NeurIPS 2025 Papers — Page 33
Conference on Neural Information Processing Systems · 5275 papers
On topological descriptors for graph products
Mattie Ji (University of Pennsylvania), Vikas K Garg
Graph Neural NetworkGraph
🎯 What it does: This paper studies the Euler angles and persistent homology of graphs under color-based filtering, and presents an efficient algorithm for computing these topological descriptors on the Cartesian product of graphs.
On Traceability in $\ell_p$ Stochastic Convex Optimization
Sasha Voitovych (Massachusetts Institute of Technology), Daniel M. Roy (University of Toronto)
Optimization
🎯 What it does: This paper studies the relationship between traceability and overfitting risk in stochastic convex optimization (SCO) under ℓp geometry, proving a fundamental traceability-risk trade-off and providing the corresponding thresholds.
On Transferring Transferability: Towards a Theory for Size Generalization
Eitan Levin (California Institute of Technology), Soledad Villar (Johns Hopkins University)
Domain AdaptationRepresentation LearningGraph Neural NetworkPoint CloudGraph
🎯 What it does: A unified theoretical framework is proposed to characterize and verify the transferability of models under different input sizes (graphs, sets, point clouds) and the corresponding size generalization.
On Union-Closedness of Language Generation
Steve Hanneke (Purdue University), Grigoris Velegkas (Google Research)
GenerationText
🎯 What it does: This paper proposes and addresses three key open problems in language generation at the limits, proving that generation does not satisfy finite union and closure, constructing a non-uniformly generable but not satisfying EUC property uncountable class, and further demonstrating that the union of uniformly and non-uniformly generable classes is not necessarily generable.
On Universality Classes of Equivariant Networks
Marco Pacini (University of Trento), Shubhendu Trivedi
🎯 What it does: This study investigates the approximation capability of equivariant neural networks beyond their separation ability, revealing that networks with the same separation ability may exhibit significant differences in approximation performance.
On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning
Alvaro Arroyo, Pierre Vandergheynst (AITHYRA EPFL)
Graph Neural NetworkGraph
🎯 What it does: Analyzed the issues of feature collapse, over-smoothing, and information compression caused by gradient vanishing in Graph Neural Networks (GNNs), and proposed GNN-SSM based on state space models, enhancing long-range information transmission capabilities through a combination of graph re-connection and non-dissipative dynamics.
On-Policy Optimization with Group Equivalent Preference for Multi-Programming Language Understanding
Haoyuan Wu (Chinese University of Hong Kong), Bei Yu (Chinese University of Hong Kong)
OptimizationAI Code AssistantTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Train large language models (LLMs) through code translation tasks to enhance their code understanding and generation capabilities across multiple programming languages.
Once Upon an Input: Reasoning via Per-Instance Program Synthesis
Adam Stein (University of Pennsylvania), Eric Wong (University of Pennsylvania)
OptimizationAI Code AssistantTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: A Per-Instance Program Synthesis (PIPS) framework is proposed, which can automatically generate and iteratively refine programs at the instance level to complete reasoning tasks.
One Filters All: A Generalist Filter For State Estimation
Shiqi Liu (Tsinghua University), Shengbo Eben Li (Tsinghua University)
TransformerLarge Language ModelPrompt EngineeringTime Series
🎯 What it does: A general state estimation framework called LLM-Filter is proposed, which utilizes pre-trained large language models to embed noisy observations into text and achieves cross-system estimation through contextual prompts.
One for All: Universal Topological Primitive Transfer for Graph Structure Learning
Yide Qiu (Nanjing University of Science and Technology), Zhen Cui (Beijing Normal University)
Domain AdaptationRepresentation LearningGraph Neural NetworkLarge Language ModelContrastive LearningTextGraph
🎯 What it does: This paper proposes the G²SN-Transfer framework, which utilizes dual-stream sequence alignment of topological primitives and text descriptions to achieve knowledge transfer across graph structures.
One Head to Rule Them All: Amplifying LVLM Safety through a Single Critical Attention Head
Junhao Xia (Nanjing University of Science and Technology), Jason Xue
Safty and PrivacyTransformerVision Language ModelMultimodality
🎯 What it does: A defense framework based on a single key attention head (Oh Defense) is proposed, which enhances the security of large visual language models (LVLMs) without requiring additional training.
One Prompt Fits All: Universal Graph Adaptation for Pretrained Models
Yongqi Huang (Tianjin University), Zhiyong Feng (Tianjin University)
Domain AdaptationRepresentation LearningGraph Neural NetworkPrompt EngineeringGraph
🎯 What it does: This paper studies a universal graph prompt learning method called UniPrompt, which can adapt to different downstream tasks through learnable topological prompts while keeping the pre-trained model frozen, making it particularly suitable for few-shot and cross-domain scenarios.
One Sample is Enough to Make Conformal Prediction Robust
Soroush H. Zargarbashi (CISPA Helmholtz Center for Information Security), Aleksandar Bojchevski (University of Cologne)
Object DetectionSegmentationAutonomous DrivingComputational EfficiencyTransformerGaussian SplattingImage
🎯 What it does: This paper proposes a method called RCP1 that achieves robust synthesized prediction with a single forward pass.
One SPACE to Rule Them All: Jointly Mitigating Factuality and Faithfulness Hallucinations in LLMs
Pengbo Wang (Beijing University of Posts and Telecommunications), Xi Zhang (Beijing University of Posts and Telecommunications)
TransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposes the SPACE framework, which utilizes editing of the activation subspace to simultaneously suppress the factuality and fidelity hallucinations of LLMs;
One Stone with Two Birds: A Null-Text-Null Frequency-Aware Diffusion Models for Text-Guided Image Inpainting
Haipeng Liu (Hefei University of Technology), Meng Wang (Hefei University of Technology)
RestorationGenerationDiffusion modelImageText
🎯 What it does: A null-text-null frequency domain aware diffusion model NTN-Diff is proposed for text-guided image inpainting, which can simultaneously preserve the unmasked areas from being damaged and ensure that the inpainted areas are semantically consistent with the text.
One Subgoal at a Time: Zero-Shot Generalization to Arbitrary Linear Temporal Logic Requirements in Multi-Task Reinforcement Learning
Zijian Guo (Boston University), Wenchao Li (Boston University)
Reinforcement LearningSequential
🎯 What it does: Proposes the GenZ-LTL method, achieving zero-shot satisfaction of any linear temporal logic (LTL) specifications;
One Token Embedding Is Enough to Deadlock Your Large Reasoning Model
Mohan Zhang (University of North Carolina), Tianlong Chen
Computational EfficiencyAdversarial AttackLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper studies an attack method called Deadlock Attack, which uses a single malicious embedding token to induce large-scale reasoning models (LRMs) into an infinite chain reasoning loop (i.e., 'deadlock'), leading to resource exhaustion.
One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding
Zheyu Aqa Zhang (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
RecognitionCompressionTransformerLarge Language ModelVideo
🎯 What it does: This paper proposes an extreme video token compression framework called XComp, which retains only one efficient token per frame in the final layer of the LLM for long videos, and further improves information utilization through frame-level filtering.
One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution
Yujing Sun (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RestorationSuper ResolutionDiffusion modelOptical FlowVideo
🎯 What it does: The Dual LoRA Learning (DLoRAL) framework is proposed, using a first-order diffusion model to achieve video super-resolution while considering spatial details and temporal consistency.
One-Step Diffusion-Based Image Compression with Semantic Distillation
Naifu Xue (Communication University of China), Yan Lu (Microsoft Research Asia)
GenerationCompressionKnowledge DistillationDiffusion modelAuto EncoderImage
🎯 What it does: A first-order diffusion generative image compression framework called OneDC is proposed, which achieves ultra-low bitrate image compression by combining a latent compression module and a first-order diffusion generator.
One-Step is Enough: Sparse Autoencoders for Text-to-Image Diffusion Models
Viacheslav Surkov (École Polytechnique Fédérale de Lausanne), David Bau (Northeastern University)
GenerationData SynthesisExplainability and InterpretabilityTransformerDiffusion modelAuto EncoderImageTextBenchmark
🎯 What it does: Deconstructs the intermediate representation of SDXL Turbo (a few-step text-to-image diffusion model) using Sparse Autoencoders (SAE) and performs image editing through feature activation; also proposes the RIEBench evaluation benchmark.
One-Step Offline Distillation of Diffusion-based Models via Koopman Modeling
Nimrod Berman (Ben-Gurion University of the Negev), Omri Azencot (Ben-Gurion University of the Negev)
GenerationKnowledge DistillationDiffusion modelImage
🎯 What it does: A method for offline single-step diffusion model distillation based on Koopman theory, KDM, is proposed, which can achieve high-quality image generation with a single forward pass.
Online Bilateral Trade With Minimal Feedback: Don’t Waste Seller’s Time
Francesco Bacchiocchi (Politecnico di Milano), Alberto Marchesi (Politecnico di Milano)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes a new asynchronous mechanism for online bilateral trading, aimed at reducing time waste for sellers. The mechanism only queries the seller after the buyer accepts the offer, thereby improving feedback efficiency.
Online Experimental Design With Estimation-Regret Trade-off Under Network Interference
Zhiheng Zhang (Shanghai University of Finance and Economics), Zichen Wang (University of Illinois Urbana-Champaign)
Reinforcement Learning
🎯 What it does: A unified framework for online experimental design of network interventions (MAB-N) is proposed, providing a Pareto optimal trade-off between estimation error and cumulative regret, and further designing a two-stage UCB-TSN/EXP3-TSN algorithm to achieve this goal.
Online Feedback Efficient Active Target Discovery in Partially Observable Environments
Anindya Sarkar (Washington University in St Louis), Yevgeniy Vorobeychik (Washington University in St Louis)
Object DetectionSegmentationOptimizationExplainability and InterpretabilityReinforcement LearningDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a method called Diffusion-guided Active Target Discovery (DiffATD) for efficiently discovering targets in partially observable environments under a limited sampling budget.
Online Functional Tensor Decomposition via Continual Learning for Streaming Data Completion
Xi Zhang (Xi'an Jiaotong University), Deyu Meng (Xi'an Jiaotong University)
OptimizationTime Series
🎯 What it does: An online functional tensor decomposition framework (OFTD) is proposed, which combines CP decomposition with implicit neural representations to handle missing completion in streaming tensors.
Online Inverse Linear Optimization: Efficient Logarithmic-Regret Algorithm, Robustness to Suboptimality, and Lower Bound
Shinsaku Sakaue (CyberAgent), Taihei Oki (Institute of Statistical Mathematics)
Optimization
🎯 What it does: In the online inverse linear optimization problem, a new learning algorithm is proposed that maintains low time complexity at each step and achieves a logarithmic level cumulative loss upper bound.
Online Learning in the Repeated Mediated Newsvendor Problem
Nataša Bolić (University of Ottawa), Christian Paravalos (University of Ottawa)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper studies a repeated intermediary version of the newsvendor problem, designs an online pricing strategy, and proves its revenue maximization.
Online Learning of Neural Networks
Amit Daniely, Elchanan Mossel
🎯 What it does: This paper studies the upper and lower bounds of error for feedforward neural networks using symbolic activation functions in online learning environments.
Online Learning of Pure States is as Hard as Mixed States
Maxime Meyer (National University of Singapore), Patrick Rebentrost (National University of Singapore)
OptimizationAdversarial AttackSequentialPhysics Related
🎯 What it does: This paper studies quantum state learning within the framework of online learning, proving that the difficulty of learning pure states is comparable to that of learning mixed states, with both being nearly identical in terms of sequential fat-shattering dimension, leading to the same regret scaling.
Online Locally Differentially Private Conformal Prediction via Binary Inquiries
Qiangqiang Zhang (Zhongtai Securities Institute for Financial Studies, Shandong University), Ting Li (Shanghai University of Finance and Economics)
OptimizationSafty and PrivacyTabularTime Series
🎯 What it does: An online, local differential privacy (LDP) framework is proposed, utilizing binary queries to construct prediction sets for real-time uncertainty quantification.
Online Mixture of Experts: No-Regret Learning for Optimal Collective Decision-Making
Larkin Liu (Technische Universitat Munchen), Jalal Etesami (Technische Universitat Munchen)
Recommendation SystemOptimizationLarge Language ModelMixture of ExpertsText
🎯 What it does: Proposed an Online Mixture of Experts (OMoE) framework to achieve optimal accuracy by aggregating expert outputs through voting in real-time environments;
Online Multi-Class Selection with Group Fairness Guarantee
Faraz Zargari (University of Alberta), Xiaoqi Tan (University of Alberta)
Optimization
🎯 What it does: This paper studies the online multi-class selection problem, providing integer and randomized algorithms with group fairness guarantees, and proposes a learning-enhanced version.
Online Optimization for Offline Safe Reinforcement Learning
Yassine Chemingui (Washington State University), Jana Doppa
OptimizationSafty and PrivacyReinforcement LearningTabular
🎯 What it does: This paper proposes an offline safe reinforcement learning framework O3SRL based on online optimization, which iteratively updates the Lagrange multipliers using offline RL or its stochastic approximation operators along with multi-armed bandit algorithms, ultimately obtaining a policy that maximizes rewards while satisfying cost constraints.
Online Portfolio Selection with ML Predictions
Ziliang Zhang (University of Sydney), Albert Zomaya
Recommendation SystemOptimizationTime SeriesFinance Related
🎯 What it does: This paper proposes the Rebalanced Arithmetic Mean (RAM) online portfolio algorithm, which dynamically reallocates weights based on asset return rankings provided by machine learning, ensuring safe returns even in the worst-case scenario.
Online Prediction with Limited Selectivity
Licheng Liu (Imperial College London), Mingda Qiao (University of Massachusetts Amherst)
🎯 What it does: A new online prediction model called Prediction with Limited Selectivity (PLS) is proposed, which studies the scenario where predictions can only be made at predefined time points, and provides upper and lower bounds on the error for this model.
Online robust locally differentially private learning for nonparametric regression
Chenfei Gu (Shanghai University of Finance and Economics), Niansheng Tang (Yunnan University)
OptimizationSafty and PrivacyTabular
🎯 What it does: This paper proposes an online robust local differential privacy (LDP) non-parametric regression algorithm H-FSGD and PH-FSGD, which can handle large-scale streaming data in a single pass and real-time update environment.
Online Segment Any 3D Thing as Instance Tracking
Hanshi Wang (Chinese Academy of Sciences), Zhipeng Zhang (Shanghai Jiao Tong University)
Object TrackingSegmentationTransformerVision Language ModelPoint Cloud
🎯 What it does: This paper proposes an online real-time 3D instance segmentation framework called AutoSeg3D, which directly uses 2D masks obtained from Vision Foundation Models (such as SAM) as queries in each frame, and maintains long-term and short-term memory in the temporal dimension to achieve instance identity continuity and immediate context updates.
Online Strategic Classification With Noise and Partial Feedback
Tianrun Zhao (Tsinghua University), Yong Liang (Tsinghua University)
ClassificationOptimizationReinforcement LearningTabular
🎯 What it does: This study investigates the online strategic classification problem with noise and partial feedback, proposing a phased learning algorithm and proving its convergence and sublinear loss upper bound.
Online Time Series Forecasting with Theoretical Guarantees
Zijian Li (Carnegie Mellon University), Kun Zhang (Carnegie Mellon University)
Anomaly DetectionOptimizationRecurrent Neural NetworkAuto EncoderTime Series
🎯 What it does: A theoretical framework (TOT) is proposed to reduce Bayesian risk in online time series forecasting by utilizing latent variables, and it is proven that the latent states and their causal dynamics can be uniquely identified from four consecutive observations; subsequently, a pluggable neural network architecture is designed to achieve automatic estimation of latent variables with sparse mixing constraints.
Online Two-Stage Submodular Maximization
Iasonas Nikolaou (Boston University), Evimaria Terzi
Recommendation SystemOptimizationTabular
🎯 What it does: This paper proposes the Online Two-Stage Submodular Maximization (O2SSM) problem and presents the RAOCO algorithm.
OnlineSplatter: Pose-Free Online 3D Reconstruction for Free-Moving Objects
Mark He Huang (Singapore University of Technology and Design), De Wen Soh (Max Planck Institute for Informatics)
TransformerGaussian SplattingVideo
🎯 What it does: A posture-free, real-time online monocular RGB video framework for 3D reconstruction of freely moving objects has been implemented.
OOD Detection with Relative Angles
Berker Demirel (Institute of Science and Technology Austria), Francesco Locatello (Institute of Science and Technology Austria)
Anomaly DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: A post-hoc OOD detection method (ORA) based on the angle of feature vectors relative to the decision boundary has been developed.
OOD-Barrier: Build a Middle-Barrier for Open-Set Single-Image Test Time Adaptation via Vision Language Models
Boyang Peng (Tongji University), Guang Chen (Tongji University)
Domain AdaptationTransformerPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes an open-set single-image testing adaptive framework called Open-IRT, which addresses the ID-OOD separation and adaptation issues under real-time single-sample streams.
Open Vision Reasoner: Transferring Linguistic Cognitive Behavior for Visual Reasoning
Yana Wei (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: This paper constructs a two-stage training framework based on Qwen2.5-VL-7B: first performing large-scale language cold start, followed by cross-modal reinforcement learning, thereby enhancing visual reasoning capabilities.
Open-Reasoner-Zero: An Open Source Approach to Scaling Up Reinforcement Learning on the Base Model
Jingcheng Hu (Tsinghua University), Heung-Yeung Shum
Reinforcement LearningText
🎯 What it does: Developed and publicly implemented Open-Reasoner-Zero, which directly applies large-scale reinforcement learning to enhance reasoning capabilities.
Open-Vocabulary Part Segmentation via Progressive and Boundary-Aware Strategy
Xinlong Li (Tianjin University), Qing Guo (Nankai University)
SegmentationDiffusion modelImageBenchmark
🎯 What it does: This paper proposes an untrained open vocabulary part segmentation framework PBAPS, which achieves fine segmentation of the boundaries of structurally connected parts using a hierarchical structure graph and a boundary-aware refinement module.
Open-World Drone Active Tracking with Goal-Centered Rewards
Haowei Sun (South China University of Technology), Mingkui Tan (South China University of Technology)
Object TrackingConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningVideoBenchmark
🎯 What it does: This paper proposes an open-world drone active tracking benchmark DAT and a reinforcement learning method GC-VAT based on goal-centered rewards and curriculum learning.
OpenBox: Annotate Any Bounding Boxes in 3D
In-Jae Lee (Seoul National University), Jaesik Park (Seoul National University)
Object DetectionSegmentationAutonomous DrivingPoint Cloud
🎯 What it does: Proposes OpenBox, a two-stage automatic 3D box annotation pipeline that utilizes 2D vision foundation models.
OpenCUA: Open Foundations for Computer-Use Agents
Xinyuan Wang (University of Hong Kong), Tao Yu (Stanford University)
TransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: An open computer usage agent (CUA) framework called OPENCUA is proposed, which includes cross-operating system annotation tools, a massive desktop task dataset AGENTNET, and a processing pipeline from human demonstrations to state-action pairs along with a Reflective Long Chain of Thought (CoT) method, resulting in a VISION-LANGUAGE model that can directly execute tasks in real environments.
OpenHOI: Open-World Hand-Object Interaction Synthesis with Multimodal Large Language Model
Zhenhao Zhang (ShanghaiTech University), Jingya Wang (ShanghaiTech University)
Data SynthesisRobotic IntelligenceTransformerLarge Language ModelDiffusion modelMultimodality
🎯 What it does: The OpenHOI framework is proposed to achieve open-world hand-object interaction sequence synthesis, capable of generating long-term interactions for new objects from open-ended language instructions.
OpenHype: Hyperbolic Embeddings for Hierarchical Open-Vocabulary Radiance Fields
Lisa Weijler (TU Wien), Pedro Hermosilla (TU Wien)
SegmentationGenerationNeural Radiance FieldAuto EncoderContrastive LearningPoint Cloud
🎯 What it does: This paper proposes OpenHype, a method for open vocabulary segmentation that utilizes hyperbolic space embedding to achieve a 3D scene hierarchy, training unified semantic features based on NeRF and enabling continuous hierarchical queries.
OpenMMEgo: Enhancing Egocentric Understanding for LMMs with Open Weights and Data
Hao Luo (Peking University), Zongqing Lu (Peking University)
RecognitionData SynthesisCompressionTransformerLarge Language ModelVision Language ModelVideoTextBenchmark
🎯 What it does: A large-scale self-supervised egocentric video QA dataset OME10M is constructed, and the OpenMMEgo framework is proposed to enhance LMM performance in first-person video understanding.
OpenOmni: Advancing Open-Source Omnimodal Large Language Models with Progressive Multimodal Alignment and Real-time Emotional Speech Synthesis
Run Luo (Shenzhen Key Laboratory for High Performance Data Mining, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Min Yang (Shenzhen Key Laboratory for High Performance Data Mining, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
GenerationData SynthesisTransformerLarge Language ModelMixture of ExpertsTextMultimodalityAudio
🎯 What it does: OpenOmni has been constructed and trained, which is a two-stage open-source multimodal large language model. It first uses text as a hub to achieve speech-text alignment and image-text alignment, enabling zero-shot multimodal alignment. Subsequently, a lightweight parallel speech decoder is trained, combined with DPO for emotional speech generation, achieving real-time emotional expression.
OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles
Yihe Deng (University of California), Kai-Wei Chang (University of California)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: This paper proposes a training framework based on an iterative SFT-RL cycle, creating an open-source LVLM—OpenVLThinker-7B, capable of performing complex Chain-of-Thought reasoning in visual tasks.
OpenWorldSAM: Extending SAM2 for Universal Image Segmentation with Language Prompts
Shiting Xiao (Yale University), Priyadarshini Panda (Yale University)
SegmentationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes OpenWorldSAM, an open-world text prompt semantic segmentation framework built on SAM2, which can accurately generate multi-instance, semantic, and panoptic segmentation masks based on open vocabulary text descriptions (from words to complete sentences).
OPHR: Mastering Volatility Trading with Multi-Agent Deep Reinforcement Learning
Zeting Chen (Nanyang Technological University), Bo An (Nanyang Technological University)
OptimizationReinforcement LearningAgentic AITime SeriesFinance Related
🎯 What it does: A multi-agent reinforcement learning framework for cryptocurrency options volatility trading, OPHR, is proposed, which includes options position decision-making (OP-Agent) and hedging strategy selection (HR-Agent), achieving a synergistic optimization of volatility timing and risk management.
Opinion Maximization in Social Networks by Modifying Internal Opinions
Gengyu Wang (Fudan University), Zhongzhi Zhang (Fudan University)
OptimizationGraph Neural NetworkGraph
🎯 What it does: The study maximizes the overall expressed opinion in social networks by changing the internal opinions of a fixed number of nodes.
OPMapper: Enhancing Open-Vocabulary Semantic Segmentation with Multi-Guidance Information
Xuehui Wang (Shanghai Jiao Tong University), Wei Shen (Shanghai Jiao Tong University)
SegmentationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: A lightweight pluggable module OPMapper is proposed to enhance the performance of Open-Vocabulary Semantic Segmentation (OVSS).
OPTFM: A Scalable Multi-View Graph Transformer for Hierarchical Pre-Training in Combinatorial Optimization
Hao Yuan (Lenovo Research), Yong Sun (Lenovo Research)
OptimizationGraph Neural NetworkTransformerContrastive LearningGraph
🎯 What it does: A graph-based model for combinatorial optimization, OPTFM, is proposed, which can directly generate high-quality node and instance representations without the need for fine-tuning and achieves superior performance in multiple downstream tasks.
Optical Coherence Tomography Harmonization with Anatomy-Guided Latent Metric Schrödinger Bridges
Shuwen Wei (Johns Hopkins University), Jerry L Prince
Image HarmonizationSegmentationDiffusion modelImageBiomedical Data
🎯 What it does: This paper proposes a method for unpaired optical coherence tomography (OCT) image harmonization based on reversible networks and latent Euclidean space;
Optimal Adjustment Sets for Nonparametric Estimation of Weighted Controlled Direct Effect
Ruiyang Lin (University of Science and Technology of China), Kyra Gan (Cornell University)
TabularBiomedical DataAgriculture Related
🎯 What it does: This paper proposes the identifiability conditions, influence function, and optimal adjustment set for weighted control direct effects (WCDE) in observational studies, and validates the AIPW estimator.
Optimal and Provable Calibration in High-Dimensional Binary Classification: Angular Calibration and Platt Scaling
Yufan Li (Harvard University), Pragya Sur (Harvard University)
ClassificationOptimizationTabular
🎯 What it does: Under high-dimensional Gaussian design, an angle calibration method is proposed for the probability prediction of linear binary classifiers, which can prove calibration in the proportional limit where both the sample size and feature dimensions diverge. It is also shown that this predictor has unique optimality in terms of Bregman loss among all calibratable predictors. Furthermore, it is demonstrated that the classic Platt scaling converges to this optimal calibrator in the high-dimensional limit, thus providing it with provable calibration and optimality in high-dimensional scenarios.
Optimal Best Arm Identification under Differential Privacy
Marc Jourdan (École Polytechnique Fédérale de Lausanne), Achraf Azize (ENSAE Paris)
OptimizationSafty and PrivacyTabular
🎯 What it does: A fixed-confidence best arm identification method under global differential privacy constraints is proposed, providing a tight lower bound and designing the DP-TT algorithm to achieve approximately optimal sample complexity.
Optimal community detection in dense bipartite graphs
Julien Chhor (Toulouse School of Economics), Parker Knight (Harvard University)
OptimizationGraph
🎯 What it does: Study the sparse community detection problem in high-dimensional dense bipartite graphs and provide detection limits.
Optimal Control for Transformer Architectures: Enhancing Generalization, Robustness and Efficiency
Kelvin Kan (University of California Los Angeles), Markos Katsoulakis
OptimizationTransformerSupervised Fine-TuningTextPoint Cloud
🎯 What it does: This paper studies the training and architecture of the Transformer by viewing it as a continuous-time dynamical system and proposes the OT-Transformer model using optimal control theory.
Optimal Dynamic Regret by Transformers for Non-Stationary Reinforcement Learning
Baiyuan Chen (University of Tokyo), Masaaki Imaizumi (University of Tokyo)
TransformerReinforcement LearningTabular
🎯 What it does: This study investigates the performance of Transformer in non-stationary reinforcement learning environments and demonstrates that it can achieve near-optimal dynamic loss.
Optimal Estimation of the Best Mean in Multi-Armed Bandits
Takayuki Osogami (IBM Research), Junpei Komiyama (New York University)
OptimizationReinforcement Learning from Human Feedback
🎯 What it does: This paper studies the optimal arm mean estimation problem in multi-armed bandits (MAB) and proposes a PAC algorithm called EllipsoidEst.
Optimal Graph Clustering without Edge Density Signals
Maximilien Dreveton (École Polytechnique Fédérale de Lausanne), Patrick Thiran (École Polytechnique Fédérale de Lausanne)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper studies the optimal error rate of graph clustering under the Popularity Adjusted Block Model (PABM), revealing the possibility of achieving clustering even when traditional edge density signals disappear.
Optimal kernel regression bounds under energy-bounded noise
Amon Lahr, Melanie Zeilinger
OptimizationTabular
🎯 What it does: This paper proposes a non-asymptotic, distribution-free uncertainty bound for kernel regression, which provides an optimal point estimation error bound under energy-constrained noise.
Optimal Minimum Width for the Universal Approximation of Continuously Differentiable Functions by Deep Narrow MLPs
Geonho Hwang (Gwangju Institute of Science and Technology)
Optimization
🎯 What it does: This paper studies the universal approximation properties of deep narrow multilayer perceptrons (MLPs) for C1 functions under the Sobolev norm, particularly the W1,∞ norm. We demonstrate that the optimal width can be determined in a wide range of cases in the C1 setting.
Optimal Mistake Bounds for Transductive Online Learning
Zachary Chase (Kent State University), Jonathan Shafer (Massachusetts Institute of Technology)
🎯 What it does: This paper proves that both the upper and lower bounds of error in transductive online learning are Θ(√d), and provides the corresponding concept class construction.
Optimal Neural Compressors for the Rate-Distortion-Perception Tradeoff
Eric Lei (JPMorganChase Global Technology Applied Research), Shirin Saeedi Bidokhti (University of Pennsylvania)
CompressionImagePhysics RelatedAudio
🎯 What it does: This paper designs and implements a low-complexity neural compressor that achieves optimal or near-optimal compression for the R-D-P trade-off using lattice coding and various shared randomness designs.
Optimal Nuisance Function Tuning for Estimating a Doubly Robust Functional under Proportional Asymptotics
Sean McGrath (Yale University), Zixiao Wang
OptimizationTabular
🎯 What it does: This paper explores the optimal tuning parameter selection for estimating the disturbance function of statistical functions using ridge regression under proportional asymptotic conditions, particularly the estimation of the Expected Conditional Covariance (ECC).
Optimal Online Change Detection via Random Fourier Features
Florian Kalinke (Karlsruhe Institute of Technology), Shakeel A O B Gavioli-Akilagun
Anomaly DetectionTabularTime Series
🎯 What it does: A completely online, windowless parameter multidimensional data stream nonparametric change point detection algorithm called Online RFF-MMD is proposed.
Optimal Rates for Generalization of Gradient Descent for Deep ReLU Classification
Yuanfan Li (Zhejiang University), Yiming Ying (University of Sydney)
OptimizationTabular
🎯 What it does: This study investigates the optimization and generalization performance of gradient descent on deep ReLU networks and provides an optimal risk upper bound.
Optimal Rates in Continual Linear Regression via Increasing Regularization
Ran Levinstein (Technion), Itay Evron (Technion)
Optimization
🎯 What it does: This paper studies the achievable continuous linear regression problem under random task order, proposing explicit homoscedastic ℓ2 regularization and implicit finite-step budget regularization, and reducing it to incremental gradient descent.
Optimal Regret Bounds via Low-Rank Structured Variation in Non-Stationary Reinforcement Learning
Tuan Quang Dam (Hanoi University of Science and Technology)
OptimizationReinforcement LearningTime Series
🎯 What it does: This paper proposes the SVUCRL algorithm for achieving low regret in non-stationary communication MDPs (where transition probabilities and rewards change over time) and provides three dynamic regret upper bounds.
Optimal Regret of Bandits under Differential Privacy
Achraf Azize (ENSAE Paris), Debabrota Basu (University of Lille)
OptimizationSafty and PrivacyTabular
🎯 What it does: This paper presents an optimal regret analysis for stochastic multi-armed bandits under ε-global differential privacy (DP) constraints and designs two new DP algorithms, DP KLUCB and DP IMED.
Optimal Single-Policy Sample Complexity and Transient Coverage for Average-Reward Offline RL
Matthew Zurek (University of Wisconsin Madison), Yudong Chen (University of Wisconsin Madison)
Reinforcement LearningTabular
🎯 What it does: This paper studies the sample complexity and theoretical limits under the single policy coverage condition in average-reward offline reinforcement learning (average-reward offline RL), and proposes a pessimistic discounted value iteration algorithm based on quantile clipping.
Optimal Spectral Transitions in High-Dimensional Multi-Index Models
Leonardo Defilippis (École Normale Supérieure), Bruno Loureiro (École Polytechnique Fédérale de Lausanne)
OptimizationTabular
🎯 What it does: This paper studies how to reconstruct the relevant index subspace in high-dimensional multi-index models through sample reconstruction, proposing a spectral algorithm based on a linearized message passing scheme, and proving that these methods achieve the optimal reconstruction threshold.
Optimality and NP-Hardness of Transformers in Learning Markovian Dynamical Functions
Yanna Ding (Rensselaer Polytechnic Institute), Jianxi Gao (IBM Research)
OptimizationTransformerTime SeriesSequential
🎯 What it does: This study investigates the context learning (ICL) capability of Transformers when processing time series data generated by Markov chains, analyzes the global optimal solution and reachability of the single-layer linear self-attention (LSA) model, and explores the multi-objective optimization corresponding to the forward propagation of multi-layer LSA.
Optimism Without Regularization: Constant Regret in Zero-Sum Games
John Lazarsfeld (Singapore University of Technology and Design), Stratis Skoulakis (Aarhus University)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper studies the learning performance of Optimistic Fictitious Play without regularization in 2×2 zero-sum games and proves that its total regret is a constant;
Optimistic Online-to-Batch Conversions for Accelerated Convergence and Universality
Yu-Hu Yan (Nanjing University), Zhi-Hua Zhou (Nanjing University)
OptimizationTabular
🎯 What it does: An Optimistic Online-to-Batch (O2B) conversion framework is proposed, which achieves accelerated convergence for convex smooth optimization and provides optimal algorithms in both strongly convex and smooth, as well as general (without the need for known smooth constants) scenarios.
Optimistic Query Routing in Clustering-based Approximate Maximum Inner Product Search
Sebastian Bruch (Northeastern University), Franco Maria Nardini (ISTI-CNR)
RetrievalOptimizationTextBenchmark
🎯 What it does: In clustering-based approximate maximum inner product search, a router called OPTIMIST is proposed based on the principle of 'optimism towards uncertainty' to more accurately select the shards required for queries.
Optimization Inspired Few-Shot Adaptation for Large Language Models
Boyan Gao (University of Oxford), David A. Clifton (University of Oxford)
OptimizationTransformerLarge Language ModelText
🎯 What it does: By learning the hierarchical precondition matrix, the internal optimization process of LLM is improved to achieve task adaptation with few samples.
Optimize Any Topology: A Foundation Model for Shape- and Resolution-Free Structural Topology Optimization
Amin Heyrani Nobari (Massachusetts Institute of Technology), Faez Ahmed (Massachusetts Institute of Technology)
OptimizationDiffusion modelAuto EncoderPoint Cloud
🎯 What it does: A topology optimization framework OAT based on a foundational model is proposed, which can directly generate approximately optimal material distributions under arbitrary shapes, aspect ratios, and resolutions.
Optimize the Unseen - Fast NeRF Cleanup with Free Space Prior
Leo Segre (Tel Aviv University), Shai Avidan (Tel Aviv University)
RestorationOptimizationNeural Radiance FieldImage
🎯 What it does: A fast post-processing method is proposed to remove floaters generated by NeRF using a global Free Space Prior.
Optimized Minimal 3D Gaussian Splatting
Joo Chan Lee (Sungkyunkwan University), Eunbyung Park (Yonsei University)
CompressionOptimizationGaussian SplattingPoint Cloud
🎯 What it does: An Optimized Minimal Gaussian Splatting (OMG) framework is proposed, achieving extremely low storage requirements and high frame rate rendering by minimizing the number of high-precision Gaussian primitives.
Optimizing Anytime Reasoning via Budget Relative Policy Optimization
Penghui Qi (Sea AI Lab), Min Lin (National University of Singapore)
OptimizationLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: A framework named AnytimeReasoner is proposed, which utilizes budget sampling, dense verifiable rewards, and Budget Relative Policy Optimization (BRPO) to achieve anytime reasoning for LLMs and enhance their performance.
Optimizing Chain-of-Thought Reasoners via Gradient Variance Minimization in Rejection Sampling and RL
Jiarui Yao (University of Illinois Urbana-Champaign), Tong Zhang (University of Illinois Urbana-Champaign)
OptimizationTransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: This paper proposes a dynamic sampling strategy based on gradient variance minimization (GVM) to optimize rejection sampling and reinforcement learning methods in chain-of-thought (CoT) training;
Optimizing Distributional Geometry Alignment with Optimal Transport for Generative Dataset Distillation
Xiao Cui (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
GenerationOptimizationKnowledge DistillationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a generative dataset distillation framework based on optimal transport (OT), which achieves high-fidelity transfer from large-scale real data to small-scale distilled data through three-phase synchronous optimization: OT-guided diffusion sampling, soft re-labeling for label-image alignment, and OT logical matching, while maintaining the geometric structure of the distribution.
Optimizing Retrieval for RAG via Reinforced Contrastive Learning
Jiawei Zhou (Hong Kong University of Science and Technology), Lei Chen (Hong Kong University of Science and Technology)
RetrievalOptimizationTransformerLarge Language ModelReinforcement LearningContrastive LearningTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Introducing a reinforcement learning-based contrastive learning framework R3 in the RAG environment, dynamically exploring and optimizing the relevance of the interaction between the retriever and LLM to improve retrieval quality;
Optimizing the Unknown: Black Box Bayesian Optimization with Energy-Based Model and Reinforcement Learning
RUIYAO MIAO, Ying Nian Wu
OptimizationReinforcement LearningTabular
🎯 What it does: Proposes the REBMBO framework, which combines Gaussian process local modeling, energy-based model global guidance, and PPO multi-step planning to address the single-step blindness of Bayesian optimization.
Option-aware Temporally Abstracted Value for Offline Goal-Conditioned Reinforcement Learning
Hongjoon Ahn (Seoul National University), Taesup Moon (Seoul National University)
Reinforcement LearningTabular
🎯 What it does: This paper proposes an option-based temporal abstraction value learning method called OTA, aimed at enhancing the performance of high-level policies in offline goal-oriented reinforcement learning.
OptiScene: LLM-driven Indoor Scene Layout Generation via Scaled Human-aligned Data Synthesis and Multi-Stage Preference Optimization
Yixuan Yang (Southern University of Science and Technology), Feng Zheng
GenerationData SynthesisOptimizationRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPoint Cloud
🎯 What it does: A large indoor layout dataset called 3D-SynthPlace has been constructed, and based on this, an indoor layout generation framework called OptiScene has been developed using LLM.
OptiTree: Hierarchical Thoughts Generation with Tree Search for LLM Optimization Modeling
Haoyang Liu (University of Science and Technology of China), Jianye HAO
OptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: By constructing a hierarchical 'model tree' and utilizing tree search, the complex operations research modeling task is decomposed into a series of subproblems, thereby guiding large language models to generate more accurate optimization models.
Oracle-Efficient Combinatorial Semi-Bandits
Jung-hun Kim (ENSAE Paris), Min-hwan Oh (Seoul National University)
OptimizationReinforcement LearningTabular
🎯 What it does: Two sparse oracle query frameworks (adaptive and planned) are proposed, reducing oracle calls from linear to double logarithmic in the combinatorial semi-bandit problem, while maintaining an almost optimal no-gap O(√T) scheduling reward.
OrbitZoo: Real Orbital Systems Challenges for Reinforcement Learning
Alexandre Oliveira (Nova University of Lisbon), Claudia Soares
Robotic IntelligenceReinforcement Learning
🎯 What it does: Developed OrbitZoo, an RL environment based on Orekit high-fidelity orbital dynamics and the PettingZoo multi-agent framework, supporting tasks such as satellite maneuvers, collision avoidance, and constellation coordination.
Order-Level Attention Similarity Across Language Models: A Latent Commonality
Jinglin Liang (South China University of Technology), Hanlin Gu (Hong Kong University of Science and Technology)
TransformerLarge Language ModelText
🎯 What it does: Exploring the common patterns of context aggregation in different language models, we propose Order-Level Attention (OLA) and implement a training-free cross-model adapter transfer (TOA) based on it.