ICLR 2025 Papers — Page 9
International Conference on Learning Representations · 3704 papers
Depth Any Video with Scalable Synthetic Data
Honghui Yang (Zhejiang University), Tong He (Shanghai AI Laboratory)
Depth EstimationDiffusion modelAuto EncoderImageVideo
🎯 What it does: This paper presents Depth Any Video, a foundational model for depth estimation that can simultaneously process images and videos, achieving high-quality, temporally consistent depth predictions on videos of arbitrary lengths.
Depth Pro: Sharp Monocular Metric Depth in Less Than a Second
Alexey Bochkovskiy, Vladlen Koltun (Apple)
Depth EstimationTransformerImage
🎯 What it does: A zero-shot multi-scale ViT base model called Depth Pro is proposed for generating high-resolution, absolute scale monocular depth maps without the need for camera intrinsics.
Deriving Causal Order from Single-Variable Interventions: Guarantees & Algorithm
Mathieu Chevalley (GSK), Arash Mehrjou (MPI for Intelligent Systems)
GraphTabular
🎯 What it does: This paper proposes a causal ordering inference method based on univariate intervention data and introduces a new computable scoring function that utilizes the ε-intervention credibility assumption to achieve causal ranking.
Descent with Misaligned Gradients and Applications to Hidden Convexity
Aditya Bhaskara (University of Utah), Manish Purohit (Google Research)
OptimizationStochastic Differential Equation
🎯 What it does: This paper studies the problem of minimizing convex objectives in the presence of 'unaligned' stochastic gradients and proposes an optimization algorithm that can identify ϵ-suboptimality with optimal iteration complexity.
Designing Concise ConvNets with Columnar Stages
Ashish Kumar (ScoreLabsAI), Jaesik Park (Seoul National University)
ClassificationObject DetectionConvolutional Neural NetworkImage
🎯 What it does: A concise convolutional network framework named CoSNet is proposed, which employs parallel column convolution, input replication, minimized 1×1 convolution, and single fusion designs, aiming to significantly reduce parameters, FLOPs, and inference latency while maintaining high accuracy.
Designing Mechanical Meta-Materials by Learning Equivariant Flows
Mehran Mirramezani (Princeton University), Ryan P Adams
OptimizationFlow-based ModelMeshPhysics RelatedOrdinary Differential Equation
🎯 What it does: Utilizing equivariant neural flow to reverse design cellular solid materials in two-dimensional lattice group space, ensuring volume preservation and symmetry constraints;
Detecting Backdoor Samples in Contrastive Language Image Pretraining
Hanxun Huang (University of Melbourne), James Bailey (University of Melbourne)
Anomaly DetectionTransformerContrastive LearningImage
🎯 What it does: This paper studies the sample detection problem of the CLIP model under low-rate (0.01%) backdoor attacks, proposing detection and cleaning through the analysis of the model's local sparse features in the representation space.
Determine-Then-Ensemble: Necessity of Top-k Union for Large Language Model Ensembling
Yuxuan Yao (City University of Hong Kong), Linqi Song (City University of Hong Kong)
TransformerLarge Language ModelMixture of ExpertsTextBenchmark
🎯 What it does: This paper researches and implements an efficient LLM integration method (UNITE) and proposes a model selection strategy based on model performance and response style, enhancing multi-model inference performance.
DexTrack: Towards Generalizable Neural Tracking Control for Dexterous Manipulation from Human References
Xueyi Liu (Tsinghua University), Li Yi (Shanghai Qi Zhi Institute)
Robotic IntelligenceReinforcement LearningDiffusion modelPoint Cloud
🎯 What it does: DexTrack has been developed, a globally generalizable neural tracking controller that can learn from human hand motion references for multi-object multi-grasping and complex hand manipulation.
DGQ: Distribution-Aware Group Quantization for Text-to-Image Diffusion Models
Hyogon Ryu (Korea Advanced Institute of Science and Technology), Hyunjung Shim (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisComputational EfficiencyTransformerDiffusion modelImageText
🎯 What it does: This paper studies low-bit quantization for text-to-image diffusion models and proposes a method called Distribution-aware Group Quantization (DGQ), which maintains high-quality image and text alignment at below 8-bit precision without requiring additional fine-tuning.
DICE: Data Influence Cascade in Decentralized Learning
Tongtian Zhu (Zhejiang University), Fengxiang He (University of Edinburgh)
Anomaly DetectionOptimizationFederated LearningData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: The DICE (Data Influence Cascade in Decentralized Learning) framework is proposed to quantify the impact of data on model performance in a fully decentralized learning environment, capturing influence cascades through a multi-hop propagation model.
DICE: End-to-end Deformation Capture of Hand-Face Interactions from a Single Image
Qingxuan Wu (University of Pennsylvania), Lingjie Liu (University of Pennsylvania)
Pose EstimationTransformerImageMesh
🎯 What it does: This paper proposes an end-to-end single-image hand-face interaction 3D reconstruction method called DICE, which can recover the deformations in hand and face interactions.
Diff-2-in-1: Bridging Generation and Dense Perception with Diffusion Models
Shuhong Zheng (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
SegmentationGenerationDepth EstimationDiffusion modelImageMultimodality
🎯 What it does: This paper proposes an integrated diffusion model framework named Diff-2-in-1, which can simultaneously perform multimodal data generation and dense visual perception tasks (such as surface normal estimation, semantic segmentation, etc.).
Diff-PIC: Revolutionizing Particle-In-Cell Nuclear Fusion Simulation with Diffusion Models
Chuan Liu (University of Rochester), Tong Geng (University of Rochester)
GenerationOptimizationComputational EfficiencyTransformerDiffusion modelRectified FlowGenerative Adversarial NetworkTime SeriesPhysics Related
🎯 What it does: A conditional diffusion model named Diff-PIC has been developed for efficiently generating 2D electric field snapshots in laser-plasma interaction (LPI), replacing traditional time-consuming PIC simulations.
Diff-Prompt: Diffusion-Driven Prompt Generator with Mask Supervision
Weicai Yan (Zhejiang University), Tao Jin (Zhejiang University)
SegmentationGenerationRetrievalTransformerPrompt EngineeringDiffusion modelImageMultimodality
🎯 What it does: This paper proposes a diffusion model-based prompt generator called Diff-Prompt, which generates rich and fine-grained input-specific prompts to assist in the efficient fine-tuning of pre-trained multimodal models.
Diff3DS: Generating View-Consistent 3D Sketch via Differentiable Curve Rendering
Yibo Zhang, Rui Ma
GenerationData SynthesisOptimizationDiffusion modelScore-based ModelImageText
🎯 What it does: Proposes Diff3DS, a differentiable rendering framework that generates perspective-consistent 3D sketches from text or a single image by optimizing 3D rational Bézier curves.
Difference-of-submodular Bregman Divergence
Masanari Kimura (University of Melbourne), Hideitsu Hino (Institute of Statistical Mathematics RIKEN AIP)
RetrievalOptimizationImagePoint Cloud
🎯 What it does: A Difference-of-Submodular Bregman Divergence (DBD) is proposed on discrete finite sets, and its parameterization is achieved through a learnable permutation-invariant neural network.
Differentiable and Learnable Wireless Simulation with Geometric Transformers
Thomas Hehn (Qualcomm AI Research), Johann Brehmer (Qualcomm AI Research)
TransformerDiffusion modelMesh
🎯 What it does: Designed and trained a fully learnable and differentiable wireless signal propagation simulator, Wi-GATr, which predicts radio frequency channel characteristics using 3D geometric meshes and antenna location information, and supports inverse localization and geometric reconstruction.
Differentiable Causal Discovery for Latent Hierarchical Causal Models
Parjanya Prajakta Prashant (University of California San Diego), Biwei Huang (University of California San Diego)
Auto EncoderImageTabular
🎯 What it does: A differentiable causal discovery method is proposed to recover the structure of nonlinear latent hierarchical causal models from observational data, along with identifiable theory and implementation algorithms.
Differentiable Integer Linear Programming
Zijie Geng (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
OptimizationGraph Neural Network
🎯 What it does: This paper proposes DiffILO, a differentiable integer linear programming method based on unsupervised learning, aimed at directly predicting feasible solutions for ILP.
Differentiable Optimization of Similarity Scores Between Models and Brains
Nathan Cloos (Massachusetts Institute of Technology), Christopher J Cueva
OptimizationBiomedical Data
🎯 What it does: This study investigates various commonly used model-brain similarity measures, revealing through differentiable gradient optimization that high scores do not necessarily represent the similarity of task-related information.
Differentiable Rule Induction from Raw Sequence Inputs
Kun Gao (Institute of High Performance Computing, Agency for Science Technology and Research), Yang Feng (Institute of High Performance Computing, Agency for Science Technology and Research)
ClassificationAnomaly DetectionAuto EncoderTime SeriesSequential
🎯 What it does: This study proposes NeurRL, a fully differentiable rule learning framework that can learn logical rules directly from raw sequences (time series, images unfolded into one-dimensional sequences), avoiding the label leakage problem caused by the need for pre-labeled feature tags in traditional neural-symbolic ILP.
Differential learning kinetics govern the transition from memorization to generalization during in-context learning
Alex Nguyen (Princeton University), Gautam Reddy (Princeton University)
TransformerSequential
🎯 What it does: This study investigates the mechanism of the transition from memorization to generalization in transformers within In-Context Learning (ICL), constructing a simplified single-layer Transformer and providing a theoretical explanation.
Differential Transformer
Tianzhu Ye (BNRist Tsinghua University), Furu Wei (Microsoft Research)
TransformerLarge Language ModelText
🎯 What it does: This paper proposes the Differential Transformer (DIFF Transformer), which eliminates attention noise by taking the difference of two softmax mappings in the attention mechanism, thereby enhancing the focus on important contexts.
Differentially Private Federated Learning with Time-Adaptive Privacy Spending
Shahrzad Kiani (University of Toronto), Franziska Boenisch (CISPA Helmholtz Center for Information Security)
Federated LearningSafty and PrivacyTabular
🎯 What it does: A time-adaptive differential privacy budget allocation framework (spend-as-you-go) is proposed in federated learning, allowing clients to save privacy budget in the early stages of training and increase spending later, thereby improving model accuracy while maintaining privacy constraints.
Differentially private learners for heterogeneous treatment effects
Maresa Schröder, Stefan Feuerriegel (Ludwig Maximilian University Munich)
Safty and PrivacyMeta LearningDrug DiscoveryBiomedical DataElectronic Health Records
🎯 What it does: The DP-CATE framework is proposed, achieving differential privacy estimation of heterogeneous treatment effects (CATE) under observational data, utilizing a Neyman-orthogonal meta-learner.
Differentially private optimization for non-decomposable objective functions
Weiwei Kong (Google Research), Mónica Ribero (Google Research)
OptimizationSafty and PrivacyConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A differentially private optimization algorithm Logit-DP is proposed for non-decomposable objective functions (such as contrastive loss), overcoming the issue of sensitivity increasing with n due to batch size in traditional DP-SGD.
Differentially Private Steering for Large Language Model Alignment
Anmol Goel (Ubiquitous Knowledge Processing Lab), Amartya Sanyal (University of Copenhagen)
OptimizationSafty and PrivacyLarge Language ModelPrompt EngineeringText
🎯 What it does: The research aligns the behavior of large language models using activation editing while maintaining differential privacy guarantees.
Differentiation and Specialization of Attention Heads via the Refined Local Learning Coefficient
George Wang (Timaeus), Daniel Murfet (University of Melbourne)
TransformerLarge Language ModelText
🎯 What it does: This paper quantifies and tracks the internal structural development of various attention heads in a two-layer attention model throughout the training process by proposing refined variants of the local learning coefficient (weight-refined LLC and data-refined LLC), revealing the differentiation, specialization of heads, and a new multigram circuit.
DiffGAD: A Diffusion-based Unsupervised Graph Anomaly Detector
Jinghan Li (University of Science and Technology of China), Xiang Wang (University of Science and Technology of China)
Anomaly DetectionGraph Neural NetworkDiffusion modelAuto EncoderGraph
🎯 What it does: This paper proposes an unsupervised graph anomaly detection method based on diffusion models, called DiffGAD. It utilizes a graph autoencoder to map the graph to a latent space, and then employs a diffusion model to reconstruct and discriminate the latent representations, thereby achieving node-level anomaly detection.
DiffPC: Diffusion-based High Perceptual Fidelity Image Compression with Semantic Refinement
Yichong Xia (Tsinghua University), Bin Chen (Harbin Institute of Technology)
CompressionDiffusion modelImage
🎯 What it does: A dual-stage image compression framework called DiffPC based on a pre-trained latent diffusion model is proposed for achieving high perceptual quality image reconstruction at low bit rates.
DiffPuter: Empowering Diffusion Models for Missing Data Imputation
Hengrui Zhang (University of Illinois at Chicago), Philip S. Yu (University of Illinois at Chicago)
Diffusion modelScore-based ModelTabular
🎯 What it does: An iterative missing value imputation method called DIFFPUTER is proposed, which combines the Expectation-Maximization (EM) algorithm and diffusion models to simultaneously estimate the complete data distribution and update missing values.
DiffSplat: Repurposing Image Diffusion Models for Scalable Gaussian Splat Generation
Chenguo Lin (Peking University), Yadong MU
GenerationData SynthesisDiffusion modelAuto EncoderGaussian SplattingImageTextPoint Cloud
🎯 What it does: This paper proposes a 3D generation framework called DIFFSPLAT, which utilizes a pre-trained 2D diffusion model to directly generate high-quality and efficient Gaussian splat representations in 3D. It is capable of synthesizing 3D content under multi-view consistency and text/image conditions.
Diffusing States and Matching Scores: A New Framework for Imitation Learning
Runzhe Wu (Cornell University), Wen Sun (Cornell University)
Robotic IntelligenceReinforcement LearningDiffusion modelSequential
🎯 What it does: This paper proposes a diffusion model-based imitation learning framework called SMILING, which directly compares the state distributions of experts and learners using score matching, thereby avoiding the training of adversarial discriminators.
Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning
Lequan Lin (University of Sydney), Junbin Gao (University of Sydney)
OptimizationHyperparameter SearchGraph Neural NetworkDiffusion modelAuto EncoderGraph
🎯 What it does: Utilizing a lightweight coarse-tuning search to collect sub-optimal model checkpoints, and then using a graph-conditioned latent diffusion model to generate better GNN parameters, thus achieving improved performance on various graph tasks with minimal hyperparameter tuning.
Diffusion Actor-Critic: Formulating Constrained Policy Iteration as Diffusion Noise Regression for Offline Reinforcement Learning
Linjiajie Fang (Hong Kong University of Science and Technology), Bingyi Jing
Reinforcement LearningDiffusion modelTabularBenchmark
🎯 What it does: Proposes the Diffusion Actor-Critic (DAC) framework, which utilizes diffusion models to directly learn the target policy in offline reinforcement learning and alternately trains with the Critic to complete KL-constrained policy iteration.
Diffusion Attribution Score: Evaluating Training Data Influence in Diffusion Models
Jinxu Lin (University of Sydney), Chang Xu (University of Sydney)
GenerationData SynthesisOptimizationDiffusion modelImage
🎯 What it does: This paper proposes the Diffusion Attribution Score (DAS) to evaluate the influence of each training sample on the generated results in diffusion models.
Diffusion Bridge AutoEncoders for Unsupervised Representation Learning
Yeongmin Kim (Korea Advanced Institute of Science and Technology), Il-chul Moon
GenerationRepresentation LearningDiffusion modelAuto EncoderImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposes Diffusion Bridge AutoEncoders (DBAE) for unsupervised representation learning, addressing the information splitting problem in diffusion models.
Diffusion Bridge Implicit Models
Kaiwen Zheng (Tsinghua University), Jun Zhu (Tsinghua University)
RestorationGenerationData SynthesisDiffusion modelImageOrdinary Differential Equation
🎯 What it does: Proposes Diffusion Bridge Implicit Models (DBIMs) for fast sampling of DDBM without additional training.
Diffusion Feedback Helps CLIP See Better
Wenxuan Wang (Institute of Automation Chinese Academy of Sciences), Xinlong Wang (Beijing Academy of Artificial Intelligence)
ClassificationSegmentationRetrievalOptimizationRepresentation LearningTransformerVision Language ModelDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: A post-training framework DIVA based on diffusion model feedback is proposed, which utilizes pure image data for self-supervised optimization of CLIP visual representations, significantly enhancing its fine-grained visual perception capabilities.
Diffusion Generative Modeling for Spatially Resolved Gene Expression Inference from Histology Images
Sichen Zhu (Georgia Institute of Technology), Peng Qiu (Georgia Institute of Technology)
GenerationDiffusion modelImageBiomedical Data
🎯 What it does: This paper presents Stem, a computational method based on conditional diffusion models, used to infer spatially resolved gene expression profiles from H&E tissue slice images.
Diffusion Models are Evolutionary Algorithms
Yanbo Zhang (Allen Discovery Center at Tufts University), Michael Levin (Wyss Institute for Biologically Inspired Engineering at Harvard University)
OptimizationReinforcement LearningDiffusion modelTabular
🎯 What it does: The paper proves that diffusion models are essentially a type of evolutionary algorithm, and based on this, proposes two evolutionary optimization methods: Diffusion Evolution and Latent Space Diffusion Evolution, which can discover diverse and high-quality solutions in multimodal and high-dimensional parameter spaces.
Diffusion Models Are Real-Time Game Engines
Dani Valevski (Google Research), Shlomi Fruchter (Google DeepMind)
GenerationReinforcement LearningDiffusion modelVideo
🎯 What it does: Developed GameNGen - a fully neural network-driven game engine capable of real-time simulation of the classic game DOOM, generating high-quality and interactive game visuals.
Diffusion Models as Cartoonists: The Curious Case of High Density Regions
Rafal Karczewski, Vikas Garg (Aalto University)
GenerationData SynthesisDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This study investigates high-density regions of diffusion models, proposing a theoretical model for tracking processes and a high-density sampler, demonstrating that images generated by high-density sampling are often cartoonish or blurry and yield higher likelihood values than those obtained through ordinary sampling.
Diffusion On Syntax Trees For Program Synthesis
Shreyas Kapur (University of California), Stuart Russell (University of California)
GenerationData SynthesisAI Code AssistantTransformerVision Language ModelDiffusion modelGraph
🎯 What it does: A diffusion model based on syntax trees is proposed for program synthesis, allowing for editing during the generation process and combining with search.
Diffusion Policy Policy Optimization
Allen Z. Ren (Princeton University), Max Simchowitz (Carnegie Mellon University)
OptimizationReinforcement LearningDiffusion modelSequential
🎯 What it does: Proposed and implemented the DPPO framework, which fine-tunes a pre-trained diffusion policy using policy gradients to address the instability issues of training diffusion policies in RL.
Diffusion State-Guided Projected Gradient for Inverse Problems
Rayhan Zirvi (California Institute of Technology), Anima Anandkumar (California Institute of Technology)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: Proposes the Diffusion State-Guided Projected Gradient (DiffStateGrad) method, which enhances data consistency sampling and reduces artifacts in inverse problems by projecting the measurement gradient onto the low-rank subspace of the current diffusion state.
Diffusion Transformer Captures Spatial-Temporal Dependencies: A Theory for Gaussian Process Data
Hengyu Fu (Stanford University), Minshuo Chen (Northwestern University)
TransformerDiffusion modelVideoTime Series
🎯 What it does: This study investigates the Diffusion Transformer in high-dimensional spatiotemporal sequence data, demonstrating its ability to effectively capture spatial-temporal correlations and providing a learning sample complexity.
Diffusion Transformers for Tabular Data Time Series Generation
Fabrizio Garuti (Prometeia Associazione), Rita Cucchiara (Prometeia SpA)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderTabularTime SeriesFinance Related
🎯 What it does: A Tabular data time series generation model called TabDiT based on diffusion transformers has been developed.
Diffusion-based Decoupled Deterministic and Uncertain Framework for Probabilistic Multivariate Time Series Forecasting
Qi Li (Beijing University of Posts and Telecommunications), Yong Zhang (Beijing University of Posts and Telecommunications)
Diffusion modelTime Series
🎯 What it does: A decoupled deterministic and uncertainty framework (D U 3) based on diffusion models is proposed, which extracts high-confidence components through a pre-trained point prediction model, while the remaining high-uncertainty components are modeled for probability distribution using conditional DDPM, achieving dual capabilities of point prediction and probability prediction for long-term multivariate time series forecasting.
Diffusion-based Neural Network Weights Generation
Bedionita Soro, Sung Ju Hwang (KAIST)
GenerationRetrievalMeta LearningDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a neural network weight generation framework based on diffusion models, D2NWG, which can generate network weights for specific tasks based on dataset or task descriptions without storing and searching through massive pre-trained models.
Diffusion-Based Planning for Autonomous Driving with Flexible Guidance
Yinan Zheng (Tsinghua University), Jingjing Liu (Tsinghua University)
Autonomous DrivingTransformerDiffusion modelScore-based ModelTime Series
🎯 What it does: A Transformer framework based on diffusion models, called Diffusion Planner, has been designed and implemented for closed-loop autonomous driving planning. It can jointly predict neighboring vehicle trajectories and generate multimodal driving trajectories, guiding customizable driving behaviors such as safety, comfort, and speed through an untrained classifier.
Diffusion-NPO: Negative Preference Optimization for Better Preference Aligned Generation of Diffusion Models
Fu-Yun Wang (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationOptimizationReinforcement LearningDiffusion modelImageVideo
🎯 What it does: This paper proposes and validates Negative Preference Optimization (NPO), which trains negative preference weights in diffusion models to enhance the alignment of generated results with human preferences.
Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models
Zeyu Yang (Fudan University), Li Zhang (Fudan University)
GenerationData SynthesisDiffusion modelScore-based ModelVideo
🎯 What it does: By combining a pre-trained video diffusion model and a multi-view diffusion model, dense multi-view multi-frame image matrices can be directly sampled without the need for additional training, thus achieving the generation of dynamic 3D content.
DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing
June Suk Choi (Korea Advanced Institute of Science and Technology), Kimin Lee (Korea Advanced Institute of Science and Technology)
Computational EfficiencyAdversarial AttackDiffusion modelImageBenchmark
🎯 What it does: A robust defense method for text-guided diffusion model image editing, called DiffusionGuard, is proposed.
Digi-Q: Learning VLM Q-Value Functions for Training Device-Control Agents
Hao Bai (University of Illinois Urbana-Champaign), Aviral Kumar (Carnegie Mellon University)
Robotic IntelligenceTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelTabular
🎯 What it does: The Digi-Q method is proposed, which utilizes intermediate features generated by VLM to train the action value function Q through TD learning on offline data, and implements policy extraction through Best-of-N re-ranking to build an efficient control agent for mobile devices.
Dimension Agnostic Neural Processes
Hyungi Lee (Korea Advanced Institute of Science and Technology), Juho Lee (Korea Advanced Institute of Science and Technology)
OptimizationMeta LearningTransformerImageVideo
🎯 What it does: A meta-learning model capable of directly handling different input/output dimensions is proposed—Dimension Agnostic Neural Process (DANP), which is applied to various regression, image/video inpainting, and Bayesian optimization tasks.
Direct Distributional Optimization for Provable Alignment of Diffusion Models
Ryotaro Kawata (University of Tokyo), Taiji Suzuki (University of Tokyo)
GenerationOptimizationDiffusion modelImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a distribution optimization-based diffusion model alignment method and provides rigorous convergence and sampling error theory.
Direct Post-Training Preference Alignment for Multi-Agent Motion Generation Model Using Implicit Feedback from Pre-training Demonstrations
Thomas Tian, Kratarth Goel (Waymo)
GenerationAutonomous DrivingReinforcement LearningContrastive LearningSequential
🎯 What it does: This paper proposes a method for directly fine-tuning preference alignment of multi-agent motion generation models using implicit preference feedback from pre-trained demonstrations, avoiding manual labeling and complex RL or reward learning.
Directional Gradient Projection for Robust Fine-Tuning of Foundation Models
Chengyue Huang (Georgia Institute of Technology), Zsolt Kira (Georgia Institute of Technology)
Domain AdaptationOptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A robust fine-tuning method based on gradient direction projection, DiGraP, is proposed, balancing the robustness of pre-trained models with the performance of downstream tasks.
Discovering Clone Negatives via Adaptive Contrastive Learning for Image-Text Matching
Renjie Pan (Shanghai Jiao Tong University), Hua Yang (Shanghai Jiao Tong University)
RetrievalContrastive LearningImageText
🎯 What it does: Proposes AdaCL, which improves contrastive learning for clone negative samples and enhances image-text matching performance.
Discovering Group Structures via Unitary Representation Learning
Dongsung Huh (IBM Research)
Representation LearningTransformerTabular
🎯 What it does: Through the differentiable tensor decomposition framework HyperCube, we automatically discover finite groups and their unitary representations using unit regularization, recovering the complete group operation from partial binary operation tables.
Discovering Influential Neuron Path in Vision Transformers
Yifan Wang (ShanghaiTech University), Kan Ren (ShanghaiTech University)
ClassificationExplainability and InterpretabilityTransformerImage
🎯 What it does: This paper proposes a 'Neuron Path' method based on hierarchical progressive search to identify the neuron paths that have the greatest impact on the inference results of visual Transformer models.
Discovering Temporally Compositional Neural Manifolds with Switching Infinite GPFA
Changmin Yu (University of Cambridge), Máté Lengyel (Central European University)
Time SeriesSequential
🎯 What it does: A fully Bayesian nonparametric extension model named infinite GPFA has been developed to simultaneously infer the number of latent factors and the activation states of these factors at each time point from neural spike data.
DiscoveryBench: Towards Data-Driven Discovery with Large Language Models
Bodhisattwa Prasad Majumder (Allen Institute for AI), Peter Clark (Allen Institute for AI)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes the DISCOVERYBENCH benchmark for evaluating the capabilities of large language models in automated data-driven scientific discovery (from data generation to hypothesis validation).
Discrete Codebook World Models for Continuous Control
Aidan Scannell (University of Edinburgh), Joni Pajarinen (Aalto University)
Robotic IntelligenceReinforcement LearningWorld ModelSequential
🎯 What it does: Designed and trained the DCWM (Discrete Codebook World Model) combined with MPPI decision-making planning, proposing the DC-MPC algorithm, which utilizes a trainable discrete latent space encoded with a discrete codebook to achieve model-based reinforcement learning for continuous control.
Discrete Copula Diffusion
Anji Liu (University of California), Guy Van den Broeck (University of Stuttgart)
GenerationOptimizationTransformerDiffusion modelText
🎯 What it does: This paper proposes a Discrete Copula Diffusion (DCD) method that integrates discrete diffusion models with autoregressive Copula models during the inference phase to compensate for the lack of interdependence caused by the independent treatment of variables in each step of the diffusion model.
Discrete Diffusion Schrödinger Bridge Matching for Graph Transformation
Jun Hyeong Kim (Korea Advanced Institute of Science and Technology), Woo Youn Kim (Korea Advanced Institute of Science and Technology)
GenerationOptimizationGraph Neural NetworkDiffusion modelGraph
🎯 What it does: A framework called Discrete Diffusion Schrödinger Bridge Matching (DDSBM) is proposed, which utilizes continuous-time Markov chains to solve the Schrödinger bridge problem in high-dimensional discrete spaces and applies it to molecular graph optimization.
Discrete Distribution Networks
Lei Yang (StepFun)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A Discrete Distribution Network (DDN) is proposed, which generates multiple samples layer by layer and selects the samples closest to the target to approximate the data distribution, thereby achieving gradient-free zero-shot conditional generation.
Discrete GCBF Proximal Policy Optimization for Multi-agent Safe Optimal Control
Songyuan Zhang (Massachusetts Institute of Technology), Chuchu Fan (Massachusetts Institute of Technology)
OptimizationSafty and PrivacyGraph Neural NetworkReinforcement LearningScore-based ModelTime Series
🎯 What it does: The DGPPO framework is proposed, which can learn distributed safety policies in discrete-time multi-agent systems, taking into account unknown dynamics, partial observability, neighborhood changes, and input constraints.
Discrete Latent Plans via Semantic Skill Abstractions
Haobin Jiang (Peking University), Zongqing Lu (Peking University)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerAuto EncoderContrastive LearningMultimodality
🎯 What it does: This paper proposes LADS, a hierarchical method for learning language-conditioned discrete implicit plans through semantic skill abstraction.
Discretization-invariance? On the Discretization Mismatch Errors in Neural Operators
Wenhan Gao (Stony Brook University), Yi Liu (Stony Brook University)
Convolutional Neural NetworkTabularPhysics Related
🎯 What it does: Theoretical and experimental research on discretization mismatch errors in neural operators is conducted, and a cross-resolution learning pipeline called CROP is proposed to eliminate this error.
Discriminating image representations with principal distortions
Jenelle Feather (Flatiron Institute), Eero P Simoncelli
Representation LearningConvolutional Neural NetworkImage
🎯 What it does: A new framework is proposed that uses the Fisher information matrix to describe the local geometry of image representations, and based on this metric, a 'principal distortion' is designed to compare the local geometries of multiple models;
Discriminator-Guided Embodied Planning for LLM Agent
Haofu Qian (Zhejiang University), Xuelong Li (Zhejiang University)
OptimizationRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningAgentic AIText
🎯 What it does: A discriminator-guided action optimization framework (DGAP) is proposed, which generates step-level scoring functions using a small number of demonstrations to guide LLM in generating plans that better align with expert actions.
Disentangled Representation Learning with the Gromov-Monge Gap
Théo Uscidda (CREST-ENSAE), marco cuturi
Representation LearningAuto EncoderImage
🎯 What it does: A regularization method based on Gromov-Monge distance, GMG, is proposed to achieve decomposable representation learning within an unsupervised VAE framework.
Disentangling 3D Animal Pose Dynamics with Scrubbed Conditional Latent Variables
Joshua Huang Wu, Timothy W DUNN
Pose EstimationAnomaly DetectionRepresentation LearningAuto EncoderGenerative Adversarial NetworkVideoBiomedical Data
🎯 What it does: This paper proposes a conditional variational autoencoder (SC-VAE) that removes noise variable information through 'scrubbing' and applies it to behavior representation learning, clustering, generation, and disease detection of 3D animal posture sequences.
Disentangling Representations through Multi-task Learning
Pantelis Vafidis (California Institute of Technology), Antonio Rangel (California Institute of Technology)
Representation LearningRecurrent Neural NetworkTransformerSupervised Fine-TuningSequential
🎯 What it does: This study demonstrates through multi-task learning that agents can learn decoupled representations when solving multi-task evidence accumulation classification tasks, which can capture the underlying structure of the data.
DisEnvisioner: Disentangled and Enriched Visual Prompt for Customized Image Generation
Jing He (Hong Kong University of Science and Technology), Ying-Cong Chen (Hong Kong University of Science and Technology)
GenerationData SynthesisPrompt EngineeringDiffusion modelImage
🎯 What it does: Proposes DisEnvisioner, a single-image tuning free personalized image generation framework based on visual decoupling and enhancement.
DiSK: Differentially Private Optimizer with Simplified Kalman Filter for Noise Reduction
Xinwei Zhang (University of Southern California), Vahab Mirrokni (Google Research)
OptimizationSafty and PrivacyConvolutional Neural NetworkTransformerSupervised Fine-TuningImageText
🎯 What it does: A differentially private optimizer based on simplified Kalman filtering (DiSK) is designed to significantly reduce the impact of DP noise on gradients during large-scale training.
DisPose: Disentangling Pose Guidance for Controllable Human Image Animation
Hongxiang Li (Peking University), Long Chen (Hong Kong University of Science and Technology)
GenerationPose EstimationDiffusion modelVideo
🎯 What it does: Proposes the DisPose plugin, which achieves plug-and-play control without dense input by decoupling skeletal poses into motion fields and keypoint correspondences, significantly improving the quality of human image animation.
Dissecting Adversarial Robustness of Multimodal LM Agents
Chen Henry Wu (Carnegie Mellon University), Aditi Raghunathan (Carnegie Mellon University)
Adversarial AttackTransformerLarge Language ModelAgentic AIPrompt EngineeringTextMultimodality
🎯 What it does: A systematic evaluation of the adversarial robustness of multimodal LM agents is conducted, constructing 200 VWA-Adv attack tasks and proposing the Agent Robustness Evaluation (ARE) framework, which models agents as information flow graphs to quantify attack propagation.
Dist Loss: Enhancing Regression in Few-Shot Region through Distribution Distance Constraint
Guangkun Nie (Peking University), Shenda Hong (Peking University)
ImageTime SeriesElectrocardiogram
🎯 What it does: Proposes Dist Loss, a loss function that simultaneously optimizes the distance between the predicted distribution and the label distribution in imbalanced regression tasks;
Distance-Based Tree-Sliced Wasserstein Distance
Hoang V. Tran, Tan Minh Nguyen
Image TranslationGenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a distance metric based on tree slicing—Distance-based Tree-Sliced Wasserstein (Db-TSW)—for comparing probability distributions in Euclidean space.
Distilled Decoding 1: One-step Sampling of Image Auto-regressive Models with Flow Matching
Enshu Liu (Tsinghua University), Zinan Lin (Microsoft Research)
GenerationComputational EfficiencyKnowledge DistillationFlow-based ModelImage
🎯 What it does: The researchers proposed the Distilled Decoding (DD) method, which allows pre-trained image autoregressive models to complete sampling in just one to two steps, significantly improving generation speed.
DistillHGNN: A Knowledge Distillation Approach for High-Speed Hypergraph Neural Networks
Saman Forouzandeh (RMIT University), Mahdi Jalili (RMIT University)
Computational EfficiencyKnowledge DistillationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A teacher-student knowledge distillation framework called DistillHGNN is designed and implemented, utilizing HGNN+MLP to generate soft labels and high-order structural information, training a lightweight TinyGCN+MLP student model, which significantly improves inference speed and memory efficiency while maintaining high accuracy.
Distilling Dataset into Neural Field
Donghyeok Shin (Korea Advanced Institute of Science and Technology), Il-chul Moon
Data SynthesisCompressionKnowledge DistillationNeural Radiance FieldImageVideoMultimodalityAudio
🎯 What it does: In the dataset distillation task, this paper proposes compressing large-scale datasets into small-scale synthetic data and parameterizing these synthetic instances through neural fields;
Distilling Reinforcement Learning Algorithms for In-Context Model-Based Planning
Jaehyeon Son (Seoul National University), Gunhee Kim (Seoul National University)
Knowledge DistillationMeta LearningTransformerReinforcement LearningSequential
🎯 What it does: A model is proposed that simultaneously learns environment dynamics and policies within a Transformer, and utilizes MPC for planning to achieve sample-efficient meta-RL without parameter updates.
Distilling Structural Representations into Protein Sequence Models
Jeffrey Ouyang-Zhang (University of Texas at Austin), Daniel Jesus Diaz
Knowledge DistillationProtein Structure PredictionGraph Neural NetworkTransformerAuto EncoderTextBiomedical Data
🎯 What it does: A protein language model called ISM is proposed, which only uses sequence input and can implicitly capture structural information, achieving better performance on various structure-related tasks.
Distributed Speculative Inference (DSI): Speculation Parallelism for Provably Faster Lossless Language Model Inference
Nadav Timor (Weizmann Institute of Science), David Harel (Weizmann Institute of Science)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A new distributed inference algorithm called Distributed Speculative Inference (DSI) is proposed, which breaks the sequential limitations of traditional Speculative Inference (SI) by parallelizing the verification process, achieving sustainable acceleration of language model inference in a multi-GPU environment.
Distribution Backtracking Builds A Faster Convergence Trajectory for Diffusion Distillation
Shengyuan Zhang (Zhejiang University), Lingyun Sun (Zhejiang University)
GenerationKnowledge DistillationDiffusion modelScore-based ModelImage
🎯 What it does: A distribution backtracking distillation (DisBack) method is proposed, which introduces the entire convergence trajectory of the teacher model into score distillation, thereby accelerating and improving the convergence speed and quality of single-step generative models.
Distribution-Free Data Uncertainty for Neural Network Regression
Domokos M. Kelen (HUN-REN SZTAKI), Andras A Benczur
OptimizationTabular
🎯 What it does: This paper proposes an unbiased sample approximation of the nonparametric Continuous Ranked Probability Score (CRPS) to train non-deterministic neural networks for direct sampling and learning arbitrary distributions in regression tasks, thereby achieving distribution-independent uncertainty quantification.
Distribution-Specific Agnostic Conditional Classification With Halfspaces
Jizhou Huang (Washington University in St. Louis), Brendan Juba (Washington University in St. Louis)
ClassificationOptimization
🎯 What it does: This study investigates the problem of agnostic conditional classification using half-space selectors under Gaussian feature distributions for a given finite or sparse set of linear classifiers. A polynomial-time algorithm is proposed along with an error upper bound, and it is proven that the computation is difficult in the general half-space case.
Distributional Associations vs In-Context Reasoning: A Study of Feed-forward and Attention Layers
Lei Chen (New York University), Alberto Bietti (Flatiron Institute)
TransformerLarge Language ModelText
🎯 What it does: This study investigates the different learning mechanisms of the feed-forward layer and the attention layer in Transformers regarding distributed associations and contextual reasoning, and explains their dynamics through experiments and theory.
DistRL: An Asynchronous Distributed Reinforcement Learning Framework for On-Device Control Agent
Taiyi Wang (University of Cambridge), Kun Shao (AI Centre)
Robotic IntelligenceTransformerLarge Language ModelReinforcement LearningMultimodality
🎯 What it does: Designed and implemented DISTRL, an asynchronous distributed reinforcement learning framework for efficiently fine-tuning multimodal large language models online on mobile devices, ultimately training device control agents with better control capabilities.
DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors
Keon Lee (KRAFTON), Jaewoong Cho (KRAFTON)
GenerationData SynthesisTransformerDiffusion modelAudio
🎯 What it does: This paper presents DiTTo-TTS, a text-to-speech model based on Latent Diffusion Transformer, capable of large-scale speech generation without the need for phonemes, duration, or other domain-specific factors.
Divergence of Neural Tangent Kernel in Classification Problems
Zixiong Yu (Huawei Technologies), Guhan Chen (Tsinghua University)
ClassificationTabular
🎯 What it does: This study investigates the convergence properties of the neural tangent kernel (NTK) of neural networks in binary classification tasks using cross-entropy loss, proving that the NTK is strictly positive definite but does not converge during training; instead, it diverges.
Divergence-enhanced Knowledge-guided Context Optimization for Visual-Language Prompt Tuning
Yilun Li (Capital Normal University), Wei Song (Capital Normal University)
ClassificationOptimizationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes knowledge-guided context optimization (DeKg) based on Hilbert–Schmidt independence criterion, which reduces bias towards pre-trained knowledge by regularizing the independence of learned prompts and pre-trained prompts, thereby enhancing the learning of task-specific knowledge in downstream tasks.
Divergence-Regularized Discounted Aggregation: Equilibrium Finding in Multiplayer Partially Observable Stochastic Games
Runyu Lu (University of Chinese Academy of Sciences), Dongbin Zhao (University of Chinese Academy of Sciences)
OptimizationReinforcement Learning
🎯 What it does: A multi-round learning framework DRDA based on discounted FTRL is proposed to solve partially observable stochastic games (POSG), and it is proven that its last iteration converges in a single round and that it converges to Nash equilibrium in multiple rounds.
Diverse Policies Recovering via Pointwise Mutual Information Weighted Imitation Learning
Hanlin Yang (Sun Yat-sen University), Haobo Fu (Tencent AI Lab)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: This paper proposes assigning weights to each (state, action) pair during the behavioral cloning process, using pointwise mutual information to measure its correlation with trajectory style, thereby achieving the recovery of diverse policies.
Diverse Preference Learning for Capabilities and Alignment
Stewart Slocum (Massachusetts Institute of Technology), Dylan Hadfield-Menell (Massachusetts Institute of Technology)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes Soft Preference Learning (SPL), which alleviates the mode collapse caused by RLHF/DPO by decomposing KL regularization into entropy and cross-entropy terms, enhancing the diversity and calibration of LLM outputs.