NeurIPS 2023 Papers — Page 9
Conference on Neural Information Processing Systems · 3218 papers
Directed Cyclic Graph for Causal Discovery from Multivariate Functional Data
Saptarshi Roy (Texas A&M University), Yang Ni (Texas A&M University)
Graph Neural NetworkMultimodalityTime Series
🎯 What it does: A linear structural equation model for causal structure learning of multivariate function data with loops is proposed, and dimensionality reduction is achieved through a low-dimensional causal embedding space.
Direction-oriented Multi-objective Learning: Simple and Provable Stochastic Algorithms
Peiyao Xiao (University at Buffalo), Kaiyi Ji (University at Buffalo)
OptimizationRobotic IntelligenceReinforcement LearningImage
🎯 What it does: A new direction-oriented multi-objective optimization (direction-oriented MOO) target is proposed, and based on this target, two stochastic gradient algorithms, SDMGrad and SDMGrad-OS, are designed to solve Pareto stable points in non-convex scenarios.
Directional diffusion models for graph representation learning
Run Yang (Shandong University of Finance and Economics), Qiang Sun (University of Toronto)
Representation LearningGraph Neural NetworkDiffusion modelGraph
🎯 What it does: A Directional Diffusion Model (DDM) is proposed for unsupervised graph representation learning, and its effectiveness is validated on various graph classification and node classification tasks.
Dis-inhibitory neuronal circuits can control the sign of synaptic plasticity
Julian Rossbroich (Friedrich Miescher Institute for Biomedical Research), Friedemann Zenke (Friedrich Miescher Institute for Biomedical Research)
ClassificationOptimizationSpiking Neural NetworkReinforcement LearningImageOrdinary Differential Equation
🎯 What it does: A Hebbian learning rule based on discrete inhibitory circuits is proposed, which achieves the positive and negative changes in synaptic plasticity through this rule to solve the credit assignment problem in neural networks.
Disambiguated Attention Embedding for Multi-Instance Partial-Label Learning
Wei Tang (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationRecognitionImage
🎯 What it does: A deep learning framework named DEMIPL is proposed for handling multi-instance partial label learning (MIPL) tasks, which maps each multi-instance bag to a single vector through attention embedding to identify the true labels.
Discover and Align Taxonomic Context Priors for Open-world Semi-Supervised Learning
Yu Wang (Peking University), Jie Chen (Peking University)
ClassificationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an open-source semi-supervised learning framework named TIDA, which aims to learn multi-level classification contexts (from subcategories to target categories to supercategories) through self-supervised methods in scenarios where there are insufficient known category samples and unknown categories, thereby improving the quality of feature representation and pseudo-labels.
DISCOVER: Making Vision Networks Interpretable via Competition and Dissection
Konstantinos P. Panousis (Cyprus University of Technology), Sotirios Chatzis (Cyprus University of Technology)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerVision Language ModelImageMultimodality
🎯 What it does: By introducing a random local winner-takes-all (LWTA) layer in the visual network, activation sparsity is achieved, and then using CLIP-Dissect to provide textual descriptions for each neuron, a post-hoc explanation framework (DISCOVER) is constructed, enhancing the interpretability of the network;
Discovering General Reinforcement Learning Algorithms with Adversarial Environment Design
Matthew Thomas Jackson, Jakob Nicolaus Foerster
Meta LearningReinforcement LearningTabular
🎯 What it does: The GROOVE method is proposed, which transfers Unsupervised Environment Design (UED) to Policy Meta Optimization (PMO), utilizing adversarial environment generation and adaptive curating to enhance the generalization ability of RL algorithms in unknown environments.
Discovering Hierarchical Achievements in Reinforcement Learning via Contrastive Learning
Seungyong Moon (Seoul National University), Hyun Oh Song (Seoul National University)
Reinforcement LearningContrastive Learning
🎯 What it does: This paper proposes a method called 'Achievement Distillation' that jointly uses contrastive learning during the reinforcement learning process, helping the model to learn and predict hierarchical achievements without an explicit planning module, thereby enabling efficient long-range exploration and achievement unlocking in procedurally generated environments.
Discovering Intrinsic Spatial-Temporal Logic Rules to Explain Human Actions
Chengzhi Cao (University of Science and Technology of China), Shuang Li (Chinese University of Hong Kong)
Object TrackingExplainability and InterpretabilityTransformerVideoTime Series
🎯 What it does: A knowledge-driven model based on spatio-temporal logic rules is proposed to automatically mine and interpret human behavior from motion trajectories.
Discrete-Smoothness in Online Algorithms with Predictions
Yossi Azar (Tel Aviv University), Noam Touitou (Amazon)
OptimizationTabular
🎯 What it does: This paper proposes a Discrete-Smoothness framework, providing a theory and algorithm to achieve a unified consistency, robustness, and smoothness in online algorithms with machine learning predictions; specific implementations are given for the classic problems of Facility Location and Set Cover, achieving better competitive ratios.
Discriminative Calibration: Check Bayesian Computation from Simulations and Flexible Classifier
Yuling Yao (Flatiron Institute), Justin Domke (University of Massachusetts)
ClassificationOptimization
🎯 What it does: This paper proposes a classifier-based Bayesian inference diagnostic method—Discriminative Calibration, which replaces traditional ranking statistics with learnable features, directly constructs label mappings from simulated data, and trains classifiers to estimate the error and divergence of the posterior distribution.
Discriminative Feature Attributions: Bridging Post Hoc Explainability and Inherent Interpretability
Usha Bhalla (Harvard University), Himabindu Lakkaraju (Harvard University)
Explainability and InterpretabilityKnowledge DistillationImage
🎯 What it does: The study proposes a method called Distractor Erasure Tuning (DiET), which maintains the original model's predictive performance while making the black-box model robust to the erasure of non-critical information (distractors), thereby providing more trustworthy feature importance explanations.
DisDiff: Unsupervised Disentanglement of Diffusion Probabilistic Models
Tao Yang (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
GenerationData SynthesisExplainability and InterpretabilityDiffusion modelImage
🎯 What it does: This paper proposes an unsupervised method called DisDiff, which performs factor disentanglement on pre-trained diffusion probabilistic models, automatically discovering and representing the intrinsic interpretable factors of images, and decomposing the diffusion gradient field into sub-gradient fields corresponding to each factor.
Disentangled Counterfactual Learning for Physical Audiovisual Commonsense Reasoning
Changsheng Lv (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
Recurrent Neural NetworkContrastive LearningVideoMultimodality
🎯 What it does: A Disentangled Counterfactual Learning (DCL) framework is proposed, which decomposes videos into static and dynamic factors and constructs contrastive learning and causal relationships to achieve physical audiovisual common sense reasoning.
Disentangled Wasserstein Autoencoder for T-Cell Receptor Engineering
Tianxiao Li (Yale University), Martin Renqiang Min (NEC Laboratories America)
GenerationDrug DiscoveryTransformerAuto EncoderBiomedical Data
🎯 What it does: A decoupling model based on Wasserstein autoencoder (TCR-dWAE) is proposed, which modifies the functional embedding of single-edited sequences to change their binding ability to specific peptides while keeping the T-cell receptor (TCR) structural framework unchanged.
Disentanglement via Latent Quantization
Kyle Hsu (Stanford University), Chelsea Finn (Stanford University)
Representation LearningAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the use of latent quantization and high-weight decay to achieve unsupervised decomposable representation learning.
Disentangling Cognitive Diagnosis with Limited Exercise Labels
Xiangzhi Chen (Hefei University of Technology), Meng Wang (Hefei University of Technology)
Explainability and InterpretabilityRepresentation LearningAuto EncoderTabular
🎯 What it does: A model DCD for interpretable cognitive diagnosis is proposed, which can diagnose students' mastery of knowledge concepts even with a small amount of labeled practice problems and in the absence of a complete Q matrix.
Disentangling Voice and Content with Self-Supervision for Speaker Recognition
Tianchi Liu (Agency for Science Technology and Research), Haizhou Li (Chinese University of Hong Kong)
RecognitionKnowledge DistillationConvolutional Neural NetworkContrastive LearningAudio
🎯 What it does: A RecXi framework based on Gaussian inference is proposed to decouple speaker information from content information without relying on text labels, and to further enhance the quality of content decoupling through self-supervised methods.
Dissecting Chain-of-Thought: Compositionality through In-Context Filtering and Learning
Yingcong Li (University of California), Samet Oymak (University of Michigan)
TransformerChain-of-Thought
🎯 What it does: This paper studies the mechanism of Chain of Thought (CoT) in Transformers, focusing on its impact on self-supervised learning multi-layer perceptrons (MLPs). It demonstrates through theory and experiments that CoT can be decomposed into two stages: filtering and context learning, significantly reducing sample complexity.
Distance-Restricted Folklore Weisfeiler-Leman GNNs with Provable Cycle Counting Power
Junru Zhou (Peking University), Muhan Zhang (Peking University)
Graph Neural NetworkGraph
🎯 What it does: A class of distance-restricted FWL(2) graph neural networks (d-DRFWL(2) GNN) is proposed, and it is proven to maintain efficiency while having the ability to count cycles of length 6 and above.
Distilling Out-of-Distribution Robustness from Vision-Language Foundation Models
Andy Zhou (University of Illinois), Haohan Wang (University of Illinois)
Knowledge DistillationAdversarial AttackVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: The study uses large visual language foundation models (such as CLIP) as teachers to enhance the out-of-distribution robustness of small visual models through knowledge distillation and discrete adversarial distillation (DAD).
Distributed Inference and Fine-tuning of Large Language Models Over The Internet
Alexander Borzunov (Higher School of Economics), Colin Raffel (Hugging Face)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A distributed and fault-tolerant inference and fine-tuning framework called PETALS is proposed, which can efficiently run language models with over 50 billion parameters on geographically distributed, potentially failing consumer-grade devices on the internet.
Distributed Personalized Empirical Risk Minimization
Yuyang Deng (Pennsylvania State University), Mehrdad Mahdavi (Pennsylvania State University)
OptimizationFederated LearningImage
🎯 What it does: A personalized experience risk minimization (PERM) framework based on distribution and system adaptation is proposed, which can learn exclusive models for each federated learning client while considering data heterogeneity and device resource differences.
Distribution Learnability and Robustness
Shai Ben-David (University of Waterloo), Tosca Lechner (University of Waterloo)
OptimizationComputational Efficiency
🎯 What it does: This study investigates the relationship between learnability and robust learning (additive/subtractive noise models) in distribution learning, proving that learnability is not equivalent to robust learnability, but that the two are equivalent under additive noise.
Distribution-Free Model-Agnostic Regression Calibration via Nonparametric Methods
Shang Liu (Imperial College London), Xiaocheng Li (Imperial College London)
TabularTime Series
🎯 What it does: A two-step nonparametric regression calibration method is proposed, which estimates quantiles of the residuals from a pre-trained regression model to achieve individual calibration.
Distribution-Free Statistical Dispersion Control for Societal Applications
Zhun Deng (Columbia University), Richard Zemel (Columbia University)
Recommendation SystemOptimizationTextTabularMagnetic Resonance Imaging
🎯 What it does: This paper proposes a distribution-independent statistical dispersion control framework that provides strict finite sample confidence bounds for various socially important dispersion metrics (such as the Gini coefficient, Atkinson index, group median deviation, etc.) and achieves fairer results in model selection by utilizing these bounds.
Distributional Learning of Variational AutoEncoder: Application to Synthetic Data Generation
Seunghwan An (University of Seoul), Jong-June Jeon (University of Seoul)
GenerationData SynthesisSafty and PrivacyAuto EncoderTabular
🎯 What it does: A distribution learning-based variational autoencoder (DistVAE) is proposed, which directly estimates the conditional cumulative distribution function (CDF) of continuous variables through an infinite mixture of asymmetric Laplace distributions, thus no longer being limited by traditional Gaussian decoders.
Distributional Model Equivalence for Risk-Sensitive Reinforcement Learning
Tyler Kastner (University of Toronto), Amir-massoud Farahmand (University of Toronto)
Reinforcement LearningFinance Related
🎯 What it does: This paper explores how model learning affects planning performance in risk-sensitive reinforcement learning and introduces a new concept of model equivalence.
Distributional Pareto-Optimal Multi-Objective Reinforcement Learning
Xin-Qiang Cai (University of Tokyo), Ashley Juan Llorens
OptimizationReinforcement LearningBenchmark
🎯 What it does: This paper proposes Distributed Pareto Optimal Multi-Objective Reinforcement Learning (DPMORL), which addresses distributional preferences in multi-objective decision-making by learning Distributed Pareto Optimal (DPO) policies.
Distributional Policy Evaluation: a Maximum Entropy approach to Representation Learning
Riccardo Zamboni (Politecnico di Milano), Marcello Restelli (Politecnico di Milano)
Representation LearningReinforcement LearningSequential
🎯 What it does: This paper proposes a distributed policy evaluation framework based on maximum entropy (D-Max-Ent PE) and achieves distributed representation learning through state aggregation features, introducing an algorithm for recursive refinement decomposition (D-Max-Ent Progressive Factorization).
Distributionally Robust Bayesian Optimization with $\varphi$-divergences
Hisham Husain (Amazon), Anton van den Hengel (Amazon)
OptimizationTabularTime Series
🎯 What it does: A distributionally robust Bayesian optimization framework based on ϕ-divergence is proposed, and its equivalence to a single maximization problem is proven, solving the original min-max computational challenge that was unsolvable;
Distributionally Robust Ensemble of Lottery Tickets Towards Calibrated Sparse Network Training
Hitesh Sapkota (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: Through the distributionally robust optimization (DRO) framework, multiple sparse sub-networks (Lottery Tickets) are directly searched and trained within a single randomly initialized dense network, which are then aggregated into an ensemble to enhance the model's calibration performance.
Distributionally Robust Linear Quadratic Control
Bahar Taskesen (École Polytechnique Fédérale de Lausanne), Daniel Kuhn (École Polytechnique Fédérale de Lausanne)
OptimizationTabularStochastic Differential Equation
🎯 What it does: This paper studies distributionally robust linear quadratic Gaussian (LQG) control under Wasserstein fuzzy sets, proving that the optimal control strategy remains linear output feedback, and provides an efficient numerical solution method based on the Frank-Wolfe algorithm.
Distributionally Robust Skeleton Learning of Discrete Bayesian Networks
Yeshu Li (Alibaba Group), Brian D Ziebart
OptimizationGraph Neural NetworkReinforcement LearningGraphTabularBenchmark
🎯 What it does: A distributionally robust optimization (DRO) framework is proposed to learn the undirected skeleton of discrete Bayesian networks in the presence of noise or contaminated data.
DiT-3D: Exploring Plain Diffusion Transformers for 3D Shape Generation
Shentong Mo (Mohamed bin Zayed University of Artificial Intelligence), Zhenguo Li (Huawei Noah's Ark Lab)
GenerationData SynthesisTransformerDiffusion modelPoint Cloud
🎯 What it does: Designed and trained a pure Transformer Diffusion model DiT-3D for denoising voxelized point clouds to generate 3D shapes.
Diverse Conventions for Human-AI Collaboration
Bidipta Sarkar (Stanford University), Dorsa Sadigh (Stanford University)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: The CoMeDi algorithm is proposed to generate diverse collaborative strategies (conventions) by minimizing cross-play and mixed-play to avoid gesture cheating and enhance zero-shot collaboration performance with human players.
Diverse Shape Completion via Style Modulated Generative Adversarial Networks
Wesley Khademi (Oregon State University), Li Fuxin (Oregon State University)
GenerationData SynthesisGenerative Adversarial NetworkPoint Cloud
🎯 What it does: A multimodal point cloud shape completion method based on conditional GAN is proposed, capable of generating various complete geometries that conform to partial observations.
Diversified Outlier Exposure for Out-of-Distribution Detection via Informative Extrapolation
Jianing Zhu (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
Anomaly DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposed Diversified Outlier Exposure (DivOE), which enhances OOD detection by extrapolating information from auxiliary outlier samples during the fine-tuning process to expand the OOB distribution.
Diversify \& Conquer: Outcome-directed Curriculum RL via Out-of-Distribution Disagreement
Daesol Cho (Seoul National University), H. Jin Kim (Seoul National University)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: This study proposes the D2C method, which automatically explores and plans for target states distributed arbitrarily through the diversification of multi-head classifiers.
Diversify Your Vision Datasets with Automatic Diffusion-based Augmentation
Lisa Dunlap (University of California Berkeley), Trevor Darrell (University of California Berkeley)
Data SynthesisDomain AdaptationVision Language ModelDiffusion modelImage
🎯 What it does: An automated language-guided image enhancement method called ALIA is proposed, which first generates task-independent domain descriptions using visual and language models, then edits the original image through a diffusion model, and maintains data quality through semantic and confidence filtering.
Diversifying Spatial-Temporal Perception for Video Domain Generalization
Kun-Yu Lin (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
RecognitionDomain AdaptationConvolutional Neural NetworkVideoBenchmark
🎯 What it does: This paper proposes a Spatial-Temporal Diversification Network (STDN) that learns diversified video features through spatial grouping and spatial-temporal relationship modeling to enhance video domain generalization performance.
Divide, Evaluate, and Refine: Evaluating and Improving Text-to-Image Alignment with Iterative VQA Feedback
Jaskirat Singh (Australian National University), Liang Zheng (Australian National University)
GenerationOptimizationTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a decomposition-based framework for text-to-image alignment evaluation and improvement. It first breaks down complex prompts into several mutually exclusive assertions, then uses a VQA model to assess the matching degree of each assertion with the generated image, and finally aggregates to obtain an overall alignment score. At the same time, this score is used as feedback to iteratively refine the alignment by gradually increasing the weight of the least matching assertions.
DiViNeT: 3D Reconstruction from Disparate Views using Neural Template Regularization
Aditya Vora (Simon Fraser University), Hao Zhang (Amazon)
GenerationDepth EstimationConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: A two-stage neural 3D reconstruction framework called DiViNet is proposed, which can achieve complete surface reconstruction using only three non-overlapping RGB images.
Django: Detecting Trojans in Object Detection Models via Gaussian Focus Calibration
Guangyu Shen (Purdue University), Xiangyu Zhang (Purdue University)
Object DetectionAdversarial AttackGaussian SplattingImage
🎯 What it does: This paper proposes the DJANGO framework, which detects backdoors in object detection models by inverting triggers on cleaned samples.
Do Not Marginalize Mechanisms, Rather Consolidate!
Moritz Willig (Technical University of Darmstadt), Kristian Kersting (German Research Center for Artificial Intelligence)
CompressionComputational EfficiencyTime Series
🎯 What it does: This paper studies a new structural causal model (SCM) simplification method called consolidation, which can compress the model size while maintaining a complete causal explanation for interventions.
Do SSL Models Have Déjà Vu? A Case of Unintended Memorization in Self-supervised Learning
Casey Meehan (University of California San Diego), Chuan Guo (Meta)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: The study investigates the phenomenon of 'déjà vu' memory, where self-supervised learning (SSL) models inadvertently memorize specific associations between the background and foreground of training samples during training.
Does a sparse ReLU network training problem always admit an optimum ?
TUNG QUOC LE, Elisa Riccietti (University of Lyon)
OptimizationTabular
🎯 What it does: This paper studies the existence of optimal solutions for training sparse ReLU neural networks, proving that under certain sparse structures, optimal solutions may not exist, and providing necessary and sufficient conditions; it also proves that single hidden layer sparse ReLU networks can always achieve optimal solutions on finite or continuous domains; and provides a closure determination algorithm based on real algebraic geometry.
Does Graph Distillation See Like Vision Dataset Counterpart?
Beining Yang (Beihang University), Jianxin Li (Beihang University)
Data SynthesisRecommendation SystemAnomaly DetectionGraph Neural NetworkGraph
🎯 What it does: A structural broadcast graph dataset distillation framework SGDD is proposed, which can generate extremely small-scale synthetic graphs while preserving the original graph structure information.
Does Invariant Graph Learning via Environment Augmentation Learn Invariance?
Yongqiang Chen (Chinese University of Hong Kong), James Cheng (Chinese University of Hong Kong)
Domain AdaptationRepresentation LearningDrug DiscoveryGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper studies the feasibility of achieving Invariant graph representation learning in graph data lacking environmental labels and proposes the GALA framework based on an environmental assistant model.
Does Localization Inform Editing? Surprising Differences in Causality-Based Localization vs. Knowledge Editing in Language Models
Peter Hase (Google Research), Asma Ghandeharioun (Google Research)
TransformerLarge Language ModelText
🎯 What it does: The relationship between causal tracing localization and model editing success rate was studied, finding no significant correlation between the two.
Does Visual Pretraining Help End-to-End Reasoning?
Chen Sun (Brown University), Cordelia Schmid (Google Research)
ClassificationRepresentation LearningTransformerAuto EncoderVideo
🎯 What it does: This study investigates and verifies that through self-supervised visual pre-training, general Transformer networks can achieve compositional generalization in end-to-end visual reasoning tasks.
Domain Adaptive Imitation Learning with Visual Observation
Sungho Choi (Korea Advanced Institute of Science and Technology), Youngchul Sung (Korea Advanced Institute of Science and Technology)
Domain AdaptationRobotic IntelligenceReinforcement LearningGenerative Adversarial NetworkImage
🎯 What it does: In the target domain, learn to imitate the behavior of source domain experts through visual observation;
Domain Agnostic Fourier Neural Operators
Ning Liu (Global Engineering and Materials), Yue Yu (Lehigh University)
TabularTime SeriesPhysics Related
🎯 What it does: A domain-agnostic Fourier neural operator (DAFNO) is proposed for learning PDE solvers for arbitrary geometries and topological changes.
Domain Re-Modulation for Few-Shot Generative Domain Adaptation
Yi Wu (University of Science and Technology of China), Dacheng Tao (JD Explore Academy)
GenerationDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: This paper studies a generative model for transferring from the source domain to the target domain under very few samples, namely few-shot generative domain adaptation (GDA).
Domain Watermark: Effective and Harmless Dataset Copyright Protection is Closed at Hand
Junfeng Guo (University of Maryland), Bo Li (Northwestern University)
ClassificationOptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new method for dataset ownership verification, utilizing domain watermarking to ensure that models trained on protected datasets perform correctly on specific 'hard' samples, rather than using traditional backdoor attack methods.
Don't be so Monotone: Relaxing Stochastic Line Search in Over-Parameterized Models
Leonardo Galli (RWTH Aachen University), Mark Schmidt (University of British Columbia)
OptimizationConvolutional Neural NetworkTransformerImageText
🎯 What it does: A non-monotonic stochastic line search method (PoNoS) is proposed, combining Polyak's stochastic step size and a new reset technique for training over-parameterized deep learning models.
Don’t blame Dataset Shift! Shortcut Learning due to Gradients and Cross Entropy
Aahlad Manas Puli (New York University), Rajesh Ranganath (New York University)
ClassificationOptimizationImageText
🎯 What it does: This study investigates why traditional cross-entropy optimization (default ERM) tends to learn shortcut features in perception tasks (such as image classification) and demonstrates that its inherent maximum margin preference leads to models overly relying on these unstable features. A unified margin-inducing loss function (MARG CTRL) is proposed, which is theoretically and experimentally validated to suppress shortcut learning, achieving better performance even in a nuisance-free setting where both training and validation lack shortcut labels.
Don’t just prune by magnitude! Your mask topology is a secret weapon
Duc N.M Hoang (University of Texas at Austin), Zhangyang Wang (University of Texas at Austin)
Image
🎯 What it does: This study investigates the spectral properties of sparse neural networks, proposes a correlation between weighted spectral distance and performance, constructs a full spectral coordinate, and based on this, designs a sparsification method PAGS/PEGS under initialization or lightweight pre-training.
Don’t Stop Pretraining? Make Prompt-based Fine-tuning Powerful Learner
Zhengxiang Shi (University College London), Aldo Lipani (University College London)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes Prompt-based Continued Pre-training (PCP), which introduces task-related text and prompt templates during the continued pre-training phase, improving traditional Task Adaptive Pre-training (TAPT).
DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining
Sang Michael Xie (Google DeepMind), Adams Wei Yu (Google DeepMind)
OptimizationComputational EfficiencyLarge Language ModelText
🎯 What it does: This paper proposes the DoReMi algorithm, which automatically finds the optimal domain weights through distributed robust optimization (DRO) using a small model, thereby improving the pre-training efficiency of large language models.
DOSE: Diffusion Dropout with Adaptive Prior for Speech Enhancement
Wenxin Tai (University of Electronic Science and Technology of China), Ting Zhong (University of Electronic Science and Technology of China)
RestorationGenerationDiffusion modelAudio
🎯 What it does: A speech enhancement method based on diffusion models, DOSE, is proposed, which significantly improves the utilization of conditional information and addresses the issue of conditional collapse by randomly dropping intermediate samples during training and using adaptive priors.
Double and Single Descent in Causal Inference with an Application to High-Dimensional Synthetic Control
Jann Spiess (Stanford University), Amar Venugopal (Stanford University)
Tabular
🎯 What it does: This study investigates the interpolation behavior of over-parameterized linear regression and synthetic control in causal inference, demonstrating the double descent and single descent curves.
Double Auctions with Two-sided Bandit Feedback
Soumya Basu (Google), Abishek Sankararaman (Amazon Web Services)
Reinforcement LearningTabular
🎯 What it does: This study investigates a double auction market with unknown valuations for both buyers and sellers under bidirectional bandwidth feedback, and designs a bidding strategy based on confidence intervals to achieve price discovery under an average mechanism.
Double Gumbel Q-Learning
David Yu-Tung Hui (Mila Université de Montréal), Pierre-Luc Bacon (Mila Université de Montréal)
Reinforcement Learning
🎯 What it does: A Deep Q-Learning algorithm for discrete and continuous control based on a dual Gumbel noise model is proposed—DoubleGum.
Double Pessimism is Provably Efficient for Distributionally Robust Offline Reinforcement Learning: Generic Algorithm and Robust Partial Coverage
Jose Blanchet (Stanford University), Han Zhong (Peking University)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes a general double pessimism model-based policy optimization algorithm (P MPO 2) that can learn optimal robust policies in a robust offline reinforcement learning environment, and provides theoretical proof.
Double Randomized Underdamped Langevin with Dimension-Independent Convergence Guarantee
Yuanshi Liu (Peking University), Tong Zhang (Hong Kong University of Science and Technology)
OptimizationStochastic Differential Equation
🎯 What it does: This paper proposes a Double Randomized Underdamped Langevin algorithm (DRUL), which achieves sampling of high-dimensional log-concave distributions through random step sizes and random midpoints;
Doubly Constrained Fair Clustering
John P Dickerson, Claire Jie Zhang (University of Washington)
OptimizationTabular
🎯 What it does: This paper studies the simultaneous satisfaction of two fairness constraints in k-center clustering: group fairness (GF) and center diversity (DS), and proposes a post-processing method to convert solutions that satisfy a single constraint into solutions that satisfy both;
Doubly Robust Augmented Transfer for Meta-Reinforcement Learning
Yuankun Jiang (Shanghai Jiao Tong University), Hongkai Xiong (Shanghai Jiao Tong University)
Meta LearningReinforcement LearningSequential
🎯 What it does: A dual robust enhanced transfer (DRaT) method is proposed to address the sample transfer problem in meta reinforcement learning when there is a discrepancy between dynamics and reward functions in sparse reward settings.
Doubly-Robust Self-Training
Banghua Zhu (University of California Berkeley), Jiantao Jiao (University of California Berkeley)
Object DetectionDomain AdaptationImagePoint Cloud
🎯 What it does: A Doubly Robust Self-Training method is proposed, which improves the self-training loss function so that the model can revert to training only on labeled data when the teacher model is inaccurate, regardless of the quality of the pseudo-labels, and fully utilize pseudo-label data when the teacher model is accurate.
DoWG Unleashed: An Efficient Universal Parameter-Free Gradient Descent Method
Ahmed Khaled (Princeton University), Chi Jin (Princeton University)
OptimizationImage
🎯 What it does: A new non-parametric gradient descent optimizer called DoWG (Distance-based Weighted Gradient) is proposed, which matches the convergence rate of optimally tuned gradient descent in convex optimization without the need to adjust any parameters.
DP-HyPO: An Adaptive Private Framework for Hyperparameter Optimization
Hua Wang (University of Pennsylvania), Milan Shen (Meta Platforms, Inc.)
OptimizationFederated LearningSafty and PrivacyHyperparameter SearchImage
🎯 What it does: An adaptive differential privacy hyperparameter optimization framework DP-HyPO is proposed, which can dynamically select hyperparameters based on previous experimental results under confidentiality constraints.
DP-Mix: Mixup-based Data Augmentation for Differentially Private Learning
Wenxuan Bao (University of Florida), Vincent Bindschaedler (University of Florida)
Safty and PrivacyDiffusion modelImage
🎯 What it does: The study explores the feasibility of using the data augmentation technique mixup with multi-sample data in differential privacy learning and proposes two mixup data augmentation methods based on self-augmentation and diffusion models, namely DP-MIXSELF and DP-MIXDIFF.
DPM-Solver-v3: Improved Diffusion ODE Solver with Empirical Model Statistics
Kaiwen Zheng (Tsinghua University), Jun Zhu (Tsinghua University)
GenerationData SynthesisDiffusion modelImageOrdinary Differential Equation
🎯 What it does: This paper presents a novel fast diffusion ODE solver DPM-Solver-v3, which significantly reduces first-order discretization errors and further improves the quality of few-step sampling by reparameterizing the ODE through the introduction of Empirical Model Statistics (EMS).
DPOK: Reinforcement Learning for Fine-tuning Text-to-Image Diffusion Models
Ying Fan (Google Research), Kimin Lee (KAIST)
GenerationReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningDiffusion modelImageText
🎯 What it does: This paper proposes an online fine-tuning method based on reinforcement learning called DPOK, which fine-tunes text-to-image diffusion models according to human feedback reward functions.
DRAUC: An Instance-wise Distributionally Robust AUC Optimization Framework
Siran Dai (Institute of Information Engineering, Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
OptimizationImage
🎯 What it does: An instance-level distributionally robust AUC (DRAUC) optimization framework is proposed to address the computational intractability issues arising from the combination of AUC and distributionally robust optimization.
Dream the Impossible: Outlier Imagination with Diffusion Models
Xuefeng Du (University of Wisconsin), Yixuan Li (University of Wisconsin)
GenerationAnomaly DetectionDiffusion modelImage
🎯 What it does: This paper proposes the DREAM-OOD framework, which utilizes diffusion models to sample low-likelihood embeddings in a text-conditioned latent space and decode them into high-resolution, realistic out-of-distribution (OOD) images, thereby providing automated OOD data for training; the method can also be extended to generate in-distribution (ID) images.
DreamHuman: Animatable 3D Avatars from Text
Nikos Kolotouros (Google Research), Cristian Sminchisescu (Google Research)
GenerationData SynthesisPose EstimationDiffusion modelScore-based ModelNeural Radiance FieldImageText
🎯 What it does: DreamHuman automatically generates animatable 3D human avatars with realistic textures and geometric details using only text prompts, supporting pose re-targeting.
DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data
Stephanie Fu, Phillip Isola
GenerationRetrievalDiffusion modelImage
🎯 What it does: A NIGHTS dataset was constructed and the DreamSim metric was trained to better align with human visual similarity judgments.
DreamSparse: Escaping from Plato’s Cave with 2D Diffusion Model Given Sparse Views
Paul Yoo (University of Tokyo), Shixiang Shane Gu (University of Tokyo)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: A framework named DreamSparse has been developed, utilizing a frozen pre-trained 2D diffusion model and a sparse view geometry module to generate high-quality new view images from sparse perspectives.
DreamWaltz: Make a Scene with Complex 3D Animatable Avatars
Yukun Huang (International Digital Economy Academy), Lei Zhang (University of Science and Technology of China)
GenerationData SynthesisPose EstimationDiffusion modelScore-based ModelNeural Radiance FieldImageText
🎯 What it does: This paper proposes a framework based on text and human morphology/pose priors that can generate and animate high-quality 3D avatars and supports scene interaction with other 3D objects.
Drift doesn't Matter: Dynamic Decomposition with Diffusion Reconstruction for Unstable Multivariate Time Series Anomaly Detection
Chengsen Wang (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
Anomaly DetectionTransformerDiffusion modelTime Series
🎯 What it does: A heterogeneous multivariate time series anomaly detection framework named D3R is proposed, which first achieves dynamic decomposition of long-period unstable sequences through data-time mixed attention and offset subtraction, and then constructs an information bottleneck using noise diffusion for direct reconstruction, obtaining anomaly scores from the reconstruction error.
DropCompute: simple and more robust distributed synchronous training via compute variance reduction
Niv Giladi (Technion - Israel Institute of Technology), Daniel Soudry (Technion - Israel Institute of Technology)
OptimizationComputational EfficiencyTransformerImageText
🎯 What it does: Proposes the DropCompute method, which reduces computational variance and enhances robustness through threshold clipping in distributed synchronous training.
DropPos: Pre-Training Vision Transformers by Reconstructing Dropped Positions
Haochen Wang (Chinese Academy of Sciences), Zhaoxiang Zhang (Chinese Academy of Sciences)
ClassificationObject DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: This paper proposes DropPos, a self-supervised pre-training task based on position recovery, allowing the Vision Transformer to predict the positions of visible image patches solely based on visual information.
DrugCLIP: Contrastive Protein-Molecule Representation Learning for Virtual Screening
Bowen Gao (Tsinghua University), Yanyan Lan (Tsinghua University)
RetrievalRepresentation LearningDrug DiscoveryTransformerContrastive LearningBiomedical Data
🎯 What it does: We propose DrugCLIP, which rephrases virtual screening as a dense retrieval task, using contrastive learning to learn a shared representation of protein pockets and small molecules, and enabling rapid retrieval on pre-encoded molecular embeddings.
DSR: Dynamical Surface Representation as Implicit Neural Networks for Protein
Daiwen Sun (Renmin University of China), Qiwei Ye (Beijing Academy of Artificial Intelligence)
Protein Structure PredictionPoint Cloud
🎯 What it does: In protein dynamics, an implicit neural network is used to learn a continuous representation of protein surfaces in three-dimensional space and time, directly predicting the signed distance function (SDF) from point clouds, enabling the reconstruction and prediction of large-scale protein trajectories.
Dual Mean-Teacher: An Unbiased Semi-Supervised Framework for Audio-Visual Source Localization
Yuxin Guo (University of Chinese Academy of Sciences), Yun Zheng (Alibaba Group)
RecognitionObject DetectionContrastive LearningVideoMultimodalityAudio
🎯 What it does: The Dual Mean-Teacher (DMT) framework is proposed, utilizing a dual teacher-student structure for semi-supervised learning in audio-video source localization tasks, enhancing localization accuracy through noise filtering and pseudo-label cross-validation.
Dual Self-Awareness Value Decomposition Framework without Individual Global Max for Cooperative MARL
Zhiwei Xu (University of Chinese Academy of Sciences), Guoliang Fan (University of Chinese Academy of Sciences)
Reinforcement LearningAuto Encoder
🎯 What it does: This paper proposes a multi-agent value decomposition framework called DAVE, which is completely independent of Individual Global Max (IGM). It utilizes each agent's ego policy and alter ego value network to achieve joint optimal action exploration and learning through explicit search.
DYffusion: A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting
Salva Rühling Cachay (University of California San Diego), Rose Yu (University of California San Diego)
Diffusion modelTime SeriesPhysics Related
🎯 What it does: A dynamic information diffusion model named DYffusion is proposed for probabilistic prediction of high-dimensional spatiotemporal sequences.
Dynamic Context Pruning for Efficient and Interpretable Autoregressive Transformers
Sotiris Anagnostidis (ETH Zurich), Thomas Hofmann (ETH Zurich)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Dynamic context pruning has been implemented on autoregressive Transformers like GPT-2, by learning differentiable sparse masks to remove unimportant historical tokens in real-time;
Dynamic Non-monotone Submodular Maximization
Kiarash Banihashem (University of Maryland), Morteza Monemizadeh (Eindhoven University of Technology)
OptimizationVideo
🎯 What it does: A dynamic algorithm is proposed and implemented to solve the non-monotone submodular function maximization (under cardinality constraint k) in dynamic environments, providing an (8+ε) approximation guarantee.
Dynamic Personalized Federated Learning with Adaptive Differential Privacy
Xiyuan Yang (Wuhan University), Mang Ye (Wuhan University)
Federated LearningSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: A method called FedDPA is proposed, which achieves dynamic Fisher information personalization and adaptive constraints in a personalized federated learning framework with user-level differential privacy, thereby improving model performance and convergence speed.
Dynamic Pricing and Learning with Bayesian Persuasion
Shipra Agrawal (Columbia University), Wei Tang (Columbia University)
OptimizationReinforcement LearningFinance Related
🎯 What it does: This study investigates the incorporation of advertising information design (Bayesian Persuasion) into a dynamic pricing and learning framework, designing online algorithms to learn optimal pricing and advertising strategies to maximize seller revenue.
Dynamic Prompt Learning: Addressing Cross-Attention Leakage for Text-Based Image Editing
Kai Wang (Computer Vision Center), Joost van de Weijer (Universitat Autonoma de Barcelona)
Image TranslationGenerationTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper proposes a Dynamic Prompt Learning (DPL) method to address the cross-attention leakage problem in text-to-image editing, achieving precise target area editing by optimizing noun embeddings at each time step.
Dynamic Regret of Adversarial Linear Mixture MDPs
Long-Fei Li (Nanjing University), Zhi-Hua Zhou (Nanjing University)
OptimizationReinforcement Learning
🎯 What it does: This study investigates how to perform reinforcement learning in adversarial linear mixed Markov decision processes (MDPs) under unknown transition kernels, proposing a new algorithm to optimize dynamic regret.
Dynamic Sparsity Is Channel-Level Sparsity Learner
Lu Yin (Eindhoven University of Technology), Shiwei Liu (University of Texas at Austin)
OptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: The Chase method is proposed, seamlessly transforming the advantages of unstructured dynamic sparse training (DST) into GPU-friendly channel-level sparse networks while maintaining or even improving model accuracy.
Dynamic Tensor Decomposition via Neural Diffusion-Reaction Processes
Zheng Wang (University of Utah), Shandian Zhe (University of Utah)
Graph Neural NetworkTime SeriesOrdinary Differential Equation
🎯 What it does: A dynamic tensor decomposition model based on neural diffusion-reaction processes (DEMOTE) is proposed to learn embeddings of entities that change over time and predict entry values of sparse multidimensional time series data.
Dynamically Masked Discriminator for GANs
Wentian Zhang (King Abdullah University of Science and Technology), Bernard Ghanem (AI Initiative, King Abdullah University of Science and Technology)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a Dynamic Masking Discriminator (DMD) that adaptively adjusts the discriminator during GAN training through an online continuous learning mechanism, enhancing the learning effectiveness of the generator.
Dynamics Generalisation in Reinforcement Learning via Adaptive Context-Aware Policies
Michael Beukman (University of Oxford), Benjamin Rosman (University of Witwatersrand)
Reinforcement LearningOrdinary Differential Equation
🎯 What it does: This paper studies the zero-shot generalization problem of reinforcement learning under dynamic changes (i.e., different responses of the environment to actions) and proposes a Decision Adapter, which dynamically adjusts the baseline policy network by generating adapter weights based on context through a hypernetwork, thereby achieving adaptive control for different dynamics.
Dynamics of Finite Width Kernel and Prediction Fluctuations in Mean Field Neural Networks
Blake Bordelon (Harvard University), Cengiz Pehlevan (Harvard University)
Convolutional Neural NetworkImage
🎯 What it does: This paper studies the dynamics of wide but finite-width neural networks during the feature learning phase, providing an analysis of finite-size fluctuations based on dynamical mean field theory (DMFT);