NeurIPS 2023 Papers — Page 20
Conference on Neural Information Processing Systems · 3218 papers
Natural Language Instruction-following with Task-related Language Development and Translation
Jing-Cheng Pang (Nanjing University), Yang Yu (Nanjing University)
Large Language ModelReinforcement LearningText
🎯 What it does: This paper proposes and implements an Inside-Out Learning (IOL) framework called TALAR, which uses Task Language (TL) instead of Natural Language (NL) for reinforcement learning instruction following.
Navigating Data Heterogeneity in Federated Learning: A Semi-Supervised Federated Object Detection
Taehyeon Kim (Korea Advanced Institute of Science and Technology), Vaikkunth Mugunthan (DynamoFL)
Object DetectionAutonomous DrivingFederated LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A semi-supervised federated learning-based object detection framework SSFOD is proposed, advocating for achieving high-precision detection in extreme scenarios where the server has labels and clients have none.
Navigating the Pitfalls of Active Learning Evaluation: A Systematic Framework for Meaningful Performance Assessment
Carsten Tim Lüth, Paul F Jaeger
ClassificationData-Centric LearningConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: A systematic improvement of the evaluation of active learning methods was conducted, and a large-scale experimental benchmark was established.
NCDL: A Framework for Deep Learning on non-Cartesian Lattices
Joshua John Horacsek, Usman Alim
Image TranslationRestorationObject DetectionSegmentationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a new data structure called the lattice tensor and implements a general framework NCDL for performing common deep learning operations such as convolution, pooling, and downsampling on non-Cartesian grid points.
Near Optimal Reconstruction of Spherical Harmonic Expansions
Amir Zandieh (Independent Researcher), Haim Avron (Tel Aviv University)
Point CloudPhysics Related
🎯 What it does: This paper proposes an algorithm that combines uniform random sampling with kernel regression, capable of recovering the low-order spherical harmonic expansion of functions defined on the d-dimensional unit sphere with nearly optimal sample size.
Near-Linear Time Algorithm for the Chamfer Distance
Ainesh Bakshi (Massachusetts Institute of Technology), Erik Waingarten (University of Pennsylvania)
RetrievalOptimizationComputational EfficiencyPoint Cloud
🎯 What it does: This paper proposes a near-linear time approximation algorithm for estimating Chamfer distance between two sets of points, providing a 1+ε approximation guarantee.
Near-Optimal $k$-Clustering in the Sliding Window Model
David Woodruff, Samson Zhou (Texas A&M University)
SegmentationOptimizationImage
🎯 What it does: An approximate (k,z)-clustering algorithm with a 1+ε guarantee is provided under the sliding window model, utilizing online core samples to achieve approximately optimal space complexity.
Near-Optimal Algorithms for Gaussians with Huber Contamination: Mean Estimation and Linear Regression
Ilias Diakonikolas (University of Wisconsin Madison), Thanasis Pittas (University of Wisconsin Madison)
OptimizationTabular
🎯 What it does: This paper proposes a near-optimal sample complexity (~O(d/ε²)) and a near-linear time algorithm for mean estimation and linear regression problems under the Huber contamination model for high-dimensional Gaussian distributions, which can recover the target parameters within ℓ₂ error O(ε√log(1/ε)) (for linear regression, it is O(σ ε√log(1/ε))).
Near-Optimal Bounds for Learning Gaussian Halfspaces with Random Classification Noise
Ilias Diakonikolas (University of Wisconsin Madison), Nikos Zarifis (University of Wisconsin Madison)
OptimizationComputational Efficiency
🎯 What it does: This study investigates the problem of learning general half-spaces with random classification noise under Gaussian distribution, establishing nearly matching algorithms and statistical query (SQ) lower bound results, revealing the information-computation gap of this fundamental problem.
Near-optimal learning with average Hölder smoothness
Guy Kornowski (Weizmann Institute of Science), Aryeh Kontorovich (Ben-Gurion University of the Negev)
🎯 What it does: This paper proposes an average measure of Hölder smoothness and provides optimal generalization bounds and learning algorithms for both realizable and non-realizable regression.
Nearest Neighbour with Bandit Feedback
Stephen Pasteris (Alan Turing Institute), Vasilios Mavroudis (Alan Turing Institute)
Recommendation SystemOptimizationComputational EfficiencySequential
🎯 What it does: This paper adapts the classic nearest neighbor rule to the contextual bandit problem, developing an efficient algorithm that can operate in a fully adversarial environment without making any assumptions about the data generation process.
Nearly Optimal Bounds for Cyclic Forgetting
William Joseph Swartworth (Carnegie Mellon University), Halyun Jeong (University of California Los Angeles)
🎯 What it does: This paper presents a theoretical upper bound on the amount of forgetting in linear task cyclical learning, proving that the upper bound is O(T^2/m) (where T is the number of tasks and m is the number of cycles), and this upper bound is independent of the dimension;
Nearly Optimal VC-Dimension and Pseudo-Dimension Bounds for Deep Neural Network Derivatives
Yahong Yang (Pennsylvania State University), Yang Xiang (Hong Kong University of Science and Technology)
🎯 What it does: This paper derives the approximate optimal upper bounds for the VC-dimension and pseudo-dimension of the derivative functions of deep neural networks (DNNs), and proves that under Sobolev spaces, a ReLU network with width O(N log N) and depth O(L log L) has an approximation rate of O(N^{−2(n−1)/d} L^{−2(n−1)/d}), while also providing the corresponding generalization error upper bound O((NL log NL log LL)^{1/2}/√M).
Nearly Tight Bounds For Differentially Private Multiway Cut
Mina Dalirrooyfard (Morgan Stanley), Yuriy Nevmyvaka (Morgan Stanley)
OptimizationSafty and PrivacyGraph Neural NetworkGraph
🎯 What it does: This paper proposes a differential privacy-based minimum s-t cut and multi-way k-cut algorithm, providing nearly optimal upper and lower bounds on error, achieving privacy protection while maintaining the time complexity of the fastest non-private algorithms.
Necessary and Sufficient Conditions for Optimal Decision Trees using Dynamic Programming
Jacobus G.M. van der Linden, Emir Demirović (Delft University of Technology)
OptimizationTabular
🎯 What it does: A general dynamic programming framework STreeD is proposed for globally optimal search of binary decision trees, supporting various objectives and constraints.
NEO-KD: Knowledge-Distillation-Based Adversarial Training for Robust Multi-Exit Neural Networks
Seokil Ham (KAIST), Jaekyun Moon (KAIST)
Knowledge DistillationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: To address the robustness issue of multi-output neural networks, a knowledge distillation-based adversarial training method called NEO-KD is proposed.
NeRF Revisited: Fixing Quadrature Instability in Volume Rendering
Mikaela Angelina Uy (Stanford University), Ke Li (Simon Fraser University)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: A body rendering formula based on piecewise linear opacity is proposed to address the sampling instability (quadrature instability) problem in NeRF.
NeRF-IBVS: Visual Servo Based on NeRF for Visual Localization and Navigation
Yuanze Wang (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
Pose EstimationRobotic IntelligenceNeural Radiance FieldSimultaneous Localization and MappingImage
🎯 What it does: A visual localization and navigation framework based on NeRF, called NeRF-IBVS, is proposed, which can achieve high-precision localization using only a small number of pose images and implement markerless, depth sensor-free navigation through image-based visual servoing.
NetHack is Hard to Hack
Ulyana Piterbarg (New York University), Rob Fergus (New York University)
Recurrent Neural NetworkTransformerSupervised Fine-TuningReinforcement LearningMultimodality
🎯 What it does: This paper constructs a HiHack dataset with hierarchical labels to study and implement a neural strategy for NetHack based on hierarchical behavior cloning, a Transformer-LSTM structure, and RL fine-tuning, significantly improving the performance of the neural model.
Networks are Slacking Off: Understanding Generalization Problem in Image Deraining
Jinjin Gu (University of Sydney), Chao Dong (Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper analyzes the fundamental reasons for the poor generalization of image de-raining networks in real environments through systematic experiments and provides corresponding training strategies.
Neural (Tangent Kernel) Collapse
Mariia Seleznova (Ludwig-Maximilians-Universität München), Hung-Hsu Chou (Ludwig-Maximilians-Universität München)
Convolutional Neural NetworkImage
🎯 What it does: This study investigates the relationship between the neural tangent kernel (NTK) and neural collapse (NC) during the training process of deep networks, deriving the training dynamics under a block-structured NTK and proving that NTK alignment leads to the emergence of NC.
Neural Algorithmic Reasoning Without Intermediate Supervision
Gleb Rodionov (Yandex Research), Liudmila Prokhorenkova (Yandex Research)
Graph Neural NetworkContrastive LearningGraphBenchmark
🎯 What it does: A method is proposed to train neural algorithm reasoning models without using intermediate supervision (hints), significantly improving performance through architectural improvements and self-supervised regularization.
Neural approximation of Wasserstein distance via a universal architecture for symmetric and factorwise group invariant functions
Samantha Chen (University of California), Yusu Wang
OptimizationPoint Cloud
🎯 What it does: A general neural network framework is designed to approximate symmetric and factor-level group invariant (SFGI) functions, particularly providing a provably low-parameter approximation method for the p-Wasserstein distance, with the number of network parameters not varying with the size of the input point set.
Neural Circuits for Fast Poisson Compressed Sensing in the Olfactory Bulb
Jacob A Zavatone-Veth, Cengiz Pehlevan (Harvard University)
RecognitionCompressionOptimizationSpiking Neural NetworkTime SeriesStochastic Differential Equation
🎯 What it does: This paper proposes a rate-type neural circuit model based on Poisson compressed sensing for simulating the early signal processing of mammalian olfactory bulbs, capable of accurately identifying and quantifying multiple odors within a single breath (approximately 100 ms), and supports Bayesian posterior sampling to estimate concentration uncertainty.
Neural Combinatorial Optimization with Heavy Decoder: Toward Large Scale Generalization
Fu Luo (Southern University of Science and Technology), Zhenkun Wang (Southern University of Science and Technology)
OptimizationTransformerSupervised Fine-TuningTabular
🎯 What it does: A neural combinatorial optimization model with a light encoding-heavy decoding (LEHD) structure is proposed, achieving constructive solutions for large-scale TSP/CVRP problems through a data-efficient 'Learn to construct partial solution' training scheme and a Random Reconstruction (RRC) mechanism.
Neural Data Transformer 2: Multi-context Pretraining for Neural Spiking Activity
Joel Ye (University of Pittsburgh), Robert Gaunt (University of Pittsburgh)
Spiking Neural NetworkTransformerPrompt EngineeringTime SeriesBiomedical Data
🎯 What it does: This paper presents NDT2, a multi-context pre-trained Transformer model designed for decoding neural spike activity in brain-machine interfaces.
Neural Fields with Hard Constraints of Arbitrary Differential Order
Fangcheng Zhong (University of Cambridge), Cengiz Oztireli (University of Cambridge)
OptimizationNeural Radiance FieldPoint Cloud
🎯 What it does: A method based on a learnable basis function called Constrained Neural Field (CNF) is proposed, which can strictly satisfy linear operator (including differential) constraints during training and achieve high-quality results in multiple domains.
Neural Foundations of Mental Simulation: Future Prediction of Latent Representations on Dynamic Scenes
Aran Nayebi (McGovern Institute for Brain Research), Guangyu Robert Yang (McGovern Institute for Brain Research)
Representation LearningRecurrent Neural NetworkContrastive LearningVideo
🎯 What it does: Construct and evaluate a multi-class perception-cognition network for predicting future states in natural dynamic scenes, and compare it with monkey neural recordings and human behavior.
Neural Frailty Machine: Beyond proportional hazard assumption in neural survival regressions
Ruofan Wu (Ant Group), Weiqiang Wang (Ant Group)
TabularBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: Proposed the Neural Frailty Machine (NFM), a survival regression framework that utilizes multiple random effects to extend CoxPH and combines neural networks.
Neural Functional Transformers
Allan Zhou (Stanford University), Chelsea Finn (Stanford University)
ClassificationRepresentation LearningTransformerImage
🎯 What it does: Designed and implemented a Neural Function Transformer (NFT) based on attention mechanisms to handle weight space and applied it to tasks such as implicit space learning (INR2ARRAY), INR classification, editing, and CNN generalization prediction.
Neural Graph Generation from Graph Statistics
Kiarash Zahirnia (Simon Fraser University), Oliver Schulte (McGill University)
GenerationData SynthesisSafty and PrivacyGraph Neural NetworkAuto EncoderGraph
🎯 What it does: We propose GenStat—a deep graph generation model that is based solely on graph statistics rather than a complete adjacency matrix, designed to generate realistic synthetic graphs under the premise of local differential privacy (LDP).
Neural Harmonics: Bridging Spectral Embedding and Matrix Completion in Self-Supervised Learning
Marina Munkhoeva (Max Planck Institute for Intelligent Systems), Ivan Oseledets (Artificial Intelligence Research Institute)
ClassificationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper views self-supervised learning as a Lagrangian dual problem of low-rank matrix completion on sparse similarity matrices generated by data augmentation, and connects it to spectral embedding through the Laplacian operator and heat kernel, providing theoretical convergence and inconsistency analysis.
Neural Ideal Large Eddy Simulation: Modeling Turbulence with Neural Stochastic Differential Equations
Anudhyan Boral (Google Research), Fei Sha (Google Research)
TransformerTime SeriesStochastic Differential Equation
🎯 What it does: This paper proposes a closed model based on neural stochastic differential equations (niLES), which combines the statistical ideas of ideal large eddy simulation (ideal LES) with data-driven generative models, enabling high-precision predictions of large-scale turbulent evolution on low-resolution grids.
Neural Image Compression: Generalization, Robustness, and Spectral Biases
Kelsey Lieberman (Duke University), Bhavya Kailkhura (Lawrence Livermore National Laboratory)
CompressionAuto EncoderImageBenchmark
🎯 What it does: This paper proposes a benchmark dataset (CLIC-C, Kodak-C) and a spectral inspection tool for evaluating neural image compression (NIC) in out-of-distribution (OOD) scenarios. It investigates the generalization and robustness performance of traditional codecs versus NIC models under different compression rates, noise types, and intensities through experimental and theoretical analysis.
Neural Injective Functions for Multisets, Measures and Graphs via a Finite Witness Theorem
Tal Amir (Technion - Israel Institute of Technology), Nadav Dym (Technion - Israel Institute of Technology)
Graph Neural NetworkGraph
🎯 What it does: Proves that neural networks (using analytic non-polynomial activations) can achieve injective mappings on multisets and measures, and introduces the finite witness theorem to compress infinite constraints into a finite number;
Neural Lad: A Neural Latent Dynamics Framework for Times Series Modeling
Ting Li (Ant Group), Zhanxing Zhu (Peking University)
Time SeriesElectronic Health RecordsOrdinary Differential Equation
🎯 What it does: This paper proposes Neural Lad, a time series prediction framework based on Neural Ordinary Differential Equations (Neural ODE), which can simultaneously capture local changes in observed signals, seasonal trends, and multivariate spatial correlations.
Neural Latent Geometry Search: Product Manifold Inference via Gromov-Hausdorff-Informed Bayesian Optimization
Haitz Sáez de Ocáriz Borde (University of Oxford), Xiaowen Dong (University of Oxford)
OptimizationImage
🎯 What it does: A new problem setting is proposed, aimed at searching for the optimal latent geometric shape to fit the model and downstream tasks, referred to as Neural Latent Geometry Search (NLGS).
Neural Lighting Simulation for Urban Scenes
Ava Pun (University of Toronto), Raquel Urtasun (University of Toronto)
Object DetectionData SynthesisAutonomous DrivingNeural Radiance FieldImage
🎯 What it does: LightSim is proposed, a neural network-based urban scene lighting simulation system that can construct a lighting-aware digital twin from real sensor data, supporting light transformations, shadow editing, and dynamic actor insertion.
Neural Lyapunov Control for Discrete-Time Systems
Junlin Wu (Washington University in St. Louis), Yevgeniy Vorobeychik (Washington University in St. Louis)
OptimizationReinforcement Learning from Human FeedbackTime Series
🎯 What it does: A neural Lyapunov control learning framework for discrete-time nonlinear systems (DITL) is proposed, which can derive provably stable control strategies.
Neural Modulation for Flash Memory: An Unsupervised Learning Framework for Improved Reliability
Jonathan Zedaka (Samsung Semiconductor), Amit Berman (Samsung Semiconductor)
Auto EncoderGenerative Adversarial Network
🎯 What it does: The research designs a flash memory writing scheme through unsupervised learning, utilizing a neural modulator to reduce error rates and extend lifespan.
Neural Multi-Objective Combinatorial Optimization with Diversity Enhancement
Jinbiao Chen (Sun Yat-sen University), Jiahai Wang (Sun Yat-sen University)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper proposes a neural heuristic method based on indicator enhancement (NHDE) for multi-objective combinatorial optimization problems, significantly improving the diversity and convergence of the Pareto front.
Neural Oscillators are Universal
Samuel Lanthaler (California Institute of Technology), Siddhartha Mishra (ETH Zurich)
Ordinary Differential Equation
🎯 What it does: An abstract framework for Neural Oscillators is proposed, and its universality (the ability to approximate any continuous causal operator) is demonstrated.
Neural Polarizer: A Lightweight and Effective Backdoor Defense via Purifying Poisoned Features
Mingli Zhu (Chinese University of Hong Kong Shenzhen), Baoyuan Wu (Chinese University of Hong Kong Shenzhen)
OptimizationAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A lightweight post-processing defense method called Neural Polarizer is proposed, which filters trigger features and retains normal features by inserting a learnable linear transformation layer into the model under backdoor attack, thereby purifying contaminated samples.
Neural Priming for Sample-Efficient Adaptation
Matthew Wallingford (University of Washington), Ali Farhadi (University of Washington)
RetrievalDomain AdaptationPrompt EngineeringVision Language ModelImageText
🎯 What it does: Proposes the Neural Priming method, which utilizes data retrieval from the pre-trained model itself and fine-tuning (or dynamic retrieval during inference) to improve adaptability to distribution shifts and downstream tasks, without the need for additional labeled data.
Neural Processes with Stability
Huafeng Liu (Beijing Jiaotong University), Jian Yu (Beijing Jiaotong University)
ImageTime Series
🎯 What it does: This paper proposes a stabilization method for Neural Processes (NP) based on algorithmic stability theory, which actively removes easily predictable samples during training to construct a hard prediction subset, thereby enhancing the model's robustness to noisy data.
Neural Relation Graph: A Unified Framework for Identifying Label Noise and Outlier Data
Jang-Hyun Kim (Seoul National University), Hyun Oh Song (Seoul National University)
Anomaly DetectionGraph Neural NetworkImageTextAudio
🎯 What it does: This paper proposes a Neural Relation Graph based on the relationships of data in the feature embedding space to unify the identification of label noise and anomalous/out-of-domain samples, providing a visualization tool for interactive data diagnosis.
Neural Sampling in Hierarchical Exponential-family Energy-based Models
Xingsi Dong (Peking University), Si Wu (Peking University)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImageStochastic Differential Equation
🎯 What it does: A hierarchical exponential family energy model (HEE) is proposed, achieving Bayesian inference through neural sampling and utilizing local Hebbian rules for model learning.
Neural Sculpting: Uncovering hierarchically modular task structure in neural networks through pruning and network analysis
Shreyas Malakarjun Patil (Georgia Institute of Technology), Constantine Dovrolis (Cyprus Institute)
ImageTabular
🎯 What it does: The Neural Sculpting method is proposed, revealing the hierarchical modular structure of hidden tasks in neural networks through iterative unit and edge pruning training.
Neural-Logic Human-Object Interaction Detection
Liulei Li (University of Technology Sydney), Yi Yang (Zhejiang University)
RecognitionObject DetectionTransformerVision Language ModelImage
🎯 What it does: Proposes the LOGICHOI framework, which rewrites self-attention as triplet reasoning attention in the interactive decoder of the Transformer, allowing the model to autonomously combine people, actions, and objects during the decoding phase and predict interactions; simultaneously embedding feasibility constraints (affordances and proxemics) expressed in first-order logic into continuous space to guide the learning and reasoning of the Transformer.
NeuralGF: Unsupervised Point Normal Estimation by Learning Neural Gradient Function
Qing Li (Tsinghua University), Zhizhong Han (Wayne State University)
OptimizationPoint Cloud
🎯 What it does: An unsupervised neural gradient function is proposed to directly estimate consistently oriented normal vectors from raw point clouds, achieving global surface fitting through multi-step point movement.
Neuro-symbolic Learning Yielding Logical Constraints
Zenan Li (Nanjing University), Jian Lu
Autonomous DrivingOptimizationImage
🎯 What it does: An end-to-end neural symbolic learning framework is proposed, capable of simultaneously training neural network perception and explicit logical constraints under weak supervision, and enabling interaction between the two through symbolic binding.
NeuroGF: A Neural Representation for Fast Geodesic Distance and Path Queries
Qijian Zhang (City University of Hong Kong), Ying He (Nanyang Technological University)
OptimizationComputational EfficiencyPoint CloudMesh
🎯 What it does: This paper proposes the NeuroGF, an implicit representation based on neural networks for fast and accurate querying of geodesic distances and shortest paths between arbitrary points; it also unifies the encoding of mesh geometry and geodesic information; further extended to a general framework that supports various inputs such as point clouds and meshes.
New Bounds for Hyperparameter Tuning of Regression Problems Across Instances
Nina Balcan, Dravyansh Sharma (Carnegie Mellon University)
OptimizationHyperparameter Search
🎯 What it does: The study investigates how to effectively tune the regularization parameters in Elastic Net and Regularized Logistic Regression in a cross-instance data-driven setting, providing theoretical sample complexity and generalization guarantees.
New Complexity-Theoretic Frontiers of Tractability for Neural Network Training
Cornelius Brand (Vienna University of Technology), Mathis Rocton (Vienna University of Technology)
Optimization
🎯 What it does: A new algorithmic upper bound is proposed, proving that optimal training can be completed in polynomial time for ReLU networks with hidden layer neuron out-degree not exceeding 1 and linear networks that satisfy the 'untangling' condition.
Newton–Cotes Graph Neural Networks: On the Time Evolution of Dynamic Systems
Lingbing Guo (Zhejiang University), Huajun Chen (Zhejiang University)
Graph Neural NetworkGraphTime SeriesPhysics Related
🎯 What it does: A graph neural network (NC) based on Newton-Cotes numerical integration is proposed, which directly predicts the time evolution of dynamic systems through multi-step velocity estimation and integration formulas.
NICE: NoIse-modulated Consistency rEgularization for Data-Efficient GANs
Yao Ni (Australian National University), Piotr Koniusz (Data61 CSIRO)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A training method is proposed that enhances the generalization and stability of GANs under limited data by incorporating adaptive multiplicative noise into the discriminator and using consistency regularization.
No Change, No Gain: Empowering Graph Neural Networks with Expected Model Change Maximization for Active Learning
Zixing Song (Chinese University of Hong Kong), Irwin King (Chinese University of Hong Kong)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes an active learning method for graph neural networks based on a Bayesian framework and provides a closed-form computable EMCM acquisition function.
No Representation Rules Them All in Category Discovery
Sagar Vaze (Visual Geometry Group University of Oxford), Andrew Zisserman (Visual Geometry Group University of Oxford)
ClassificationRepresentation LearningContrastive LearningImageBenchmark
🎯 What it does: A new benchmark dataset Clevr-4 is proposed, and the GCD method is improved.
No Train No Gain: Revisiting Efficient Training Algorithms For Transformer-based Language Models
Jean Kaddour (University College London), Matt Kusner
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper evaluates three types of efficient training algorithms (dynamic architecture, batch selection, efficiency optimizers) under a fixed computational budget (reference system time) and explores their impact on the pre-training and downstream performance of BERT and T5.
No-regret Algorithms for Fair Resource Allocation
Abhishek Sinha (Tata Institute of Fundamental Research), Mohammad Hajiesmaili (University of Massachusetts)
OptimizationTabularSequential
🎯 What it does: An online fair allocation algorithm (OFA) is designed to achieve approximately regret-free α-fairness in the NOFRA problem.
No-Regret Learning in Dynamic Competition with Reference Effects Under Logit Demand
Mengzi Amy Guo (University of California Berkeley), Zuo-Jun Shen
Optimization
🎯 What it does: An online project gradient ascent algorithm (OPGA) is proposed to address the issues of price competition and reference price effects between two companies in an opaque market environment. It is proven that the algorithm globally converges to a unique steady-state Nash equilibrium (SNE) from any initial state, with a convergence rate of O(1/t);
No-Regret Learning with Unbounded Losses: The Case of Logarithmic Pooling
Eric Neyman (Columbia University), Tim Roughgarden (Columbia University)
OptimizationReinforcement Learning
🎯 What it does: This paper studies regret-free learning under unbounded loss, with a particular focus on aggregation methods for expert predictions, specifically logarithmic pooling. A new algorithm based on Online Mirror Descent is proposed, which can learn optimal expert weights in an online adversarial environment and achieve an expected regret of O(√T log T).
No-Regret Online Prediction with Strategic Experts
Omid Sadeghi (University of Washington), Maryam Fazel (University of Washington)
Reinforcement LearningTabular
🎯 What it does: This paper studies the online binary prediction problem where m experts can be selected in each round, and designs an algorithm that can incentivize truthful reporting while achieving no-regret even when experts may cheat.
No-Regret Online Reinforcement Learning with Adversarial Losses and Transitions
Tiancheng Jin (University of Southern California), Haipeng Luo (University of Southern California)
Reinforcement Learning
🎯 What it does: An algorithm for achieving no-regret learning in a reinforcement learning environment with both adversarial loss functions and adversarial state transition functions is proposed, along with an asymptotically optimal regret upper bound.
Noether Embedding: Efficient Learning of Temporal Regularities
Chi Gao (Tsinghua University), Luping Shi (Tsinghua University)
RetrievalAnomaly DetectionComputational EfficiencyTime SeriesSequential
🎯 What it does: This paper proposes and implements Noether Embedding (NE), which directly learns and retrieves temporal patterns (TR) through the embedding of event samples, and defines two tasks: TR detection and TR querying, which are experimentally evaluated on ICEWS14, ICEWS18, and GDELT.
Noise-Adaptive Thompson Sampling for Linear Contextual Bandits
Ruitu Xu (Yale University), Tianhao Wang (Yale University)
OptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: Two linear contextual bandit sampling (TS) algorithms are proposed, achieving noise-adaptive decision-making and theoretically interpretable convergence for the linear contextual multi-armed bandit problem with unknown heteroscedastic noise.
Nominality Score Conditioned Time Series Anomaly Detection by Point/Sequential Reconstruction
Chih-Yu Lai (Massachusetts Institute of Technology), Duane S Boning
Anomaly DetectionAuto EncoderTime SeriesSequential
🎯 What it does: An unsupervised time series anomaly detection framework based on point reconstruction and sequence reconstruction (NPSR) is proposed, which captures both point anomalies and contextual anomalies by introducing nominality scores and induced anomaly scores.
Non-adversarial training of Neural SDEs with signature kernel scores
Zacharia Issa (King's College London), Cristopher Salvi (Imperial College London)
GenerationData SynthesisAnomaly DetectionOptimizationGenerative Adversarial NetworkTime SeriesSequentialFinance RelatedStochastic Differential Equation
🎯 What it does: This paper proposes a non-adversarial training method based on signature kernels for continuous-time generative models, specifically Neural SDE.
Non-Asymptotic Analysis of a UCB-based Top Two Algorithm
Marc Jourdan, Rémy Degenne (University of Lille)
Tabular
🎯 What it does: A Top Two algorithm TTUCB based on UCB is proposed, and its non-asymptotic sample complexity upper bound is provided for any confidence level.
Non-autoregressive Machine Translation with Probabilistic Context-free Grammar
Shangtong Gui (Institute of Computing Technology, Chinese Academy of Sciences), Yang Feng (Institute of Computing Technology, Chinese Academy of Sciences)
Knowledge DistillationTransformerTextMultimodality
🎯 What it does: This paper proposes the PCFG-NAT model, which embeds probabilistic context-free grammar (right-heavy PCFG) into a non-autoregressive Transformer, utilizing hierarchical structures to capture semantic dependencies of the target language, addressing multimodal issues and improving translation quality.
Non-Convex Bilevel Optimization with Time-Varying Objective Functions
Sen Lin (Ohio State University), Ness Shroff (Ohio State University)
OptimizationRepresentation LearningHyperparameter SearchTextTabular
🎯 What it does: This paper studies online time-varying objective functions in non-convex bilevel optimization and proposes a single-loop online bilevel optimizer called SOBOW.
Non-Rigid Shape Registration via Deep Functional Maps Prior
Puhua Jiang (Tsinghua University), Ruqi Huang (Tsinghua University)
Pose EstimationOptimizationGraph Neural NetworkPoint CloudMesh
🎯 What it does: This paper proposes a hybrid framework based on Deep Functional Maps (DFM) and non-rigid shape registration, which can non-rigidly align the source mesh to the target point cloud under the condition of no correspondence supervision, achieving high-quality point cloud correspondences.
Non-Smooth Weakly-Convex Finite-sum Coupled Compositional Optimization
Quanqi Hu (Texas A&M University), Tianbao Yang (Texas A&M University)
OptimizationTabularBiomedical DataStochastic Differential Equation
🎯 What it does: This paper proposes two-layer (FCCO) and three-layer (TCCO) finite sum coupling optimization problems in the non-smooth weakly convex case, and presents single-cycle stochastic algorithms SONX and SONT, analyzing their convergence to the Moreau envelope. The algorithm is then applied to maximize the bidirectional partial AUC (TPAUC) in deep learning.
Non-Stationary Bandits with Auto-Regressive Temporal Dependency
Qinyi Chen (Massachusetts Institute of Technology), Djallel Bouneffouf (IBM Research)
Reinforcement LearningTime Series
🎯 What it does: A non-stationary multi-armed bandit model is proposed, where the expected reward of each arm evolves over time according to an autoregressive (AR) structure, and an algorithm AR2 is provided that can achieve nearly optimal dynamic steady-state returns within this framework.
Non-stationary Experimental Design under Linear Trends
David Simchi-Levi (Massachusetts Institute of Technology), Zeyu Zheng (University of California)
Optimization
🎯 What it does: Designed and analyzed a non-stationary experimental scheme under linear trends.
Nonparametric Boundary Geometry in Physics Informed Deep Learning
Scott Alexander Cameron (Oxford University), Stephen J. Roberts (Oxford University)
TransformerMeshPhysics Related
🎯 What it does: Designed and trained a neural operator capable of receiving triangular mesh boundary geometry and outputting corresponding PDE solutions, addressing the issue of fixed geometric parameterization in traditional PINNs.
Nonparametric Identifiability of Causal Representations from Unknown Interventions
Julius von Kügelgen (Max Planck Institute for Intelligent Systems), Bernhard Schölkopf (Max Planck Institute for Intelligent Systems)
Flow-based ModelAuto Encoder
🎯 What it does: This paper studies the identifiability problem of non-parametric causal representation learning under multiple environments with unknown intervention targets, proving that potential causal variables and their causal graphs can be theoretically recovered solely through perfect single-node interventions.
Nonparametric Teaching for Multiple Learners
Chen Zhang (Jilin University), James Kwok
OptimizationMeta LearningImage
🎯 What it does: A multi-learner non-parametric teaching framework MINT is proposed, which can provide teaching samples for multiple learners (each learner learning a scalar target function) at once, addressing the issue that single-learner teaching cannot be directly extended to multi-learner scenarios.
Norm-based Generalization Bounds for Sparse Neural Networks
Tomer Galanti (Massachusetts Institute of Technology), Tomaso Poggio (Massachusetts Institute of Technology)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper derives a norm-based generalization upper bound for sparse ReLU neural networks (including convolutional networks), focusing on the norm of convolution kernels rather than the norm of convolution layer matrices.
Norm-guided latent space exploration for text-to-image generation
Dvir Samuel (Bar-Ilan University), Gal Chechik (Bar-Ilan University)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A seed space interpolation and centroid calculation method based on a normal distribution χ distribution prior is proposed, which generates rare concept images and performs data augmentation in text-to-image diffusion models using this method.
Normalization Layers Are All That Sharpness-Aware Minimization Needs
Maximilian Mueller, Matthias Hein (University of Tübingen)
OptimizationTransformerImage
🎯 What it does: This paper explores the application of Sharpness-Aware Minimization (SAM) by perturbing only the parameters of normalization layers (SAM-ON) to enhance model generalization performance.
Normalization-Equivariant Neural Networks with Application to Image Denoising
Sébastien Herbreteau (Inria), Charles Kervrann (Inria)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper modifies existing neural networks by introducing affine convolution (where the sum of the weights of each convolution kernel equals 1) and channel-level sorted pooling (taking pairwise minimum/maximum) to maintain equivariance in both scale and shift, and applies it to the image denoising task.
Normalizing flow neural networks by JKO scheme
Chen Xu (Georgia Tech), Yao Xie (Georgia Tech)
GenerationData SynthesisOptimizationComputational EfficiencyGraph Neural NetworkFlow-based ModelAuto EncoderImageTabularOrdinary Differential Equation
🎯 What it does: A neural ODE flow network JKO-iFlow based on the JKO (Jordan–Kinderlehrer–Otto) scheme is proposed, which achieves a reversible mapping of data distribution to a standard normal distribution by discretizing the continuous Wasserstein gradient flow into a series of reversible residual blocks, and employs block-level training for efficient utilization of memory and computational resources.
Not All Neuro-Symbolic Concepts Are Created Equal: Analysis and Mitigation of Reasoning Shortcuts
Emanuele Marconato (University of Trento), Andrea Passerini (University of Trento)
Autonomous DrivingTabular
🎯 What it does: This paper analyzes the reasoning shortcuts present in neural-symbolic models and proposes various mitigation strategies.
Not All Out-of-Distribution Data Are Harmful to Open-Set Active Learning
Yang Yang (Nanjing University of Science and Technology), Yi Xu (Dalian University of Technology)
ClassificationObject DetectionData-Centric LearningMeta LearningImage
🎯 What it does: A Progressive Active Learning (PAL) method is proposed, which actively selects valuable pseudo-ID and pseudo-OOD instances to simultaneously improve the performance of ID classifiers and OOD detectors.
NPCL: Neural Processes for Uncertainty-Aware Continual Learning
Saurav Jha (University of New South Wales Sydney), Lina Yao (CSIRO Data61)
TransformerImage
🎯 What it does: Proposes Neural Processes for Continual Learning (NPCL), which maps continual learning tasks to a hierarchical latent variable model and combines experience replay to achieve knowledge sharing and uncertainty estimation between tasks.
NU-MCC: Multiview Compressive Coding with Neighborhood Decoder and Repulsive UDF
Stefan Lionar (Sea AI Lab), Gim Hee Lee (National University of Singapore)
RestorationCompressionTransformerPoint Cloud
🎯 What it does: This study investigates how to achieve high-quality, texture-rich 3D reconstruction models using single-view RGB-D input.
NuTrea: Neural Tree Search for Context-guided Multi-hop KGQA
Hyeong Kyu Choi (University of Wisconsin Madison), Hyunwoo J. Kim (Korea University)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a tree search-based graph neural network, NuTrea, for multi-hop knowledge graph question answering, which can simultaneously consider forward information and backward subtree context during the search path, addressing the limitations of traditional GNNs that only focus on past information.
NVFi: Neural Velocity Fields for 3D Physics Learning from Dynamic Videos
Jinxi Li (Hong Kong Polytechnic University), Bo Yang (Hong Kong Polytechnic University)
SegmentationGenerationData SynthesisNeural Radiance FieldVideoPhysics RelatedOrdinary Differential Equation
🎯 What it does: The NVFi framework is proposed, which utilizes multi-view videos to simultaneously learn the geometry, appearance, and decoupled velocity fields of 3D scenes, achieving future frame extrapolation, unsupervised semantic decomposition, and motion transfer.
OBJECT 3DIT: Language-guided 3D-aware Image Editing
Oscar Michel (Allen Institute for Artificial Intelligence), Tanmay Gupta (Allen Institute for Artificial Intelligence)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a language prompt-based 3D friendly image editing model, 3DIT, which can perform object translation, rotation, insertion, and deletion on a single image while maintaining consistency in geometry, lighting, and shadows.
Object-Centric Learning for Real-World Videos by Predicting Temporal Feature Similarities
Andrii Zadaianchuk (Max Planck Institute for Intelligent Systems), Georg Martius (Max Planck Institute for Intelligent Systems)
Object DetectionSegmentationTransformerContrastive LearningVideo
🎯 What it does: A completely unsupervised video object-centric learning framework called VideoSAUR is proposed, which can automatically discover and segment objects in diverse real-world videos.
Object-centric Learning with Cyclic Walks between Parts and Whole
Ziyu Wang (National University of Singapore), Mengmi Zhang (Agency for Science Technology and Research)
Object DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: Proposes an unsupervised object-centric learning method that conducts bidirectional cyclic visits (W-P-W and P-W-P) between features extracted by a visual Transformer and object representations generated by Slot-Attention; directly supervises the learning of slots through cyclic consistency loss, eliminating the need for decoding reconstruction.
Object-Centric Slot Diffusion
Jindong Jiang (Rutgers University), Sungjin Ahn (KAIST)
SegmentationGenerationTransformerDiffusion modelAuto EncoderImage
🎯 What it does: Latent Slot Diffusion (LSD) is proposed in object-centric learning, combining diffusion models with slot attention to achieve unsupervised segmentation, attribute prediction, combinatorial generation, and editing of complex scenes.
ODE-based Recurrent Model-free Reinforcement Learning for POMDPs
Xuanle Zhao (Institute of Automation Chinese Academy of Sciences), Bo XU
Recurrent Neural NetworkReinforcement LearningSequentialOrdinary Differential Equation
🎯 What it does: This paper proposes a framework that combines a recursive model based on ODE (GRU-ODE) with model-free reinforcement learning to address continuous control problems under partially observable Markov decision processes (POMDP).
Off-Policy Evaluation for Human Feedback
Qitong Gao (Duke University), Miroslav Pajic (Duke University)
Reinforcement Learning from Human FeedbackAuto EncoderSequential
🎯 What it does: This paper proposes an Off-Policy Evaluation for Human Feedback (OPEHF) framework that can reconstruct immediate human rewards from offline trajectories containing only final human feedback, and utilize existing OPE methods to evaluate the total human return of the target policy.
Offline Imitation Learning with Variational Counterfactual Reasoning
Zexu Sun (Renmin University of China), Shuai Zhang (Tianjin University)
Robotic IntelligenceReinforcement LearningAuto EncoderGenerative Adversarial NetworkTabular
🎯 What it does: This paper proposes an offline imitation learning framework OILCA, which uses identifiable variational autoencoders for counterfactual reasoning of exogenous variables, automatically generating high-quality expert data for enhancement, thereby improving the performance and generalization ability of the policy without the need for online interaction.
Offline Minimax Soft-Q-learning Under Realizability and Partial Coverage
Masatoshi Uehara (Genentech), Wen Sun (Cornell University)
OptimizationReinforcement Learning
🎯 What it does: Two value-based offline reinforcement learning algorithms, MSQP and MQP, are proposed, which can provide PAC performance guarantees under the conditions of single-policy partial coverage and function approximation feasibility.
Offline Multi-Agent Reinforcement Learning with Implicit Global-to-Local Value Regularization
Xiangsen Wang (Beijing Jiaotong University), Xianyuan Zhan (Tsinghua University)
Reinforcement Learning
🎯 What it does: The OMIGA algorithm is proposed, which utilizes implicit global-to-local value regularization to address the offline multi-agent reinforcement learning problem.
Offline Reinforcement Learning for Mixture-of-Expert Dialogue Management
Dhawal Gupta (University of Massachusetts), Craig Boutilier (Google Research)
Large Language ModelReinforcement LearningMixture of ExpertsText
🎯 What it does: A framework for offline reinforcement learning dialogue management based on the Mixture-of-Expert (MoE) language model is proposed, utilizing a pre-trained expert LM to generate candidate responses and performing high-level planning based on this.
Offline Reinforcement Learning with Differential Privacy
Dan Qiao (University of California Santa Barbara), Yu-Xiang Wang (University of California Santa Barbara)
Safty and PrivacyReinforcement LearningTabular
🎯 What it does: Two offline reinforcement learning algorithms (DP-APVI and DP-VAPVI) are designed to achieve approximately optimal policies while maintaining differential privacy, proving that their instance-dependent errors only increase by lower-order terms.