NeurIPS 2023 Papers — Page 2
Conference on Neural Information Processing Systems · 3218 papers
A Trichotomy for Transductive Online Learning
Steve Hanneke (Purdue University), Jonathan Shafer (UC Berkeley)
🎯 What it does: Under the framework of 'pass-through online learning', the author presents a trichotomy of misclassification counts: depending on the VC dimension and Littlestone dimension, the misclassification count is either n, Θ(log n), or Θ(1).
A U-turn on Double Descent: Rethinking Parameter Counting in Statistical Learning
Alicia Curth (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
Image
🎯 What it does: Investigate and explain the double descent phenomenon observed in non-deep learning methods (trees, boosted trees, linear regression), proving that it does not contradict the traditional U-shaped complexity-generalization curve, but is instead caused by implicit multiple complexity axes and effective parameter counting;
A Unified Algorithm Framework for Unsupervised Discovery of Skills based on Determinantal Point Process
Jiayu Chen (Purdue University), Tian Lan (George Washington University)
OptimizationReinforcement Learning
🎯 What it does: The ODPP framework is proposed for unsupervised option discovery, utilizing DPP to jointly optimize diversity and coverage of skills.
A Unified Approach for Maximizing Continuous DR-submodular Functions
Mohammad Pedramfar (Purdue University), Vaneet Aggarwal (Purdue University)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes a unified Frank-Wolfe foundational algorithm framework for maximizing continuous DR-submodular functions under various constraints, monotonicity, and different oracle access types, covering 16 offline and online settings.
A Unified Approach to Count-Based Weakly Supervised Learning
Vinay Shukla (University of California), Guy Van den Broeck (University of California)
ClassificationOptimizationTabular
🎯 What it does: A unified counting constraint method (Count Loss) is proposed, which constrains weakly supervised labels (such as label proportions, multiple instance learning, and positive-negative unlabeled learning) by calculating the counting distribution of model-predicted labels, directly optimizing probabilities rather than approximations.
A Unified Approach to Domain Incremental Learning with Memory: Theory and Algorithm
Haizhou Shi (Rutgers University), Hao Wang (Rutgers University)
Domain AdaptationKnowledge DistillationImage
🎯 What it does: A unified framework called UDIL is proposed to address the problem of domain incremental learning with limited memory, theoretically unifying existing methods under the same error upper bound and achieving the tightest error bounds through adaptive coefficients.
A Unified Conditional Framework for Diffusion-based Image Restoration
Yi Zhang (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
RestorationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A unified conditional diffusion framework is proposed for image restoration, which includes a lightweight UNet to predict initial guidance and learns residuals through a diffusion model; spatial guidance and scalar conditions are integrated in each diffusion block to achieve adaptive generation; a step-wise patch-splitting strategy is also proposed to handle high-resolution images, avoiding grid artifacts.
A Unified Detection Framework for Inference-Stage Backdoor Defenses
Xun Xian (University of Minnesota), Jie Ding (University of Minnesota)
Anomaly DetectionImageText
🎯 What it does: A unified framework for backdoor detection in the inference phase is proposed, along with a provable upper bound on the false positive rate.
A Unified Discretization Framework for Differential Equation Approach with Lyapunov Arguments for Convex Optimization
Kansei Ushiyama (University of Tokyo), Takayasu Matsuo (University of Tokyo)
OptimizationOrdinary Differential Equation
🎯 What it does: A unified discretization framework is proposed, which maps continuous differential equation (DE) methods to discrete optimization methods using weak discrete gradients (wDG), along with corresponding convergence rate analysis.
A Unified Fast Gradient Clipping Framework for DP-SGD
Weiwei Kong (Google Research), Andres Munoz medina
OptimizationComputational EfficiencyTransformer
🎯 What it does: A unified fast gradient clipping framework is proposed, allowing DP-SGD to efficiently compute gradient norms in neural networks that include arbitrary (even nonlinear) intermediate operations.
A unified framework for information-theoretic generalization bounds
Yifeng Chu (University of Illinois), Maxim Raginsky (University of Illinois)
🎯 What it does: A unified framework is proposed, utilizing decorrelation techniques and chaining methods in measure spaces to derive upper bounds on information-theoretic generalization errors.
A Unified Framework for Rank-based Loss Minimization
Rufeng Xiao (Fudan University), Yifan Yan
OptimizationTabular
🎯 What it does: This paper proposes a unified ADMM framework for minimizing rank-based loss, capable of handling both convex and non-convex losses as well as weakly convex regularization;
A Unified Framework for U-Net Design and Analysis
Christopher Williams (University of Oxford), Saifuddin Syed (University of Oxford)
SegmentationGenerationConvolutional Neural NetworkDiffusion modelImageBenchmark
🎯 What it does: A unified theoretical framework for U-Net is proposed, and based on this framework, various improved U-Net architectures (such as Multi-ResNet, U-Net based on boundary conditions and geometric structures) are designed and applied to tasks such as image segmentation, PDE surrogate modeling, and diffusion models.
A Unified Framework for Uniform Signal Recovery in Nonlinear Generative Compressed Sensing
Junren Chen (University of Hong Kong), Zhaoqiang Liu (University of Electronic Science and Technology of China)
GenerationCompressionImage
🎯 What it does: A unified framework is proposed for achieving unified recovery of all signals that satisfy generative model constraints in nonlinear generative compressive sensing (GCS).
A Unified Generalization Analysis of Re-Weighting and Logit-Adjustment for Imbalanced Learning
Zitai Wang (Institute of Information Engineering, Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
ClassificationImage
🎯 What it does: A unified analysis of the generalization performance of reweighting and logit adjustment in imbalanced learning is presented, along with a data-dependent shrinkage technique based on local Lipschitz and the corresponding training algorithm.
A Unified Model and Dimension for Interactive Estimation
Nataly Brukhim (Princeton University), Robert E. Schapire
🎯 What it does: This paper studies a general interactive learning framework—interactive estimation—and introduces a new complexity measure called dissimilarity dimension. It provides a general algorithm based on least squares and proves polynomial upper bounds for sublinear regret and PAC generalization.
A Unified Solution for Privacy and Communication Efficiency in Vertical Federated Learning
Ganyu Wang (Western University), Charles Ling
OptimizationFederated LearningSafty and PrivacyImage
🎯 What it does: A hybrid optimization framework is proposed for vertical federated learning, using zero-order optimization for the output layer and first-order optimization for the remaining layers, combined with compression to achieve a unified solution for privacy protection and communication efficiency.
A Unified, Scalable Framework for Neural Population Decoding
Mehdi Azabou (Georgia Tech), Eva L Dyer (Georgia Tech)
TransformerTime Series
🎯 What it does: A unified and scalable Transformer framework (POYO) has been established for neural spike decoding across sessions, animals, and tasks, enabling rapid adaptation to new data through a small amount of labeling or single-unit mapping after large-scale pre-training.
A Unifying Perspective on Multi-Calibration: Game Dynamics for Multi-Objective Learning
Nika Haghtalab (University of California), Eric Zhao (University of California)
OptimizationReinforcement LearningTabular
🎯 What it does: A unified framework based on game dynamics is proposed for designing and analyzing multicalibration predictors, covering various scenarios such as multi-class, online, unbiased, and conditional cases.
A Variational Perspective on High-Resolution ODEs
Hoomaan Maskan (Umea University), Alp Yurtsever (Umea University)
OptimizationConvolutional Neural NetworkImageTabularOrdinary Differential Equation
🎯 What it does: This paper proposes using a variational perspective of the forced Euler-Lagrange equations to construct high-resolution ODEs, achieving a faster convergence rate of gradient norms for NAG, and interprets NAG as a discretization that matches the rate; based on this, a stochastic algorithm for noisy gradients is provided along with convergence analysis.
A-NeSI: A Scalable Approximate Method for Probabilistic Neurosymbolic Inference
Emile van Krieken (University of Edinburgh Vrije Universiteit Amsterdam), Annette Ten Teije (Vrije Universiteit Amsterdam)
Explainability and InterpretabilityComputational EfficiencyImage
🎯 What it does: A scalable approximate reasoning framework named A-NESI is proposed, which combines neural networks and symbolic reasoning to achieve probabilistic neural-symbolic learning.
A*Net: A Scalable Path-based Reasoning Approach for Knowledge Graphs
Zhaocheng Zhu (University of Montréal), Jian Tang (Mila - Québec AI Institute)
Computational EfficiencyKnowledge DistillationGraph Neural NetworkGraph
🎯 What it does: This paper proposes A*Net, a scalable path-based knowledge graph reasoning method;
A3FL: Adversarially Adaptive Backdoor Attacks to Federated Learning
Hangfan Zhang (Pennsylvania State University), Dinghao Wu (Pennsylvania State University)
Federated LearningAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an A3FL backdoor attack for federated learning, utilizing adversarial adaptive triggers to maintain efficiency and persistence of the backdoor in the global training dynamics.
AbDiffuser: full-atom generation of in-vitro functioning antibodies
Karolis Martinkus (Genentech), Andreas Loukas (Genentech)
GenerationData SynthesisDrug DiscoveryProtein Structure PredictionGraph Neural NetworkTransformerDiffusion modelGraphBiomedical Data
🎯 What it does: We propose AbDiffuser, a diffusion model based on SE(3) equivariance that can simultaneously generate antibody sequences and complete 3D structures, with its performance validated through computer simulations and experiments.
Abide by the law and follow the flow: conservation laws for gradient flows
Sibylle Marcotte (École normale supérieure Paris Sciences et Lettres University), Gabriel Peyré (Centre National de la Recherche Scientifique)
OptimizationRecurrent Neural NetworkOrdinary Differential Equation
🎯 What it does: This paper provides a formal definition of conservation laws in gradient flows and presents an algorithm to determine the maximum set of independent functions conserved by the gradient flow of any model (especially linear and ReLU networks) under arbitrary data, using reparameterization and Lie-algebra methods.
Accelerated On-Device Forward Neural Network Training with Module-Wise Descending Asynchronism
Xiaohan Zhao (Nanjing University of Information Science and Technology), Bin Gu (Mohamed bin Zayed University of Artificial Intelligence)
OptimizationComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes an Asynchronous Forward Gradient Descent (AsyncFGD) method that utilizes modular sparse parameters to break the hierarchical locking of traditional Forward Gradient Descent (FGD), enabling parallel computation, thereby significantly reducing memory usage and improving hardware utilization on edge devices.
Accelerated Quasi-Newton Proximal Extragradient: Faster Rate for Smooth Convex Optimization
Ruichen Jiang (University of Texas at Austin), Aryan Mokhtari (University of Texas at Austin)
OptimizationTabular
🎯 What it does: An accelerated quasi-Newton proximal gradient method (A-QNPE) is proposed to solve unconstrained smooth convex optimization problems.
Accelerated Training via Incrementally Growing Neural Networks using Variance Transfer and Learning Rate Adaptation
Xin Yuan (University of Chicago), Michael Maire (University of Chicago)
Computational EfficiencyImageText
🎯 What it does: The paper proposes an efficient method for incremental growth neural networks, achieving training acceleration and performance improvement by retaining functionality during network growth, using variance transfer, and adaptive learning rates.
Accelerated Zeroth-order Method for Non-Smooth Stochastic Convex Optimization Problem with Infinite Variance
Nikita Kornilov (Moscow Institute of Physics and Technology), Samuel Horváth (Mohammed Bin Zayed University of Artificial Intelligence)
OptimizationTabular
🎯 What it does: A zero-order acceleration method for non-smooth stochastic convex optimization with infinite variance noise (ZO-clipped-SSTM and R-ZO-clipped-SSTM) is proposed, along with high-probability convergence and optimal operator complexity.
Accelerating Exploration with Unlabeled Prior Data
Qiyang Li (University of California Berkeley), Sergey Levine (University of California Berkeley)
Reinforcement LearningImage
🎯 What it does: By utilizing unlabeled prior data, we estimate the upper confidence bound (UCB) of rewards through a reward model and Random Network Distillation (RND), optimistically re-label the prior data with rewards, and input this along with online collected data into an offline-online RL algorithm (such as RLPD) for training, significantly accelerating exploration and learning in sparse reward tasks.
Accelerating Molecular Graph Neural Networks via Knowledge Distillation
Filip Ekström Kelvinius (Linköping University), Johannes Gasteiger (Google Research)
Computational EfficiencyKnowledge DistillationDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: This paper enhances the inference speed and prediction accuracy of Molecular Graph Neural Networks (Molecular GNN) through the Knowledge Distillation (KD) method, without altering the student model structure, thus maintaining the high throughput of the lightweight model.
Accelerating Monte Carlo Tree Search with Probability Tree State Abstraction
Yangqing Fu (Shanghai Jiao Tong University), Yue Gao (Shanghai Jiao Tong University)
Reinforcement Learning
🎯 What it does: A Probability Tree State Abstraction (PTSA) algorithm is proposed to accelerate the search process by aggregating similar paths in the MCTS search tree.
Accelerating Motion Planning via Optimal Transport
An Thai Le, Jan Peters (Technische Universitat Darmstadt)
OptimizationRobotic Intelligence
🎯 What it does: A zero-order batch trajectory optimization method MPOT based on Sinkhorn steps is proposed to generate smooth and executable paths for high-dimensional, non-convex motion planning tasks.
Accelerating Reinforcement Learning with Value-Conditional State Entropy Exploration
Dongyoung Kim (KAIST), Younggyo Seo (KAIST)
Reinforcement LearningMultimodalityBenchmark
🎯 What it does: A value-conditioned state entropy (VCSE) exploration method is proposed, which accelerates the learning process in reinforcement learning by maximizing the state entropy within different value partitions.
Accelerating Value Iteration with Anchoring
Jongmin Lee (Seoul National University), Ernest K. Ryu (Seoul National University)
OptimizationReinforcement Learning
🎯 What it does: An accelerated value iteration based on an anchoring mechanism (Anc-VI) is proposed, achieving faster convergence speed on Bellman error;
Accessing Higher Dimensions for Unsupervised Word Translation
Sida Wang
Text
🎯 What it does: A new unsupervised word translation method called coocmap is proposed, which utilizes high-dimensional signals instead of low-dimensional word vectors for translation.
Accountability in Offline Reinforcement Learning: Explaining Decisions with a Corpus of Examples
Hao Sun (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
Explainability and InterpretabilityReinforcement LearningTabularBiomedical Data
🎯 What it does: This paper proposes an Accountable Offline Controller (AOC), which utilizes an offline dataset as a decision corpus and employs a subset of representative examples for traceable offline control decisions.
Accurate Interpolation for Scattered Data through Hierarchical Residual Refinement
Shizhe Ding (Chinese Academy of Sciences), Dongbo Bu (Chinese Academy of Sciences)
TransformerPoint CloudBenchmark
🎯 What it does: A hierarchical residual refinement framework HINT is proposed, which recursively estimates the main function and residuals through multiple lightweight interpolation blocks to achieve more accurate scattered interpolation results.
Achieving $\mathcal{O}(\epsilon^{-1.5})$ Complexity in Hessian/Jacobian-free Stochastic Bilevel Optimization
Yifan Yang (University at Buffalo), Kaiyi Ji (University at Buffalo)
OptimizationHyperparameter SearchConvolutional Neural NetworkImage
🎯 What it does: A fully single-loop, Hessian/Jacobian-free bi-level optimization algorithm FdeHBO is proposed, which derives FMBO to achieve a sample complexity of O(ε⁻⁵⁄²).
Achieving Cross Modal Generalization with Multimodal Unified Representation
Yan Xia (Zhejiang University), Zhou Zhao (Zhejiang University)
ClassificationSegmentationRetrievalAuto EncoderContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Proposes the Cross Modal Generalization (CMG) task and constructs the Uni-Code framework to learn a unified discrete representation across modalities, achieving zero-shot generalization to other modalities with only one modality labeled.
Act As You Wish: Fine-Grained Control of Motion Diffusion Model with Hierarchical Semantic Graphs
Peng Jin (Peking University), Li Yuan (Peking University)
GenerationData SynthesisPose EstimationGraph Neural NetworkTransformerDiffusion modelTextMultimodality
🎯 What it does: A text-driven human motion generation model called GraphMotion is proposed, which is based on a hierarchical semantic graph. By breaking down action descriptions into three layers of nodes: movement, action, and details, it achieves fine-grained control over motion details.
Action Inference by Maximising Evidence: Zero-Shot Imitation from Observation with World Models
Xingyuan Zhang (Volkswagen Group), Maximilian Karl (Volkswagen Group)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningWorld ModelVideo
🎯 What it does: This paper proposes an algorithm named AIME, which first learns the dynamics of the agent using a variational world model, and then treats the policy as an inference model, achieving zero-shot imitation learning from observations by maximizing evidence (ELBO).
Active Bipartite Ranking
James Cheshire (Telecom ParisTech), Stephan Clémençon (ENS Paris Saclay)
OptimizationTabular
🎯 What it does: An active learning framework for bipartite ranking problems is proposed, along with the active-ranking algorithm, which can approximate the optimal ROC curve within a limited number of queries.
Active Learning for Semantic Segmentation with Multi-class Label Query
Sehyun Hwang (Pohang University of Science and Technology), Suha Kwak (Pohang University of Science and Technology)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an active learning framework for semantic segmentation, with the core idea being multi-class label querying: within a selected local area, all occurring categories are manually annotated, and then a two-stage training strategy is employed to address label uncertainty.
Active Learning-Based Species Range Estimation
Christian Lange (University of Edinburgh), Oisin Mac Aodha (University of Edinburgh)
ClassificationOptimizationTabularAgriculture Related
🎯 What it does: This paper proposes an active learning method that efficiently estimates the distribution range of unknown species using fewer field observation points by selecting the most uncertain geographical locations for sampling after online updates. This is achieved through weighted averaging in the trained multi-species range model space and using spatial implicit features obtained from transfer learning.
Active Negative Loss Functions for Learning with Noisy Labels
Xichen Ye (Shanghai University), Weiqin Tong (Shanghai University)
ClassificationOptimizationImage
🎯 What it does: A new noise-robust learning framework based on 'negative' loss (ANL) is proposed, and various new noise-robust loss functions are constructed based on it.
Active Observing in Continuous-time Control
Samuel Holt (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
OptimizationRobotic IntelligenceReinforcement LearningTime SeriesBiomedical Data
🎯 What it does: The study actively decides when to conduct expensive observations in continuous-time control tasks and proposes an active observation control method based on uncertainty thresholds.
Active Reasoning in an Open-World Environment
Manjie Xu (Beijing Institute of Technology), Yixin Zhu (Peking University)
Robotic IntelligenceTransformerReinforcement LearningVision Language ModelMultimodality
🎯 What it does: Designed and implemented an interactive open-world environment 🔍 Conan to evaluate the active reasoning capabilities of the model (active exploration + multi-round causal reasoning) and proposed the AfD method to transform causation into deduction.
Active representation learning for general task space with applications in robotics
Yifang Chen (University of Washington), Guanya Shi (Carnegie Mellon University)
Representation LearningRobotic IntelligenceTabularTime Series
🎯 What it does: This paper proposes a general active representation learning framework that can adaptively select the most informative source tasks in any discrete or continuous source task space to enhance the few-shot learning performance of the target task.
Active Vision Reinforcement Learning under Limited Visual Observability
Jinghuan Shang (Stony Brook University), Michael S Ryoo
Robotic IntelligenceReinforcement LearningMultimodality
🎯 What it does: This research focuses on active visual reinforcement learning, allowing agents to simultaneously learn action policies and visual perception strategies to complete tasks in partially observable environments.
Actively Testing Your Model While It Learns: Realizing Label-Efficient Learning in Practice
Dayou Yu (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
ClassificationData-Centric LearningConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: An ATL (Active Learning-Testing) framework is proposed, which performs active evaluation and active feedback in real-time during the active learning process to achieve more efficient learning and evaluation of labels.
Activity Grammars for Temporal Action Segmentation
Dayoung Gong (Pohang University of Science and Technology), Minsu Cho (Pohang University of Science and Technology)
SegmentationOptimizationTransformerVideoTime SeriesSequentialBenchmark
🎯 What it does: This paper proposes a temporal action segmentation framework based on active grammar, which uses the KARI algorithm to automatically induce probabilistic context-free grammar with recursive rules from training sequences. It converts frame-level predictions from neural networks into grammatically compliant action sequences using a BEP (Breadth-First Earley) parser, and subsequently optimizes action lengths through dynamic programming.
AD-PT: Autonomous Driving Pre-Training with Large-scale Point Cloud Dataset
Jiakang Yuan (Fudan University), Yu Qiao (Shanghai Artificial Intelligence Laboratory)
Object DetectionAutonomous DrivingContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a self-driving pre-training framework based on large-scale point cloud data, AD-PT, which can generate unified feature representations and transfer to various downstream detection tasks.
AdANNS: A Framework for Adaptive Semantic Search
Aniket Rege (University of Washington), Ali Farhadi (University of Washington)
RetrievalComputational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImageText
🎯 What it does: A framework named AdANNS is proposed, which utilizes adaptive multi-dimensional embeddings represented by Matryoshka. It employs different dimensional representations at various stages of approximate nearest neighbor search (ANNS), such as clustering, linear search, and quantization, significantly reducing computational costs while maintaining accuracy.
AdaPlanner: Adaptive Planning from Feedback with Language Models
Haotian Sun (Georgia Institute of Technology), Chao Zhang (Georgia Institute of Technology)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes AdaPlanner, a framework that adaptively generates and refines action plans using large language models (LLMs) under closed-loop environmental feedback;
Adapting Fairness Interventions to Missing Values
Raymond Feng (Harvard University), Hao Wang (MIT IBM Watson AI Lab)
ClassificationOptimizationTabular
🎯 What it does: This paper proposes an adaptive algorithm that can be combined with existing fairness intervention algorithms to train fair and accurate classifiers in the presence of missing values. Theoretical and empirical evidence shows that retaining missing pattern information can significantly improve the fairness-accuracy balance.
Adapting Neural Link Predictors for Data-Efficient Complex Query Answering
Erik Arakelyan (University of Copenhagen), Isabelle Augenstein (University of Copenhagen)
Knowledge DistillationData-Centric LearningGraph Neural NetworkSupervised Fine-TuningGraph
🎯 What it does: Predicting answers to complex logical queries on knowledge graphs and achieving more accurate reasoning through score calibration of a pre-trained neural link predictor.
Adapting to Continuous Covariate Shift via Online Density Ratio Estimation
Yu-Jie Zhang (University of Tokyo), Masashi Sugiyama (RIKEN AIP)
Domain AdaptationMeta LearningTabularTime Series
🎯 What it does: For the scenario of continuous covariate drift, an online density ratio estimation framework is proposed, which combines importance-weighted learning to achieve an adaptive predictor.
Adaptive Algorithms for Relaxed Pareto Set Identification
Cyrille Kone (Univ. Lille), Laura Richert (Univ. Bordeaux)
OptimizationBiomedical Data
🎯 What it does: A self-adaptive multi-objective multi-armed bandit sampling strategy is proposed to address the fixed confidence problem in Pareto set identification and supports different relaxation conditions.
Adaptive Contextual Perception: How To Generalize To New Backgrounds and Ambiguous Objects
Zhuofan Ying (Columbia University), Mohit Bansal (University of North Carolina at Chapel Hill)
Object DetectionDomain AdaptationImage
🎯 What it does: This paper studies how visual models adaptively utilize background information to generalize to new backgrounds and ambiguous objects, and proposes two types of OOD settings (BACKGROUND-INVARIANCE and OBJECT-DISAMBIGUATION), analyzing the performance differences of the model in these two contexts.
Adaptive Data Analysis in a Balanced Adversarial Model
Kobbi Nissim (Georgetown University), Eliad Tsfadia (Georgetown University)
OptimizationSafty and Privacy
🎯 What it does: An adaptive data analysis lower bound under the 'balanced adversary' model is proposed, and an efficient balanced adversary is constructed using public key cryptography;
Adaptive Linear Estimating Equations
Mufang Ying (Rutgers University), Cun-Hui Zhang (Rutgers University)
Time Series
🎯 What it does: A method based on Adaptive Linear Estimation Equation (ALEE) is proposed for parameter estimation in adaptive linear models, addressing the non-normal asymptotic behavior of ordinary least squares (OLS) estimators in adaptive data collection.
Adaptive Normalization for Non-stationary Time Series Forecasting: A Temporal Slice Perspective
Zhiding Liu (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
TransformerTime Series
🎯 What it does: An adaptive normalization framework based on time slicing, SAN, is proposed as a model-agnostic plugin to normalize/denormalize various time series forecasting models, significantly improving prediction accuracy.
Adaptive Online Replanning with Diffusion Models
Siyuan Zhou (Hong Kong University of Science and Technology), Chuang Gan (UMass Amherst)
Robotic IntelligenceReinforcement LearningDiffusion modelTabularBenchmark
🎯 What it does: An adaptive online replanning method RDM based on diffusion models is proposed, which automatically decides whether to replan during execution based on the trajectory likelihood estimated by the model, and combines two strategies during replanning: replanning from scratch and local correction based on future context.
Adaptive Principal Component Regression with Applications to Panel Data
Anish Agarwal (Columbia University), Steven Wu
TabularTime Series
🎯 What it does: This paper proposes an adaptive principal component regression (PCR) method and provides the first time-uniform finite sample guarantees for online (regularized) PCR, particularly in cases where data is collected adaptively.
Adaptive Privacy Composition for Accuracy-first Mechanisms
Ryan Rogers, Aaditya Ramdas (Carnegie Mellon University)
Safty and PrivacyTextStochastic Differential Equation
🎯 What it does: A privacy filter is proposed that can adaptively combine ex-post privacy mechanisms (such as the Brownian noise reduction mechanism) with zCDP mechanisms, thereby achieving more flexible interactive privacy analysis while maintaining overall (ε,δ)-DP.
Adaptive recurrent vision performs zero-shot computation scaling to unseen difficulty levels
Vijay Veerabadran (University of California), Virginia R. de Sa (University of California)
RecognitionSegmentationComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningImage
🎯 What it does: This paper studies the combination of the Adaptive Computation Time (ACT) mechanism with Convolutional Recurrent Networks (ConvRNN), constructing an Adaptive Recurrent Network (AdRNN) that can dynamically allocate computation steps based on input difficulty in static visual tasks, and verifies its zero-shot generalization capability under unknown higher difficulties.
Adaptive Selective Sampling for Online Prediction with Experts
Rui M. Castro (Eindhoven University of Technology), Tim van Erven (University of Amsterdam)
Reinforcement LearningSequential
🎯 What it does: This paper proposes an adaptive selective sampling algorithm for online binary sequence prediction under expert advice, aiming to reduce the number of labels collected while maintaining good predictive performance.
Adaptive SGD with Polyak stepsize and Line-search: Robust Convergence and Variance Reduction
Xiaowen Jiang (CISPA Helmholtz Center for Information Security), Sebastian U Stich
OptimizationConvolutional Neural NetworkImageTabular
🎯 What it does: A robust adaptive stochastic gradient descent algorithm, AdaSPS and AdaSLS, is proposed, along with convergence proofs for both interpolation and non-interpolation, as well as strong convexity and convexity cases; simultaneously, AdaSVRPS and AdaSVRLS are designed, incorporating variance reduction to eliminate dependence on interpolation.
Adaptive Test-Time Personalization for Federated Learning
Wenxuan Bao (University of Illinois Urbana-Champaign), Jingrui He (University of Illinois Urbana-Champaign)
Domain AdaptationFederated LearningImage
🎯 What it does: This paper proposes an algorithm called ATP (Adaptive Test-Time Personalization) aimed at addressing the model personalization problem for unlabeled test clients in a federated learning environment.
Adaptive Topological Feature via Persistent Homology: Filtration Learning for Point Clouds
Naoki Nishikawa (University of Tokyo), Kenji Yamanishi (University of Tokyo)
ClassificationProtein Structure PredictionPoint Cloud
🎯 What it does: A learnable weighted filter network architecture is proposed, which adaptively generates weights for point clouds through neural networks, calculates persistent homology, and vectorizes it for classification.
Adaptive Uncertainty Estimation via High-Dimensional Testing on Latent Representations
Tsai Hor Chan (University of Hong Kong), Lequan Yu (University of Hong Kong)
Object DetectionAnomaly DetectionConvolutional Neural NetworkContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A framework for posterior uncertainty estimation based on high-dimensional hypothesis testing is proposed, utilizing a Bayesian neural network encoder to extract latent representations, and then performing OOD detection on test samples using an adjustable regularized Hotelling T² (ARHT) statistic.
Adaptive whitening with fast gain modulation and slow synaptic plasticity
Lyndon Duong, David Lipshutz (Flatiron Institute)
Image
🎯 What it does: Design and implement a multi-timescale adaptive whitening neural network model that combines synaptic plasticity and rapid gain modulation to achieve fast whitening for different contexts.
AdaptSSR: Pre-training User Model with Augmentation-Adaptive Self-Supervised Ranking
Yang Yu (University of Science and Technology of China), Sanshi Lei Yu
Recommendation SystemTransformerContrastive LearningSequential
🎯 What it does: A self-supervised ranking-based pre-training task (AdaptSSR) is proposed to learn representations of user behavior sequences, reducing reliance on semantically consistent data augmentation.
AdaVAE: Bayesian Structural Adaptation for Variational Autoencoders
Paribesh Regmi (Rochester Institute of Technology), Rui Li (Rochester Institute of Technology)
GenerationData SynthesisAuto EncoderImageGraph
🎯 What it does: This paper proposes AdaVAE, which uses Bayesian non-parametric methods (beta process + Bernoulli process) to adaptively infer the depth and width of the VAE encoding/decoding network during training, and achieves joint inference of structure and latent variables through a scalable MIWAE estimator.
Add and Thin: Diffusion for Temporal Point Processes
David Lüdke (Technical University of Munich), Stephan Günnemann (Technical University of Munich)
GenerationData SynthesisOptimizationConvolutional Neural NetworkRecurrent Neural NetworkDiffusion modelTime SeriesSequential
🎯 What it does: A diffusion model named ADD-THIN is proposed for temporal point processes, capable of parallel generation of complete event sequences and achieving non-autoregressive long-term predictions.
Additive Decoders for Latent Variables Identification and Cartesian-Product Extrapolation
Sebastien Lachapelle (Mila), Simon Lacoste-Julien (Mila)
GenerationData SynthesisAuto EncoderImage
🎯 What it does: This paper proposes an additive decoder and demonstrates its ability to achieve identifiability of latent variables under unsupervised conditions, as well as to generate reasonable images beyond the training set through Cartesian-product extrapolation.
Addressing Negative Transfer in Diffusion Models
Hyojun Go (Twelvelabs), Seungtaek Choi (Yanolja)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper studies the problem of negative transfer in the training of diffusion models and mitigates negative transfer through task clustering combined with multi-task learning methods, thereby improving generation quality.
Addressing the speed-accuracy simulation trade-off for adaptive spiking neurons
Luke Taylor (University of Oxford), Nicol Spencer Harper
OptimizationComputational EfficiencySpiking Neural NetworkTime Series
🎯 What it does: A block-based algorithm reconstruction is proposed, utilizing the Absolute Refractory Period (ARP) of neurons to convert the stepwise simulation of the ALIF model into O(1) parallel computation per block, significantly accelerating simulation and training on GPUs.
Adjustable Robust Reinforcement Learning for Online 3D Bin Packing
Yuxin Pan (Hong Kong University of Science and Technology), Fangzhen Lin (Hong Kong University of Science and Technology)
OptimizationTransformerReinforcement LearningTabular
🎯 What it does: This paper proposes an Adjustable Robust Reinforcement Learning (AR2L) framework for the online 3D Bin Packing Problem (3D-BPP), which achieves more robust and performance-acceptable packing strategies by balancing expected and worst-case returns.
Advancing Bayesian Optimization via Learning Correlated Latent Space
Seunghun Lee (Korea University), Hyunwoo J. Kim (Korea University)
OptimizationDrug DiscoveryAuto EncoderTabular
🎯 What it does: This paper proposes a new Bayesian optimization method called CoBO, specifically designed for optimization tasks involving structured or discrete inputs in the latent space.
Adversarial Attacks on Online Learning to Rank with Click Feedback
Jinhang Zuo (University of Massachusetts Amherst), Adam Wierman (California Institute of Technology)
Recommendation SystemAdversarial AttackTabular
🎯 What it does: This paper proposes an adversarial attack scheme for UCB-based algorithms in online learning to rank (OLTR) that can induce the learner to frequently select the target item at sublinear cost.
Adversarial Counterfactual Environment Model Learning
Xiong-Hui Chen (Nanjing University), Huang Fangsheng (Meituan)
Adversarial AttackReinforcement LearningGenerative Adversarial NetworkTabular
🎯 What it does: Proposed the Adversarial Weighted Risk Minimization (AWRM) framework and implemented the GALILEO algorithm for offline environment model learning to enhance adversarial robustness and generalization ability.
Adversarial Examples Are Not Real Features
Ang Li (Peking University), Yisen Wang (Peking University)
ClassificationAdversarial AttackDiffusion modelContrastive LearningImage
🎯 What it does: This paper evaluates the true usability and robustness of robust and non-robust features within a broader learning paradigm (supervised learning, contrastive learning, masked image modeling, diffusion models).
Adversarial Examples Exist in Two-Layer ReLU Networks for Low Dimensional Linear Subspaces
Odelia Melamed (Weizmann Institute of Science), Gal Vardi (TTI-Chicago and the Hebrew University of Jerusalem)
OptimizationAdversarial AttackImage
🎯 What it does: This study investigates the vulnerability of two-layer ReLU networks on low-dimensional linear subspaces to orthogonal perturbations after gradient descent training, demonstrating that the gradient is large in the orthogonal direction and that universal perturbations exist.
Adversarial Examples Might be Avoidable: The Role of Data Concentration in Adversarial Robustness
Ambar Pal (Johns Hopkins University), Rene Vidal
Adversarial AttackData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: This study investigates the impact of data distribution concentration on the existence of adversarially robust classifiers and constructs classifiers that can obtain polygonal robust certificates based on low-dimensional linear subspace clustering.
Adversarial Learning for Feature Shift Detection and Correction
Míriam Barrabés (Stanford University), Alexander G Ioannidis
Anomaly DetectionData-Centric LearningGenerative Adversarial NetworkTabular
🎯 What it does: A framework called DataFix based on tree models and adversarial learning is proposed to locate and correct feature distribution drift in datasets.
Adversarial Model for Offline Reinforcement Learning
Mohak Bhardwaj (University of Washington), Ching-An Cheng (Microsoft Research)
Reinforcement LearningTabular
🎯 What it does: A model-based offline reinforcement learning framework called ARMOR is designed, utilizing an adversarial training MDP model to achieve relatively pessimistic policy improvement.
Adversarial Resilience in Sequential Prediction via Abstention
Surbhi Goel (University of Pennsylvania), Abhishek Shetty (University of California)
OptimizationAdversarial AttackSequential
🎯 What it does: This study investigates the sequence prediction problem with clean label adversarial injection, introducing a rejection mechanism to avoid errors on adversarial samples, and provides error and rejection upper bounds for VC classes.
Adversarial Robustness in Graph Neural Networks: A Hamiltonian Approach
Kai Zhao (Nanyang Technological University), Wee Peng Tay (Nanyang Technological University)
Adversarial AttackGraph Neural NetworkGraphOrdinary Differential Equation
🎯 What it does: This paper proposes and evaluates a class of Hamiltonian-based graph neural flows (HANG and HANG-quad) aimed at enhancing the robustness of graph neural networks against adversarial attacks.
Adversarial Robustness through Random Weight Sampling
Yanxiang Ma (University of Sydney), Chang Xu (University of Sydney)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A defense mechanism named Constrained Trainable Random Weights (CTRW) is proposed, which improves the adversarial robustness of deep neural networks by introducing random weights during the optimization process.
Adversarial Self-Training Improves Robustness and Generalization for Gradual Domain Adaptation
Lianghe Shi (Wuhan University), Weiwei Liu (Wuhan University)
Domain AdaptationAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes applying Adversarial Self-Training (AST) to Gradual Domain Adaptation (GDA) to enhance the adversarial robustness and clean accuracy of the target domain.
Adversarial Training for Graph Neural Networks: Pitfalls, Solutions, and New Directions
Lukas Gosch (Technical University of Munich), Stephan Günnemann (Technical University of Munich)
Adversarial AttackGraph Neural NetworkGraph
🎯 What it does: This paper points out and corrects the memory-based perfect robustness and evaluation bias caused by previous transductive learning settings in adversarial training of graph neural networks. It reintroduces adversarial training under a fully inductive setting and achieves robust diffusion through learnable diffusion models (GPRGNN/ChebNetII);
Adversarial Training from Mean Field Perspective
Soichiro Kumano (University of Tokyo), Toshihiko Yamasaki (University of Tokyo)
OptimizationAdversarial AttackImage
🎯 What it does: A full network probability framework based on mean field theory is proposed to theoretically analyze the learning dynamics, trainability, robustness, and capacity degradation of random deep networks during adversarial training.
Adversarially Robust Distributed Count Tracking via Partial Differential Privacy
Zhongzheng Xiong (Fudan University), Zengfeng Huang (Fudan University)
OptimizationSafty and PrivacyStochastic Differential Equation
🎯 What it does: This paper studies a distributed counting tracking model and proposes a new stochastic algorithm that achieves robustness in the presence of adaptive adversaries while maintaining near-optimal communication complexity.
Adversarially Robust Learning with Uncertain Perturbation Sets
Tosca Lechner (University of Waterloo), Ruth Urner (York University)
OptimizationAdversarial Attack
🎯 What it does: This paper proposes and studies an adversarial robust learning framework under uncertain perturbation sets, exploring how to achieve robust learning when the types of perturbations are unknown, given that the perturbation set belongs to a known class U.
Advice Querying under Budget Constraint for Online Algorithms
Ziyad Benomar (CREST, ENSAE, Ecole polytechnique), Vianney Perchet (CREST, ENSAE and Criteo AI LAB)
Optimization
🎯 What it does: In the framework of budget-constrained learning-enhanced online algorithms, an adaptive query strategy is proposed for three classic problems: ski rental, secretary, and preemptive task scheduling. This strategy achieves a balance between consistency and robustness by incorporating prediction accuracy information, along with corresponding competitive analysis and experimental validation.
Affinity-Aware Graph Networks
Ameya Velingker (Google Research), Sreenivas Gollapudi (Google Research)
Graph Neural NetworkGraph
🎯 What it does: This study implements the integration of affinity metrics based on random walks (such as effective resistance, hitting time, commuting time, etc.) as node/edge features into the Message Passing Neural Network (MPNN) to enhance the model's expressive power and predictive performance.
AGD: an Auto-switchable Optimizer using Stepwise Gradient Difference for Preconditioning Matrix
Yun Yue (Ant Group), Ke Zhang (Ant Group)
Recommendation SystemOptimizationImageText
🎯 What it does: A new adaptive optimizer AGD is proposed, which constructs a preprocessing matrix using adjacent gradient differences and achieves automatic switching between SGD and adaptive optimizers.
Aggregating Capacity in FL through Successive Layer Training for Computationally-Constrained Devices
Kilian Pfeiffer (Karlsruhe Institute of Technology), Joerg Henkel
Federated LearningComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: The Successive Layer Training (SLT) method is proposed to implement federated learning training on resource-constrained edge devices, gradually freezing the front layers and expanding the back head to significantly reduce memory requirements while maintaining parameter cooperation.