NeurIPS 2023 Papers — Page 8
Conference on Neural Information Processing Systems · 3218 papers
Deciphering Spatio-Temporal Graph Forecasting: A Causal Lens and Treatment
Yutong Xia (National University of Singapore), Roger Zimmermann (National University of Singapore)
Graph Neural NetworkGraphTime Series
🎯 What it does: The CaST framework is proposed to address the temporal out-of-distribution (OoD) and dynamic spatial causal issues in spatiotemporal graph prediction using causal inference methods.
Decision Stacks: Flexible Reinforcement Learning via Modular Generative Models
Siyan Zhao (University of California), Aditya Grover (University of California)
TransformerReinforcement LearningDiffusion modelTabularSequential
🎯 What it does: Proposes the Decision Stacks framework, which decomposes goal-oriented RL into three modules: observation, reward, and action generative models, enabling offline RL and long-term planning.
Decision Tree for Locally Private Estimation with Public Data
Yuheng Ma (Renmin University of China), Hanfang Yang (Renmin University of China)
Safty and PrivacyReinforcement LearningTabular
🎯 What it does: This paper proposes a locally differentially private regression algorithm called LPDT (Locally Differentially Private Decision Tree) that utilizes public data to assist in the process. It constructs decision tree partitions based on public data and then applies local randomization and Laplace noise injection to private data for regression estimation.
Decision-Aware Actor-Critic with Function Approximation and Theoretical Guarantees
Sharan Vaswani (Simon Fraser University), Nicolas Le Roux (Microsoft Research)
OptimizationReinforcement Learning
🎯 What it does: A decision-aware Actor-Critic (AC) framework is proposed, designing a joint objective and providing a general AC algorithm that can be used with any function approximation;
Decompose a Task into Generalizable Subtasks in Multi-Agent Reinforcement Learning
Zikang Tian (Institute of Computing Technology, Chinese Academy of Sciences), Yunji Chen (Institute of Computing Technology, Chinese Academy of Sciences)
Recurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: A framework called DT2GS is proposed to decompose tasks into generalizable subtasks, enabling zero-shot generalization and transfer of multi-agent reinforcement learning models across tasks.
Decompose Novel into Known: Part Concept Learning For 3D Novel Class Discovery
Tingyu Weng (University of Chinese Academy of Sciences), Haiyong Jiang (University of Chinese Academy of Sciences)
ClassificationRecognitionKnowledge DistillationTransformerContrastive LearningPoint Cloud
🎯 What it does: For the task of 3D Novel Category Discovery (NCD), a component-based framework DNIK is proposed, which uses component concepts and spatial relationships between components to alleviate the feature bias caused by learning known categories, thereby better identifying unlabeled new categories.
Deconstructing Data Reconstruction: Multiclass, Weight Decay and General Losses
Gon Buzaglo (Weizmann Institute of Science), michal Irani
ClassificationRestorationImage
🎯 What it does: This study investigates the reconstruction capability of neural networks for training samples and achieves sample reconstruction under multi-class classification, weight decay, and general loss functions.
Decorate3D: Text-Driven High-Quality Texture Generation for Mesh Decoration in the Wild
Yanhui Guo (McMaster University), Xiaofei Wu (Noah's Ark Lab)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImageMesh
🎯 What it does: Text-based texture generation and editing for 3D models of the real world.
Deductive Verification of Chain-of-Thought Reasoning
Zhan Ling (University of California San Diego), Hao Su (University of California San Diego)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: A 'Natural Program' format based on natural language is designed to enable large language models to generate decomposable and verifiable deductive reasoning chains, enhancing the credibility and interpretability of reasoning through step-by-step self-verification.
Deep Contract Design via Discontinuous Networks
Tonghan Wang (Harvard University), David C. Parkes (Harvard University)
Optimization
🎯 What it does: This paper automatically designs optimal contracts using deep learning methods, proposing a Discontinuous ReLU (DeLU) network that can assign different biases in activation modes to approximate the utility function of the dominant party, and presents parallel inference techniques to find the optimal contract.
Deep Equilibrium Based Neural Operators for Steady-State PDEs
Tanya Marwah (Carnegie Mellon University), Andrej Risteski (Carnegie Mellon University)
Ordinary Differential Equation
🎯 What it does: A Fourier Neural Operator (FNO-DEQ) based on a Deep Equilibrium Model (DEQ) is proposed for learning solvers of steady-state PDEs.
Deep Fractional Fourier Transform
Hu Yu (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
ClassificationRestorationObject DetectionSuper ResolutionImage
🎯 What it does: This paper proposes a fast implementation of the two-dimensional fractional Fourier transform (FRFT) and designs a multi-order fractional convolution operator (MFRFC) for processing images from spatial, frequency, and unified space-frequency perspectives.
Deep Gaussian Markov Random Fields for Graph-Structured Dynamical Systems
Fiona Lippert (University of Amsterdam), Patrick Forré (University of Amsterdam)
Graph Neural NetworkTime Series
🎯 What it does: This paper proposes ST-DGMRF, a framework that reconstructs graph-structured state space models into multi-layer spatiotemporal Gaussian Markov random fields (GMRF) for state estimation and learning of spatiotemporal systems with high dimensions, partially unknown dynamics, and limited historical data.
Deep Insights into Noisy Pseudo Labeling on Graph Data
Botao WANG, Fugee Tsung (Hong Kong University of Science and Technology)
ClassificationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A cautious pseudo-labeling (CPL) method is proposed, which conducts a theoretical analysis of the negative impact of pseudo-label noise on training in graph data and provides practical solutions.
Deep learning with kernels through RKHM and the Perron-Frobenius operator
Yuka Hashimoto (NTT Network Service Systems Laboratories), Hachem Kadri (Aix-Marseille University)
Convolutional Neural NetworkImage
🎯 What it does: This paper proposes a deep kernel learning framework called Deep RKHM, which combines Reproducing Kernel Hilbert Modules (RKHM) based on C* algebras with the Peron–Frobenius operator. It provides upper bounds on generalization error, representation theorems, and theoretical connections with convolutional neural networks and neural tangent kernels.
Deep Momentum Multi-Marginal Schrödinger Bridge
Tianrong Chen (Georgia Institute of Technology), Evangelos Theodorou (Georgia Institute of Technology)
OptimizationBiomedical DataStochastic Differential Equation
🎯 What it does: Proposes Deep Momentum Multi-Marginal Schrödinger Bridge (DMSB) to solve the multi-marginal trajectory inference problem in phase space.
Deep Neural Collapse Is Provably Optimal for the Deep Unconstrained Features Model
Peter Súkeník (Institute of Science and Technology Klosterneuburg), Christoph H Lampert
ClassificationOptimizationImage
🎯 What it does: This paper proposes and analyzes the Deep Unconstrained Features Model (DUFM), proving that its global optimal solution must satisfy the three main properties of Deep Neural Collapse (DNC) in binary classification tasks.
Deep Non-line-of-sight Imaging from Under-scanning Measurements
Yue Li (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationDepth EstimationOptimizationConvolutional Neural NetworkImage
🎯 What it does: An end-to-end deep learning method is proposed for reconstructing non-line-of-sight (NLOS) images from under-sampled measurements (USM), which includes an Instantaneous Recovery Network (TRN) and a Volume Reconstruction Network (VRN);
Deep Optimal Transport: A Practical Algorithm for Photo-realistic Image Restoration
Theo Joseph Adrai, Tomer Michaeli (Technion Israel Institute of Technology)
RestorationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A framework is proposed that allows for optimal linear transport of outputs from existing recovery models during the testing phase with only a small number of samples, enhancing perceptual quality while preserving MSE.
Deep Patch Visual Odometry
Zachary Teed (Princeton University), Jia Deng (Princeton University)
Pose EstimationOptimizationComputational EfficiencyRecurrent Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a deep learning visual odometry system DPVO based on sparse image patches, achieving real-time monocular visual odometry using recurrent networks and differentiable bundle adjustment.
Deep Recurrent Optimal Stopping
NIRANJAN DAMERA VENKATA, Chiranjib Bhattacharyya (Indian Institute of Science)
OptimizationReinforcement Learning from Human FeedbackRecurrent Neural NetworkReinforcement LearningTime SeriesSequential
🎯 What it does: This paper proposes an offline optimal stopping policy gradient algorithm (OSPG) based on RNN, which can efficiently learn optimal stopping decisions in non-Markovian environments.
Deep Reinforcement Learning with Plasticity Injection
Evgenii Nikishin (DeepMind), Andre Barreto
Computational EfficiencyReinforcement LearningSequential
🎯 What it does: Designed and evaluated an intervention method called 'plasticity injection' to increase the plasticity of deep reinforcement learning networks, aimed at diagnosing and alleviating the loss of plasticity during the learning process, and achieving dynamic network expansion to enhance computational efficiency.
Deep Stochastic Processes via Functional Markov Transition Operators
Jin Xu (University of Oxford), Yee Whye Teh (University of Oxford)
GenerationData SynthesisOptimizationFlow-based ModelTime SeriesSequentialStochastic Differential Equation
🎯 What it does: This paper proposes Markov Neural Processes (MNPs), which gradually transform a simple initial stochastic process into a more expressive generative stochastic process by stacking reversible Markov transition operators (i.e., reversible transformations) in function space, while maintaining exchangeability and consistency.
DeepACO: Neural-enhanced Ant Systems for Combinatorial Optimization
Haoran Ye (Soochow University), Yong Li (Tsinghua University)
OptimizationGraph Neural NetworkReinforcement LearningTabular
🎯 What it does: A deep reinforcement learning-based ant colony optimization framework, DeepACO, is designed to automate the heuristic design of the ant colony algorithm by learning to generate a global heuristic heatmap, and it is combined with ant colony search to improve the solution quality of various combinatorial optimization problems.
DeepPCR: Parallelizing Sequential Operations in Neural Networks
Federico Danieli (Apple), Luca Zappella (Apple)
GenerationComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkDiffusion modelImage
🎯 What it does: The DeepPCR algorithm is proposed, which utilizes Parallel Cyclic Reduction (PCR) to transform the denoising steps of traditional serialized forward/backward propagation and diffusion models into block bidiagonal linear/nonlinear equation systems that can be solved in parallel, significantly reducing time complexity.
DeepSimHO: Stable Pose Estimation for Hand-Object Interaction via Physics Simulation
Rong Wang (Australian National University), Hongdong Li (Australian National University)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an end-to-end hand-object interaction 3D pose estimation framework called DeepSimHO, which can simultaneously obtain the poses of the hand and the object from a single image, ensuring grip stability through physical simulation.
Defending against Data-Free Model Extraction by Distributionally Robust Defensive Training
Zhenyi Wang (University of Maryland), Mingchen Gao (University at Buffalo)
Computational EfficiencyKnowledge DistillationAdversarial AttackImage
🎯 What it does: A preventive defense method called MeCo, based on distributionally robust defense training, has been developed to resist data-independent model extraction attacks (DFME) while maintaining the inference quality of the target model.
Defending Pre-trained Language Models as Few-shot Learners against Backdoor Attacks
Zhaohan Xi (Pennsylvania State University), Ting Wang (Stony Brook University)
ClassificationAdversarial AttackTransformerPrompt EngineeringText
🎯 What it does: A backdoor defense method based on mask sensitivity, MDP, is proposed, specifically designed for pre-trained language models in few-shot prompt learning scenarios.
Delayed Algorithms for Distributed Stochastic Weakly Convex Optimization
Wenzhi Gao (Stanford University), Qi Deng (Shanghai University of Finance and Economics)
RetrievalOptimizationTabular
🎯 What it does: This paper studies delayed stochastic algorithms in distributed weakly convex optimization, analyzes traditional Delayed Stochastic Subgradient (DSGD), and proposes a new Delayed Stochastic Proximal Linear (DSPL) method, providing robust safety strategies in environments with arbitrary delays.
Delegated Classification
Eden Saig (Technion - Israel Institute of Technology), Nir Rosenfeld (Technion - Israel Institute of Technology)
ClassificationOptimizationTabular
🎯 What it does: This study proposes a delegation classification framework for machine learning outsourcing, utilizing contract design to incentivize agents to provide high-quality models.
DELIFFAS: Deformable Light Fields for Fast Avatar Synthesis
YoungJoong Kwon, Christian Theobalt (Max Planck Institute for Informatics)
GenerationData SynthesisPose EstimationComputational EfficiencyNeural Radiance FieldVideoMesh
🎯 What it does: A deformable two-surface parameterized light field model (DELIFFAS) is proposed, capable of rendering high-quality digital puppets in real-time and under controllable conditions.
DELTA: Diverse Client Sampling for Fasting Federated Learning
Lin Wang (Chinese University of Hong Kong), Xiaoying Tang (Shenzhen Institute of Artificial Intelligence and Robotics for Society)
OptimizationFederated LearningImage
🎯 What it does: A novel unbiased diversity client sampling scheme named DELTA is proposed to accelerate the convergence speed of federated learning;
Demo2Code: From Summarizing Demonstrations to Synthesizing Code via Extended Chain-of-Thought
Huaxiaoyue Wang (Cornell University), Sanjiban Choudhury (Cornell University)
Robotic IntelligenceAI Code AssistantTransformerLarge Language ModelPrompt EngineeringVideoTextChain-of-Thought
🎯 What it does: The Demo2Code framework is proposed, which recursively summarizes demonstration generation task specifications and expands them into executable robot code.
Demographic Parity Constrained Minimax Optimal Regression under Linear Model
Kazuto Fukuchi (University of Tsukuba), Jun Sakuma (Tokyo Institute of Technology)
OptimizationTabular
🎯 What it does: Analyzes and solves the optimal mean square error of the regression problem that satisfies demographic fairness constraints in a one-dimensional linear model, and provides the corresponding optimal estimator;
Demystifying Oversmoothing in Attention-Based Graph Neural Networks
Xinyi Wu (Massachusetts Institute of Technology), Ali Jadbabaie (Massachusetts Institute of Technology)
Graph Neural NetworkTransformerGraph
🎯 What it does: The study quantitatively demonstrates that the attention mechanism-based graph neural networks (such as GAT, transformer, etc.) inevitably exhibit oversmoothing phenomena in deep layers, leading to a conclusion of exponential loss of expressive power.
Demystifying Softmax Gating Function in Gaussian Mixture of Experts
Huy Nguyen (University of Texas at Austin), Nhat Ho (University of Grenoble Alpes)
Mixture of Experts
🎯 What it does: This paper studies the convergence rate of maximum likelihood estimation in Gaussian mixture expert models with softmax gating, proposing a new Voronoi loss function and proving the parameter estimation rates under both exact fitting and overfitting settings.
Demystifying Structural Disparity in Graph Neural Networks: Can One Size Fit All?
Haitao Mao (Michigan State University), Jiliang Tang (Michigan State University)
Graph Neural NetworkGraph
🎯 What it does: This paper systematically studies the performance differences of Graph Neural Networks (GNN) on node subgroups with homogeneous and heterogeneous structural differences, and proposes a performance differentiation mechanism caused by the aggregation mechanism. It also provides a PAC-Bayes generalization bound based on the differences in feature distance and homogeneity ratio after aggregation, and validates its effectiveness through experiments. Furthermore, it explores the advantages of deeper GNNs on a few structural nodes and proposes a novel graph OOD scenario based on structural differences.
Demystifying the Optimal Performance of Multi-Class Classification
Minoh Jeong (University of Minnesota), Alex Dytso (Qualcomm Flarion Technology)
ClassificationRecommendation SystemTabular
🎯 What it does: This paper proposes a Bayes Error Rate (BER) estimator based on soft labels and designs denoising and robust methods for noisy labels and outliers, capable of directly estimating the theoretical optimal error rate from datasets in multi-class classification problems.
Dense and Aligned Captions (DAC) Promote Compositional Reasoning in VL Models
Sivan Doveh (IBM Research), Leonid Karlinsky (MIT-IBM Watson AI Lab)
GenerationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This paper enhances the combinatorial reasoning ability of visual language models by automatically improving the quality and density of image-text pairing data titles.
Dense-Exponential Random Features: Sharp Positive Estimators of the Gaussian Kernel
Valerii Likhosherstov (University of Cambridge), Adrian Weller (University of Cambridge)
OptimizationTransformerImageTextAudio
🎯 What it does: A new Dense Index Random Feature (DERFs) is proposed to approximate Gaussian and softmax kernels, and the estimation variance is reduced through closed-form optimization of parameters.
Density of States Prediction of Crystalline Materials via Prompt-guided Multi-Modal Transformer
Namkyeong Lee (Korea Advanced Institute of Science and Technology), Chanyoung Park (Korea Advanced Institute of Science and Technology)
Graph Neural NetworkTransformerPrompt EngineeringMultimodalityPhysics Related
🎯 What it does: This study investigates how to predict the density of states (DOS) of crystal materials by combining crystal structure and energy information through a multimodal Transformer.
Depth-discriminative Metric Learning for Monocular 3D Object Detection
Wonhyeok Choi (DGIST), Sunghoon Im (DGIST)
Object DetectionDepth EstimationAutonomous DrivingConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a metric learning scheme based on (K,B,ϵ)-quasi-isometric loss, encouraging the feature extractor to learn discriminative features that are only related to depth, and improving the quality of depth estimation in monocular 3D object detection through an auxiliary depth head.
Derandomized novelty detection with FDR control via conformal e-values
Meshi Bashari (Technion Israel Institute of Technology), Matteo Sesia (University of Southern California)
Anomaly DetectionTabular
🎯 What it does: This paper proposes a derandomized novelty detection method based on conformal e-values, utilizing multiple data splits and e-value aggregation to achieve FDR control in anomaly detection and reduce algorithm randomness.
DesCo: Learning Object Recognition with Rich Language Descriptions
Liunian Harold Li (University of California Los Angeles), Kai-Wei Chang (University of California Los Angeles)
RecognitionObject DetectionTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a language-supervised object detection method that utilizes rich language descriptions to train a vision-language model for better object recognition and localization.
Describe, Explain, Plan and Select: Interactive Planning with LLMs Enables Open-World Multi-Task Agents
Zihao Wang (Peking University), Yitao Liang (Peking University)
TransformerLarge Language ModelAgentic AIPrompt EngineeringTextBenchmark
🎯 What it does: An interactive planning framework named DEPS is proposed to improve the execution of open-world tasks through description, explanation, planning, and selection.
Described Object Detection: Liberating Object Detection with Flexible Expressions
Chi Xie (Tongji University), Shuang Liang (Tongji University)
Object DetectionTransformerVision Language ModelImageMultimodality
🎯 What it does: This paper proposes and implements the Described Object Detection (DOD) task, constructs a fully annotated D3 dataset, evaluates existing OVD, REC, and dual-function models, and proposes an improved OFA-DOD baseline based on the OFA architecture.
Design from Policies: Conservative Test-Time Adaptation for Offline Policy Optimization
Jinxin Liu (Westlake University), Bin Wang (Huawei Noah's Ark Lab)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: This paper proposes a non-iterative two-layer offline reinforcement learning framework called DROP (Design from Policies), which separates value estimation and policy extraction into training and testing phases. It utilizes embedded task partitioning and a conservative score model to achieve safe policy search, and adapts the embedding online during testing.
Designing Robust Transformers using Robust Kernel Density Estimation
Xing Han (University of Texas at Austin), Nhat Ho (University of Texas at Austin)
Adversarial AttackTransformerImageTextTime Series
🎯 What it does: This paper proposes three robust self-attention mechanisms (RKDE, SPKDE, MoM) by viewing the self-attention of the Transformer as a non-parametric kernel density estimation, aiming to enhance the model's resistance to attacked or contaminated data.
DESSERT: An Efficient Algorithm for Vector Set Search with Vector Set Queries
Joshua Engels (ThirdAI), Anshumali Shrivastava (Rice University)
RetrievalComputational EfficiencyText
🎯 What it does: The DESSERT algorithm is proposed and implemented for approximate search in set-to-set queries on vector collections, significantly improving retrieval speed.
Detecting Any Human-Object Interaction Relationship: Universal HOI Detector with Spatial Prompt Learning on Foundation Models
Yichao Cao (Southeast University), Chang Xu (University of Sydney)
RecognitionObject DetectionTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageText
🎯 What it does: Developed UniHOI, a general human-object interaction (HOI) detection framework that utilizes visual-language foundation models and large language models for spatial prompt learning, enabling the recognition of any interaction relationship.
Detecting hidden confounding in observational data using multiple environments
Rickard Karlsson, JH Krijthe
Tabular
🎯 What it does: Using observational data from multiple environments, a statistical testing method based on the principle of independent causal mechanisms was developed to identify potential hidden confounding between treatment and outcome.
Detection Based Part-level Articulated Object Reconstruction from Single RGBD Image
Yuki Kawana (University of Tokyo), Tatsuya Harada (RIKEN AIP)
Object DetectionPose EstimationTransformerImage
🎯 What it does: End-to-end detection and reconstruction of multi-instance everyday manipulable objects at the component level in terms of shape, pose, and kinematics from a single RGBD image;
DeWave: Discrete Encoding of EEG Waves for EEG to Text Translation
Yiqun Duan (University of Technology Sydney), Chin-teng Lin
GenerationData SynthesisTransformerAuto EncoderContrastive LearningTextMultimodalityTime SeriesSequential
🎯 What it does: The DeWave framework is proposed, which directly translates EEG signals into text using discrete encoding without the need for eye movement markers or event triggers;
DFRD: Data-Free Robustness Distillation for Heterogeneous Federated Learning
Kangyang Luo (East China Normal University), Ming Gao (East China Normal University)
Federated LearningKnowledge DistillationImage
🎯 What it does: A data-independent robust distillation method DFRD is proposed to address the issues of data and model heterogeneity in federated learning.
Diff-Foley: Synchronized Video-to-Audio Synthesis with Latent Diffusion Models
Simian Luo (Tsinghua University), Hang Zhao (Tsinghua University)
GenerationData SynthesisDiffusion modelContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes DIFF-FOLEY, a video-to-audio (V2A) synthesis framework based on latent diffusion models, capable of generating high-quality audio that is highly synchronized and semantically relevant to the video content.
Diff-Instruct: A Universal Approach for Transferring Knowledge From Pre-trained Diffusion Models
Weijian Luo (Peking University), Zhihua Zhang (Huawei Noah's Ark Lab)
GenerationKnowledge DistillationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Proposes the Diff-Instruct framework, which utilizes the knowledge of pre-trained diffusion models to guide the training of any implicit generative model in a data-free manner.
DiffAttack: Evasion Attacks Against Diffusion-Based Adversarial Purification
Mintong Kang (University of Illinois Urbana-Champaign), Bo Li (University of Illinois Urbana-Champaign)
Adversarial AttackDiffusion modelScore-based ModelImage
🎯 What it does: This paper proposes DiffAttack, an adversarial purification defense based on diffusion models;
DiffComplete: Diffusion-based Generative 3D Shape Completion
Ruihang Chu, Jiaya Jia
RestorationGenerationData SynthesisConvolutional Neural NetworkDiffusion modelPoint CloudMesh
🎯 What it does: This paper presents DiffComplete, a 3D shape completion method based on diffusion models, capable of recovering complete, realistic, and multi-modal 3D shapes from sparse depth scans.
DIFFER:Decomposing Individual Reward for Fair Experience Replay in Multi-Agent Reinforcement Learning
Xunhan Hu (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
Reinforcement LearningTabularBenchmark
🎯 What it does: A framework called DIFFER is constructed to decompose team rewards into individual rewards, utilizing individual TD-error to achieve fair experience replay, thereby enhancing the learning efficiency and fairness of multi-agent reinforcement learning.
Differentiable and Stable Long-Range Tracking of Multiple Posterior Modes
Ali Younis (University of California), Erik B. Sudderth (University of California)
Object TrackingConvolutional Neural NetworkReinforcement LearningTime Series
🎯 What it does: A differentiable and stable multimodal particle filter (MDPF/A-MDPF) is proposed, which achieves unbiased low-variance gradients for continuous mixed distributions through importance sampling gradient estimation, enabling end-to-end training.
Differentiable Blocks World: Qualitative 3D Decomposition by Rendering Primitives
Tom Monnier (Ecole des Ponts), Mathieu Aubry
GenerationOptimizationImage
🎯 What it does: Given a set of calibrated multi-view images, this paper proposes Differentiable Blocks World (DBW), which directly optimizes renderable textured superquadric primitives from random initialization to achieve interpretable, editable, and physically simulative 3D scene decomposition.
Differentiable Clustering with Perturbed Spanning Forests
Lawrence Stewart (École Normale Supérieure), Quentin Berthet (Google DeepMind)
OptimizationImage
🎯 What it does: A differentiable clustering method is proposed, utilizing a minimum spanning forest with random perturbations for end-to-end training.
Differentiable Neuro-Symbolic Reasoning on Large-Scale Knowledge Graphs
CHEN SHENGYUAN, Mingming Sun (Baidu Research)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper presents DiffLogic, a differentiable neural symbolic reasoning framework that combines knowledge graph embedding models with rule reasoning. It adaptively selects important base formulas and utilizes PSL (Probabilistic Soft Logic) for end-to-end joint optimization, achieving efficient reasoning on large-scale knowledge graphs.
Differentiable Random Partition Models
Thomas M. Sutter (ETH Zurich), Julia E Vogt
Image
🎯 What it does: A differentiable random partition model (DRPM) is proposed, which can end-to-end partition set elements into an unknown number of subsets and is applied to variational clustering, weakly supervised generative factor separation, and multi-task learning.
Differentiable Registration of Images and LiDAR Point Clouds with VoxelPoint-to-Pixel Matching
Junsheng Zhou (Tsinghua University), Zhizhong Han (Wayne State University)
Autonomous DrivingOptimizationConvolutional Neural NetworkSupervised Fine-TuningImagePoint Cloud
🎯 What it does: Proposes a VoxelPoint-to-Pixel matching framework to achieve cross-modal registration between images and LiDAR point clouds.
Differentiable Sampling of Categorical Distributions Using the CatLog-Derivative Trick
Lennert De Smet (KU Leuven), Pedro Zuidberg Dos Martires (Örebro University)
OptimizationReinforcement LearningTabular
🎯 What it does: The CatLog-Derivative trick is proposed, and based on it, an unbiased gradient estimator for independent multivariate categorical distributions, called IndeCateR, is developed.
Differentiable sorting for censored time-to-event data.
Andre Vauvelle (University College London), Spiros Denaxas (University College London)
TabularBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes Diffsurv, a differentiable ranking method capable of handling censored time-to-event data for risk prediction in survival analysis.
Differentially Private Approximate Near Neighbor Counting in High Dimensions
Alexandr Andoni (Columbia University), Shyam Narayanan (Massachusetts Institute of Technology)
Safty and PrivacyComputational Efficiency
🎯 What it does: A differential privacy neighbor counting ((c,r)-neighbor counting) data structure is designed in high-dimensional Euclidean space, which can ensure an approximation error of 1±o(1) while the amount of noise is only related to the data size and the neighbor error factor c, and is independent of the dimension.
Differentially Private Decoupled Graph Convolutions for Multigranular Topology Protection
Eli Chien (University of Illinois at Urbana-Champaign), Olgica Milenkovic (University of Illinois at Urbana-Champaign)
Safty and PrivacyGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Graph Differential Privacy (GDP) framework for graph learning and designs a Decoupled Graph Convolution (DPDGC) based on it to achieve multi-granularity graph topology protection.
Differentially Private Image Classification by Learning Priors from Random Processes
Xinyu Tang (Princeton University), Prateek Mittal (Princeton University)
ClassificationSafty and PrivacyRepresentation LearningGenerative Adversarial NetworkContrastive LearningImageBiomedical Data
🎯 What it does: Using synthetic images generated by stochastic processes to learn visual priors, combined with the three-stage differential privacy training framework DP-RandP, enhances the performance of private image classification.
Differentially Private Statistical Inference through $\beta$-Divergence One Posterior Sampling
Jack Jewson (Universitat Pompeu Fabra), Christopher C. Holmes
Safty and PrivacyComputational EfficiencyTabular
🎯 What it does: A posterior sampling method based on β-divergence (β D-Bayes) is proposed to achieve different private statistical inferences.
DiffKendall: A Novel Approach for Few-Shot Learning with Differentiable Kendall's Rank Correlation
Kaipeng Zheng (Shanghai Jiao Tong University), Weiran Huang (Microsoft)
ClassificationMeta LearningImage
🎯 What it does: This paper replaces geometric similarity with Kendall's rank correlation and proposes a differentiable approximation for training and inference in few-shot learning frameworks such as Meta-Baseline, thereby improving classification performance.
DiffPack: A Torsional Diffusion Model for Autoregressive Protein Side-Chain Packing
Yangtian Zhang (Mila - Quebec AI Institute), Jian Tang (HEC Montreal)
Protein Structure PredictionGraph Neural NetworkDiffusion modelBiomedical Data
🎯 What it does: We propose DiffPack, a method based on an autoregressive torsional diffusion model to predict protein side chain conformations, modeling distributions directly in the torsional angle space.
DiffSketcher: Text Guided Vector Sketch Synthesis through Latent Diffusion Models
XiMing Xing, Dong Xu
GenerationData SynthesisDiffusion modelScore-based ModelImageText
🎯 What it does: Based on a pre-trained text-image diffusion model, this work utilizes a differentiable vectorized renderer and an improved Score Distillation Sampling, along with attention-driven initialization, to directly generate high-quality, adjustable abstract-level vector hand-drawn sketches from text.
DiffTraj: Generating GPS Trajectory with Diffusion Probabilistic Model
Yuanshao Zhu (Southern University of Science and Technology), James Yu
GenerationSafty and PrivacyConvolutional Neural NetworkDiffusion modelTime Series
🎯 What it does: This paper proposes a GPS trajectory generation method called DiffTraj based on a diffusion probability model, aiming to achieve privacy protection while maintaining the practicality of the trajectories.
DiffuseBot: Breeding Soft Robots With Physics-Augmented Generative Diffusion Models
Tsun-Hsuan Wang (Massachusetts Institute of Technology), Daniela Rus (Massachusetts Institute of Technology)
OptimizationRobotic IntelligenceDiffusion modelPhysics Related
🎯 What it does: This paper presents DiffuseBot, a co-design framework that combines physical simulation with diffusion models, capable of automatically generating morphology and control strategies based on task requirements.
Diffused Redundancy in Pre-trained Representations
Vedant Nanda (University of Maryland), Adrian Weller (Alan Turing Institute and University of Cambridge)
Representation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes and verifies the existence of a diffusion redundancy phenomenon in the hidden layers of pre-trained models, meaning that randomly selecting a sufficient number of neurons can approximate the complete representation of the layer and downstream performance.
Diffused Task-Agnostic Milestone Planner
Mineui Hong (Seoul National University), Songhwai Oh (Seoul National University)
Robotic IntelligenceReinforcement LearningDiffusion modelSequential
🎯 What it does: This paper proposes a Diffused Task-Agnostic Milestone Planner (DTAMP) that generates a sequence of milestones in a low-dimensional latent space using a diffusion model, and trains an encoder, actor, and critic through goal-conditioned imitation learning, enabling the agent to accomplish target tasks along these milestones.
Diffusion Hyperfeatures: Searching Through Time and Space for Semantic Correspondence
Grace Luo (University of California Berkeley), Trevor Darrell (University of California Berkeley)
RecognitionSegmentationRetrievalDiffusion modelImage
🎯 What it does: This paper proposes Diffusion Hyperfeatures, which utilize intermediate features from all time steps and scales of the diffusion model to generate a single pixel-level descriptor through an aggregation network for semantic correspondence.
Diffusion Model for Graph Inverse Problems: Towards Effective Source Localization on Complex Networks
Xin Yan (Northeastern University), Qiang He (Northeastern University)
Graph Neural NetworkDiffusion modelGraph
🎯 What it does: A discrete diffusion denoising model DDMSL is proposed to simultaneously achieve source node localization and information propagation path recovery in graph inverse problems.
Diffusion Model is an Effective Planner and Data Synthesizer for Multi-Task Reinforcement Learning
Haoran He (Shanghai Jiao Tong University), Xuelong Li (Shanghai Artificial Intelligence Laboratory)
Data SynthesisTransformerReinforcement LearningPrompt EngineeringDiffusion modelSequential
🎯 What it does: This paper proposes MTDIFF, a GPT-based diffusion model that can simultaneously perform generative planning and data synthesis in multi-task offline reinforcement learning.
Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labels
Zebin You (Renmin University of China), Jun Zhu (Tsinghua University)
ClassificationGenerationTransformerDiffusion modelImage
🎯 What it does: A three-stage Dual Pseudo Training (DPT) method is proposed, which utilizes a semi-supervised classifier to generate pseudo-labels, trains a conditional diffusion model to generate pseudo-images, and then retrains the classifier using these pseudo-images, thereby achieving few-label conditional generation and semi-supervised classification.
Diffusion Probabilistic Models for Structured Node Classification
Hyosoon Jang (POSTECH), Sungsoo Ahn (POSTECH)
ClassificationGraph Neural NetworkDiffusion modelGraph
🎯 What it does: This paper proposes a structured node classification framework DPM-SNC based on the Diffusion Probability Model (DPM) for predicting unknown node labels on partially labeled graphs, capable of capturing dependencies between labels.
Diffusion Representation for Asymmetric Kernels via Magnetic Transform
Mingzhen He (Shanghai Jiao Tong University), Xiaolin Huang (Shanghai Jiao Tong University)
Representation LearningDiffusion modelGraph
🎯 What it does: Proposes the MagDM framework, which uses magnetic transformation to map asymmetric kernels to Hermitian kernels, and defines corresponding integral operators, diffusion distances, and mappings to achieve dimensionality reduction for asymmetric data.
Diffusion Schrödinger Bridge Matching
Yuyang Shi (University of Oxford), Arnaud Doucet (University of Oxford)
GenerationData SynthesisDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes a new Schrödinger Bridge (SB) solving method—Iterative Markovian Fitting (IMF), and based on this method, introduces the Diffusion Schrödinger Bridge Matching (DSBM) algorithm, which can efficiently learn SB and generate samples;
Diffusion Self-Guidance for Controllable Image Generation
Dave Epstein (University of California Berkeley), Aleksander Holynski (Google Research)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A zero-shot self-guidance method based on the internal representations of pre-trained diffusion models is proposed, achieving fine control over attributes such as shape, position, size, and appearance of objects in generated images, as well as enabling cross-image attribute stitching and real image editing.
Diffusion with Forward Models: Solving Stochastic Inverse Problems Without Direct Supervision
Ayush Tewari (Massachusetts Institute of Technology), Vincent Sitzmann (Massachusetts Institute of Technology)
RestorationGenerationDiffusion modelGenerative Adversarial NetworkImageVideo
🎯 What it does: This paper proposes a method to directly embed differentiable forward models (such as rendering, deformation, GAN generation, etc.) into the conditional denoising diffusion process, training the model using only observed data without the need for direct observation of the signal, enabling consistent signal distribution sampling from partial observations during testing.
Diffusion-Based Adversarial Sample Generation for Improved Stealthiness and Controllability
Haotian Xue (Georgia Institute of Technology), Yongxin Chen (Georgia Institute of Technology)
Adversarial AttackConvolutional Neural NetworkTransformerDiffusion modelImageStochastic Differential Equation
🎯 What it does: A method for generating adversarial examples based on diffusion models, called Diff-PGD, is proposed, which maintains a high attack success rate while making adversarial examples more aligned with the distribution of natural images.
Diffusion-Based Probabilistic Uncertainty Estimation for Active Domain Adaptation
Zhekai Du (University of Electronic Science and Technology of China), Jingjing Li (University of Electronic Science and Technology of China)
Domain AdaptationKnowledge DistillationDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a probabilistic uncertainty estimation framework based on diffusion models for Active Domain Adaptation.
Diffusion-SS3D: Diffusion Model for Semi-supervised 3D Object Detection
Cheng-Ju Ho (National Yang Ming Chiao Tung University), Yi-Hsuan Tsai (Google)
Object DetectionDiffusion modelPoint Cloud
🎯 What it does: This paper proposes Diffusion-SS3D, a method that utilizes diffusion models to generate high-quality pseudo-labels in semi-supervised 3D object detection. It is integrated into a teacher-student framework, using noise perturbation of object size and category distribution for denoising to obtain more reliable 3D bounding boxes.
Diffusion-TTA: Test-time Adaptation of Discriminative Models via Generative Feedback
Mihir Prabhudesai (Carnegie Mellon University), Katerina Fragkiadaki (Carnegie Mellon University)
ClassificationSegmentationDepth EstimationDomain AdaptationConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: Adaptive testing of pre-trained discriminative models (classification, segmentation, depth prediction) is performed, utilizing the likelihood feedback from the pre-trained diffusion generative model to perform gradient updates on the weights of each unlabeled test sample.
DiffUTE: Universal Text Editing Diffusion Model
Haoxing Chen (Ant Group), Weiqiang Wang (Ant Group)
RecognitionImage TranslationGenerationLarge Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes DiffUTE, a universal text editing method based on diffusion models, which can replace or modify text in any image while maintaining a realistic background.
DiffVL: Scaling Up Soft Body Manipulation using Vision-Language Driven Differentiable Physics
Zhiao Huang (University of California, San Diego), Chuang Gan (Massachusetts Institute of Technology)
OptimizationRobotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: The DiffVL framework is proposed, which represents multi-stage tasks through visual and natural language descriptions, utilizing large language models to compile task instructions into differentiable optimization programs, and solving long-term soft manipulation problems under differentiable physical simulation.
DIFUSCO: Graph-based Diffusion Solvers for Combinatorial Optimization
Zhiqing Sun (Carnegie Mellon University), Yiming Yang (Carnegie Mellon University)
OptimizationGraph Neural NetworkDiffusion modelGraph
🎯 What it does: A combinatorial optimization solver DIFUSCO based on graph diffusion models has been designed and implemented, capable of solving NP-complete problems such as TSP and MIS without relying on manual heuristics.
DIN-SQL: Decomposed In-Context Learning of Text-to-SQL with Self-Correction
Mohammadreza Pourreza (University of Alberta), Davood Rafiei (University of Alberta)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: This paper proposes a few-shot prompting method that decomposes the natural language to SQL task into four modules (schema linking, query classification and decomposition, SQL generation, and automatic error correction), significantly improving the execution accuracy of LLMs.
DinoSR: Self-Distillation and Online Clustering for Self-supervised Speech Representation Learning
Alexander H. Liu (Massachusetts Institute of Technology), James R. Glass
RecognitionRepresentation LearningTransformerContrastive LearningBenchmarkAudio
🎯 What it does: Developed DinoSR, which combines masked language modeling, self-supervised distillation, and online clustering to learn speech representations and generate discrete phoneme units.
Direct Diffusion Bridge using Data Consistency for Inverse Problems
Hyungjin Chung (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
RestorationSuper ResolutionDiffusion modelScore-based ModelImage
🎯 What it does: This paper proposes an inference framework that directly uses the Direct Diffusion Bridge (DDB) in inverse problems, and incorporates data consistency correction to form the CDDB method.
Direct Preference Optimization: Your Language Model is Secretly a Reward Model
Rafael Rafailov (Stanford University), Chelsea Finn (Stanford University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: A direct preference optimization (DPO) method is proposed that does not require reinforcement learning, using binary cross-entropy to directly train language models to meet human preferences.
Direct Preference-based Policy Optimization without Reward Modeling
Gaon An (Seoul National University), Hyun Oh Song (Seoul National University)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningContrastive LearningSequential
🎯 What it does: An algorithm for learning strategies directly from human preference labels (DPPO) is proposed, eliminating the need to construct a reward model and optimizing the strategy directly through contrastive learning.
Direct Training of SNN using Local Zeroth Order Method
Bhaskar Mukhoty (Mohamed bin Zayed University of Artificial Intelligence), Bin Gu (Jilin University)
OptimizationSpiking Neural NetworkImage
🎯 What it does: A direct training algorithm for SNN based on the local zero-order method (LOCALZO) is proposed, utilizing random sampling to approximate the gradient of the spike function, achieving gradient estimation and sparsification in backpropagation;