NeurIPS 2025 Papers — Page 42
Conference on Neural Information Processing Systems · 5275 papers
Seeing through Uncertainty: Robust Task-Oriented Optimization in Visual Navigation
Yiyuan Pan (Shanghai Jiao Tong University), Hesheng Wang (Shanghai Jiao Tong University)
OptimizationRobotic IntelligenceRecurrent Neural NetworkReinforcement LearningImage
🎯 What it does: The NEURO framework is proposed, which combines visual perception networks with robust optimization to achieve end-to-end training of visual navigation agents.
Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation
Riccardo Corvi (NVIDIA), Luisa Verdoliva (University Federico II of Naples)
GenerationAuto EncoderVideo
🎯 What it does: A Forensic-Oriented Augmentation based on beam transformation is proposed for training AI-generated video detectors.
Seemingly Redundant Modules Enhance Robust Odor Learning in Fruit Flies
HaiYang Li, Yunliang Zang (Xiamen Intretech Inc)
Spiking Neural NetworkReinforcement LearningTime Series
🎯 What it does: This study investigates the effects of lateral inhibition (LI) and spike frequency adaptation (SFA) as two sparsification mechanisms on olfactory discrimination learning under different noise levels in the fruit fly olfactory circuit, and compares their performance under noise-enhanced conditions through computational modeling.
SeerAttention: Self-distilled Attention Gating for Efficient Long-context Prefilling
Yizhao Gao (University of Hong Kong), Mao Yang (Microsoft Research)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: This paper proposes SeerAttention, a self-supervised sparse attention mechanism that learns block-level sparse gating (AttnGate) in each attention head of LLMs, and achieves efficient long text pre-filling inference by combining block-sparse FlashAttention kernels.
Seg-VAR:Image Segmentation with Visual Autoregressive Modeling
rongkun Zheng, Hengshuang Zhao (University of Hong Kong)
SegmentationTransformerAuto EncoderImage
🎯 What it does: Redefines image segmentation as a conditional autoregressive mask generation task and proposes the Seg-VAR framework;
Seg2Any: Open-set Segmentation-Mask-to-Image Generation with Precise Shape and Semantic Control
Danfeng Li (Fudan University), Zuxuan Wu (Fudan University)
SegmentationGenerationTransformerVision Language ModelDiffusion modelImage
🎯 What it does: Proposes the Seg2Any framework, which enables the generation of segmentation masks to images and supports precise shape and semantic control.
Seg4Diff: Unveiling Open-Vocabulary Semantic Segmentation in Text-to-Image Diffusion Transformers
Chaehyun Kim, Seungryong Kim
SegmentationTransformerDiffusion modelRectified FlowImageTextMultimodality
🎯 What it does: This paper proposes the Seg4Diff framework, which conducts a systematic analysis of the cross-modal attention within the Multi-modal Diffusion Transformer (MM-DiT). It finds that the 'semantic grounding expert' layer can naturally achieve alignment between text and image regions; using the I2T attention of this layer, zero-shot open vocabulary segmentation masks are directly generated, and a lightweight LoRA fine-tuning scheme called MAGNET is further designed to enhance segmentation accuracy and the consistency between images and text without sacrificing generation quality.
SEGA: Shaping Semantic Geometry for Robust Hashing under Noisy Supervision
Yiyang Gu (Peking University), Ming Zhang (Peking University)
RetrievalRepresentation LearningContrastive LearningImage
🎯 What it does: In multi-label hash learning, a SEGA framework is proposed to address the scenario where training labels contain noise, enhancing the robustness of hash codes by actively shaping the semantic geometric structure of the hash space.
SegGraph: Leveraging Graphs of SAM Segments for Few-Shot 3D Part Segmentation
Yueyang Hu (University of Chinese Academy of Sciences), Hao Pan (Tsinghua University)
SegmentationGraph Neural NetworkPoint Cloud
🎯 What it does: Utilizing the segmentation information generated by SAM, a segmentation graph is constructed to achieve few-shot 3D part segmentation.
SegMASt3R: Geometry Grounded Segment Matching
Rohit Jayanti (International Institute of Information Technology Hyderabad), Madhava Krishna (International Institute of Information Technology Hyderabad)
Object DetectionSegmentationTransformerContrastive LearningImagePoint CloudBenchmark
🎯 What it does: Utilizing the spatial prior of the 3D foundational model MASt3R, combined with a lightweight segment feature head and a differentiable Sinkhorn matching layer, to achieve image segment matching under extreme viewpoint changes (up to 180°).
Segment Anything Model Meets Semi-supervised Medical Image Segmentation: A Novel Perspective
Haifeng Zhao (Anhui University), Dengdi Sun (Anhui University)
SegmentationKnowledge DistillationImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A semi-supervised medical image segmentation framework is proposed, fully based on an efficient SAM, utilizing default embedding for prompt-free segmentation, and designing a hierarchical knowledge distillation and dynamic loss weighting strategy.
Segment Policy Optimization: Effective Segment-Level Credit Assignment in RL for Large Language Models
Yiran Guo (Institute of Software, Chinese Academy of Sciences), Shuang Qiu (City University of Hong Kong)
OptimizationTransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: In response to the reinforcement learning training of large language models (LLM) in reasoning tasks, the Segment Policy Optimization (SPO) framework is proposed, utilizing intermediate-granularity segment-level advantage estimation to achieve more accurate credit allocation. The SPO-chain is designed for short chain-of-thought (CoT) scenarios, while the SPO-tree is designed for long CoT scenarios.
Segment then Splat: Unified 3D Open-Vocabulary Segmentation via Gaussian Splatting
Yiren Lu (Case Western Reserve University), Yu Yin (Case Western Reserve University)
Object TrackingSegmentationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes the Segment then Splat framework, which implements 3D open vocabulary semantic segmentation by first segmenting and then reconstructing.
Selective Learning for Deep Time Series Forecasting
Yisong Fu (Institute of Computing Technology Chinese Academy of Sciences), Fei Wang (Institute of Computing Technology Chinese Academy of Sciences)
Anomaly DetectionOptimizationTransformerTime Series
🎯 What it does: A new learning strategy called Selective Learning is proposed, which suppresses the model's overfitting to noise and anomalous moments by only calculating loss at moments that can generalize in deep time series forecasting.
Selective Omniprediction and Fair Abstention
Sílvia Casacuberta (Stanford University), Varun Kanade (University of Oxford)
ClassificationOptimizationTabular
🎯 What it does: This study investigates a selective classifier that allows for the abandonment of predictions, capable of achieving optimal performance under various loss functions while making fair abandonment decisions across multiple fair settings.
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion
Xun Huang (Adobe Research), Eli Shechtman (Adobe Research)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkVideo
🎯 What it does: Proposes the Self Forcing training paradigm, using autoregressive self-replay and KV caching to train video diffusion models, eliminating exposure bias and improving generation quality through full video-level distribution matching loss.
Self Iterative Label Refinement via Robust Unlabeled Learning
Hikaru Asano (University of Tokyo), Yukino Baba (University of Tokyo)
ClassificationOptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes a self-iterative label improvement framework that uses two groups of unlabeled data with different positive-negative ratios to iteratively denoise and refine the pseudo-labels generated by LLM through robust UU learning.
Self supervised learning for in vivo localization of microelectrode arrays using raw local field potential
Tianxiao He (New York University), Erdem Varol (New York University)
ClassificationRepresentation LearningTransformerContrastive LearningBiomedical DataAlzheimer's Disease
🎯 What it does: A self-supervised learning framework Lfp2vec has been developed to achieve real-time localization of brain regions using raw local field potential (LFP) signals from multi-channel microelectrode arrays, which can be transferred to downstream tasks such as disease classification.
Self-Adapting Language Models
Adam Zweiger (Massachusetts Institute of Technology), Pulkit Agrawal (Massachusetts Institute of Technology)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: The SEAL framework is proposed, allowing large language models to achieve adaptation by generating their own self-editing (synthetic fine-tuning data and update instructions);
Self-alignment of Large Video Language Models with Refined Regularized Preference Optimization
Pritam Sarkar (Queen's University), Ali Etemad (Queen's University)
OptimizationTransformerReinforcement LearningVision Language ModelVideoMultimodality
🎯 What it does: An adaptive alignment framework is constructed through self-generated pairs of good/bad answers, allowing large video language models (LVLM) to self-correct and enhance video understanding capabilities.
Self-Assembling Graph Perceptrons
Jialong Chen (Sun Yat-sen University), Chuan Chen (Sun Yat-sen University)
ClassificationRecognitionGraph Neural NetworkImageTextAudio
🎯 What it does: A self-assembling graph perceptron (SAGP) is proposed, which dynamically adjusts the network topology during training through neuron growth, competition, and apoptosis, achieving a model size of only about 5% of an MLP while maintaining equal or even superior performance.
Self-Boost via Optimal Retraining: An Analysis via Approximate Message Passing
Adel Javanmard (University of Southern California), Vahab Mirrokni (Google Research)
ClassificationOptimizationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: For the binary classification task with noisy labels, this study investigates how to optimally utilize the combination of model predictions and original labels through iterative retraining (self-boost), proposing a theoretical framework based on Approximate Message Passing (AMP) and deriving a Bayesian optimal aggregator.
Self-Calibrating BCIs: Ranking and Recovery of Mental Targets Without Labels
Jonathan Grizou (GrizAI University of Glasgow), Tuukka Ruotsalo (LUT University University of Copenhagen)
GenerationRetrievalOptimizationGenerative Adversarial NetworkImageMultimodalityTime Series
🎯 What it does: This paper proposes an unsupervised self-calibrating brain-computer interface framework named CURSOR, which can recover the target image that the user is thinking of solely through paired EEG and image embeddings, without the need for labels or a pre-trained decoder.
Self-Challenging Language Model Agents
Yifei Zhou (Meta), Sainbayar Sukhbaatar (Meta)
TransformerLarge Language ModelReinforcement LearningAgentic AIText
🎯 What it does: A self-challenge framework is proposed, allowing large language models to generate high-quality 'Code-as-Task' tasks by first interacting with tools in the environment, and then performing reinforcement learning on these tasks to enhance the capabilities of multi-turn tool-using agents.
Self-diffusion for Solving Inverse Problems
Guanxiong Luo, Shoujin Huang
RestorationSuper ResolutionConvolutional Neural NetworkDiffusion modelImageMagnetic Resonance ImagingStochastic Differential Equation
🎯 What it does: A self-diffusion framework is proposed, utilizing a single randomly initialized neural network to solve inverse problems without the aid of pre-trained models or external data, through an iterative process of adding noise at each step and denoising in reverse.
Self-Evolving Pseudo-Rehearsal for Catastrophic Forgetting with Task Similarity in LLMs
Jun Wang (Wuhan University), Bo Du (Wuhan University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A Self-Evolving Pseudo-Rehearsal framework (SERS) is proposed to address the catastrophic forgetting problem in large language models through self-evolving pseudo-sample replay and task similarity-driven dynamic regularization.
Self-Generated In-Context Examples Improve LLM Agents for Sequential Decision-Making Tasks
Vishnu Sarukkai (Stanford University), Kayvon Fatahalian (Stanford University)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelAgentic AITextSequentialRetrieval-Augmented Generation
🎯 What it does: This paper proposes a self-generating trajectory database construction and refinement method, allowing LLM agents to improve their performance on sequential decision-making tasks by learning from their own successful experiences, without the need for manual prompts or task-specific engineering.
Self-Guided Hierarchical Exploration for Generalist Foundation Model Web Agents
Qianlan Yang (University of Illinois Urbana Champaign), Yu-Xiong Wang (University of Illinois Urbana Champaign)
Large Language ModelReinforcement LearningAgentic AIVision Language ModelBenchmark
🎯 What it does: This paper proposes a self-guided multi-level exploration framework called SAGE, designed to train a general web agent from scratch, enabling it to autonomously discover, generate, and complete tasks on real websites.
Self-Improving Embodied Foundation Models
Seyed Kamyar Seyed Ghasemipour (Generalist AI), Igor Mordatch (Google DeepMind)
Robotic IntelligenceTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodality
🎯 What it does: A two-stage post-training method is proposed, first optimizing the base model using supervised fine-tuning (behavior cloning + steps-to-go prediction), and then utilizing the self-predicted rewards and success detection from steps-to-go for online self-improvement to enhance the robot's low-level control performance.
Self-Perturbed Anomaly-Aware Graph Dynamics for Multivariate Time-Series Anomaly Detection
Jinyu Cai (National University of Singapore), See-Kiong Ng (National University of Singapore)
Anomaly DetectionGraph Neural NetworkTransformerTime Series
🎯 What it does: An end-to-end multivariate time series anomaly detection framework named SPAGD is proposed, which generates auxiliary anomaly samples through a self-disturbance module, dynamically constructs a graph structure, and combines spatiotemporal convolution for anomaly discrimination.
Self-Refining Language Model Anonymizers via Adversarial Distillation
Kyuyoung Kim (KAIST), Jinwoo Shin (KAIST)
Safty and PrivacyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Train small language models to achieve text anonymization without relying on external large models, and implement iterative improvements through self-assessment.
Self-supervised Blending Structural Context of Visual Molecules for Robust Drug Interaction Prediction
Tengfei Ma (Hunan University), xiangxiang Zeng
Drug DiscoveryTransformerAuto EncoderImageGraph
🎯 What it does: A self-supervised visual pre-training framework S²VM is proposed for jointly encoding molecular images of drug pairs and reconstructing the original images, thereby enhancing drug-drug interaction (DDI) prediction.
Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning
Achleshwar Luthra (Texas A&M University), Tomer Galanti (Texas A&M University)
OptimizationRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper studies the theoretical relationship between self-supervised contrastive learning (CL) and its supervised counterpart (the supervised contrastive loss with only negative samples, NSCL). It proves that when the number of classes is sufficiently large, the losses of the two are similar, and provides a geometric characterization of the global minimization solution of NSCL. Furthermore, it presents a new upper bound on the few-shot error and validates the theoretical predictions on various datasets.
Self-Supervised Direct Preference Optimization for Text-to-Image Diffusion Models
Liang Peng (FABU Inc), Xiaofei He (Zhejiang University)
GenerationOptimizationDiffusion modelContrastive LearningImageText
🎯 What it does: A self-supervised direct preference optimization method called Self-DPO is proposed for post-training alignment of text-to-image diffusion models.
Self-Supervised Discovery of Neural Circuits in Spatially Patterned Neural Responses with Graph Neural Networks
Kijung Yoon (Hanyang University)
Graph Neural NetworkTime SeriesSequential
🎯 What it does: This paper proposes a self-supervised framework based on graph neural networks, which learns potential synaptic connections and predicts future synaptic firing by utilizing the temporal activity of neurons.
Self-supervised Learning of Echocardiographic Video Representations via Online Cluster Distillation
Divyanshu Mishra (University of Oxford), Alison Noble
ClassificationSegmentationAnomaly DetectionKnowledge DistillationRepresentation LearningTransformerVideoBiomedical DataUltrasound
🎯 What it does: This paper presents DISCOVR, a self-supervised learning framework that combines video self-distillation and online image guidance, capable of learning rich spatiotemporal representations in unlabeled cardiac ultrasound videos and performing downstream tasks such as anomaly detection, classification, and segmentation.
Self-Supervised Learning of Graph Representations for Network Intrusion Detection
Lorenzo Guerra (Telecom Paris), Van-Tam Nguyen (Ampere Software Technology)
Anomaly DetectionGraph Neural NetworkTransformerAuto EncoderGraphTabular
🎯 What it does: A joint model called GraphIDS, which combines self-supervised graph neural networks and Transformer-based masked autoencoders, is proposed for network traffic anomaly detection.
Self-Supervised Learning of Motion Concepts by Optimizing Counterfactuals
Stefan Stojanov (Stanford University), Daniel LK Yamins
Object TrackingOptimizationRepresentation LearningContrastive LearningOptical FlowVideo
🎯 What it does: Proposes Opt-CWM, a technique that utilizes a self-supervised video model to extract optical flow and occlusion through an optimized counterfactual detector;
Self-Supervised Selective-Guided Diffusion Model for Old-Photo Face Restoration
Wenjie Li (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
RestorationDiffusion modelImage
🎯 What it does: To address the problem of restoring faces in old photographs, a self-supervised selective guidance diffusion model (SSDiff) is proposed, which utilizes weak guidance to generate pseudo-reference facial images and conducts staged structural and color guidance during the reverse diffusion process.
Self-Training with Dynamic Weighting for Robust Gradual Domain Adaptation
Zixi Wang (University of Electronic Science and Technology of China), Xin Lai (New Jersey Institute of Technology)
Domain AdaptationImage
🎯 What it does: A self-training with dynamic weighting (STDW) method is proposed for progressive domain adaptation, addressing issues such as unsmooth knowledge transfer and instability during the transition phase.
Self-Verification Provably Prevents Model Collapse in Recursive Synthetic Training
Shi Fu (Nanyang Technological University), Dacheng Tao (Nanyang Technological University)
TransformerLarge Language Model
🎯 What it does: This paper theoretically studies the model collapse problem in the recursive synthesis training process, proposes a self-verification mechanism, and provides an upper bound on finite sample error, proving that stable training can be achieved in Transformer-LLMs even when using entirely synthetic data.
Self-Verifying Reflection Helps Transformers with CoT Reasoning
Zhongwei Yu (Hong Kong University of Science and Technology), Jun Wang (King's College London)
TransformerSupervised Fine-TuningReinforcement LearningTabularChain-of-Thought
🎯 What it does: This study investigates the effectiveness of small Transformers in enhancing multi-step Chain-of-Thought reasoning through self-verification reflection in non-natural language environments, providing both theoretical and experimental analysis.
Selftok-Zero: Reinforcement Learning for Visual Generation via Discrete and Autoregressive Visual Tokens
Bohan Wang (Huawei Central Media Technology Institute), Hanwang Zhang (Huawei Central Media Technology Institute)
GenerationTransformerReinforcement LearningVision Language ModelImageText
🎯 What it does: The Selftok self-consistent tokenizer encodes images into 1D autoregressive visual tokens, and based on this, reinforcement learning is used for post-training of the visual generation model to achieve unsupervised text-to-image generation.
Semantic and Visual Crop-Guided Diffusion Models for Heterogeneous Tissue Synthesis in Histopathology
Saghir Alfasly (Mayo Clinic), Hamid Tizhoosh (Mayo Clinic)
SegmentationGenerationData SynthesisDiffusion modelImageBiomedical DataAgriculture Related
🎯 What it does: A dual conditional latent diffusion model is proposed that combines semantic segmentation maps and real tissue slices to generate high-fidelity pathological images with mixed tissues, achieving self-supervised expansion of unlabeled whole slide data.
Semantic Representation Attack against Aligned Large Language Models
Jiawei Lian (Northwestern Polytechnical University), Lap-Pui Chau (Hong Kong Polytechnic University)
Adversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A novel attack method aimed at aligned large language models is proposed—Semantic Representation Attack, which efficiently generates semantically consistent and natural attack prompts.
Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models
Lexiang Xiong (National University of Singapore), Yuecong Xu (National University of Singapore)
GenerationData SynthesisAdversarial AttackDiffusion modelImage
🎯 What it does: A zero-shot, inference-time text embedding vector subtraction framework called Semantic Surgery is proposed, which can dynamically remove unwanted concepts from text prompts without changing the model parameters.
Semantic-guided Diverse Decoding for Large Language Model
Weijie Shi (Hong Kong University of Science and Technology), Xiaofang Zhou (Hong Kong University of Science and Technology)
GenerationOptimizationTransformerLarge Language ModelReinforcement LearningTextBiomedical Data
🎯 What it does: Proposes the SemDiD method, which achieves semantic-level diverse decoding in large language models, directly guiding generation in the embedding space;
SemCoT: Accelerating Chain-of-Thought Reasoning through Semantically-Aligned Implicit Tokens
Yinhan He (University of Virginia), Jundong Li (University of Virginia)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelContrastive LearningTextChain-of-Thought
🎯 What it does: Proposes the SemCoT framework, which utilizes semantically aligned implicit Chain-of-Thought (CoT) tokens to accelerate the reasoning process of large language models (LLMs).
Semi-infinite Nonconvex Constrained Min-Max Optimization
Cody Melcher (University of Arizona), Erfan Yazdandoost Hamedani (University of Arizona)
OptimizationTabular
🎯 What it does: This paper studies semi-infinite non-convex min-max optimization problems and proposes an incomplete dynamic barrier primal-dual algorithm (iDB-PD) to solve non-convex min-max problems with infinite constraints.
Semi-off-Policy Reinforcement Learning for Vision-Language Slow-Thinking Reasoning
Junhao Shen (Shanghai Jiao Tong University), Kai Chen (Shanghai AI Laboratory)
TransformerReinforcement LearningPrompt EngineeringVision Language ModelTextMultimodality
🎯 What it does: Enhancing the slow thinking reasoning ability of large visual language models through a semi-offline reinforcement learning framework (SOPHIA), combining the visual understanding of the policy with the slow thinking of the offline language model.
Semi-supervised Graph Anomaly Detection via Robust Homophily Learning
GuoguoAi, Guansong Pang (Singapore Management University)
Anomaly DetectionGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A semi-supervised graph anomaly detection method called RHO is proposed, which learns anomalous nodes in the graph using a small number of labeled normal nodes.
Semi-Supervised Regression with Heteroscedastic Pseudo-Labels
Xueqing Sun (Xi'an Jiaotong University), Deyu Meng (Xi'an Jiaotong University)
OptimizationImage
🎯 What it does: A pseudo-labeling framework for uncertainty-aware semi-supervised regression is proposed, which can dynamically adjust the heteroscedastic uncertainty of pseudo-labels through bi-level optimization, thereby mitigating the negative impact of incorrect pseudo-labels on the model.
Semi-supervised Vertex Hunting, with Applications in Network and Text Analysis
Yicong Jiang (Harvard University), Tracy Ke
Anomaly DetectionOptimizationTextGraph
🎯 What it does: A new semi-supervised vertex extraction (SSVH) method is proposed for estimating simple shapes from noisy data, applied to network and text analysis.
SEMPO: Lightweight Foundation Models for Time Series Forecasting
Hui He (Beijing Institute of Technology), Guansong Pang (Singapore Management University)
TransformerPrompt EngineeringTime Series
🎯 What it does: A lightweight foundational model SEMPO is proposed, utilizing energy-aware spectral decomposition and a mixture of prompts Transformer to achieve strong generalization in time series prediction with a small amount of pre-training data.
SensorLM: Learning the Language of Wearable Sensors
Yuwei Zhang (Google Research), Yuzhe Yang (Google Research)
RecognitionRetrievalTransformerContrastive LearningMultimodalityTime Series
🎯 What it does: Proposes SensorLM, a foundational model that aligns wearable sensor data with natural language;
Separating the 'what' and 'how' of compositional computation to enable reuse and continual learning
Haozhe Shan (Columbia University), Lea Duncker (Columbia University)
Recurrent Neural NetworkGenerative Adversarial NetworkSequential
🎯 What it does: A two-system (what/how) framework is proposed, which uses a probabilistic generative model to infer the context of the time period (epoch) for online tasks, and then uses this context as gating for a low-rank RNN to achieve continuous learning and combinatorial computation for multiple cognitive tasks.
seq-JEPA: Autoregressive Predictive Learning of Invariant-Equivariant World Models
Hafez Ghaemi (Universite de Montreal), Shahab Bakhtiari (Universite de Montreal)
Representation LearningConvolutional Neural NetworkTransformerContrastive LearningWorld ModelImage
🎯 What it does: This paper proposes seq-JEPA, a world model that learns visual representations with both invariance and equivariance by autoregressively predicting action-observation sequences.
Sequence Modeling with Spectral Mean Flows
Jinwoo Kim, Seunghoon Hong
GenerationData SynthesisComputational EfficiencyFlow-based ModelTime SeriesSequentialFinance RelatedPhysics Related
🎯 What it does: A sequence modeling method based on operator theory is proposed—Spectral Mean Flows, which embeds sequence distributions into Hilbert space using the linear operator mapping of HMM, and defines MMD gradient flows for generation in this space;
Sequential Attention-based Sampling for Histopathological Analysis
Tarun G (Indian Institute of Science), Devarajan Sridharan (Indian Institute of Science)
Computational EfficiencyTransformerReinforcement LearningImageBiomedical Data
🎯 What it does: The SASHA framework is proposed, combining multi-attention multi-instance learning with reinforcement learning to achieve efficient sequential sampling and diagnosis of panoramic tissue sections.
Sequential Monte Carlo for Policy Optimization in Continuous POMDPs
Hany Abdulsamad (University of Amsterdam), Simo Särkkä (Aalto University)
OptimizationReinforcement LearningSequential
🎯 What it does: A continuous POMDP policy optimization algorithm (P3O) based on the Feynman-Kac model and nested particle filtering is proposed, which can directly model and learn the joint Bayesian inference and decision-making process.
Sequential Multi-Agent Dynamic Algorithm Configuration
Chen Lu (Nanjing University), Chao Qian (Nanjing University)
OptimizationReinforcement LearningSequential
🎯 What it does: In response to the inherent dependencies of multiple parameters in Dynamic Algorithm Configuration (DAC), the Seq-MADAC framework is proposed, and a Sequence Advantage Decomposition Network (SADN) is designed to achieve multi-agent dynamic configuration.
Sequentially Auditing Differential Privacy
Tomás González (Carnegie Mellon University), Mónica Ribero (Google Research)
Safty and Privacy
🎯 What it does: A sequential auditing framework based on Maximum Mean Discrepancy (MMD) is proposed for real-time detection of black-box differential privacy mechanisms;
SeRL: Self-play Reinforcement Learning for Large Language Models with Limited Data
Wenkai Fang (Zhejiang University), Dacheng Tao (Nanyang Technological University)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Proposes the SeRL framework, which achieves unsupervised reinforcement learning through LLM self-generated instructions and self-evaluated rewards in data-scarce situations;
Set Smoothness Unlocks Clarke Hyper-stationarity in Bilevel Optimization
He Chen (Chinese University of Hong Kong), Anthony Man-Cho So (Chinese University of Hong Kong)
Optimization
🎯 What it does: This paper proposes a feasible algorithmic framework for computing approximate Clarke stationary points in non-convex-PŁ bilevel optimization problems.
Set-LLM: A Permutation-Invariant LLM
Beni Egressy (Heidelberg Institute for Theoretical Studies), Jan Stühmer (Karlsruhe Institute of Technology)
TransformerLarge Language ModelSupervised Fine-TuningTextMultimodality
🎯 What it does: A decoder-type language model named Set-LLM is designed to achieve permutation invariance for mixed sets and text inputs.
SGAR: Structural Generative Augmentation for 3D Human Motion Retrieval
Jiahang Zhang (Peking University), Jiaying Liu (Peking University)
RetrievalTransformerLarge Language ModelContrastive LearningVideoText
🎯 What it does: This paper proposes SGAR, which utilizes body part action descriptions generated by large language models to enhance structured generation for 3D action and text retrieval.
SGCD: Stain-Guided CycleDiffusion for Unsupervised Domain Adaptation of Histopathology Image Classification
Hsi-Ling Chen (National Cheng Kung University), Pau-Choo Chung (National Cheng Kung University)
ClassificationDomain AdaptationDiffusion modelImage
🎯 What it does: A dye-guided cyclic diffusion (SGCD) method based on a bidirectional diffusion model is proposed for unsupervised domain adaptation in tissue pathology image classification.
SGN: Shifted Window-Based Hierarchical Variable Grouping for Multivariate Time Series Classification
Zenan Ying (University of Science and Technology of China), Wei Chen (University of Science and Technology of China)
ClassificationConvolutional Neural NetworkTransformerTime Series
🎯 What it does: This paper proposes SwinGroupNet (SGN) for multivariate time series classification, achieving structured modeling of inter-variable and temporal dependencies through modules such as variable grouping embedding, cross-scale window mixing, and periodic window shifting and merging.
Shallow Diffuse: Robust and Invisible Watermarking through Low-Dim Subspaces in Diffusion Models
Wenda Li (University of Michigan), Qing Qu (University of Michigan)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A watermarking method called Shallow Diffuse is proposed, which embeds a robust and invisible watermark in images generated by diffusion models.
Shallow Flow Matching for Coarse-to-Fine Text-to-Speech Synthesis
Dong Yang (University of Tokyo), Hiroshi Saruwatari (University of Tokyo)
GenerationData SynthesisFlow-based ModelOrdinary Differential EquationAudio
🎯 What it does: Designed and evaluated a shallow flow matching (SFM) mechanism for coarse-to-fine text-to-speech synthesis models, allowing the flow model to sample from the intermediate state beyond noise, enhancing generation stability and naturalness.
SHAP Meets Tensor Networks: Provably Tractable Explanations with Parallelism
Reda Marzouk (University of Montpellier), Guy Katz (Hebrew University of Jerusalem)
Explainability and InterpretabilityComputational Efficiency
🎯 What it does: The first algorithm for provably exact computation of SHAP values under Tensor Networks (TN), especially Tensor Train (TT) models, is proposed and parallelized to the NC complexity class.
SHAP values via sparse Fourier representation
Ali Gorji (ETH Zurich), Andreas Krause (ETH Zurich)
Explainability and InterpretabilityComputational EfficiencyProtein Structure PredictionConvolutional Neural NetworkTabularBiomedical Data
🎯 What it does: A two-stage SHAP value computation method is proposed, which first obtains a compact representation of black-box or tree models through sparse Fourier transform, and then directly and in parallel computes SHAP values using a closed-form formula of the Fourier basis.
SHAP zero Explains Biological Sequence Models with Near-zero Marginal Cost for Future Queries
Darin Tsui (Georgia Institute of Technology), Amirali Aghazadeh (Georgia Institute of Technology)
Explainability and InterpretabilityComputational EfficiencyDrug DiscoveryBiomedical Data
🎯 What it does: Proposes the SHAP zero algorithm, which utilizes sparse Fourier folding techniques to provide Shapley explanations for black-box biological sequence models, and achieves near-zero incremental cost for subsequent queries by constructing a global Fourier sketch in one go;
Shape it Up! Restoring LLM Safety during Finetuning
ShengYun Peng (Georgia Tech), Duen Horng Chau (Georgia Tech)
Safty and PrivacyReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: A dynamic safety shaping framework (DSS) is proposed, utilizing the token-level safety scoring STAR from the guardrail model to guide reinforcement learning on safe content during LLM fine-tuning, significantly enhancing safety while maintaining capabilities.
Shape-Informed Clustering of Multi-Dimensional Functional Data via Deep Functional Autoencoders
Samuel Singh (Trinity College Dublin), Mimi Zhang (Trinity College Dublin)
Auto EncoderTime Series
🎯 What it does: Introducing the FAEclust deep functional autoencoder framework for unsupervised clustering of multidimensional functional data.
ShapeCraft: LLM Agents for Structured, Textured and Interactive 3D Modeling
Shuyuan Zhang (Imperial College London), Jiankang Deng (Imperial College London)
GenerationData SynthesisTransformerLarge Language ModelAgentic AIPrompt EngineeringScore-based ModelTextMesh
🎯 What it does: The multi-agent LLM framework ShapeCraft transforms text descriptions into structured GPS programs, generating structured, textureable, and interactive 3D models.
ShapeEmbed: a self-supervised learning framework for 2D contour quantification
Anna Foix Romero (European Bioinformatics Institute), Virginie Uhlmann (European Bioinformatics Institute)
ClassificationRepresentation LearningConvolutional Neural NetworkAuto EncoderContrastive LearningImage
🎯 What it does: A self-supervised framework called ShapeEmbed is proposed, which encodes two-dimensional closed contours using a distance matrix and learns shape descriptors that are invariant to geometric transformations (translation, rotation, scaling, reflection, point indexing).
ShapeLLM-Omni: A Native Multimodal LLM for 3D Generation and Understanding
Junliang Ye (Tsinghua University), Jun Zhu (Peking University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityPoint Cloud
🎯 What it does: Developed ShapeLLM-Omni, a multimodal large language model capable of generating, understanding, and editing text, images, and 3D content within the same model.
ShapeX: Shapelet-Driven Post Hoc Explanations for Time Series Classification Models
Bosong Huang (Griffith University), Shirui Pan (Griffith University)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTime SeriesElectrocardiogram
🎯 What it does: A post-hoc time series classification model interpretation framework SHAPEX is proposed, which utilizes shape subsequences for segmentation and evaluates their contributions using Shapley values, thereby providing more causal and interpretable explanations.
Shaping Sequence Attractor Schema in Recurrent Neural Networks
Zhikun Chu (Chongqing University), Yuanyuan Mi (Tsinghua University)
RecognitionRecurrent Neural NetworkSequentialAudio
🎯 What it does: By implementing a 'shaping' process in recurrent neural networks, this study simulates schema learning in animals/humans, validates its neurodynamics, and applies it to keyword recognition tasks;
Shapley-Based Data Valuation for Weighted $k$-Nearest Neighbors
Guangyi Zhang (Shenzhen Technology University), Aristides Gionis (KTH Royal Institute of Technology)
Data-Centric LearningTabular
🎯 What it does: This paper proposes to transform the calculation of Shapley values for weighted k-nearest neighbor models into an unweighted k-nearest neighbor form through data replication, and provides an efficient approximation algorithm.
Shapley-Coop: Credit Assignment for Emergent Cooperation in Self-Interested LLM Agents
Yun Hua (Shanghai Jiao Tong University), Jun Luo (Shanghai Jiao Tong University)
Large Language ModelTextChain-of-Thought
🎯 What it does: The Shapley-Coop framework is proposed, utilizing Shapley values and structured negotiation protocols to achieve spontaneous collaboration and fair credit distribution among self-interested LLM agents in open environments.
Sharp Analysis for KL-Regularized Contextual Bandits and RLHF
Heyang Zhao (University of California Los Angeles), Tong Zhang (University of Illinois Urbana Champaign)
Reinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper provides a theoretical analysis of inverse KL regularization in contextual bandits and RLHF, and presents a two-stage mixed sampling algorithm.
Sharp Gap-Dependent Variance-Aware Regret Bounds for Tabular MDPs
Shulun Chen (Tsinghua University), Simon Shaolei Du
Reinforcement LearningTabular
🎯 What it does: Research and provide a gap-based variance-aware cumulative regret upper bound, proving the superiority of the MVP algorithm in tabular MDPs.
Sharp Gaussian approximations for Decentralized Federated Learning
Soham Bonnerjee, Wei Biao Wu
Anomaly DetectionFederated LearningSafty and Privacy
🎯 What it does: This paper presents, for the first time, the Berry-Esseen bound and full-period Gaussian approximation for the iterative process of local SGD in decentralized federated learning (DFL), and based on this, constructs a multiplier bootstrap method for model parameter inference and timely detection of malicious attacks.
Sharp Matrix Empirical Bernstein Inequalities
Hongjian Wang (Carnegie Mellon University), Aaditya Ramdas (Carnegie Mellon University)
🎯 What it does: Two precise empirical Bernstein inequalities for symmetric random matrices are proposed, applicable to independent samples and Markov-dependent cases, respectively; closed-form explicit expressions are provided.
Sharper Convergence Rates for Nonconvex Optimisation via Reduction Mappings
Evan Markou (Australian National University), Stephen Gould (Australian National University)
Optimization
🎯 What it does: This paper proposes a reduction mapping that utilizes the structure of known optimal solutions to reformulate non-convex optimization problems. By analyzing improvements in curvature and sharpness, it proves that gradient descent can achieve faster local linear convergence within this framework.
SharpZO: Hybrid Sharpness-Aware Vision Language Model Prompt Tuning via Forward-Only Passes
Yifan Yang (University of California), Zheng Zhang (University of California)
ClassificationObject DetectionSegmentationDomain AdaptationOptimizationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Under the condition of using only forward propagation, a two-stage hybrid zero-order optimization method called SharpZO is proposed for fine-tuning prompts of visual language models such as CLIP.
Sherlock: Self-Correcting Reasoning in Vision-Language Models
Yi Ding (Purdue University), Ruqi Zhang (Purdue University)
TransformerSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: The Sherlock framework was constructed and trained, enabling visual language models to self-correct and self-improve during the reasoning process, significantly enhancing multimodal reasoning performance.
SHF: Symmetrical Hierarchical Forest with Pretrained Vision Transformer Encoder for High-Resolution Medical Segmentation
Enzhi Zhang (Hokkaido University), Mohamed Wahib (RIKEN Center for Computational Science)
SegmentationTransformerImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes the Symmetrical Hierarchical Forest (SHF) scheme, which utilizes an adaptive hierarchical forest structure to segment and de-segment high-resolution medical images, thereby compressing the sequence length and eliminating the convolutional decoder without altering the Vision Transformer structure, achieving efficient long-sequence processing and fine segmentation.
SHGR: A Generalized Maximal Correlation Coefficient
Samuel Stocksieker (CNRS I2M Aix Marseille University), Denys Pommeret (CNRS I2M Aix Marseille University)
OptimizationSupervised Fine-TuningTabular
🎯 What it does: This paper proposes the maximum correlation coefficient SHGR based on Spearman rank correlation, and provides a differentiable neural network estimator and its cross-encoder architecture for one-time estimation of one-to-one, multivariate to univariate, and complete correlation matrices of two sets of variables; it also verifies its robustness under noise, outliers, and independence assumptions, and applies it to feature selection.
Shift Before You Learn: Enabling Low-Rank Representations in Reinforcement Learning
Bastien Dubail (KTH Royal Institute of Technology), Alexandre Proutiere (KTH Royal Institute of Technology)
Reinforcement LearningTabular
🎯 What it does: This study investigates the use of shifted successor measures in reinforcement learning for low-rank approximation and provides finite sample upper bounds for sampling error and estimation error.
ShiQ: Bringing back Bellman to LLMs
Pierre Clavier (Cohere), Matthieu Geist (Earth Species Project)
TransformerLarge Language ModelReinforcement LearningSequential
🎯 What it does: Proposes the ShiQ algorithm, which utilizes Bellman consistency to transfer Q-learning to LLM fine-tuning, supporting offline, token-level, and importance-sampling-free reinforcement learning;
ShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-On
Yuhan Li (Shanghai Jiao Tong University), Bingbing Ni (Alibaba Group)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: The first multi-view shoe fitting dataset MVShoes has been constructed, and a ShoeFit framework based on a dual-stream Diffusion Transformer has been proposed to achieve high-fidelity virtual shoe fitting.
Short-length Adversarial Training Helps LLMs Defend Long-length Jailbreak Attacks: Theoretical and Empirical Evidence
Shaopeng Fu (King Abdullah University of Science and Technology), Di Wang (King Abdullah University of Science and Technology)
Computational EfficiencyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper studies prison break attacks on large language models (LLMs) and proposes a method to effectively defend against long-length prison break attacks through short-length adversarial training (AT).
Shortcut Features as Top Eigenfunctions of NTK: A Linear Neural Network Case and More
Jinwoo Lim (Seoul National University), Soo-Mook Moon (Seoul National University)
Convolutional Neural NetworkImage
🎯 What it does: This study investigates how linear neural networks produce shortcut learning under biased data distributions within the NTK framework, providing theoretical analysis and experimental validation.
Shortcuts and Identifiability in Concept-based Models from a Neuro-Symbolic Lens
Samuele Bortolotti (University of Trento), Stefano Teso (University of Trento)
Autonomous DrivingExplainability and InterpretabilityContrastive LearningImage
🎯 What it does: This paper studies the Conceptual Basis Models (CBMs) and their neural-symbolic variants in the context of Joint Reasoning Shortcuts (JRSs) that arise when learning interpretable concepts and reasoning layers, providing both theoretical and experimental validation of recognizability.
Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch
Xu Cai, Hongkai Wen (University of Warwick)
GenerationKnowledge DistillationDiffusion modelFlow-based ModelContrastive LearningImage
🎯 What it does: This paper proposes an efficient post-training method called SCFM, which utilizes velocity field self-distillation to quickly distill a large-scale pre-trained flow-matching diffusion model into a high-quality sampler that requires only 3 steps;
ShorterBetter: Guiding Reasoning Models to Find Optimal Inference Length for Efficient Reasoning
Jingyang Yi (University of Chicago), Sida Li (University of Chicago)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: This paper proposes an adaptive reasoning length optimization method based on reinforcement learning called ShorterBetter, which allows large language models to automatically find the optimal Chain-of-Thought length during inference.
ShortListing Model: A Streamlined Simplex Diffusion for Discrete Variable Generation
Yuxuan Song (Tsinghua University), Wei-Ying Ma (Tsinghua University)
GenerationData SynthesisDrug DiscoveryDiffusion modelTextBiomedical Data
🎯 What it does: A simplified Simplex Diffusion Model (SLM) is proposed for discrete variable generation, transforming the generation process into a process from all candidates to a single category using a stepwise candidate pruning approach.