounds for Non-Gaussian Component Analysis with Weaker Assumptions
Ilias Diakonikolas (University of Wisconsin Madison), Yuxin Sun (University of Wisconsin Madison)
🎯 What it does: This paper studies the complexity of Non-Gaussian Component Analysis (NGCA) under the Statistical Query (SQ) model and proves a near-optimal SQ lower bound that only satisfies moment matching conditions.
Squared Neural Families: A New Class of Tractable Density Models
Russell Tsuchida (Data61 CSIRO), Dino Sejdinovic (University of Adelaide)
GenerationOptimizationFlow-based ModelTabular
🎯 What it does: A class of interpretable probability density models called Squared Neural Family (SNEFY) is proposed, which constructs distributions by normalizing the squared 2-norm of a single hidden layer neural network with respect to a base measure.
Squeeze, Recover and Relabel: Dataset Condensation at ImageNet Scale From A New Perspective
Zeyuan Yin (Mohamed bin Zayed University of AI), Zhiqiang Shen (Mohamed bin Zayed University of AI)
Data SynthesisOptimizationImage
🎯 What it does: Through a three-stage Squeeze–Recover–Relabel process, large-scale image datasets are compressed into a small number of synthetic samples.
Stability and Generalization of the Decentralized Stochastic Gradient Descent Ascent Algorithm
Miaoxi Zhu (Wuhan University), Dacheng Tao (University of Sydney)
OptimizationGenerative Adversarial NetworkTabularSequential
🎯 What it does: This paper studies the algorithm stability and generalization performance of Distributed Stochastic Gradient Descent Ascent (D-SGDA) in decentralized min-max optimization.
Stability Guarantees for Feature Attributions with Multiplicative Smoothing
Anton Xue (University of Pennsylvania), Eric Wong (University of Pennsylvania)
Explainability and InterpretabilityVision Language ModelImageText
🎯 What it does: This paper analyzes the stability of feature attribution methods and proposes a method called Multiplicative Smoothing (MuS) to ensure model stability when features are masked, providing theoretical guarantees.
Stability of Random Forests and Coverage of Random-Forest Prediction Intervals
Yan Wang (Wayne State University), Dan Nettleton (Iowa State University)
Tabular
🎯 What it does: This study investigates the stability of Random Forests (RF) under practical implementations (such as the R language randomForest package) and constructs prediction intervals (PI) with coverage guarantees based on this stability.
Stability-penalty-adaptive follow-the-regularized-leader: Sparsity, game-dependency, and best-of-both-worlds
Taira Tsuchiya (University of Tokyo), Junya Honda (Kyoto University)
OptimizationReinforcement Learning from Human FeedbackTabular
🎯 What it does: This paper proposes a new adaptive learning rate framework called Stability-Penalty Adaptive (SPA) learning rate, aimed at enhancing the adaptability of the FTRL algorithm in multi-armed bandit and partially monitored problems.
Stabilized Neural Differential Equations for Learning Dynamics with Explicit Constraints
Alistair White (Technical University of Munich), Niklas Boers (University of Exeter)
Time SeriesSequentialPhysics RelatedStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A Stabilized Neural Differential Equation (SNDE) is proposed, which enforces the solution trajectory to remain on any explicit constraint manifold by adding a penalty term based on the constrained Jacobian to the neural ODE.
Stable and low-precision training for large-scale vision-language models
Mitchell Wortsman (University of Washington), Ludwig Schmidt (Allen Institute for AI)
OptimizationComputational EfficiencyTransformerVision Language ModelImage
🎯 What it does: This paper proposes the SwitchBack linear layer for implementing int8/float8 low-precision training and develops the StableAdamW hybrid optimizer to reduce loss bursts during the training of CLIP ViT-Huge.
Stable Diffusion is Unstable
Chengbin Du (University of Sydney), Chang Xu (University of Sydney)
GenerationAdversarial AttackTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper studies the instability of Stable Diffusion under slight perturbations of text prompts and proposes an automatic attack method called ATM based on Gumbel-Softmax, which can generate perturbations similar to the original prompts but lead to generation failures.
Stable Nonconvex-Nonconcave Training via Linear Interpolation
Thomas Pethick (École Polytechnique Fédérale de Lausanne), Volkan Cevher (École Polytechnique Fédérale de Lausanne)
GenerationOptimizationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes and analyzes the stability of linear interpolation techniques in non-convex and non-concave optimization, introduces a new double-loop algorithm RAPP, and incorporates the Lookahead algorithm into the Krasnosel'skiĭ–Mann iterative framework, providing the final iteration convergence theory under the co-hypomonotone problem.
Stable Vectorization of Multiparameter Persistent Homology using Signed Barcodes as Measures
David Loiseaux (Inria d'Université Côte d'Azur), Steve Oudot (Inria d'Université Côte d'Azur)
ClassificationDrug DiscoveryGraphTime Series
🎯 What it does: This paper presents a complete feature generation pipeline that treats the signature barcode of multiparameter persistent homology as a signature measure, and provides two stable vectorization methods (convolutional images and sliced Wasserstein kernels) to achieve efficient vectorization of multiparameter persistent homology.
StableFDG: Style and Attention Based Learning for Federated Domain Generalization
Jungwuk Park (Korea Advanced Institute of Science and Technology), Jaekyun Moon (Korea Advanced Institute of Science and Technology)
Domain AdaptationFederated LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A domain generalization method called StableFDG is proposed for the federated learning scenario, which can enhance the model's prediction performance on unknown domains in clients with insufficient data.
StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learners
Yonglong Tian (Google Research), Dilip Krishnan (Google Research)
ClassificationSegmentationGenerationRepresentation LearningTransformerDiffusion modelContrastive LearningImageText
🎯 What it does: Using synthetic images generated by the text-to-image model Stable Diffusion for visual representation learning.
Star-Shaped Denoising Diffusion Probabilistic Models
Andrey Okhotin (Higher School of Economics University), Dmitry P. Vetrov
RestorationGenerationData SynthesisDiffusion modelImageText
🎯 What it does: A star-shaped denoising diffusion probabilistic model (SS-DDPM) is proposed, which can implement diffusion models on any exponential family distribution through a non-Markovian forward process that relies only on the marginal distribution.
State Regularized Policy Optimization on Data with Dynamics Shift
Zhenghai Xue (Nanyang Technology University), Bo An (Nanyang Technology University)
OptimizationReinforcement LearningGenerative Adversarial NetworkSequential
🎯 What it does: This paper proposes a reinforcement learning method that utilizes data from different dynamic environments—State Regularized Policy Optimization (SRPO)—which enhances learning efficiency on data with dynamic shifts by regularizing the stationary state distribution of the policy.
State Sequences Prediction via Fourier Transform for Representation Learning
Mingxuan Ye (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
Representation LearningReinforcement LearningTime SeriesSequential
🎯 What it does: An auxiliary self-supervised task is proposed that uses Fourier transform prediction in the frequency domain for infinite step state sequences to learn high-quality representations.
State-Action Similarity-Based Representations for Off-Policy Evaluation
Brahma S Pavse, Josiah P. Hanna (University of Wisconsin - Madison)
Reinforcement LearningTabular
🎯 What it does: A state-action similarity metric ROPE based on OPE is proposed, which learns an encoder from offline data and improves data efficiency in FQE.
State-space models with layer-wise nonlinearity are universal approximators with exponential decaying memory
Shida Wang (National University of Singapore), Beichen Xue (National University of Singapore)
Recurrent Neural NetworkSequential
🎯 What it does: It is proven that the state space model (SSM) with nonlinear activation added between layers has universal approximation properties, and its memory decay characteristics are analyzed.
State2Explanation: Concept-Based Explanations to Benefit Agent Learning and User Understanding
Devleena Das (Georgia Institute of Technology), Been Kim (Google Research)
Explainability and InterpretabilityReinforcement LearningContrastive LearningTabular
🎯 What it does: Proposes the State2Explanation framework, which jointly embeds RL state actions with conceptual explanations for reward shaping and interpretation.
StateMask: Explaining Deep Reinforcement Learning through State Mask
Zelei Cheng (Northwestern University), Xinyu Xing (Northwestern University)
Explainability and InterpretabilityReinforcement LearningSequential
🎯 What it does: The StateMask explanation method is proposed, which learns a state masking network to randomize actions at non-critical moments without significantly affecting the final reward of the target DRL agent, thereby identifying and quantifying which states are most important for the final reward.
Static and Sequential Malicious Attacks in the Context of Selective Forgetting
CHENXU ZHAO, Mengdi Huai (Iowa State University)
OptimizationAdversarial AttackConvolutional Neural NetworkReinforcement LearningImageTabular
🎯 What it does: This paper studies the attack framework in the process of selective forgetting (unlearning) in machine learning models, where an attacker manipulates the model by submitting malicious data update requests, proposing two types of attacks: static (all requests at once) and sequential (step-by-step requests).
Statistical Analysis of Quantum State Learning Process in Quantum Neural Networks
Hao-Kai Zhang, Xin Wang (Hong Kong University of Science and Technology)
Physics Related
🎯 What it does: This study investigates the learnability of quantum neural networks (QNN) in learning unknown quantum state processes and provides a statistical upper bound on the probability of local minima.
Statistical and Computational Trade-off in Multi-Agent Multi-Armed Bandits
Filippo Vannella (KTH Royal Institute of Technology), Jaeseong Jeong (Ericsson Research)
OptimizationReinforcement Learning from Human FeedbackGraph
🎯 What it does: This paper studies the regret minimization problem based on factor graph reward structures in multi-agent multi-armed bandits (MAMAB), deriving instance-specific regret lower bounds and designing a scalable efficient sampling algorithm (ESM) based on these bounds.
Statistical Guarantees for Variational Autoencoders using PAC-Bayesian Theory
Sokhna Diarra Mbacke (Laval University), Pascal Germain (Laval University)
GenerationData SynthesisAuto EncoderTabular
🎯 What it does: This paper constructs PAC-Bayes statistical generalization guarantees for Variational Autoencoders (VAE), proving upper bounds on its reconstruction loss, regeneration, and the Wasserstein distance of generation.
Statistical Insights into HSIC in High Dimensions
Tao Zhang (Shanghai University of Finance and Economics), Tingyou Zhou (Zhejiang University of Finance and Economics)
TabularTime SeriesFinance Related
🎯 What it does: This study investigates the statistical properties of the Hilbert–Schmidt Independence Criterion (HSIC) in high-dimensional samples, deriving the asymptotic normal distribution under the null hypothesis and providing conditions for testing non-independence in high dimensions. It further analyzes the different types of nonlinear associations that HSIC can capture.
Statistical Knowledge Assessment for Large Language Models
Qingxiu Dong (Peking University), Lei Li (Carnegie Mellon University)
Large Language ModelPrompt EngineeringText
🎯 What it does: A statistical method called KaRR is proposed to evaluate the reliability of factual answers generated by large language models under different prompts;
Statistical Limits of Adaptive Linear Models: Low-Dimensional Estimation and Inference
Licong Lin (University of California), Cun-Hui Zhang (Rutgers University)
🎯 What it does: This paper studies the statistical limits of adaptive sampling for low-dimensional parameter estimation and inference in high-dimensional linear models, and provides corresponding estimation and confidence interval methods.
Statistically Valid Variable Importance Assessment through Conditional Permutations
Ahmad Chamma (Inria), Bertrand Thirion (Inria)
Biomedical DataMagnetic Resonance Imaging
🎯 What it does: A conditional permutation importance method (CPI) is proposed and implemented to provide statistically valid variable importance assessments in the presence of correlated features.
StEik: Stabilizing the Optimization of Neural Signed Distance Functions and Finer Shape Representation
Huizong Yang (Georgia Institute of Technology), Anthony Yezzi (Georgia Institute of Technology)
OptimizationPoint CloudMeshBenchmark
🎯 What it does: This study investigates the instability caused by the eikonal loss in neural SDF learning and proposes a new regularization method and a network structure using quadratic layers to stabilize optimization and improve detail reconstruction quality.
Stein $\Pi$-Importance Sampling
Congye Wang (Newcastle University), Chris J. Oates (Newcastle University)
TabularBenchmark
🎯 What it does: The Stein Π-Importance Sampling method is proposed, which approximates the target distribution P by sampling from a Π-invariant Markov chain and post-processing through Stein discrepancy.
STEVE-1: A Generative Model for Text-to-Behavior in Minecraft
Shalev Lifshitz (University of Toronto), Sheila A. McIlraith
GenerationRobotic IntelligenceTransformerPrompt EngineeringVision Language ModelVideoTextMultimodality
🎯 What it does: In the Minecraft game, STEVE-1 is constructed using a pre-trained VPT behavior model and the MineCLIP visual-text alignment model, enabling the generation of low-level control behaviors based on text or visual instructions.
Stochastic Approximation Algorithms for Systems of Interacting Particles
Mohammad Reza Karimi Jaghargh, Andreas Krause (ETH Zurich)
OptimizationPhysics RelatedStochastic Differential Equation
🎯 What it does: A unified framework is proposed to analyze the convergence relationship between discrete-time particle systems and their continuous-time mean field limits.
Stochastic Approximation Approaches to Group Distributionally Robust Optimization
Lijun Zhang (Nanjing University), Zhi-Hua Zhou (Nanjing University)
OptimizationTabularStochastic Differential Equation
🎯 What it does: This paper proposes and analyzes various algorithms based on Stochastic Approximation (SA) to solve the Group Distributionally Robust Optimization (GDRO) problem, providing distribution-dependent convergence rates in practical scenarios with different sample budgets.
Stochastic Collapse: How Gradient Noise Attracts SGD Dynamics Towards Simpler Subnetworks
Feng Chen (Stanford University), Surya Ganguli (Stanford University)
CompressionOptimizationConvolutional Neural NetworkImageStochastic Differential Equation
🎯 What it does: This paper discusses the implicit bias of stochastic gradient descent (SGD) in training deep neural networks, revealing that SGD dynamically attracts parameters to the invariant set of sparse or low-rank sub-networks caused by 'reversible symmetry', thereby achieving network compression and better generalization.
Stochastic Distributed Optimization under Average Second-order Similarity: Algorithms and Analysis
Dachao Lin (Peking University), Zhihua Zhang (Peking University)
OptimizationTabular
🎯 What it does: This paper proposes two new distributed optimization algorithms, SVRS and its direct accelerated version AccSVRS, for solving finite and distributed optimization problems under the conditions of average second-order similarity and strong convexity.
Stochastic Multi-armed Bandits: Optimal Trade-off among Optimality, Consistency, and Tail Risk
David Simchi-Levi (Massachusetts Institute of Technology), Feng Zhu (Massachusetts Institute of Technology)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningTabularTime Series
🎯 What it does: A new randomized Successive Elimination algorithm is proposed, which can achieve an optimal trade-off between optimality, instance consistency, and tail risk under given conditions, and extends the theory to cases with non-stationary baseline rewards.
Stochastic Optimal Control for Collective Variable Free Sampling of Molecular Transition Paths
Lars Holdijk (University of Oxford), Max Welling (University of Amsterdam)
OptimizationReinforcement LearningSequentialPhysics RelatedStochastic Differential Equation
🎯 What it does: A path integral stochastic optimal control method (PIPS) is proposed, which does not require prior specification of collective variables (CVs), for efficiently sampling the transition paths of molecular systems between two steady states.
STORM: Efficient Stochastic Transformer based World Models for Reinforcement Learning
Weipu Zhang (Beijing Institute of Technology), Gao Huang (Tsinghua University)
TransformerReinforcement LearningAuto EncoderWorld ModelSequential
🎯 What it does: A Transformer-based random world model STORM is proposed for sample-efficient reinforcement learning.
Strategic Apple Tasting
Keegan Harris (Carnegie Mellon University), Steven Wu
OptimizationReinforcement Learning
🎯 What it does: This paper studies the strategic classification problem under apple tasting (one-sided) feedback, proposing a greedy OLS algorithm and an exploration-commitment algorithm, and implements a feasible online learning strategy in both random and adversarial environments.
Strategic Behavior in Two-sided Matching Markets with Prediction-enhanced Preference-formation
Stefania Ionescu (University of Zürich), Aniko Hannak
TabularSequential
🎯 What it does: This paper proposes a framework for embedding predictive models into two-sided matching markets and defines the concept of 'adversarial interaction attack' for the first time, which refers to the returning party interacting with the matching object in a non-optimal way in the current round to disrupt future predictions and matching results. Through economic models and simplified analysis, it is demonstrated that such attacks can yield benefits for the returning party in certain situations, and it further illustrates the inequality and overall welfare decline caused by these attacks.
Strategic Classification under Unknown Personalized Manipulation
Han Shao (Toyota Technological Institute Chicago), Omar Montasser (Toyota Technological Institute Chicago)
ClassificationOptimization
🎯 What it does: This paper analyzes the online error upper bounds and PAC sampling complexity under a framework of individualized and unknown manipulation strategies, proposing various algorithms and providing corresponding theoretical limits.
Strategic Data Sharing between Competitors
Nikita Tsoy (Sofia University), Nikola Konstantinov (Sofia University)
🎯 What it does: A modular framework for evaluating data sharing incentives among market competitors is proposed and studied, which includes market models, data impact models, and collaboration schemes;
Strategic Distribution Shift of Interacting Agents via Coupled Gradient Flows
Lauren E Conger, Lillian J Ratliff
Stochastic Differential EquationOrdinary Differential Equation
🎯 What it does: The study investigates the distribution shift caused by feedback loops, proposes a coupled PDE gradient flow model, and proves its exponential convergence under both alignment and competition objectives; numerical simulations validate the phenomena of polarization and multimodal distributions.
Strategyproof Voting under Correlated Beliefs
Daniel Halpern (Harvard University), Ariel D. Procaccia (Harvard University)
🎯 What it does: The study investigates the feasibility of voting strategies under related beliefs, proving that the plurality voting method is ordinal Bayesian incentive compatible (OBIC) under specific classes of beliefs.
STREAMER: Streaming Representation Learning and Event Segmentation in a Hierarchical Manner
Ramy Mounir (University of South Florida), Sudeep Sarkar (University of South Florida)
SegmentationRetrievalRepresentation LearningTransformerVideo
🎯 What it does: A fully self-supervised, layer-wise trainable hierarchical event segmentation and representation learning framework called STREAMER is proposed.
Streaming Algorithms and Lower Bounds for Estimating Correlation Clustering Cost
Sepehr Assadi (University of Waterloo), Chen Wang (Rutgers University)
Graph
🎯 What it does: This paper proposes two streaming algorithms that can estimate the optimal cost of related clustering in polynomial logarithmic space;
Streaming Factor Trajectory Learning for Temporal Tensor Decomposition
Shikai Fang (University of Utah), Shandian Zhe (University of Utah)
OptimizationExplainability and InterpretabilityComputational EfficiencyTime SeriesStochastic Differential Equation
🎯 What it does: A streaming time tensor decomposition method SFTL is proposed, which can learn and update the factor (i.e., object representation) time trajectories online without re-accessing historical data.
Streaming PCA for Markovian Data
Syamantak Kumar (University of Texas at Austin), Purnamrita Sarkar (University of Texas at Austin)
OptimizationTime Series
🎯 What it does: This paper studies the principal component analysis (PCA) problem of sampling data streams from irreducible, non-periodic, and reversible Markov chains, with the goal of estimating the principal eigenvector of an unknown covariance matrix.
StreamNet: Memory-Efficient Streaming Tiny Deep Learning Inference on the Microcontroller
Hong Sheng Zheng, Tsung Tai Yeh (National Yang-Ming Chiao Tung University)
Computational EfficiencyTabular
🎯 What it does: This paper presents StreamNet, which addresses the redundant computation problem in patch-based inference when running TinyML models on MCUs by designing 1D and 2D streaming processing with automatic parameter selection, significantly reducing the number of MACs and improving inference speed.
Strong and Precise Modulation of Human Percepts via Robustified ANNs
Guy Gaziv (Massachusetts Institute of Technology), James J. DiCarlo (Massachusetts Institute of Technology)
ClassificationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: The researchers utilized an adversarially trained ResNet50 model to generate low ‖p‖₂ norm image perturbations and conducted nine-class visual classification experiments via MTurk to explore the impact of these perturbations on human perception.
Structural Pruning for Diffusion Models
Gongfan Fang (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationCompressionComputational EfficiencyDiffusion modelImage
🎯 What it does: This paper proposes a structural pruning method for diffusion models called Diff-Pruning, which achieves the generation of lightweight models without retraining by retaining the core parameters of the pre-trained model and removing redundant weights.
Structure from Duplicates: Neural Inverse Graphics from a Pile of Objects
Tianhang Cheng (University of Illinois Urbana-Champaign), Shenlong Wang (University of Illinois Urbana-Champaign)
Pose EstimationOptimizationSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes inverse rendering in a single image containing repeated objects, utilizing multi-view information of the repeated objects to recover the geometry, material, and environmental lighting of the same object.
Structure Learning with Adaptive Random Neighborhood Informed MCMC
Xitong Liang (University College London), Jim Griffin (University College London)
Reinforcement LearningGraphBiomedical Data
🎯 What it does: A new MCMC sampler PARNI-DAG has been developed for complete Bayesian structure learning under observed data, sampling directly in the DAG space and achieving efficient mixing.
Structure of universal formulas
Dmitry Yarotsky (Skolkovo Institute of Science and Technology)
🎯 What it does: Analyzed and classified the expressiveness hierarchy of general formulas (fixed-size approximate functions), proving the membership of various function families under different approximation capabilities and providing their limit point structures.
Structure-free Graph Condensation: From Large-scale Graphs to Condensed Graph-free Data
Xin Zheng (Monash University), Shirui Pan (Griffith University)
CompressionKnowledge DistillationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Structure-Free Graph Compression (SFGC) method that compresses large-scale graph data into an unstructured set of nodes while implicitly encoding the graph structure information into node attributes.
Structured Federated Learning through Clustered Additive Modeling
Jie Ma (University of Technology Sydney), Chengqi Zhang (University of Technology Sydney)
Federated LearningImageBiomedical Data
🎯 What it does: A structured federated learning method called Clustered Additive Modeling (CAM) is proposed, which simultaneously trains a global model and a clustering model on the server side and uses additive combinations for predictions, thereby alleviating clustering collapse and enhancing cross-cluster knowledge sharing.
Structured Neural Networks for Density Estimation and Causal Inference
Asic Q Chen (University of Toronto), Rahul G Krishnan (University of Toronto)
Flow-based ModelTabular
🎯 What it does: A structured neural network (StrNN) is proposed, which implements explicit constraints on conditional independence through weight masking and integrates it into autoregressive flows (StrAF) and continuous flows (StrCNF) for density estimation and causal inference.
Structured Neural-PI Control with End-to-End Stability and Output Tracking Guarantees
Wenqi Cui (University of Washington), Yuanyuan Shi (University of California San Diego)
Autonomous DrivingOptimizationTabular
🎯 What it does: This paper proposes a structured neural PI controller for multi-input multi-output systems, which optimizes transient performance while ensuring system stability and zero steady-state tracking error.
Structured Prediction with Stronger Consistency Guarantees
Anqi Mao (New York University), Yutao Zhong (New York University)
Optimization
🎯 What it does: Theoretical analysis of surrogate loss functions in structured prediction is conducted, proving that existing mainstream loss functions do not satisfy Bayes consistency. Two new classes of structured surrogate losses (structured comp-sum and structured constrained) are proposed, along with their H-consistency bounds, demonstrating that Bayes consistency can be achieved under the assumption of a symmetric complete hypothesis set. Efficient methods for gradient computation and optimization for these losses are also provided.
Structured Semidefinite Programming for Recovering Structured Preconditioners
Arun Jambulapati (Simons Institute), Kevin Tian (University of Texas at Austin)
Optimization
🎯 What it does: A general 'matrix dictionary recovery' framework is proposed to construct approximately optimal preconditioners for various structured preprocessing problems (such as diagonal preconditioning, semi-random regression, perturbed Laplacian matrices, inverse M-matrices, and Laplacian pseudoinverses) in nearly linear time for linear system solving; corresponding efficient algorithms and runtime analyses are provided.
Structured State Space Models for In-Context Reinforcement Learning
Chris Lu (University of Oxford), Feryal Behbahani (DeepMind)
Meta LearningRecurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: This paper designs and evaluates an S5 structured state space model that can reset hidden states through parallel scanning during training, aimed at addressing the intrinsic learning problem of long sequences in reinforcement learning.
Structured Voronoi Sampling
Afra Amini (ETH Zurich), Ryan Cotterell (Johns Hopkins University)
GenerationOptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper studies a gradient-based sampling method called Structural Voronoi Sampling (SVS) for sampling from discrete language models and achieving controlled text generation.
Students Parrot Their Teachers: Membership Inference on Model Distillation
Matthew Jagielski (Google DeepMind), Florian Tramèr (ETH Zurich)
Safty and PrivacyKnowledge DistillationAdversarial AttackImageText
🎯 What it does: By constructing LiRA-based attacks (Transfer LiRA, End-to-End LiRA, and indirect attacks using only student queries), the system evaluates and reveals the privacy leakage of membership inference in knowledge distillation on teacher and student datasets.
STXD: Structural and Temporal Cross-Modal Distillation for Multi-View 3D Object Detection
Sujin Jang (Samsung Advanced Institute of Technology), Daehyun Ji (Samsung Advanced Institute of Technology)
Object DetectionAutonomous DrivingKnowledge DistillationImageMultimodalityPoint Cloud
🎯 What it does: A structure and time cross-modal distillation framework named STXD is proposed, which transfers 3D detection knowledge from a LiDAR teacher model to a multi-view camera student model.
StyleDrop: Text-to-Image Synthesis of Any Style
Kihyuk Sohn (Google Research), Daniel Castro Chin (Google Research)
GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: The StyleDrop method is proposed, which utilizes a minimal number of style reference images (even just one) combined with text descriptions, enabling text-to-image models to generate new images with specified visual styles.
StyleGAN knows Normal, Depth, Albedo, and More
Anand Bhattad (University of Illinois Urbana Champaign), David Forsyth
SegmentationGenerationDepth EstimationGenerative Adversarial NetworkImage
🎯 What it does: The study found that the pre-trained StyleGAN can directly generate various intrinsic images (surface normals, depth, albedo, shadows, and semantic segmentation) by adding fixed offsets in the latent space.
StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models
Yinghao Aaron Li (Columbia University), Nima Mesgarani (Columbia University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkAudio
🎯 What it does: Developed StyleTTS 2, an end-to-end TTS system that utilizes a style diffusion model to generate speaker styles from unreferenced audio, directly outputting waveforms from a speech encoder and undergoing adversarial training with a large speech language model to achieve human-level synthetic speech.
Sub-optimality of the Naive Mean Field approximation for proportional high-dimensional Linear Regression
Jiaze Qiu (Harvard University)
OptimizationTabular
🎯 What it does: In high-dimensional linear regression (where the ratio of feature dimensions to sample size increases), the asymptotic behavior of the naive mean field (NMF) variational approximation is precisely derived, and it is proven that there are significant biases in the log normalization constant and uncertainty quantification.
Subclass-Dominant Label Noise: A Counterexample for the Success of Early Stopping
Yingbin Bai (University of Sydney), Tongliang Liu (University of Sydney)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper studies a new type of label noise called Subclass Dominant Noise (SDN) and proposes a correction method called NoiseCluster for this noise.
Subject-driven Text-to-Image Generation via Apprenticeship Learning
Wenhu Chen, William W. Cohen
GenerationData SynthesisLarge Language ModelDiffusion modelImageText
🎯 What it does: We propose SuTI, a single model trained through apprenticeship learning that can instantly complete text-driven image generation for any visual subject after being given 3-5 demonstration images, without the need for additional optimization.
SUBP: Soft Uniform Block Pruning for 1$\times$N Sparse CNNs Multithreading Acceleration
Jingyang Xiang (Zhejiang University), Yong Liu (Zhejiang University)
Object DetectionSegmentationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A zero-training soft unified block pruning (SUBP) method is proposed for constructing 1×N sparse convolutional neural networks and achieving inference acceleration in a multi-threaded CPU environment.
Subspace Identification for Multi-Source Domain Adaptation
Zijian Li (Carnegie Mellon University), Kun Zhang (Mohamed bin Zayed University of Artificial Intelligence)
Domain AdaptationAuto EncoderImage
🎯 What it does: This paper proposes a subspace identification theory and a SIG model based on variational inference for multi-source domain adaptation tasks, significantly reducing the strict assumptions of traditional methods regarding the number of domains, monotonic transformations, and label distribution invariance.
Successor-Predecessor Intrinsic Exploration
Changmin Yu (University College London), Samuel Gershman
Reinforcement LearningSequential
🎯 What it does: A localized exploration reward that combines prospective and retrospective information is proposed, called Successor-Predecessor Intrinsic Exploration (SPIE).
Suggesting Variable Order for Cylindrical Algebraic Decomposition via Reinforcement Learning
Fuqi Jia (Chinese Academy of Sciences), Jian Zhang (Chinese Academy of Sciences)
OptimizationGraph Neural NetworkReinforcement LearningTabular
🎯 What it does: Two variants based on reinforcement learning and graph neural networks (GRL-SVO(UP) and GRL-SVO(NUP)) have been designed and implemented to automatically suggest variable ordering in Cylindrical Algebraic Decomposition (CAD).
Supervised Pretraining Can Learn In-Context Reinforcement Learning
Jonathan Lee, Emma Brunskill (Stanford University)
TransformerSupervised Fine-TuningReinforcement LearningSequential
🎯 What it does: Designed and trained a Decision-Pretrained Transformer (DPT), which achieves reinforcement learning by allowing the Transformer to predict the optimal action during multi-task pre-training given a contextual dataset.
Supply-Side Equilibria in Recommender Systems
Meena Jagadeesan (University of California Berkeley), Jacob Steinhardt (University of California Berkeley)
Recommendation SystemTabular
🎯 What it does: Construct and analyze a supply-side equilibrium model in recommendation systems, exploring the impact of multi-dimensional content vectors and production costs on content specialization and market competition.
Supported Value Regularization for Offline Reinforcement Learning
Yixiu Mao (Tsinghua University), Xiangyang Ji (Tsinghua University)
Reinforcement Learning
🎯 What it does: A Support Value Regularization (SVR) method is proposed, which penalizes the Q-values of out-of-distribution (OOD) actions that exceed the support of the behavior dataset in offline reinforcement learning, while retaining the standard Bellman update in the in-distribution (ID) region, thus addressing the issues of overestimation and extrapolation error.
Survival Instinct in Offline Reinforcement Learning
Anqi Li (University of Washington), Ching-An Cheng (Microsoft Research)
Reinforcement Learning
🎯 What it does: This paper studies the phenomenon of offline reinforcement learning (Offline RL) being able to learn excellent policies even when reward labels are incorrect, and provides a theoretical explanation for it.
Survival Permanental Processes for Survival Analysis with Time-Varying Covariates
Hideaki Kim (NTT Corporation)
OptimizationTime SeriesBiomedical Data
🎯 What it does: This paper proposes a non-parametric Bayesian survival model based on permanent processes—SurvPP—to analyze survival data with time-varying covariates. It transforms infinite-dimensional optimization into finite-dimensional parameterization through a representation theorem, achieving MAP inference with linear time complexity.
SutraNets: Sub-series Autoregressive Networks for Long-Sequence, Probabilistic Forecasting
Shane Bergsma (Huawei Cloud), Lei Guo (Huawei Cloud)
Recurrent Neural NetworkTransformerTime SeriesSequential
🎯 What it does: The SutraNets method is proposed, which splits long sequences into low-frequency subsequences and uses subsequence autoregressive generation to address the issues of error accumulation and signal path in long sequence prediction.
Swap Agnostic Learning, or Characterizing Omniprediction via Multicalibration
Parikshit Gopalan (Apple Inc.), Omer Reingold (Stanford University)
🎯 What it does: This paper proposes and analyzes a new learning framework called 'Swap Agnostic Learning' and proves its equivalence to Swap Omniprediction and Swap Multicalibration. It also provides an efficient algorithm for the Swap Agnostic Learner under any convex loss using MCBoost.
SwapPrompt: Test-Time Prompt Adaptation for Vision-Language Models
Xiaosong Ma (Hong Kong Polytechnic University), Wenchao Xu (Hong Kong Polytechnic University)
Domain AdaptationRepresentation LearningTransformerPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: This paper studies a method called SwapPrompt for prompt adaptation during testing, which dynamically adjusts prompt words using a pre-trained vision-language model on an unlabeled target domain to enhance the model's generalization ability.
Swarm Reinforcement Learning for Adaptive Mesh Refinement
Niklas Freymuth (Karlsruhe Institute of Technology), Gerhard Neumann (Karlsruhe Institute of Technology)
Graph Neural NetworkReinforcement LearningMesh
🎯 What it does: An adaptive grid refinement method based on adaptive swarm reinforcement learning (ASMR) is proposed, treating each grid cell as an agent and utilizing graph neural networks to learn refinement strategies without the need for error estimation during inference.