NeurIPS 2024 Papers — Page 25
Conference on Neural Information Processing Systems · 4035 papers
Near-Optimal Distributionally Robust Reinforcement Learning with General $L_p$ Norms
Pierre Clavier (Ecole Polytechnique), Matthieu Geist (Cohere)
OptimizationReinforcement Learning
🎯 What it does: This study explores Robust Markov Decision Processes (RMDPs), aiming to optimize worst-case performance within certain uncertainty sets to address the simulation-reality gap and sample efficiency issues in reinforcement learning.
Near-Optimal Dynamic Regret for Adversarial Linear Mixture MDPs
Long-Fei Li (Nanjing University), Zhi-Hua Zhou (Nanjing University)
OptimizationReinforcement Learning
🎯 What it does: This study investigates the dynamic regret problem of linear mixed MDPs with unknown transitions and adversarial rewards under full information feedback, proposing a new algorithm that combines the advantages of occupancy measure-based and policy-based methods.
Near-Optimal Streaming Heavy-Tailed Statistical Estimation with Clipped SGD
Aniket Das (Stanford University), Prateek Varshney (Stanford University)
🎯 What it does: This paper studies the problem of high-dimensional heavy-tailed statistical estimation in a streaming setting, proposing a near-optimal algorithm based on Clipped Stochastic Gradient Descent (Clipped SGD) that can achieve statistical estimation of heavy-tailed distributions under memory constraints.
Near-Optimality of Contrastive Divergence Algorithms
Pierre Glaser (Gatsby Computational Neuroscience Unit University College London), Arthur Gretton (Gatsby Computational Neuroscience Unit University College London)
OptimizationContrastive Learning
🎯 What it does: This paper conducts a non-asymptotic analysis of the convergence properties of the Contrastive Divergence (CD) algorithm in unnormalized exponential family models, proving that under suitable conditions, its parameter estimation can achieve a parameter rate of O(n⁻¹/²) and can realize a variance close to the Cramér–Rao lower bound.
Nearest Neighbor Speculative Decoding for LLM Generation and Attribution
Minghan Li (Cohere), Xi Victoria Lin (Meta)
GenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: NEST is proposed, a semi-parametric language model that performs nearest neighbor speculative decoding during inference to enhance the factuality and source attribution of large language models.
Nearly Minimax Optimal Regret for Multinomial Logistic Bandit
Joongkyu Lee (Seoul National University), Min-hwan Oh (Seoul National University)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: This paper studies the contextual multinomial logistic (MNL) Bandit problem, proving the optimal lower and upper bounds under both uniform and non-uniform rewards, and proposes the OFU-MNL+ algorithm with constant time computational complexity, achieving near-optimal cumulative returns.
Nearly Minimax Optimal Submodular Maximization with Bandit Feedback
Artin Tajdini (University of Washington), Kevin Jamieson (University of Washington)
OptimizationReinforcement LearningGraph
🎯 What it does: This paper studies the problem of maximizing an unknown monotone submodular function under random feedback and proposes an algorithm called Sub-UCB, aimed at minimizing the learner's regret.
Nearly Optimal Approximation of Matrix Functions by the Lanczos Method
Noah Amsel (New York University), Christopher Musco (University of Massachusetts Amherst)
Optimization
🎯 What it does: This paper studies the application of the Lanczos method in matrix function approximation, particularly proving the approximate optimality of Lanczos-FA when dealing with a class of rational functions.
Nearly Tight Black-Box Auditing of Differentially Private Machine Learning
Meenatchi Sundaram Muthu Selva Annamalai (University College London), Emiliano De Cristofaro (University of California)
Safty and PrivacyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposes an approximate tight auditing method for DP-SGD under a black-box threat model, significantly amplifying privacy leakage using 'worst-case' initial model parameters;
Neglected Hessian component explains mysteries in sharpness regularization
Yann Dauphin, Hossein Mobahi (Google DeepMind)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: This study investigates the regularization effect of second-order information in deep learning, focusing on the non-positive definite part of the Hessian (NME), and explains the performance differences of methods such as gradient penalty and weight noise.
NeoRL: Efficient Exploration for Nonepisodic RL
Bhavya Sukhija (ETH Zurich), Andreas Krause (ETH Zurich)
OptimizationReinforcement Learning
🎯 What it does: A model-based episodic exploration algorithm NEORL is proposed for episodic RL in continuous nonlinear dynamics, optimizing average cost and achieving robust learning.
Nesterov acceleration despite very noisy gradients
Kanan Gupta (University of Pittsburgh), Stephan Wojtowytsch (University of Pittsburgh)
OptimizationConvolutional Neural NetworkImageTabularStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: An improved accelerated gradient descent algorithm AGNES is proposed, which can achieve convergence acceleration in the presence of multiplicative noise.
Neuc-MDS: Non-Euclidean Multidimensional Scaling Through Bilinear Forms
Chengyuan Deng (Rutgers University), Cheng Xin (Rutgers University)
ImageTabular
🎯 What it does: Developed Neuc-MDS, which provides a method to extend traditional Multidimensional Scaling (MDS) to non-Euclidean, non-metric similarity using symmetric bilinear forms and negative eigenvalues for low-dimensional embedding.
NeuMA: Neural Material Adaptor for Visual Grounding of Intrinsic Dynamics
Junyi Cao (Shanghai Jiao Tong University), Chao Ma (Shanghai Jiao Tong University)
Object DetectionGenerationData SynthesisGaussian SplattingVideo
🎯 What it does: This paper studies a model called NeuMA that can directly learn the intrinsic dynamics of objects from video observations, and implements differentiable simulation and image generation through a particle-driven 3D Gaussian splatting renderer called Particle-GS.
Neur2BiLO: Neural Bilevel Optimization
Justin Dumouchelle (University of Toronto), Elias Boutros Khalil
OptimizationGraph Neural NetworkTabular
🎯 What it does: A dual-layer optimization framework NEUR2BILO based on neural networks is proposed, which transforms the original mixed-integer nonlinear bilevel problem into a single-layer problem that can be solved using MIP, providing a fast heuristic solution method.
Neural Assets: 3D-Aware Multi-Object Scene Synthesis with Image Diffusion Models
Ziyi Wu (University of Toronto), Thomas Kipf (Google DeepMind)
GenerationData SynthesisAutonomous DrivingTransformerDiffusion modelImageVideo
🎯 What it does: This paper proposes Neural Assets, a learnable representation that decouples object appearance from 3D pose, enabling multi-object 3D spatial control and compositional generation in a pre-trained diffusion model.
Neural Characteristic Activation Analysis and Geometric Parameterization for ReLU Networks
Wenlin Chen (University of Cambridge), Hong Ge (University of Cambridge)
OptimizationImageTabular
🎯 What it does: This paper proposes and analyzes the feature activation boundary of ReLU networks, discovering that common parameterization/normalization can lead to instability in stochastic optimization, and introduces a geometric parameterization GmP in spherical coordinates.
Neural Collapse Inspired Feature Alignment for Out-of-Distribution Generalization
Zhikang Chen (Tsinghua University), Kun Zhang (Mohamed bin Zayed University of Artificial Intelligence)
ClassificationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: By utilizing the phenomenon of Neural Collapse, semantic features are aligned to a Simplex ETF (simple orthogonal tight frame), and combined with automatic environment partitioning and learnable masks, the suppression of background spurious features is achieved, thereby enhancing generalization ability across different environments.
Neural Collapse To Multiple Centers For Imbalanced Data
Hongren Yan (Shanxi University), Feijiang Li (Shanxi University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper studies how to improve classification performance under imbalanced data by designing the output features of neural networks and the multi-center geometric structure of classifiers.
Neural collapse vs. low-rank bias: Is deep neural collapse really optimal?
Peter Súkeník (Institute of Science and Technology Austria), Marco Mondelli
OptimizationImage
🎯 What it does: This study investigates the optimality of Deep Neural Collapse (DNC) under the framework of a multi-layer multi-class unconstrained feature model (DUFM), finding that low-rank bias leads to DNC no longer being a global optimal solution, and proposes a graph-theoretic construction of low-rank solutions to prove this conclusion.
Neural Combinatorial Optimization for Robust Routing Problem with Uncertain Travel Times
Pei Xiao (Sun Yat-sen University), Zhenzhen Zhang (Tongji University)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes a neural network-based combinatorial optimization framework for solving robust path planning with uncertain travel times (RTSP and RVCRP). It extracts features from upper and lower bound information through a dual-head cross-attention mechanism and employs a reinforcement learning (REINFORCE + POMO) training strategy to quickly compute maximum returns using a pre-trained TSP model.
Neural Concept Binder
Wolfgang Stammer (Technical University of Darmstadt), Kristian Kersting (Technical University of Darmstadt)
ClassificationExplainability and InterpretabilityRepresentation LearningTransformerImageRetrieval-Augmented Generation
🎯 What it does: Proposes the Neural Concept Binder (NCB) framework, which learns interpretable and revisable object factor concepts under unsupervised conditions and integrates them into symbolic and sub-symbolic computation.
Neural Conditional Probability for Uncertainty Quantification
Vladimir R Kostic, Massimiliano Pontil (University College London)
Tabular
🎯 What it does: A new Neural Conditional Probability (NCP) method is proposed for learning conditional distributions, with a particular focus on statistical inference tasks.
Neural Cover Selection for Image Steganography
Karl Chahine (University of Texas at Austin), Hyeji Kim (University of Texas at Austin)
GenerationOptimizationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A framework is proposed that utilizes pre-trained generative models (DDIM or GAN) to perform gradient optimization in the latent space, thereby generating cover images that are most suitable for hidden information;
Neural decoding from stereotactic EEG: accounting for electrode variability across subjects
Georgios Mentzelopoulos (University of Pennsylvania), Flavia Vitale (University of Pennsylvania)
TransformerTime SeriesBiomedical Data
🎯 What it does: Proposed and implemented the seegnificant framework, which utilizes multi-channel, cross-subject sEEG data to train a Transformer-based neural decoding model aimed at predicting the reaction time of subjects.
Neural Embeddings Rank: Aligning 3D latent dynamics with movements
Chenggang Chen (Johns Hopkins University), Xiaoqin Wang (Johns Hopkins University)
Explainability and InterpretabilityRepresentation LearningContrastive LearningTime Series
🎯 What it does: This paper proposes the Neural Embeddings Rank (NER) method, which embeds neural dynamics into a three-dimensional latent space and aligns continuous motion labels through ranking.
Neural Experts: Mixture of Experts for Implicit Neural Representations
Yizhak Ben-Shabat (Roblox, Australian National University), Stephen Gould (Australian National University)
RestorationMixture of ExpertsImagePoint CloudAudio
🎯 What it does: This paper proposes a method that applies the Mixture of Experts (MoE) architecture to Implicit Neural Representation (INR), called Neural Experts, which achieves automatic segmentation of the input domain and local continuous function modeling.
Neural Flow Diffusion Models: Learnable Forward Process for Improved Diffusion Modelling
Grigory Bartosh (University of Amsterdam), Christian A. Naesseth (University of Amsterdam)
GenerationData SynthesisDiffusion modelFlow-based ModelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposes Neural Flow Diffusion Models (NFDM), a framework that can learn the forward process and enhance the performance of diffusion models.
Neural Gaffer: Relighting Any Object via Diffusion
Haian Jin (Cornell University), Noah Snavely (Cornell University)
RestorationGenerationDiffusion modelImage
🎯 What it does: An end-to-end 2D image relighting model called Neural Gaffer is proposed, which can perform high-quality relighting on any single object image under any HDR environmental lighting.
Neural Isometries: Taming Transformations for Equivariant ML
Thomas Mitchel, Vincent Sitzmann (Massachusetts Institute of Technology)
Pose EstimationAuto EncoderImageVideo
🎯 What it does: A self-supervised autoencoder framework named Neural Isometries is proposed, which learns to map the observation space to the latent space such that geometric transformations in the world space correspond to decomposable isometric linear mappings in the latent space, thereby transforming the complex symmetries of the observation space into structured linear transformations in the latent space.
Neural Krylov Iteration for Accelerating Linear System Solving
Jian Luo (MoE Key Laboratory of Brain-inspired Intelligent Perception and Cognition University of Science and Technology of China), Yufei Kuang (MoE Key Laboratory of Brain-inspired Intelligent Perception and Cognition University of Science and Technology of China)
OptimizationComputational Efficiency
🎯 What it does: The NeurKItt method is proposed, which uses neural operators to predict the invariant subspace of linear systems and utilizes this subspace to accelerate Krylov subspace iterations, significantly reducing the number of iterations and computation time.
Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation
István Sárándi (University of Tübingen), Gerard Pons-Moll (Max Planck Institute for Informatics)
Pose EstimationImage
🎯 What it does: This paper proposes a Neural Localizer Fields (NLF) that learns a continuous point localization function field, capable of predicting the position of any point (surface or internal) of the human body from a single RGB image, achieving unified training and inference across multiple datasets and annotation formats.
Neural Model Checking
Mirco Giacobbe (University of Birmingham), Michael Tautschnig (Amazon Web Services)
OptimizationComputational EfficiencyTabular
🎯 What it does: A model checking method based on neural networks is proposed, utilizing the quantized neural ranking function obtained from training as a formal proof certificate to formally verify the linear temporal logic (LTL) properties of hardware designs.
Neural network learns low-dimensional polynomials with SGD near the information-theoretic limit
Jason D. Lee (Princeton University), Denny Wu (New York University)
OptimizationSupervised Fine-TuningTabular
🎯 What it does: Introducing batch reuse in SGD training for two-dimensional neural networks, utilizing the non-related terms in the gradient to achieve polynomial label transformation, thereby breaking through the CSQ lower bound caused by the information index of single exponential models, reaching a sample complexity that is nearly at the information-theoretic limit;
Neural Network Reparametrization for Accelerated Optimization in Molecular Simulations
Nima Dehmamy (IBM Research), Tommi Jaakkola (Massachusetts Institute of Technology)
OptimizationDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: A reparameterization method based on neural networks is proposed to replace traditional coarse-graining, which can both reduce and increase degrees of freedom, directly minimizing energy in the CG space.
Neural P$^3$M: A Long-Range Interaction Modeling Enhancer for Geometric GNNs
Yusong Wang (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
Graph Neural NetworkGraph
🎯 What it does: The Neural P3M framework is proposed to introduce grid points in geometric GNNs to model long-range interactions and facilitate information exchange between atoms and grids.
Neural Persistence Dynamics
Sebastian Zeng (University of Salzburg), Roland Kwitt (University of Salzburg)
OptimizationRepresentation LearningPoint CloudSequentialOrdinary Differential Equation
🎯 What it does: A framework for learning latent dynamics of point cloud topological features using time evolution (Neural Persistence Dynamics) is proposed, and it is used to infer the parameters of collective behavior models.
Neural Pfaffians: Solving Many Many-Electron Schrödinger Equations
Nicholas Gao (Technical University of Munich), Stephan Günnemann (Technical University of Munich)
TabularPhysics Related
🎯 What it does: This paper proposes a neural network wave function based on Pfaffian (Neural Pfaffian) that achieves generalizable high-precision quantum chemical calculations for many-body electronic systems without the need for discrete orbital selection or Hartree–Fock pre-calculation.
Neural Pose Representation Learning for Generating and Transferring Non-Rigid Object Poses
Seungwoo Yoo (Korea Advanced Institute of Science and Technology), Minhyuk Sung (Korea Advanced Institute of Science and Technology)
GenerationPose EstimationRepresentation LearningTransformerDiffusion modelPoint CloudMesh
🎯 What it does: An unsupervised neural pose representation learning method is proposed, which can learn the pose space from multiple pose variants of a single object and transfer the learned poses to any new identity's 3D non-rigid model, while also supporting the generation of diverse poses through a diffusion model.
Neural Residual Diffusion Models for Deep Scalable Vision Generation
Zhiyuan Ma (Tsinghua University), Bowen Zhou (Tsinghua University)
GenerationData SynthesisDiffusion modelImageVideoOrdinary Differential Equation
🎯 What it does: A learnable gated residual mechanism (Neural-RDM) is proposed, unifying manifolds and U-shaped residual networks, replacing the manually designed mean-variance scheduler to achieve deep scalable diffusion models.
Neural Signed Distance Function Inference through Splatting 3D Gaussians Pulled on Zero-Level Set
Wenyuan Zhang (Tsinghua University), Zhizhong Han (Wayne State University)
Gaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a neural SDF inference method based on 3D Gaussian splitting, which aligns 3D Gaussians to the SDF zero isosurface through differentiable stretching, and gradually optimizes the SDF and Gaussian parameters using joint rendering and geometric constraints.
NeuralClothSim: Neural Deformation Fields Meet the Thin Shell Theory
Navami Kairanda (Max Planck Institute for Informatics), Vladislav Golyanik (Max Planck Institute for Informatics)
MeshBenchmarkPhysics Related
🎯 What it does: A differentiable thin shell theory simulator called NeuralClothSim is proposed, based on a neural deformation field, for continuous, mesh-free reconstruction of static equilibrium calculations for fabrics.
NeuralFluid: Nueral Fluidic System Design and Control with Differentiable Simulation
Yifei Li (Massachusetts Institute of Technology), Wojciech Matusik (Massachusetts Institute of Technology)
OptimizationRobotic IntelligenceReinforcement LearningBenchmarkPhysics Related
🎯 What it does: A fully differentiable neural fluid system has been designed and implemented for fluid control and geometric collaborative design in dynamic solid boundary environments.
NeuralFuse: Learning to Recover the Accuracy of Access-Limited Neural Network Inference in Low-Voltage Regimes
Hao-Lun Sun (National Tsing Hua University), Tsung-Yi Ho (Chinese University of Hong Kong)
OptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper presents NeuralFuse, a pluggable input transformation module that can restore the inference accuracy of deep networks when low voltage causes random bit flips in SRAM.
NeuralSolver: Learning Algorithms For Consistent and Efficient Extrapolation Across General Tasks
Bernardo Esteves (Instituto Superior Tecnico, Universidade de Lisboa), Francisco S. Melo (Instituto Superior Tecnico, Universidade de Lisboa)
Recurrent Neural NetworkSequential
🎯 What it does: We propose NeuralSolver, a recursive solver that achieves efficient and robust extrapolation on tasks of both the same and different sizes.
NeuralSteiner: Learning Steiner Tree for Overflow-avoiding Global Routing in Chip Design
Ruizhi Liu (Chinese Academy of Sciences), Dongbo Bu (Chinese Academy of Sciences)
OptimizationConvolutional Neural NetworkGraph Neural NetworkGraph
🎯 What it does: This paper proposes NeuralSteiner, a two-stage learning-based global routing method that first uses a neural network to predict Steiner points and then constructs a non-overflow RST using augmented graph post-processing.
Neuro-Symbolic Data Generation for Math Reasoning
Zenan Li (Nanjing University), Xiaoxing Ma (Nanjing University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Automatically generate high-quality, multi-difficulty mathematical datasets through a neuro-symbolic framework, and fine-tune LLMs to enhance mathematical reasoning abilities.
Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction
Guobin Shen (Chinese Academy of Sciences), Yi Zeng (Chinese Academy of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityMagnetic Resonance Imaging
🎯 What it does: This paper proposes a general brain signal decoding framework that combines a 3D Vision Transformer (ViT3D) with fMRI signals, integrating it with large language models (LLM) to achieve visual reconstruction, language interaction, and concept localization from a single fMRI experiment.
NeuroBOLT: Resting-state EEG-to-fMRI Synthesis with Multi-dimensional Feature Mapping
Yamin Li (Vanderbilt University), Catie Chang (Vanderbilt University)
Data SynthesisTransformerTime SeriesBiomedical DataMagnetic Resonance Imaging
🎯 What it does: The NeuroBOLT model is proposed, which predicts whole-brain BOLD fMRI signals from raw EEG using multi-dimensional feature mapping, supporting any number of channels and capable of reconstructing deep structures in a resting state;
NeuroClips: Towards High-fidelity and Smooth fMRI-to-Video Reconstruction
Zixuan Gong (Tongji University), Yu Zhang (University of Technology Sydney)
GenerationData SynthesisDiffusion modelContrastive LearningVideoMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This study proposes the NeuroClips framework, which reconstructs high-fidelity, smooth, and continuous videos from non-invasive fMRI.
NeuRodin: A Two-stage Framework for High-Fidelity Neural Surface Reconstruction
Yifan Wang (Shanghai Jiao Tong University), Tong He (Shanghai Artificial Intelligence Laboratory)
GenerationOptimizationNeural Radiance FieldMesh
🎯 What it does: A two-stage neural surface reconstruction framework called NeuRodin is proposed, capable of generating high-fidelity, detail-rich 3D meshes using only calibrated RGB images.
NeuroGauss4D-PCI: 4D Neural Fields and Gaussian Deformation Fields for Point Cloud Interpolation
Chaokang Jiang (PhiGent Robotics), Jie Zhou (Tsinghua University)
Data SynthesisAutonomous DrivingGraph Neural NetworkNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes NeuroGauss4D-PCI, which combines 4D neural fields with Gaussian deformation fields to achieve point cloud frame interpolation, capable of generating continuous and detailed 3D point cloud sequences in sparse and dynamic scenes.
Neuronal Competition Groups with Supervised STDP for Spike-Based Classification
Gaspard Goupy (University of Lille), Ioan Marius Bilasco (University of Lille)
ClassificationRecognitionSpiking Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A neural competition group (NCG) architecture is proposed, which implements intra-class Winner-Take-All competition using a supervised STDP classification layer with first spike coding, thereby enhancing the learning capability of different category patterns.
Newton Informed Neural Operator for Solving Nonlinear Partial Differential Equations
Wenrui Hao (Pennsylvania State University), Yahong Yang (Pennsylvania State University)
TabularPhysics Related
🎯 What it does: Proposed the 'Newton Informed Neural Operator' (NINO) — a framework that combines classical Newton methods with neural operator learning to solve nonlinear partial differential equations (PDEs) with multiple solutions and can capture multiple solutions in a single training session;
Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms
Felix Petersen (Stanford University), Stefano Ermon (Stanford University)
OptimizationReinforcement LearningImageBenchmark
🎯 What it does: By constructing the Newton Loss that utilizes second-order information from the loss function, the training of differentiable algorithm losses that are difficult to optimize in weak supervision has been improved.
Nimbus: Secure and Efficient Two-Party Inference for Transformers
Zhengyi Li (Shanghai Jiao Tong University), Jingwen Leng (Shanghai Jiao Tong University)
Safty and PrivacyComputational EfficiencyTransformerTextBenchmark
🎯 What it does: A two-party secure inference framework named Nimbus is proposed for Transformer models, providing efficient implementations for both linear and nonlinear layers.
No "Zero-Shot" Without Exponential Data: Pretraining Concept Frequency Determines Multimodal Model Performance
Vishaal Udandarao (University of Cambridge), Matthias Bethge (University of Tübingen)
ClassificationGenerationRetrievalTransformerVision Language ModelDiffusion modelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: By systematically evaluating the zero-shot performance of multimodal models (such as CLIP, Stable Diffusion, etc.) on various downstream tasks, this paper analyzes the relationship between the frequency of concept occurrence in pre-training data and model performance, and based on this, constructs a long-tail benchmark set called 'Let It Wag!'.
No Filter: Cultural and Socioeconomic Diversity in Contrastive Vision-Language Models
Angéline Pouget (ETH Zurich), Ibrahim Alabdulmohsin (Google DeepMind)
TransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: The study examines the impact of cultural and socioeconomic diversity on contrastive visual-text models (VLM), revealing biases in using only English data for low-income and non-Western communities, and proposes a geographic localization task as an evaluation metric.
No Free Delivery Service: Epistemic limits of passive data collection in complex social systems
Maximilian Nickel (Meta)
Graph Neural NetworkTabular
🎯 What it does: This paper proves that under passive data collection in complex social systems, the training-testing paradigm cannot ensure the validity of model validation.
No Free Lunch in LLM Watermarking: Trade-offs in Watermarking Design Choices
Qi Pang (Carnegie Mellon University), Virginia Smith (Carnegie Mellon University)
TransformerLarge Language ModelText
🎯 What it does: This paper studies and demonstrates three types of attacks (de-watermarking, forgery, API attacks) against watermarking for large language models (LLMs) and their impacts on robustness, usability, and security, proposing corresponding defense measures and design guidelines.
No Free Lunch Theorem and Black-Box Complexity Analysis for Adversarial Optimisation
Per Kristian Lehre (University of Birmingham), Shishen Lin (University of Birmingham)
OptimizationAdversarial Attack
🎯 What it does: This paper studies the No Free Lunch theorem and black-box complexity for the unique pure Nash equilibrium in black-box adversarial optimization (maximin) in two-player zero-sum games. It proves that all structure-independent black-box algorithms have the same average performance and provides a general lower bound on query complexity.
No Regrets: Investigating and Improving Regret Approximations for Curriculum Discovery
Alexander Rutherford (University of Oxford), Jakob Nicolaus Foerster (University of Oxford)
Robotic IntelligenceReinforcement Learning
🎯 What it does: This paper studies and improves the task prioritization method in Unsupervised Environment Design (UED), proposing to select training environments directly based on the 'learnability' of tasks to enhance the robustness of RL agents.
No Representation, No Trust: Connecting Representation, Collapse, and Trust Issues in PPO
Skander Moalla (Ecole Polytechnique Federale de Lausanne), Caglar Gulcehre (Ecole Polytechnique Federale de Lausanne)
OptimizationReinforcement LearningSequential
🎯 What it does: This paper studies the relationship between feature representation degradation and performance collapse of PPO in non-stationary training environments, and proposes a Proximal Feature Optimization (PFO) auxiliary loss based on feature pre-activation to alleviate this issue.
No Train, all Gain: Self-Supervised Gradients Improve Deep Frozen Representations
Walter Simoncini (University of Amsterdam), Yuki M Asano
ClassificationSegmentationRetrievalTransformerContrastive LearningImageTextAudio
🎯 What it does: The FUNGI method is proposed, which combines the output embeddings of pre-trained models with self-supervised gradients to improve k-nearest neighbor retrieval.
No-Regret Bandit Exploration based on Soft Tree Ensemble Model
Shogo Iwazaki (LY Corporation), Shinya Suzumura (LY Corporation)
OptimizationReinforcement LearningTabular
🎯 What it does: A no-regret UCB-style stochastic bandit algorithm ST-UCB based on a soft tree ensemble model is proposed, which achieves precise estimation of the reward function using gradient descent training of soft trees and the tree neural tangent kernel (TNTK) theory.
No-Regret Learning for Fair Multi-Agent Social Welfare Optimization
Mengxiao Zhang (University of Iowa), Haipeng Luo (University of Southern California)
OptimizationReinforcement Learning
🎯 What it does: This study investigates the online multi-agent Nash social welfare (NSW) maximization problem, providing algorithms and lower bounds with bandit and full-information feedback in random and adversarial environments.
No-regret Learning in Harmonic Games: Extrapolation in the Face of Conflicting Interests
Davide Legacci (University of Grenoble Alpes), Bary Pradelski (CNRS)
OptimizationReinforcement Learning
🎯 What it does: This paper studies the long-term behavior of no-regret learning algorithms in harmonic games, proving that the trajectory of continuous-time FTRL is a Poincaré cycle, while discrete-time FTRL may diverge; it introduces extrapolated FTRL+ (including optimistic and extra step variants) and proves that it achieves O(1) constant regret and converges to Nash equilibrium in all harmonic games.
No-Regret M${}^{\natural}$-Concave Function Maximization: Stochastic Bandit Algorithms and NP-Hardness of Adversarial Full-Information Setting
Taihei Oki (Hokkaido University), Shinsaku Sakaue (University of Tokyo)
Optimization
🎯 What it does: An algorithm for online maximization of M-convex functions in a stochastic bandit environment is proposed, achieving O(T^{-1/2}) simple regret and O(T^{2/3}) cumulative regret; it is also proven that in an adversarial full-information setting, any polynomial-time learner aiming for sublinear regret must lead to P=NP.
Noether's Razor: Learning Conserved Quantities
Tycho F. A. van der Ouderaa (Imperial College London), Pim De Haan
OptimizationRepresentation LearningReinforcement LearningTime SeriesPhysics Related
🎯 What it does: In this paper, the authors propose a method that utilizes Noether's theorem to parameterize symmetries as learnable conserved quantities, and through Bayesian model selection combined with variational inference, automatically learns the symmetries of Hamiltonian dynamics from trajectory data.
Noise Contrastive Alignment of Language Models with Explicit Rewards
Huayu Chen (Tsinghua University), Jun Zhu (Tsinghua University)
Recommendation SystemReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningContrastive LearningText
🎯 What it does: Two language model alignment methods based on Noise Contrastive Estimation (NCE / InfoNCE) are proposed—InfoNCA and NCA, which can handle both explicit reward data and preference data simultaneously.
Noise-Aware Differentially Private Regression via Meta-Learning
Ossi Räisä (University of Helsinki), Richard E. Turner (University of Cambridge)
Safty and PrivacyMeta LearningConvolutional Neural NetworkTabular
🎯 What it does: This paper proposes a differential privacy convolutional conditional neural process (DPConvCNP) based on meta-learning, which incorporates a functional DP mechanism during meta-training on non-sensitive simulated data. Ultimately, it achieves calibrated and accurate predictions with a single forward inference on real private data while satisfying DP privacy guarantees.
NoiseGPT: Label Noise Detection and Rectification through Probability Curvature
Haoyu Wang (Beijing Institute of Technology), Tongliang Liu (University of Sydney)
ClassificationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringImageMultimodality
🎯 What it does: Utilizing a multimodal large language model (MLLM) to detect and correct label noise in image datasets through probability curvature differences, a zero-shot noise detection and correction framework based on In-Context Discrepancy (ICD) is proposed.
Noisy Dual Mirror Descent: A Near Optimal Algorithm for Jointly-DP Convex Resource Allocation
Du Chen (Nanyang Technological University), Geoffrey A. Chua (Nanyang Technological University)
OptimizationSafty and PrivacyTabular
🎯 What it does: This paper studies the convex resource allocation problem under Joint Differential Privacy (JDP) constraints and proposes a Noisy Dual Mirror Descent algorithm that adds noise in the dual space.
Noisy Label Learning with Instance-Dependent Outliers: Identifiability via Crowd Wisdom
Tri Nguyen (Oregon State University), Xiao Fu (Oregon State University)
ClassificationAnomaly DetectionContrastive LearningImage
🎯 What it does: This paper proposes a noise label learning method based on multiple annotators and sparse constraints, COINNet, which can identify and correct noisy labels in the presence of instance-related noise (considered as outliers), achieving identifiable learning for noise-free classifiers.
NoMAD-Attention: Efficient LLM Inference on CPUs Through Multiply-add-free Attention
Tianyi Zhang (Rice University), Anshumali Shrivastava (Rice University)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes NoMAD-Attention, which replaces traditional multiply-accumulate (MAD) operations with low-latency lookups using CPU SIMD registers to achieve efficient inference for large language models.
Non-asymptotic Analysis of Biased Adaptive Stochastic Approximation
Sobihan Surendran (Sorbonne Université), Sylvain Le Corff (LOPF Califrais Machine Learning Lab)
OptimizationAuto EncoderImage
🎯 What it does: This paper conducts a non-asymptotic convergence analysis of stochastic gradient descent (SGD) with biased gradients and adaptive step sizes, providing a linear convergence rate under the PL condition and an O(log n / √n + b_n) convergence rate in general non-convex smooth scenarios. It proves that Adagrad, RMSProp, and AMSGRAD can still converge to critical points in the presence of biased gradients.
Non-asymptotic Approximation Error Bounds of Parameterized Quantum Circuits
Zhan Yu (Wuhan University), Jerry Zhijian Yang (Wuhan University)
🎯 What it does: The paper studies the approximation capability of parameterized quantum circuits (PQC) for multivariable continuous functions, providing explicit constructions and non-asymptotic error bounds.
Non-asymptotic Convergence of Training Transformers for Next-token Prediction
Ruiquan Huang (Penn State University), Jing Yang (Penn State University)
TransformerText
🎯 What it does: This paper studies the non-asymptotic convergence properties of a single-layer Transformer in the next word prediction (NTP) task, proposing a two-stage training algorithm based on normalized gradient descent, and provides theoretical proof for the training dynamics.
Non-asymptotic Global Convergence Analysis of BFGS with the Armijo-Wolfe Line Search
Qiujiang Jin (University of Texas at Austin), Aryan Mokhtari (University of Texas at Austin)
Optimization
🎯 What it does: This paper conducts a rigorous global non-asymptotic convergence analysis of the BFGS method combined with the Armijo-Wolfe line search, providing linear, condition number independent linear, and global superlinear convergence rates, and deriving the overall iteration and line search complexity.
Non-Asymptotic Uncertainty Quantification in High-Dimensional Learning
Frederik Hoppe (RWTH Aachen University), Holger Rauhut (Ludwig Maximilian University of Munich)
OptimizationData-Centric LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningImageMagnetic Resonance Imaging
🎯 What it does: This paper proposes a non-asymptotic uncertainty quantification method based on debiasing estimation and data-driven correction, providing confidence intervals with provable coverage under finite samples.
Non-convolutional graph neural networks.
Yuanqing Wang (New York University), Kyunghyun Cho (Genetech)
Representation LearningDrug DiscoveryRecurrent Neural NetworkGraph Neural NetworkGraphBiomedical Data
🎯 What it does: A convolution-free graph neural network RUM is proposed, which constructs node representations by integrating semantic and topological features through random walks and RNN.
Non-Euclidean Mixture Model for Social Network Embedding
Roshni Iyer, Yizhou Sun (University of California Los Angeles)
Recommendation SystemRepresentation LearningGraph Neural NetworkAuto EncoderGraph
🎯 What it does: A non-Euclidean mixed model (NMM) that combines homogeneity and social influence factors is proposed to explain and generate social networks.
Non-geodesically-convex optimization in the Wasserstein space
Hoang Phuc Hau Luu (University of Helsinki), Arto Klami (University of Helsinki)
Optimization
🎯 What it does: A 'semi-forward-backward Euler' discretization method is proposed and analyzed for solving non-geodesically convex non-convex optimization problems with a differential convex (DC) form objective function in Wasserstein space;
Non-parametric classification via expand-and-sparsify representation
Kaushik Sinha (Wichita State University)
ClassificationTabular
🎯 What it does: Two non-parametric classification algorithms based on the Expand-Sparse (EaS) representation are proposed, and their consistency and optimal convergence rate are proven.
Non-Stationary Learning of Neural Networks with Automatic Soft Parameter Reset
Alexandre Galashov (Google DeepMind), Maneesh Sahani (Gatsby Unit UCL)
OptimizationRobotic IntelligenceReinforcement LearningImageSequential
🎯 What it does: A Soft Reset algorithm is proposed, which achieves a soft reset of neural network parameters through an adaptive Ornstein-Uhlenbeck parameter drift model, thereby addressing the issue of plasticity loss in non-stationary learning.
Nonconvex Federated Learning on Compact Smooth Submanifolds With Heterogeneous Data
Jiaojiao Zhang (KTH Royal Institute of Technology), Mikael Johansson (KTH Royal Institute of Technology)
OptimizationFederated LearningComputational EfficiencyImage
🎯 What it does: In a heterogeneous data environment, an algorithm has been designed to perform non-convex federated learning on compact smooth submanifolds. This algorithm enhances computational and communication efficiency through random Riemannian gradients and manifold projections, effectively eliminating client drift.
Nonlinear dynamics of localization in neural receptive fields
Leon Lufkin (Yale University), Erin Grant (Gatsby Unit and Sainsbury Wellcome Centre University College London)
Image
🎯 What it does: The study investigates the theoretical dynamics of single neuron and multi-neuron models by learning to produce localized receptive fields in feedforward networks without explicit sparse constraints, and validates this through simulation experiments.
Nonlocal Attention Operator: Materializing Hidden Knowledge Towards Interpretable Physics Discovery
Yue Yu (Lehigh University), Stewart A Silling
OptimizationExplainability and InterpretabilityComputational EfficiencyAuto EncoderTabularPhysics Related
🎯 What it does: A Nonlocal Attention Operator (NAO) based on attention is proposed, achieving unified learning for forward prediction and inverse mechanism discovery of physical systems.
Nonparametric Classification on Low Dimensional Manifolds using Overparameterized Convolutional Residual Networks
Zixuan Zhang (Georgia Institute of Technology), Yu-Xiang Wang (University of California San Diego)
ClassificationOptimizationConvolutional Neural Network
🎯 What it does: This paper conducts a non-parametric classification theoretical analysis of over-parameterized ConvResNeXt with weight decay, proving that it achieves nearly optimal learning rates for Besov functions on low-dimensional manifolds.
Nonparametric Evaluation of Noisy ICA Solutions
Syamantak Kumar (University of Texas at Austin), Purnamrita Sarkar (University of Texas at Austin)
OptimizationMeta LearningImage
🎯 What it does: A non-parametric independence score is proposed to evaluate noisy ICA solutions, and this score is used to design a meta-algorithm for adaptive selection of the best algorithm, while introducing new logarithmic feature functions and cumulative generating function comparison functions.
Nonparametric Instrumental Variable Regression through Stochastic Approximate Gradients
Yuri Fonseca, Yuri Saporito
OptimizationTabular
🎯 What it does: A non-parametric instrumental variable regression algorithm based on Stochastic Approximate Gradient Descent (SAGD-IV) is proposed, which directly minimizes the overall risk;
Nonstationary Sparse Spectral Permanental Process
Zicheng Sun (Renmin University of China), Feng Zhou (Renmin University of China)
Point CloudTime Series
🎯 What it does: This paper proposes a Non-Stationary Sparse Spectral Poisson Process (NSSPP) and its deep variant (DNSSPP) for flexible modeling of point process intensity functions.
Normal-GS: 3D Gaussian Splatting with Normal-Involved Rendering
Meng Wei (Monash University), Jianfei Cai (Monash University)
GenerationOptimizationGaussian SplattingPoint Cloud
🎯 What it does: A rendering method involving normals (surface normals) is proposed within the 3D Gaussian Splatting framework, achieving a unification of real-time rendering and high-precision normal estimation by combining low-frequency lighting vectors and self-supervised normal regularization.
Normalization and effective learning rates in reinforcement learning
Clare Lyle (Google DeepMind), Will Dabney (Google DeepMind)
TransformerReinforcement LearningImageSequential
🎯 What it does: Proposes the Normalize-and-Project (NaP) training protocol, which normalizes before each nonlinear layer and periodically projects weights to a fixed norm to maintain a constant effective learning rate, thereby enhancing the plasticity and performance in non-stationary tasks (continual learning, reinforcement learning).
Normalization Layer Per-Example Gradients are Sufficient to Predict Gradient Noise Scale in Transformers
Gavia Gray (Cerebras Systems), Joel Hestness (Cerebras Systems)
OptimizationComputational EfficiencyTransformerText
🎯 What it does: This paper studies the per-sample gradient norm in the Transformer model and proposes an efficient method to accurately estimate the gradient noise scale (GNS) using only the LayerNorm layer.
Not All Diffusion Model Activations Have Been Evaluated as Discriminative Features
Benyuan Meng (Institute of Information Engineering, Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
RecognitionSegmentationTransformerDiffusion modelImage
🎯 What it does: Based on the extended diffusion U-Net activation pool, a qualitative filtering and quantitative comparison is proposed based on three unique characteristics (diffusion noise, fine-grained changes within resolution, and locality without positional embedding) to select the most discriminative activations as features.
Not All Tokens Are What You Need for Pretraining
Zhenghao Lin (Xiamen University), Weizhu Chen (Microsoft)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed and implemented the Selective Language Modeling (SLM) technique, training the RHO-1 language model by applying loss only to 'useful and high-information' tokens during training, thereby achieving more efficient data utilization;
Not Just Object, But State: Compositional Incremental Learning without Forgetting
Yanyi Zhang (Dalian University of Technology), Ran He (Institute of Automation Chinese Academy of Sciences)
ClassificationObject DetectionRepresentation LearningTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes the Composition-Incremental Learning (Composition-IL) task and designs the CompILer model to achieve incremental learning of state-object combinations.
Not so griddy: Internal representations of RNNs path integrating more than one agent
William T Redman, Nina Miolane (University of California)
Recurrent Neural NetworkSequential
🎯 What it does: This paper trains a recurrent neural network (RNN) to perform a dual-agent path integration task and compares its internal representations with those of a single-agent RNN, exploring the possible computational mechanisms of the central nervous system (MEC) under multi-agent navigation.
Novel Object Synthesis via Adaptive Text-Image Harmony
Zeren Xiong (Nanjing University of Science and Technology), Jun Li (Nanjing University of Science and Technology)
Image HarmonizationGenerationData SynthesisHyperparameter SearchTransformerDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: An Adaptive Text-Image Harmony (ATIH) framework is proposed to fuse object text with object images, generating novel object images.