NeurIPS 2025 Papers — Page 34
Conference on Neural Information Processing Systems · 5275 papers
OrdShap: Feature Position Importance for Sequential Black-Box Models
Davin Hill (Northeastern University), Jennifer Dy (Northeastern University)
Explainability and InterpretabilityTransformerTextSequentialBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes a new feature attribution method called OrdShap, specifically designed for sequence deep learning models, which can separate the impact of feature values and their positions in the sequence on model predictions.
Orient Anything V2: Unifying Orientation and Rotation Understanding
Zehan Wang (Zhejiang University), Zhou Zhao (Sea AI Lab)
Pose EstimationTransformerVision Language ModelImage
🎯 What it does: Proposes Orient Anything V2, an integrated model for single-view absolute pose estimation and multi-frame relative rotation prediction, supporting rotation symmetry recognition.
Orientation Matters: Making 3D Generative Models Orientation-Aligned
Yichong Lu (Zhejiang University), Yiyi Liao (Zhejiang University)
GenerationPose EstimationVision Language ModelDiffusion modelPoint CloudMesh
🎯 What it does: This study investigates the directional consistency of 3D generative models, constructs the Objaverse-OA dataset, and fine-tunes existing 3D generative models (Trellis, Wonder3D) to generate 3D models with consistent orientations. It also proposes zero-shot pose estimation and arrow-based rotation manipulation.
Orientation-anchored Hyper-Gaussian for 4D Reconstruction from Casual Videos
Junyi Wu (University of Illinois Chicago), Yan Yan (University of Illinois Chicago)
RestorationDepth EstimationGaussian SplattingVideo
🎯 What it does: A framework named Orientation-anchored Gaussian Splatting (OriGS) is proposed for high-quality reconstruction of four-dimensional (time + space) scenes from casually captured monocular videos.
ORIGAMISPACE: Benchmarking Multimodal LLMs in Multi-Step Spatial Reasoning with Mathematical Constraints
Rui Xu, Xu Yinghui
AI Code AssistantTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextMultimodalityBenchmark
🎯 What it does: This paper designs and releases the ORIGAMISPACE dataset along with four multimodal LLM evaluation tasks (pattern prediction, multi-step spatial reasoning, spatial relationship prediction, and CP code generation), and conducts experiments and evaluations based on an improved origami compiler.
ORIGEN: Zero-Shot 3D Orientation Grounding in Text-to-Image Generation
Yunhong Min (KAIST), Minhyuk Sung (KAIST)
GenerationPose EstimationReinforcement LearningFlow-based ModelImageBenchmark
🎯 What it does: We propose ORIGEN, a zero-shot method for achieving 3D orientation (pose) localization in text-to-image generation, supporting multi-object and multi-category scenes.
Orochi: Versatile Biomedical Image Processor
Gaole Dai (Peking University), Shanghang Zhang (Peking University)
RestorationSuper ResolutionSupervised Fine-TuningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A general foundational model for multi-task, low-level medical image processing called Orochi is proposed, which achieves unified processing of tasks such as registration, fusion, denoising, and super-resolution through self-supervised techniques like random multi-scale sampling and Task-related Joint-embedding Pre-Training.
Orthogonal Contrastive Learning for Multi-Representation fMRI Analysis
Tony Yousefnezhad
Domain AdaptationRepresentation LearningTransformerContrastive LearningBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes an Orthogonal Contrastive Learning (OCL) framework for aligning and feature learning of multi-subject and cross-site task fMRI data without the need for temporal alignment and uniform sequence lengths.
Orthogonal Survival Learners for Estimating Heterogeneous Treatment Effects from Time-to-Event Data
Dennis Frauen (Ludwig Maximilian University of Munich), Stefan Feuerriegel (Ludwig Maximilian University of Munich)
OptimizationDrug DiscoveryTime SeriesBiomedical Data
🎯 What it does: A set of Neyman-orthogonal survival learners is proposed to estimate heterogeneous treatment effects (HTE) in the presence of censored time-to-event data.
Oryx: a Scalable Sequence Model for Many-Agent Coordination in Offline MARL
Juan Claude Formanek, Arnu Pretorius (Stellenbosch University)
Reinforcement LearningSequential
🎯 What it does: This paper presents Oryx, an autoregressive sequence model for offline multi-agent reinforcement learning, focusing on large-scale multi-agent collaboration and long-term temporal dependencies.
OSCAR: One-Step Diffusion Codec Across Multiple Bit-rates
Jinpei Guo (Carnegie Mellon University), Yulun Zhang (Shanghai Jiao Tong University)
CompressionDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A first-order diffusion code (OSCAR) is proposed, which achieves multi-bitrate image compression in a single-step inference and requires training only one unified model.
OSKAR: Omnimodal Self-supervised Knowledge Abstraction and Representation
Mohamed O Abdelfattah, Alexandre Alahi (Ecole Polytechnique Federale de Lausanne)
RecognitionRetrievalRepresentation LearningTransformerVideoTextMultimodality
🎯 What it does: We propose OSKAR, a self-supervised multimodal foundation model that leverages cross-modal latent space prediction to learn rich cross-modal representations.
OSTAR: Optimized Statistical Text-classifier with Adversarial Resistance
Yuhan Yao (Beijing University of Posts and Telecommunications), LI Haisheng
ClassificationAdversarial AttackTransformerLarge Language ModelContrastive LearningText
🎯 What it does: The OSTAR framework is proposed, which achieves robust detection of machine-generated text through statistical feature profiling and multi-faceted contrastive learning.
OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation
Raktim Gautam Goswami (New York University), Farshad Khorrami (New York University)
Robotic IntelligenceConvolutional Neural NetworkTransformerWorld ModelVideo
🎯 What it does: Proposes the OSVI-WM framework, which utilizes single-segment expert videos and agent initial observations to generate potential state trajectories through a world model and decode them into physical waypoints, achieving one-shot visual imitation for unseen tasks.
Out-of-Distribution Generalized Graph Anomaly Detection with Homophily-aware Environment Mixup
Sibo Tian (Tsinghua University), Wenwu Zhu (Tsinghua University)
Anomaly DetectionGraph Neural NetworkGraphFinance Related
🎯 What it does: A framework named HEM is proposed, which combines a self-space decoupled ego-neighborhood decoder and a homogeneity attention-based environmental mixing technique to address the structural distribution drift problem in graph anomaly detection.
Outcome-Based Online Reinforcement Learning: Algorithms and Fundamental Limits
Fan Chen (Massachusetts Institute of Technology), Tengyang Xie (University of Wisconsin)
Reinforcement Learning
🎯 What it does: This paper proposes an online reinforcement learning framework that only observes endpoint rewards (outcome-based feedback) and provides a sample-efficient algorithm under general function approximation.
Over-squashing in Spatiotemporal Graph Neural Networks
Ivan Marisca (Università della Svizzera italiana), Michael M. Bronstein (University of Oxford)
Graph Neural NetworkGraphTime Series
🎯 What it does: This paper formally defines and analyzes the 'spatiotemporal over-squashing' problem in spatiotemporal graph neural networks (STGNNs), revealing its distinction from traditional static GNN over-squashing, and proves that convolutional STGNNs are more susceptible to the influence of distant temporal information along the time axis.
Overcoming Challenges of Long-Horizon Prediction in Driving World Models
Arian Mousakhan (University of Freiburg), Thomas Brox (University of Freiburg)
GenerationAutonomous DrivingTransformerWorld ModelVideo
🎯 What it does: This paper researches and implements a continuous spatial world model called Orbis, based on flow matching, for long-term prediction of driving scenarios and generating more realistic and controllable trajectories.
Overcoming Long Context Limitations of State Space Models via Context Dependent Sparse Attention
Zhihao Zhan (Mila - Québec AI Institute), Jian Tang (Mila - Québec AI Institute)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Designed and validated a Joint Recall task to assess long context modeling capabilities, and proposed the HAX architecture that combines SSM with context-aware sparse attention.
Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning
Julian Minder (Ecole Polytechnique Fédérale de Lausanne), Neel Nanda (Ecole Polytechnique Fédérale de Lausanne)
Explainability and InterpretabilityRepresentation LearningTransformerSupervised Fine-TuningText
🎯 What it does: Analyze the representation differences of the Gemma model due to chat fine-tuning, using a crosscoder to compare the intermediate layer representations of the base and chat versions, detecting and explaining newly emerged concepts.
OVS Meets Continual Learning: Towards Sustainable Open-Vocabulary Segmentation
Dongjun Hwang (Sogang University), Junsuk Choe (Sogang University)
Object DetectionSegmentationSupervised Fine-TuningMixture of ExpertsImage
🎯 What it does: We propose ConOVS, an incremental learning framework for Open Vocabulary Segmentation (OVS) that can gradually expand the model's recognition capabilities in an environment where new data is continuously added.
OWL: Optimized Workforce Learning for General Multi-Agent Assistance in Real-World Task Automation
Mengkang Hu (University of Hong Kong), Guohao Li (Eigent.AI)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextMultimodality
🎯 What it does: This paper presents WORKFORCE, a hierarchical multi-agent framework that separates task planning, coordination, and domain-specific execution; it also introduces the OPTIMIZED WORKFORCE LEARNING (OWL) training paradigm, specifically designed to enhance the cross-domain generalization ability of general planners.
OWMM-Agent: Open World Mobile Manipulation With Multi-modal Agentic Data Synthesis
Junting Chen (Shanghai AI Laboratory), Lin Shao (National University of Singapore)
Data SynthesisRobotic IntelligenceTransformerLarge Language ModelVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: This paper presents OWMM-Agent, a mobile manipulation intelligent agent that utilizes a multimodal VLM for global scene understanding, state tracking, and end-to-end action generation.
P-Law: Predicting Quantitative Scaling Law with Entropy Guidance in Large Recommendation Models
Tingjia Shen (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
Recommendation SystemTransformerSequential
🎯 What it does: This paper proposes the Performance Law, which evaluates data quality using minimum encoding length and true entropy, and introduces a decay term to quantitatively predict the performance of large sequence recommendation models.
PAC-Bayes Bounds for Multivariate Linear Regression and Linear Autoencoders
Ruixin Guo, Yang Zhou
Recommendation SystemAuto EncoderTabular
🎯 What it does: This paper presents the PAC-Bayes generalization bound for multiple linear regression and linear autoencoders (LAE), theoretically explaining the excellent performance of LAE in recommendation systems.
PaceLLM: Brain-Inspired Large Language Models for Long-Context Understanding
Kangcong Li (Fudan University), Tao Chen (Fudan University)
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsTextBenchmark
🎯 What it does: A long text understanding framework PaceLLM based on brain neural mechanisms is proposed, which includes an Activation Memory Pool (AMB) and Cortical Expert Clustering (CE).
PAID: Pairwise Angular-Invariant Decomposition for Continual Test-Time Adaptation
Kunyu Wang (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Domain AdaptationAutonomous DrivingImage
🎯 What it does: A diagonal-invariant decomposition method based on pre-trained weights, PAID, is proposed for adaptive continuous testing.
PairEdit: Learning Semantic Variations for Exemplar-based Image Editing
Haoguang Lu (Sun Yat-sen University), Xudong Mao (Sun Yat-sen University)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: This paper proposes PairEdit, a continuous image editing method that learns complex semantic transformations based on a small number of image pairs without relying on text prompts.
Pairwise Calibrated Rewards for Pluralistic Alignment
Daniel Halpern (Harvard University), Itai Shapira (Harvard University)
Recommendation SystemReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper proposes achieving diversified artificial preference alignment through learning a mixed model (ensemble) that contains a small number of reward functions, without relying on the identity of evaluators or predefined groups;
Pairwise Optimal Transports for Training All-to-All Flow-Based Condition Transfer Model
Kotaro Ikeda (University of Tokyo), Kenji Fukumizu (Institute of Statistical Mathematics)
Data SynthesisOptimizationFlow-based ModelTabularSequentialOrdinary Differential Equation
🎯 What it does: This paper proposes an All-to-All Flow-based Transfer Model (A2A-FM) based on flow matching, which can learn approximate optimal transport mappings between all pairs of conditions under any conditional distribution, achieving conditional transfer.
PALQO: Physics-informed model for Accelerating Large-scale Quantum Optimization
Yiming Huang (Peking University), Xiaoting Wang (University of Electronic Science and Technology of China)
OptimizationTime SeriesPhysics Related
🎯 What it does: A new method based on Physics-Informed Neural Networks (PINN) called PALQO is proposed to predict parameter updates in Variational Quantum Algorithms (VQA) for large-scale tasks, significantly reducing quantum resource consumption.
Pan-LUT: Efficient Pan-sharpening via Learnable Look-Up Tables
Zhongnan Cai (Xiamen University), Xinghao Ding
Super ResolutionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A Pan-LUT framework based on learnable lookup tables (LUT) is proposed for efficient panchromatic fusion of high-resolution remote sensing images while maintaining high quality.
Panacea: Mitigating Harmful Fine-tuning for Large Language Models via Post-fine-tuning Perturbation
Yibo Wang (Shenzhen Campus of Sun Yat-sen University), Dacheng Tao (Nanyang Technological University)
OptimizationAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes an adaptive perturbation method (Panacea) for the post-fine-tuning phase, aimed at restoring the safety alignment of large language models after harmful fine-tuning attacks while maintaining downstream task performance.
Pancakes: Consistent Multi-Protocol Image Segmentation Across Biomedical Domains
Marianne Rakic (Massachusetts Institute of Technology), Adrian V Dalca
SegmentationData SynthesisConvolutional Neural NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A framework named Pancakes is proposed, which can automatically generate multiple feasible segmentation protocols in unseen biomedical imaging domains, ensuring semantic consistency of labels across images within the same set.
PANDA: Towards Generalist Video Anomaly Detection via Agentic AI Engineer
Zhiwei Yang (Xidian University), Mike Zheng Shou (National University of Singapore)
Anomaly DetectionLarge Language ModelAgentic AIVision Language ModelVideoMultimodalityRetrieval-Augmented Generation
🎯 What it does: A universal video anomaly detection framework called PANDA is proposed, which is training-free and requires no human intervention, and can adapt to different scenes and types of anomalies.
PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space
Jinghong Zheng (Huazhong University of Science and Technology), Joey Tianyi Zhou (Agency for Science, Technology and Research)
Pose EstimationTransformerImage
🎯 What it does: PandaPose is proposed, which enhances 3D human pose from a single frame RGB image by propagating 2D priors into the 3D anchor space.
PANGEA: Projection-Based Augmentation with Non-Relevant General Data for Enhanced Domain Adaptation in LLMs
Seungyoo Lee, Juho Lee (Kookmin University)
Data SynthesisDomain AdaptationTransformerLarge Language ModelPrompt EngineeringTextFinance Related
🎯 What it does: A three-stage framework called PANGEA is proposed, which utilizes unrelated general large-scale data to generate structured prompts through a Prompt Writer, and then projects them into high-quality synthetic samples in the target domain, achieving domain adaptation in extremely data-scarce scenarios.
Panoptic Captioning: An Equivalence Bridge for Image and Text
Kun-Yu Lin, Kai Han
Object DetectionGenerationData SynthesisTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the Panoptic Captioning task, which generates minimal text equivalent descriptions that include entities, locations, attributes, relationships, and global states.
PanoWan: Lifting Diffusion Video Generation Models to 360$^\circ$ with Latitude/Longitude-aware Mechanisms
Yifei Xia (Peking University), Boxin Shi (Peking University)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: Proposes the PanoWan framework, which transfers pre-trained text-to-video diffusion models to 360° panoramic video generation, and introduces low-cost modules to achieve seamless long-term temporal and spatial consistency;
PANTHER: Generative Pretraining Beyond Language for Sequential User Behavior Modeling
Guilin Li (Tencent), Matthias Hwai Yong Tan (City University of Hong Kong)
Recommendation SystemAnomaly DetectionTransformerContrastive LearningTime SeriesSequentialFinance Related
🎯 What it does: Proposes the PANTHER framework, which combines generative pre-training and real-time discrimination to achieve multi-task applications such as fraud detection and next transaction prediction in payment scenarios.
Parallel Scaling Law for Language Models
Mouxiang Chen (Zhejiang University), Zhongxin Liu (Zhejiang University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A scale expansion method named PAR-SCALE is proposed, which enhances model performance and improves inference efficiency by executing multiple streams in parallel on the same set of model parameters (each stream uses learnable prefix transformations and dynamically weighted aggregation) without increasing the number of parameters.
Parallelizing MCMC Across the Sequence Length
David M. Zoltowski (Stanford University), Scott Linderman
OptimizationComputational EfficiencyTabularSequential
🎯 What it does: This paper proposes a parallel MCMC method based on the parallel Newton iteration (DEER) framework, which enables parallelization along the sequence length dimension. The framework is applied to three commonly used samplers: Gibbs, MALA, and HMC, while also providing memory-friendly techniques such as random approximate Jacobian, sliding window, and block quantum Newton.
Parameter Dynamics of Online Machine Learning and Test-time Adaptation
Jae-Hong Lee (Hankuk University of Foreign Studies)
Domain AdaptationTransformerImageStochastic Differential Equation
🎯 What it does: A probability framework based on SDE is proposed to model the dynamics of parameters over time in online learning, and the SIGMA algorithm is designed based on this framework to stabilize the Test-Time Adaptation (TTA) process by maintaining the alignment of the parameter transfer distribution with the Inverse-Gamma distribution.
Parameter Efficient Fine-tuning via Explained Variance Adaptation
Fabian Paischer (Johannes Kepler University Linz), Sepp Hochreiter (Johannes Kepler University Linz)
OptimizationComputational EfficiencySupervised Fine-TuningReinforcement LearningText
🎯 What it does: A PEFT method (EVA) for optimal initialization of explanatory variance and adaptive rank allocation in LoRA is proposed, achieving faster and more efficient model fine-tuning.
Parameter-free Algorithms for the Stochastically Extended Adversarial Model
Shuche Wang (National University of Singapore), Vincent Y. F. Tan (National University of Singapore)
OptimizationAdversarial AttackMixture of Experts
🎯 What it does: Under the Stochastically Extended Adversarial (SEA) model, two types of parameter-free algorithms are proposed: Comparator Adaptive based on Optimistic Online Newton Step (CA-OONS) and Comparator and Lipschitz Adaptive (CLA-OONS).
Parameter-Free Hypergraph Neural Network for Few-Shot Node Classification
Chaewoon Bae (Korea Advanced Institute of Science and Technology), Jaemin Yoo (Korea Advanced Institute of Science and Technology)
ClassificationGraph Neural NetworkGraph
🎯 What it does: A parameter-free hypergraph neural network ZEN is proposed for few-shot node classification;
ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation
Pengcheng Huang (Northeastern University), Chenyan Xiong (Carnegie Mellon University)
GenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Proposes the ParamMute framework to suppress the activation of specific FFNs to reduce the LLM's reliance on internal memory and enhance the credibility of retrieval-augmented generation.
PARCO: Parallel AutoRegressive Models for Multi-Agent Combinatorial Optimization
Federico Berto (KAIST), Jinkyoo Park (Brandenburg University of Technology)
OptimizationTransformerReinforcement LearningTabularBenchmark
🎯 What it does: Proposes the PARCO framework, which uses a parallel autoregressive model to solve multi-agent combinatorial optimization problems.
Pareto Optimal Risk-Agnostic Distributional Bandits with Heavy-Tail Rewards
Kyungjae Lee (Korea University), Sungbin Lim (Korea University)
Recommendation SystemAnomaly DetectionOptimizationTabularTime SeriesFinance Related
🎯 What it does: A multi-risk Pareto optimal multi-armed bandit framework is proposed in heavy-tailed reward environments, along with the DistLCB and MR-DistLCB algorithms and their theoretical convergence.
Pareto-Optimal Energy Alignment for Designing Nature-Like Antibodies
Yibo Wen (Northwestern University), Han Liu (Northwestern University)
OptimizationDrug DiscoveryLarge Language ModelDiffusion modelText
🎯 What it does: This paper proposes a three-stage framework: first, a large-scale antibody sequence pre-trained language model is used, then its knowledge is transferred to a diffusion model for sequence-structure co-design, and finally, multi-objective energy alignment of antibodies is performed through Pareto-Optimal Energy Alignment to generate naturally styled antibodies.
ParetoQ: Improving Scaling Laws in Extremely Low-bit LLM Quantization
Zechun Liu (Meta AI), Vikas Chandra (Meta AI)
Large Language ModelText
🎯 What it does: A unified ultra-low bit quantization framework called ParetoQ is proposed, which enables fair and systematic comparisons between 1, 1.58, 2, 3, and 4 bits;
PAROAttention: Pattern-Aware ReOrdering for Efficient Sparse and Quantized Attention in Visual Generation Models
Tianchen Zhao (Tsinghua University), Yu Wang (Tsinghua University)
GenerationComputational EfficiencyTransformerImageVideo
🎯 What it does: This paper proposes a pattern-aware token reordering technique (PARO) that unifies diverse visual attention into block patterns by rearranging attention patterns, and based on this, designs efficient sparse and quantization strategies that significantly reduce the attention computation cost of video/image generation models.
Parsimonious Predictions for Strategyproof Scheduling
Richard Cole (New York University), Pranav Jangir (New York University)
Optimization
🎯 What it does: A learning-enhanced policy-independent scheduling mechanism is proposed, achieving feasibility with a linear number of predictions.
Part-Aware Bottom-Up Group Reasoning for Fine-Grained Social Interaction Detection
Dongkeun Kim (Pohang University of Science and Technology), Suha Kwak (Pohang University of Science and Technology)
RecognitionObject DetectionTransformerVideo
🎯 What it does: A bottom-up group reasoning framework based on part perception is proposed for fine-grained social interaction detection.
PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers
Yuchen Lin (Peking University), Katerina Fragkiadaki (Carnegie Mellon University)
GenerationData SynthesisTransformerDiffusion modelRectified FlowMesh
🎯 What it does: A model called PARTCRAFTER has been developed to directly generate structured 3D meshes from a single RGB image, capable of generating multi-part objects or scenes in one go without the need for pre-segmentation.
Partial Correlation Network Estimation by Semismooth Newton Methods
DongWon Kim, Joong-Ho Won (Seoul National University)
OptimizationComputational EfficiencyReinforcement LearningTabularBiomedical Data
🎯 What it does: A semi-smooth Newton algorithm based on the ACCROD framework is proposed for high-dimensional partial correlation network estimation, and row-separable parallel computation is implemented.
Partial Information Decomposition via Normalizing Flows in Latent Gaussian Distributions
Wenyuan Zhao (Texas A&M University), Paul Pu Liang (Massachusetts Institute of Technology)
OptimizationFlow-based ModelMultimodalityBiomedical Data
🎯 What it does: A framework for partial information decomposition (PID) based on normalized flows under a latent Gaussian distribution is proposed, which includes an efficient Thin-PID algorithm and a scalable Flow-PID encoder.
Partial Physics Informed Diffusion Model for Ocean Chlorophyll Concentration Reconstruction
Qianxun Xu (Duke Kunshan University), Zuchuan Li (Duke Kunshan University)
Diffusion modelTime SeriesPhysics Related
🎯 What it does: This study investigates a framework that incorporates partial physical knowledge into diffusion models for reconstructing ocean chlorophyll concentration.
Partition to Evolve: Niching-enhanced Evolution with LLMs for Automated Algorithm Discovery
Qinglong Hu (City University of Hong Kong), Qingfu Zhang (City University of Hong Kong)
OptimizationTransformerLarge Language ModelPrompt Engineering
🎯 What it does: A framework for evolutionary search (LES) that combines large language models is proposed, utilizing feature-assisted partitioning to construct niches and achieving algorithmic auto-discovery (PartEvo) based on this.
Partition-Then-Adapt: Combating Prediction Bias for Reliable Multi-Modal Test-Time Adaptation
Guowei Wang (South China University of Technology), Changxing Ding (Chinese Academy of Sciences)
Domain AdaptationTransformerVideoMultimodalityAudio
🎯 What it does: To address the bias prediction problem of multimodal data under temporal shifts during testing, we propose the Partition-Then-Adapt (PTA) method, which first divides samples into reliable and unreliable subsets based on predicted bias, and then adapts through quantized reweighting and attention alignment.
Partner Modelling Emerges in Recurrent Agents (But Only When It Matters)
Ruaridh Mon-Williams (University of Edinburgh), Christopher G. Lucas (University of Edinburgh)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: This paper investigates and verifies whether an agent can spontaneously encode and utilize partner capability information in its hidden state, relying solely on reward signals without explicit modeling modules, by training a simple GRU-type RNN agent to interact with diverse partners in the Overcooked-AI collaborative environment.
PARTONOMY: Large Multimodal Models with Part-Level Visual Understanding
Ansel Blume (University of Illinois Urbana-Champaign), Heng Ji (University of Illinois Urbana-Champaign)
RecognitionObject DetectionSegmentationTransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This study investigates the capabilities of multimodal models in pixel-level part recognition and reasoning, proposing the Explanatory Part Segmentation task and the PARTONOMY benchmark.
PASS: Path-selective State Space Model for Event-based Recognition
Jiazhou Zhou (Hong Kong University of Science and Technology), Lin Wang (Nanyang Technological University)
RecognitionObject DetectionComputational EfficiencyConvolutional Neural NetworkTime SeriesSequential
🎯 What it does: The PASS framework is proposed for object/action recognition in event cameras, supporting event lengths from 10⁶ to 10⁹ while maintaining good generalization at different inference frequencies.
Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems
Christian Walder (Google DeepMind), Deep Tejas Karkhanis (Google DeepMind)
OptimizationReinforcement LearningText
🎯 What it does: A Pass@k Policy Optimization (PKPO) framework is proposed, which directly optimizes the best performance in a multi-sample set by transforming the RL reward from a single sample to the maximum expected reward over multiple samples.
PaTH Attention: Position Encoding via Accumulating Householder Transformations
Songlin Yang (Massachusetts Institute of Technology), Yoon Kim (Massachusetts Institute of Technology)
TransformerTextSequential
🎯 What it does: A data-driven cumulative Householder transformation-based polynomial position encoding (PaTH) is proposed for the attention mechanism of Transformers;
Path Gradients after Flow Matching
Lorenz Vaitl, Leon Klein (Freie Universität Berlin)
OptimizationComputational EfficiencyGraph Neural NetworkFlow-based ModelSequentialStochastic Differential Equation
🎯 What it does: A hybrid training method that combines Flow Matching pre-training with Path Gradients fine-tuning is proposed for Boltzmann Generators to enhance sampling efficiency.
Path-Enhanced Contrastive Learning for Recommendation
Haoran Sun (Beijing Jiaotong University), Liang Wang (Northwestern Polytechnical University)
Recommendation SystemGraph Neural NetworkContrastive LearningTabular
🎯 What it does: Introduce path-enhanced contrastive learning in collaborative filtering to improve the quality of user-item embeddings by sampling multiple interaction paths.
Path-specific effects for pulse-oximetry guided decisions in critical care
Kevin Zhang (Columbia University), Shalmali Joshi (Columbia University)
Biomedical DataElectronic Health Records
🎯 What it does: This paper studies the causal impact of racial differences caused by errors in pulse oximeter readings on invasive ventilation decisions in the ICU, and proposes an analysis framework based on path-specific effects;
PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization
Honglin Li (Zhejiang University), Lin Yang (Westlake University)
CompressionRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper addresses the issues of information loss and high computational and storage costs associated with traditional methods that only use the [CLS] token by compressing spatial patch features in whole slide images (WSI) through vector quantization (VQ).
Pattern-Guided Adaptive Prior for Structure Learning
Lyuzhou Chen (University of Science and Technology of China), Huanhuan Chen (University of Science and Technology of China)
Graph
🎯 What it does: An adaptive prior framework based on structural patterns, PGAP, is proposed to improve the integration of imprecise prior knowledge in structural learning.
Pause Tokens Strictly Increase the Expressivity of Constant-Depth Transformers
Charles London (University of Oxford), Varun Kanade (University of Oxford)
TransformerPrompt EngineeringSequential
🎯 What it does: This study investigates the impact of pause tokens on the expressive power of Transformers and provides theoretical upper bounds on the expressiveness of constant-depth Transformers with and without pause tokens.
Pay Attention to Small Weights
Chao Zhou (CISPA Helmholtz Center for Information Security), Rebekka Burkholz (CISPA Helmholtz Center for Information Security)
OptimizationTransformerSupervised Fine-TuningImageText
🎯 What it does: This paper proposes a sparse fine-tuning optimizer called NANOADAM based on small weights for efficient fine-tuning of large models.
PaZO: Preconditioned Accelerated Zeroth-Order Optimization for Fine-Tuning LLMs
Hanzhen Zhao (Peking University), Zhouchen Lin (Peking University)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper presents PaZO—a preconditioned accelerated zero-order optimization algorithm for fine-tuning large language models (LLMs) in resource-constrained environments.
PBR-SR: Mesh PBR Texture Super Resolution from 2D Image Priors
Yujin Chen (Technical University of Munich), Matthias Nießner (Technical University of Munich)
RestorationSuper ResolutionDiffusion modelImageMesh
🎯 What it does: This paper proposes a zero-shot PBR texture super-resolution method called PBR-SR, which can enhance low-resolution PBR textures to high resolution without requiring any PBR-specific training data, while maintaining material consistency and real-time relighting capabilities.
PC-Net: Weakly Supervised Compositional Moment Retrieval via Proposal-Centric Network
Mingyao Zhou (Central China Normal University), Mang Ye (Wuhan University)
RetrievalContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes a weakly supervised compositional moment retrieval task and designs PC-Net to achieve retrieval using only video-text pairs without timestamp annotations.
PCA++: How Uniformity Induces Robustness to Background Noise in Contrastive Learning
Mingqi Wu (McGill University), Archer Y. Yang (McGill University)
OptimizationRepresentation LearningContrastive LearningImageBiomedical Data
🎯 What it does: A new contrastive learning method PCA++ is proposed, which achieves robust signal subspace recovery through hard unification constraints.
PDEfuncta: Spectrally-Aware Neural Representation for PDE Solution Modeling
Minju Jo (Sorbonne University), Kookjin Lee (Arizona State University)
CompressionMeta LearningAuto EncoderTime SeriesPhysics Related
🎯 What it does: Proposes two frameworks, Global Fourier Modulation (GFM) and PDEfuncta, which achieve compression, reconstruction, and bidirectional inference of PDE solutions through modulated unified implicit vectors reparameterized by Fourier.
PDPO: Parametric Density Path Optimization
Sebastian Gutierrez Hernandez (Georgia Institute of Technology), Hao-Min Zhou
OptimizationOrdinary Differential Equation
🎯 What it does: This paper proposes the Parametric Density Path Optimization (PDPO) method, which uses a parametric pushforward mapping (Neural ODE) and cubic Hermite spline approximation to solve for the path that minimizes action in probability density space, applicable to various constraints such as obstacles, interactions, and stochastic control.
Per-Architecture Training-Free Metric Optimization for Neural Architecture Search
Mingzhuo Lin (Shenzhen University), Jianping Luo (Shenzhen University)
OptimizationNeural Architecture SearchImage
🎯 What it does: This paper proposes a Per-Architecture Training-Free Metric Optimization NAS (PO-NAS), which achieves efficient and competitive performance in neural architecture search by dynamically assigning training-independent metric weights for each candidate architecture and combining proxy models with evolutionary search.
Perceive Anything: Recognize, Explain, Caption, and Segment Anything in Images and Videos
Weifeng Lin (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
RecognitionSegmentationGenerationTransformerLarge Language ModelSupervised Fine-TuningImageVideoTextMultimodality
🎯 What it does: This paper proposes the Perceive Anything Model (PAM), which combines the visual encoder based on SAM 2 with LLM to achieve segmentation, recognition, interpretation, and caption generation for specified areas in images and videos.
Perception Encoder: The best visual embeddings are not at the output of the network
Daniel Bolya (Meta), Christoph Feichtenhofer (Meta)
ClassificationObject DetectionSegmentationDepth EstimationRetrievalTransformerLarge Language ModelContrastive LearningImageVideoTextMultimodality
🎯 What it does: This paper proposes the Perception Encoder (PE) family, which combines contrastive vision-language (CLIP) pre-training with video data engines and language and spatial alignment techniques to generate a unified image and video encoder, achieving multi-task performance through short-term fine-tuning.
Perception-R1: Pioneering Perception Policy with Reinforcement Learning
En Yu (Huazhong University of Science and Technology), Wenbing Tao (Huazhong University of Science and Technology)
RecognitionObject DetectionTransformerReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: Explores the learning of perceptual strategies in multimodal LLM post-training with regularized RL, and proposes the Perception-R1 framework, achieving performance improvements across various visual perception tasks.
PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding
Jang Hyun Cho (Meta), Christoph Feichtenhofer (Meta)
RecognitionGenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageVideoTextBenchmark
🎯 What it does: A fully open and reproducible visual language model, PerceptionLM (PLM), has been constructed, along with the release of a large-scale human-annotated video question-answering and spatiotemporal subtitle dataset, as well as a dedicated benchmark for fine-grained video understanding, PLM-VideoBench.
Performative Risk Control: Calibrating Models for Reliable Deployment under Performativity
Victor Li (New York University), Zhun Deng (University of California, Los Angeles)
TabularFinance Related
🎯 What it does: A Performative Risk Control (PRC) framework is proposed for risk control of black-box machine learning models in environments with performativity.
Performative Validity of Recourse Explanations
Gunnar König (Tübingen AI Center), Ulrike von Luxburg (University of Tübingen)
Recommendation SystemExplainability and InterpretabilityTabularFinance Related
🎯 What it does: This paper proposes the concept of 'performative validity' and systematically analyzes how recursive explanations can lead to recommendation failures due to distribution shifts induced by themselves in high-risk decision-making systems.
Periodic Skill Discovery
Jonghae Park (Seoul National University), H. Jin Kim (Seoul National University)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: This paper proposes an unsupervised skill discovery framework called PSD, which learns controllable periodic skills by mapping states to a circular latent space.
PeRL: Permutation-Enhanced Reinforcement Learning for Interleaved Vision-Language Reasoning
Yizhen Zhang (Tsinghua University), Yujiu Yang (Tsinghua University)
TransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the PeRL method, utilizing a reinforcement learning framework with image sequence permutation and rollout filtering, trained on multi-image interactive visual-language reasoning tasks.
Permissioned LLMs: Enforcing Access Control in Large Language Models
Bargav Jayaraman (Oracle Corporation), Krishnaram Kenthapadi (Oracle Corporation)
Large Language ModelSupervised Fine-TuningTextBiomedical Data
🎯 What it does: Modeling the access control issues during LLM fine-tuning and proposing the concept of Permissioned LLM (PermLLM).
PermLLM: Learnable Channel Permutation for N:M Sparse Large Language Models
Lancheng Zou (Chinese University of Hong Kong), Bei Yu (Chinese University of Hong Kong)
TransformerLarge Language ModelText
🎯 What it does: This study proposes the PermLLM framework, which utilizes learnable channel permutation (LCP) for post-training pruning of N:M semi-structured sparse large language models, directly minimizing the output error between the sparse model and the original model;
Permutation Equivariant Neural Controlled Differential Equations for Dynamic Graph Representation Learning
Torben Berndt (Heidelberg Institute for Theoretical Studies), Andrey Kormilitzin (Heidelberg Institute for Theoretical Studies)
Representation LearningGraph Neural NetworkGraphTime SeriesStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposed and implemented Permutation Equivariant Neural Graph Controlled Differential Equations (PENG-CDE) for representation learning of dynamic graphs.
Personalized Bayesian Federated Learning with Wasserstein Barycenter Aggregation
Ting Wei (Renmin University of China), Yifan Sun (Renmin University of China)
Federated LearningImage
🎯 What it does: Proposes the FedWBA framework, which implements personalized Bayesian inference and Wasserstein barycenter aggregation in federated learning.
Personalized Decision Modeling: Utility Optimization or Textualized-Symbolic Reasoning
Yibo Zhao (Johns Hopkins University), Hao Frank Yang (Johns Hopkins University)
Recommendation SystemOptimizationTransformerLarge Language ModelTextTabular
🎯 What it does: The ATHENA framework is proposed, utilizing LLM for group-level symbolic utility function discovery and individual-level semantic adaptation, achieving individual decision modeling in tasks such as travel mode and vaccination.
Personalized Exercise Recommendation with Semantically-Grounded Knowledge Tracing
Yilmazcan Ozyurt (ETH Zurich), Mrinmaya Sachan (ETH Zurich)
Recommendation SystemRecurrent Neural NetworkTransformerLarge Language ModelReinforcement LearningContrastive LearningTextSequential
🎯 What it does: This paper proposes the ExRec framework for personalized exercise recommendation. The framework automatically annotates knowledge concepts (KC) using LLM, utilizes contrastive learning to obtain semantically rich exercises, problem-solving steps, and KC embeddings, trains a calibrated knowledge tracing (KT) model for instant knowledge state estimation, and uses the calibrated KT model as a reinforcement learning (RL) environment, enhancing the learning effect of continuous action RL with model-based value estimation (MVE).
Personalized Federated Conformal Prediction with Localization
Yinjie Min (Nankai University), Changliang Zou (Nankai University)
Federated LearningTabular
🎯 What it does: A personalized federated quantile prediction framework (PFCP) is proposed, which combines personalized federated learning and local compliant prediction to provide a statistically valid prediction set for the target agent and achieve instance localization.
Personalized Image Editing in Text-to-Image Diffusion Models via Collaborative Direct Preference Optimization
Connor Dunlop, Pinar Yanardag
GenerationOptimizationGraph Neural NetworkDiffusion modelImage
🎯 What it does: A personalized image editing framework based on Collaborative Direct Preference Optimization (C-DPO) is proposed.
Personalized Safety in LLMs: A Benchmark and A Planning-Based Agent Approach
Yuchen Wu (University of Washington), Jindong Wang (William and Mary)
Safty and PrivacyTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: The PENGUIN benchmark and RAISE framework are proposed to evaluate and enhance the personalized safety of large language models in high-risk scenarios.
Personalized Subgraph Federated Learning with Differentiable Auxiliary Projections
Wei Zhuo (Nanyang Technological University), Han Yu (Nanyang Technological University)
Federated LearningSafty and PrivacyGraph Neural NetworkGraph
🎯 What it does: Designed and implemented FedAux, a personalized framework in the subgraph federated learning scenario, which captures the heterogeneity of local subgraphs by learning auxiliary projection vectors (APV) at each client, thereby achieving privacy-preserving client similarity assessment and personalized model aggregation.
Personalized Visual Content Generation in Conversational Systems
Xianquan Wang (University of Science and Technology of China), Qi Liu (University of Science and Technology of China)
GenerationRecommendation SystemTransformerLarge Language ModelDiffusion modelTextMultimodality
🎯 What it does: Personalized visual content generation is implemented in dialogue systems.
Perturb a Model, Not an Image: Towards Robust Privacy Protection via Anti-Personalized Diffusion Models
Tae-Young Lee (Korea University), Gyeong-Moon Park (Korea University)
GenerationSafty and PrivacyDiffusion modelImage
🎯 What it does: A model-level protection framework called APDM has been designed and implemented to prevent diffusion models from generating personalized outputs for specified subjects while maintaining generation quality.
Perturbation Bounds for Low-Rank Inverse Approximations under Noise
Phuc Tran (Yale University), Nisheeth K. Vishnoi (Yale University)
Tabular
🎯 What it does: This study investigates the spectral norm error of low-rank inverse approximations under the influence of noise and proposes explicit non-asymptotic perturbation bounds.