NeurIPS 2024 Papers — Page 9
Conference on Neural Information Processing Systems · 4035 papers
DataStealing: Steal Data from Diffusion Models in Federated Learning with Multiple Trojans
Yuan Gan (Zhejiang University), Yi Yang (Zhejiang University)
Federated LearningAdversarial AttackDiffusion modelImage
🎯 What it does: In the federated learning framework, multiple backdoors are implanted in the diffusion model using combination triggers (ComboTs) to achieve data stealing of a large amount of private image data from local clients.
DC-Gaussian: Improving 3D Gaussian Splatting for Reflective Dash Cam Videos
Linhan Wang (Virginia Tech), Chang-Tien Lu (Virginia Tech)
RestorationGenerationAutonomous DrivingGaussian SplattingVideo
🎯 What it does: A method called DC-Gaussian is proposed to generate high-fidelity new perspective renderings from in-vehicle dashcam videos, achieving the separation and elimination of occlusions (such as mirrors, phone mounts, stains, etc.) without reflections and obstructions.
DCDepth: Progressive Monocular Depth Estimation in Discrete Cosine Domain
Kun Wang (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Depth EstimationRecurrent Neural NetworkTransformerImage
🎯 What it does: A monocular depth estimation framework named DCDepth is proposed, which first performs staged predictions on different frequency coefficients in the DCT frequency domain of the depth map, and then reconstructs spatial domain depth through inverse DCT.
DDGS-CT: Direction-Disentangled Gaussian Splatting for Realistic Volume Rendering
Zhongpai Gao (United Imaging Intelligence), Ziyan Wu (United Imaging Intelligence)
GenerationOptimizationComputational EfficiencyGaussian SplattingBiomedical DataComputed Tomography
🎯 What it does: A direction-decoupled 3D Gaussian splatting (DDGS) model has been developed for efficiently generating realistic digitally reconstructed radiographs (DRR), balancing speed and accuracy, and applicable to tasks such as intraoperative 2D/3D image registration.
DDK: Distilling Domain Knowledge for Efficient Large Language Models
Jiaheng Liu (Taobao and Tmall Group of Alibaba), Bo Zheng (Taobao and Tmall Group of Alibaba)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelText
🎯 What it does: A dynamic distillation framework DDK based on domain knowledge is proposed for knowledge transfer between large language models (teachers) and small language models (students); by dynamically adjusting the domain mixing ratio of the distillation dataset, the student gains more training in domains with significant performance gaps.
DDN: Dual-domain Dynamic Normalization for Non-stationary Time Series Forecasting
Tao Dai (Shenzhen University), Zexuan Zhu (Tsinghua University)
Anomaly DetectionOptimizationTransformerTime Series
🎯 What it does: To address the non-stationarity problem in time series forecasting, a Dual-Domain Dynamic Normalization (DDN) module is proposed, which performs sliding normalization and denormalization on both the time domain and frequency domain sides, thereby achieving dynamic capture and correction of distribution drift.
DDR: Exploiting Deep Degradation Response as Flexible Image Descriptor
Juncheng Wu (Tongji University), Shiqi Wang (City University of Hong Kong)
RestorationSuper ResolutionPrompt EngineeringContrastive LearningImage
🎯 What it does: A deep feature-based metric for image degradation response, called DDR, is proposed as a flexible and adjustable image descriptor.
Dealing with Synthetic Data Contamination in Online Continual Learning
Maorong Wang (University of Tokyo), Toshihiko Yamasaki (University of Tokyo)
ClassificationData SynthesisDiffusion modelContrastive LearningImage
🎯 What it does: This study investigates the impact of synthetic image pollution on online continual learning and proposes a method called ESRM to mitigate this effect.
DeBaRA: Denoising-Based 3D Room Arrangement Generation
Léopold Maillard (École Polytechnique), Maks Ovsjanikov (École Polytechnique)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelScore-based ModelPoint Cloud
🎯 What it does: We propose DeBaRA, a score-based diffusion model designed to generate controllable, precise, and diverse 3D indoor layouts conditioned on a given room floor plan and a list of object categories, supporting downstream tasks such as rearrangement and completion.
Debiasing Synthetic Data Generated by Deep Generative Models
Alexander Decruyenaere (Ghent University Hospital), Stijn Vansteelandt (Ghent University)
GenerationData SynthesisGenerative Adversarial NetworkTabular
🎯 What it does: A post-processing strategy for debiasing synthetic data generated by deep generative models (DGM) is proposed, aiming to ensure that the statistical inferences (such as mean, linear regression coefficients) from the synthetic data are consistent with those from real data.
Decentralized Noncooperative Games with Coupled Decision-Dependent Distributions
Wenjing Yan (Hong Kong University of Science and Technology), Xuanyu Cao (Hong Kong University of Science and Technology)
OptimizationTabular
🎯 What it does: The study investigates distributional adaptation caused by decision-making in decentralized non-cooperative games, deriving the existence and uniqueness of PSE and NE, and providing an upper bound on their distance.
Decision Mamba: A Multi-Grained State Space Model with Self-Evolution Regularization for Offline RL
Qi Lv (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
TransformerReinforcement LearningSequential
🎯 What it does: This paper presents Decision Mamba, a multi-granularity state space model that combines self-evolutionary strategy learning for offline reinforcement learning, addressing issues such as insufficient utilization of historical information, neglect of local relationships, and overfitting to noisy trajectories.
Decision Mamba: Reinforcement Learning via Hybrid Selective Sequence Modeling
Sili Huang (Jilin University), Bo Yang (Lehigh University)
TransformerReinforcement LearningSequential
🎯 What it does: A reinforcement learning framework called Decision Mamba-Hybrid (DM-H) is proposed, which is based on a hybrid of Mamba and Transformer, achieving self-improvement in long-term memory and high-quality decision-making through sub-goals.
Decision-Focused Learning with Directional Gradients
Michael Huang (CUNY Baruch Zicklin School of Business), Vishal Gupta (USC Marshall School of Business)
OptimizationTabularFinance Related
🎯 What it does: A class of Perturbation Gradient (PG) loss is proposed for the predict-then-optimize framework, which directly optimizes decision performance through gradient descent in a data-driven manner.
Decision-Making Behavior Evaluation Framework for LLMs under Uncertain Context
Jingru Jia (University of Illinois at Urbana-Champaign), Deming Chen (University of Illinois at Urbana-Champaign)
TransformerLarge Language ModelTextFinance Related
🎯 What it does: A behavioral economics-based evaluation framework is proposed, utilizing multi-round multi-choice experiments to measure the decision-making behavior of large language models in uncertain environments, including risk preference, probability weighting, and loss aversion, and to study the impact of demographic characteristics on model behavior.
Decoding-Time Language Model Alignment with Multiple Objectives
Ruizhe Shi (Tsinghua University), Simon Shaolei Du
OptimizationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes a training-free algorithm called MOD (Multi-Objective Decoding) that can instantaneously align language models to multiple objectives during inference, achieving fine control over arbitrary objective weights through a linear combination of the predicted probabilities of different single-objective models.
Decomposable Transformer Point Processes
Aristeidis Panos (University of Cambridge)
TransformerTime SeriesSequential
🎯 What it does: A point process model that separately models event types and time intervals has been designed and implemented, using Transformer to model the label distribution and a mixed log-normal distribution to model time intervals, completely eliminating the dependence on thinning algorithms, achieving efficient and parallel inference.
Decompose, Analyze and Rethink: Solving Intricate Problems with Human-like Reasoning Cycle
Shangzi Xue (University of Science and Technology of China), Qi Liu (University of Science and Technology of China)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: The DeAR framework is proposed, which constructs a reasoning tree within large language models and generates, verifies, and updates the reasoning process layer by layer through the Decompose-Analyze-Rethink cycle, thereby achieving a more human-like solution to complex problems.
Decomposed Prompt Decision Transformer for Efficient Unseen Task Generalization
Hongling Zheng (Wuhan University), Dacheng Tao (Nanyang Technological University)
Meta LearningTransformerLarge Language ModelReinforcement LearningPrompt EngineeringSequential
🎯 What it does: A multi-task offline reinforcement learning framework DPDT based on Prompt is proposed, which uses pre-trained language model parameters for initialization and achieves efficient generalization to unseen tasks through prompt decomposition (cross-task and task-specific) and test-time alignment (TTA).
Decomposing and Interpreting Image Representations via Text in ViTs Beyond CLIP
Sriram Balasubramanian (University of Maryland), Soheil Feizi (University of Maryland)
RetrievalExplainability and InterpretabilityTransformerContrastive LearningImageText
🎯 What it does: A general framework is proposed to automatically decompose the final representation of Vision Transformers into the contributions of various components (such as attention heads and MLPs) and map these contributions to the CLIP space through linear mapping for text interpretation. A continuous scoring function is designed to evaluate the importance of components for different image features.
Decoupled Kullback-Leibler Divergence Loss
Jiequan Cui (Nanyang Technological University), Hanwang Zhang (Nanyang Technological University)
OptimizationKnowledge DistillationAdversarial AttackImage
🎯 What it does: A gradient optimization analysis of KL divergence loss is conducted, and an improved IKL loss is proposed to enhance adversarial robustness and knowledge distillation performance.
Decoupling Semantic Similarity from Spatial Alignment for Neural Networks.
Tassilo Wald (German Cancer Research Center), Klaus Maier-Hein
RetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This study investigates the metric of representational similarity in visual neural networks and proposes a semantic RSM that removes the influence of spatial alignment.
DECRL: A Deep Evolutionary Clustering Jointed Temporal Knowledge Graph Representation Learning Approach
Qian Chen (Zhejiang University), Ling Chen (Zhejiang University)
Representation LearningGraph Neural NetworkGraphTime Series
🎯 What it does: The DECRL framework is proposed, which implements event prediction of higher-order correlations evolving over time through deep evolutionary clustering combined with T-KG representation learning.
Deep Bayesian Active Learning for Preference Modeling in Large Language Models
Luckeciano Carvalho Melo, Yarin Gal (University of Oxford)
Recommendation SystemOptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A new Bayesian active learning method, BAL-PM, is proposed for efficiently collecting human preference labels in large-scale language models, significantly reducing the amount of required annotations.
Deep Correlated Prompting for Visual Recognition with Missing Modalities
Lianyu Hu (Tianjin University), Liang Wan (Tianjin University)
ClassificationRecognitionTransformerPrompt EngineeringMultimodality
🎯 What it does: A Deep Correlated Prompting method is designed to fine-tune large multimodal Transformers (CLIP) with learnable prompts to maintain model robustness in the absence of any modality.
Deep Equilibrium Algorithmic Reasoning
Dobrik Georgiev Georgiev, Pietro Lio
Graph Neural NetworkGraphBenchmark
🎯 What it does: This paper studies the application of Deep Equilibrium Networks (DEQ) to neural algorithm reasoning, directly solving the equilibrium points of graph neural networks to complete algorithm execution without needing to know the number of steps in the algorithm.
Deep Graph Mating
Yongcheng Jing (University of Sydney), Dacheng Tao (Nanyang Technological University)
Knowledge DistillationGraph Neural NetworkGraph
🎯 What it does: The Deep Graph Mating (GRAMA) task is proposed, which can generate sub-models from multiple pre-trained parent graph neural networks without retraining and without labels, integrating their knowledge.
Deep Graph Neural Networks via Posteriori-Sampling-based Node-Adaptative Residual Module
Jingbo Zhou (Westlake University), Stan Z. Li (Emory University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a posterior sampling node adaptive residual module (PSNR) to alleviate the over-smoothing problem that occurs in deep Graph Neural Networks (GNNs);
Deep Homomorphism Networks
Takanori Maehara (Roku), Hoang NT (University of Tokyo)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a Deep Homomorphism Network (DHN) based on graph homomorphism, which achieves the expression of graph structures through a learnable nonlinear transformation and aggregation of the graph homomorphism numbers of a predefined set of subgraph patterns P;
Deep Learning for Computing Convergence Rates of Markov Chains
Yanlin Qu (Stanford University), Peter Glynn (Stanford University)
Supervised Fine-TuningTabular
🎯 What it does: Using deep learning methods (Deep Contractive Drift Calculator, DCDC) to solve the Contract Drift Equation (CDE), thereby providing a computable upper bound on the convergence rate of the Wasserstein distance for general state space Markov chains.
Deep Learning in Medical Image Registration: Magic or Mirage?
Rohit Jena (University of Pennsylvania), James Gee
Domain AdaptationOptimizationImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper systematically compares the performance and generalization ability of classical optimization registration methods with deep learning registration (DLIR) methods in medical image registration, focusing on the differences between unsupervised and supervised training as well as domain transfer.
Deep Learning Through A Telescoping Lens: A Simple Model Provides Empirical Insights On Grokking, Gradient Boosting & Beyond
Alan Jeffares (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
ImageTabular
🎯 What it does: A 'telescope' model based on stepwise linear approximation has been constructed to explain various anomalous phenomena in deep learning.
Deep linear networks for regression are implicitly regularized towards flat minima
Pierre Marion (Institute of Mathematics EPFL), Lénaïc Chizat (Institute of Mathematics EPFL)
Optimization
🎯 What it does: This study investigates the optimization dynamics of deep linear networks in univariate regression, analyzing the maximum eigenvalue of the Hessian of the minimizer (sharpness) and proving that gradient flow implicitly regularizes towards flat local minima under two initializations (small scale and residual initialization).
Deep Policy Gradient Methods Without Batch Updates, Target Networks, or Replay Buffers
Gautham Vasan (University of Alberta), A. Rupam Mahmood (University of Alberta)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: An incremental deep policy gradient algorithm called AVG is proposed, which does not require experience replay, target networks, or batch updates, enabling real-time learning on resource-constrained robots.
Deep Submodular Peripteral Networks
Gantavya Bhatt (University of Washington), Jeff Bilmes
OptimizationFederated LearningKnowledge DistillationImage
🎯 What it does: Design and train a deep submodular function (DSPN) to learn submodularity through pairwise comparisons with numerical levels (GPC) and apply it to experimental design and streaming sampling;
Deep Support Vectors
Junhoo Lee (Seoul National University), Nojun Kwak (Seoul National University)
GenerationExplainability and InterpretabilityKnowledge DistillationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: By generalizing the KKT conditions of SVM to DeepKKT, deep support vectors (DSVs) are introduced, utilizing pre-trained models to extract or generate DSVs without a training set, and achieving model reconstruction, few-shot data distillation, and transforming classifiers into high-fidelity generative models; at the same time, DSVs are used for visual interpretation of model decision boundaries.
DeepDRK: Deep Dependency Regularized Knockoff for Feature Selection
Hongyu Shen (University of Illinois at Urbana Champaign), Zhizhen Zhao (University of Illinois at Urbana Champaign)
TransformerGenerative Adversarial NetworkBiomedical Data
🎯 What it does: This paper proposes DeepDRK, a deep knockoff generation framework that utilizes Transformer and multi-source adversarial training, and subsequently enhances retrieval effectiveness through regularization perturbations.
DeepITE: Designing Variational Graph Autoencoders for Intervention Target Estimation
Hongyuan Tao (Ant Group), Jianguo Li (Ant Group)
Graph Neural NetworkAuto EncoderGraph
🎯 What it does: This paper proposes DeepITE, a framework based on Variational Graph Autoencoders (VGAE) for collaboratively learning multiple causal graphs and various intervention targets, allowing for the rapid identification of new intervention targets without the need for retraining during inference.
DeepLag: Discovering Deep Lagrangian Dynamics for Intuitive Fluid Prediction
Qilong Ma (Tsinghua University), Mingsheng Long (Tsinghua University)
Convolutional Neural NetworkTransformerTime SeriesSequential
🎯 What it does: By integrating Eulerian grid features with Lagrangian particle tracking, DeepLag is proposed to achieve autoregressive prediction of fluids.
DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs
Lingchen Meng (Fudan University), Yu-Gang Jiang (Fudan University)
RecognitionGenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes DeepStack, a strategy for hierarchically injecting visual tokens into the decoder of large language models (LLMs), which maintains the context length while significantly enhancing multimodal performance.
DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution
Yang Yue (Tsinghua University), Gao Huang (Tsinghua University)
Computational EfficiencyRobotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelTextMultimodalityBenchmark
🎯 What it does: A dynamic early stopping multimodal large language model (DeeR) is designed, which adaptively adjusts the model size based on the robot operation scenario through multi-level intermediate exits, significantly reducing computational and memory requirements.
Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models
Yimeng Zhang (Michigan State University), Sijia Liu (MIT IBM Watson AI Lab)
GenerationOptimizationAdversarial AttackDiffusion modelGenerative Adversarial NetworkImageText
🎯 What it does: A robust machine unlearning framework called AdvUnlearn based on adversarial training (AT) is proposed to safely erase harmful concepts in diffusion models and resist adversarial prompt attacks.
DeformableTST: Transformer for Time Series Forecasting without Over-reliance on Patching
Donghao Luo (Tsinghua University), Xue Wang (Tsinghua University)
TransformerTime Series
🎯 What it does: DeformableTST is proposed, a Transformer structure with optional patching, achieving time series prediction through deformable attention and hierarchical structure.
DEFT: Efficient Fine-tuning of Diffusion Models by Learning the Generalised $h$-transform
Alexander Denker (University College London), Pietro Lio (University of Cambridge)
RestorationGenerationSuper ResolutionDiffusion modelImageComputed Tomography
🎯 What it does: A unified conditional diffusion framework based on Doob h-transformation is proposed, and within this framework, the DEFT method is designed, utilizing fine-tuned minimal networks for efficient conditional generation.
DeiSAM: Segment Anything with Deictic Prompting
Hikaru Shindo (Technical University of Darmstadt), Kristian Kersting (Technical University of Darmstadt)
Object DetectionSegmentationTransformerLarge Language ModelPrompt EngineeringImage
🎯 What it does: This paper proposes DeiSAM, which combines large-scale pre-trained models with differentiable logical reasoning to achieve image segmentation based on complex instructive natural language prompts.
Déjà Vu Memorization in Vision–Language Models
Bargav Jayaraman (Meta), Kamalika Chaudhuri (Meta)
Object DetectionRetrievalTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: A method is proposed that evaluates the memory of visual-language models for training image details based on the comparison between target and reference models, utilizing kNN retrieval.
DEL: Discrete Element Learner for Learning 3D Particle Dynamics with Neural Rendering
Jiaxu Wang (Hong Kong University of Science and Technology), Renjing Xu (Hong Kong University of Science and Technology)
Graph Neural NetworkNeural Radiance FieldImagePhysics Related
🎯 What it does: A Discrete Element Learner (DEL) based on discrete element analysis and graph neural networks is proposed, which learns three-dimensional particle dynamics from two-dimensional images through inverse rendering.
Delta-CoMe: Training-Free Delta-Compression with Mixed-Precision for Large Language Models
Bowen Ping (Peking University), Maosong Sun (Tsinghua University)
CompressionTransformerLarge Language ModelTextMultimodality
🎯 What it does: This paper studies an untrained mixed-precision Delta compression method called Delta-CoMe, which is used to compress the incremental weights of fine-tuned LLMs, supporting multi-model multi-tenant deployment.
DeltaDEQ: Exploiting Heterogeneous Convergence for Accelerating Deep Equilibrium Iterations
Zuowen Wang (Institute of Neuroinformatics University of Zurich and ETH Zurich), Shih-Chii Liu (Institute of Neuroinformatics University of Zurich and ETH Zurich)
OptimizationComputational EfficiencyOptical FlowImage
🎯 What it does: The DeltaDEQ method is proposed, which significantly reduces inference computation by using the delta update rule in the fixed-point iteration of the Deep Equilibrium Model (DEQ) to skip converged dimensions.
DeltaDock: A Unified Framework for Accurate, Efficient, and Physically Reliable Molecular Docking
Jiaxian Yan (University of Science and Technology of China), Qi Liu (University of Science and Technology of China)
Drug DiscoveryGraph Neural NetworkContrastive LearningGraph
🎯 What it does: The DeltaDock two-stage framework is proposed, which first selects the best pocket through pocket-ligand alignment, and then uses a dual-layer coarse-to-fine iteration to complete blind docking and pose docking.
Delving into the Reversal Curse: How Far Can Large Language Models Generalize?
Zhengkai Lin (Zhejiang University), Jieping Ye (Alibaba Cloud)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper conducts experiments on large language models across various tasks (open-ended question answering and multiple choice) to study and reveal the so-called 'reversal curse' and the model's 'thinking bias', and explores the impact of training data structure on knowledge transfer and generalization ability.
DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States
Bozhou Zhang (Fudan University), Li Zhang (Fudan University)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: The DeMo framework is proposed, which splits trajectory prediction queries into pattern queries (capturing directional intent) and state queries (tracking dynamic states), and achieves more accurate multimodal trajectory prediction through a hybrid attention mechanism and the Mamba module.
Demystify Mamba in Vision: A Linear Attention Perspective
Dongchen Han (Tsinghua University), Gao Huang (Tsinghua University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: The core mechanisms of Mamba and linear attention Transformers are theoretically compared and experimentally validated, and based on this, the Mamba-Inspired Linear Attention (MILA) model is proposed.
Dendritic Integration Inspired Artificial Neural Networks Capture Data Correlation
Chongming Liu (Shanghai Jiao Tong University), Douglas Zhou (Shanghai Jiao Tong University)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: The study is based on the second-order neuron with dendritic second integral rules, and constructs the Dit-CNNs model to enhance the generalization and few-shot learning capabilities of neural networks.
DeNetDM: Debiasing by Network Depth Modulation
Silpa Vadakkeeveetil Sreelatha (University of Surrey), Anjan Dutta (University of Surrey)
Knowledge DistillationImage
🎯 What it does: This paper proposes a debiasing method called DeNetDM that does not require bias labels or data augmentation, achieving suppression of spurious correlations by adjusting the network depth.
DenoiseRep: Denoising Model for Representation Learning
zhengrui Xu, Jitao Sang (Beijing Jiaotong University)
RecognitionObject DetectionSegmentationRepresentation LearningConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: This paper proposes a framework called DenoiseRep that integrates the denoising process of diffusion models into feature extraction to enhance the feature representation capability for discriminative tasks.
Denoising Diffusion Path: Attribution Noise Reduction with An Auxiliary Diffusion Model
Yiming Lei (Fudan University), Hongming Shan (Fudan University)
Explainability and InterpretabilityDiffusion modelImage
🎯 What it does: A path based on the denoising diffusion model (DDPath) is proposed to improve path-based attribution methods (such as Integrated Gradients) to reduce noise and enhance interpretability.
Dense Associative Memory Through the Lens of Random Features
Benjamin Hoover (IBM Research), Dmitry Krotov (IBM Research)
RetrievalOptimizationImage
🎯 What it does: This paper proposes a distributed Dense Associative Memory (DrDAM) that utilizes random features to achieve scalable memory storage and retrieval with a fixed number of parameters.
Dense Connector for MLLMs
Huanjin Yao (Baidu Inc.), Jingdong Wang (Chinese University of Hong Kong)
RecognitionGenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the Dense Connector plugin, which enhances the visual perception capabilities of multimodal large language models by utilizing multi-layer features from visual encoders.
DenseFormer: Enhancing Information Flow in Transformers via Depth Weighted Averaging
Matteo Pagliardini (Ecole Polytechnique Federale de Lausanne), Martin Jaggi (Ecole Polytechnique Federale de Lausanne)
TransformerLarge Language ModelText
🎯 What it does: DenseFormer builds on the Transformer architecture by adding a Depth-Weighted-Average (DWA) module after each Transformer block to enhance information flow and improve the model's expressive capability.
Density-based User Representation using Gaussian Process Regression for Multi-interest Personalized Retrieval
Haolun Wu (McGill University), MARYAM KARIMZADEHGAN
RetrievalRecommendation SystemTabular
🎯 What it does: A density-based user representation method GPR4DUR based on Gaussian process regression is proposed for multi-interest personalized retrieval.
DePLM: Denoising Protein Language Models for Property Optimization
Zeyuan Wang (Zhejiang University), Huajun Chen (Zhejiang University)
OptimizationDrug DiscoveryProtein Structure PredictionTransformerDiffusion modelBiomedical Data
🎯 What it does: DePLM is proposed to remove noise from evolutionary information through denoising diffusion in the rank space, thereby predicting the effects of protein mutations on function more accurately.
DEPrune: Depth-wise Separable Convolution Pruning for Maximizing GPU Parallelism
Cheonjun Park (Samsung Electronics), Won Woo Ro (Yonsei University)
OptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A sparsification method for depthwise separable convolution (DSConv) called DEPrune is proposed, which achieves acceleration through fine-grained pruning of depthwise convolution layers.
Depth Anything V2
Lihe Yang (HKU), Hengshuang Zhao (HKU)
Depth EstimationKnowledge DistillationTransformerImage
🎯 What it does: This paper presents Depth Anything V2, which constructs a multi-scale monocular depth estimation foundational model. It achieves more refined and robust depth predictions by utilizing accurate synthetic data, an expanded teacher model, and a large number of pseudo-labeled real images.
Depth Anywhere: Enhancing 360 Monocular Depth Estimation via Perspective Distillation and Unlabeled Data Augmentation
Ning-Hsu Wang, Yu-Lun Liu (National Yang Ming Chiao Tung University)
Depth EstimationKnowledge DistillationImage
🎯 What it does: This paper proposes the use of a perspective depth model to generate pseudo-labels, combined with cube projection, random rotation, and affine-invariant loss, for semi-supervised joint training of 360-degree monocular depth estimation, significantly improving depth prediction accuracy.
Derandomizing Multi-Distribution Learning
Kasper Green Larsen (Aarhus University), Nikita Zhivotovskiy (University of California)
🎯 What it does: This paper studies the problem of converting random predictors into deterministic predictors in multi-distribution learning and proposes an algorithm for achieving deterministic learning under label-consistent distributions. It also proves that such derandomization is computationally difficult in general cases.
Derivative-enhanced Deep Operator Network
Yuan Qiu (Georgia Institute of Technology), Peng Chen (Georgia Institute of Technology)
Auto EncoderPhysics Related
🎯 What it does: This paper proposes the Derivative-enhanced Deep Operator Network (DE-DeepONet), which improves the accuracy of predicting parameter functions and their derivatives by incorporating derivative information and dimensionality reduction into DeepONet.
Derivatives of Stochastic Gradient Descent in parametric optimization
Franck Iutzeler (Université Paul Sabatier), Samuel Vaiter (CNRS)
OptimizationTabular
🎯 What it does: This study investigates the evolution of the derivative of parameters during the iterations of stochastic gradient descent (SGD) in strongly convex parameterized optimization problems, proving its convergence to the derivative of the optimal solution mapping. It provides convergence rates and error bounds for three scenarios: constant step size, diminishing step size, and interpolation (σ=0).
Designing Cell-Type-Specific Promoter Sequences Using Conservative Model-Based Optimization
Aniketh Janardhan Reddy (University of California), Nilah M Ioannidis
OptimizationTransformerSupervised Fine-TuningBiomedical Data
🎯 What it does: This paper proposes an offline model-driven optimization (MBO) framework based on Conservative Objective Models (COMs) for the efficient design of cell type-specific promoters targeting similar leukemia cell lines (Jurkat, K562, THP1) under data-limited conditions.
Designs for Enabling Collaboration in Human-Machine Teaming via Interactive and Explainable Systems
Rohan R Paleja, Matthew Gombolay
Explainability and InterpretabilityReinforcement Learning
🎯 What it does: This study investigates an artificial intelligence teammate based on an interpretable tree model and achieves iterative human-machine collaboration through interactive rewriting of the tree structure.
DeSparsify: Adversarial Attack Against Token Sparsification Mechanisms
Oryan Yehezkel (Ben Gurion University of the Negev), Asaf Shabtai (Ben Gurion University of the Negev)
Computational EfficiencyAdversarial AttackTransformerImage
🎯 What it does: For visual Transformers using the token sparsification (TS) mechanism, a DeSparsify attack is proposed, which utilizes a custom loss function to disable the TS mechanism, resulting in a decrease in model usability.
DETAIL: Task DEmonsTration Attribution for Interpretable In-context Learning
Zijian Zhou (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes a task demonstration attribution method called DETAIL based on influence functions, aimed at explaining and evaluating the impact of demonstration texts on model predictions in the context of non-parametric updates.
Detecting and Measuring Confounding Using Causal Mechanism Shifts
Abbavaram Gowtham Reddy (Indian Institute of Technology Hyderabad), Vineeth N. Balasubramanian
Graph
🎯 What it does: Three metrics for measuring confounding (CNF-1, CNF-2, CNF-3) in multiple environments have been proposed, which can distinguish between observed and unobserved confounding and assess the strength of confounding. These metrics are applicable in scenarios where different environmental information and causal paths are known or unknown.
Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers
Jonas Ngnawe, Christian Gagné (Laval University)
ClassificationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: The study introduces the concept of 'margin consistency' in robustly trained deep classifiers, utilizing logit margins as an approximation of input space margins to detect the vulnerability of individual samples, and enhances the detection performance of models with insufficient margin consistency by learning pseudo-margins.
Detecting Bugs with Substantial Monetary Consequences by LLM and Rule-based Reasoning
Brian Zhang (University of Texas at Austin), ZHUO ZHANG
Anomaly DetectionOptimizationTransformerLarge Language ModelPrompt EngineeringTextFinance Related
🎯 What it does: This paper designs and implements a hybrid system called ABAUDITOR, which combines large language models (LLM) with rule-based reasoning to automatically detect accounting error vulnerabilities in smart contracts.
DeTeCtive: Detecting AI-generated Text via Multi-Level Contrastive Learning
Xun Guo (ByteDance), Chongyang Ma (University of Chinese Academy of Sciences)
ClassificationRetrievalDomain AdaptationTransformerContrastive LearningText
🎯 What it does: Proposes the DeTeCtive framework, which utilizes multi-task assisted multi-level contrastive learning and KNN retrieval to achieve AI-generated text detection.
Deterministic Policies for Constrained Reinforcement Learning in Polynomial Time
Jeremy McMahan (University of Wisconsin Madison)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes a novel FPTAS algorithm that can solve deterministic reinforcement learning (CRL) problems subject to arbitrary time-space recursive (TSR) constraints in polynomial time, addressing the long-standing approximability issues related to deterministic expected constraints, almost certain constraints, and constraints at arbitrary moments.
Deterministic Uncertainty Propagation for Improved Model-Based Offline Reinforcement Learning
Abdullah Akgül (University of Southern Denmark), Melih Kandemir (University of Southern Denmark)
Reinforcement LearningTabular
🎯 What it does: The MOMBO algorithm is proposed, which utilizes a deterministic moment matching method to propagate the uncertainty of the environment model through the Q-network, thereby enabling more accurate Bellman targets for conservative value iteration in model-based offline RL.
DeTikZify: Synthesizing Graphics Programs for Scientific Figures and Sketches with TikZ
Jonas Belouadi (University of Mannheim), Steffen Eger (University of Technology Nuremberg)
GenerationData SynthesisOptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningImageTextMultimodality
🎯 What it does: This paper presents DeTi k Zify, a multimodal language model capable of automatically generating TikZ graphic programs based on sketches or existing images, achieving high-quality reproduction of scientific graphics and sketch transformation.
DeTrack: In-model Latent Denoising Learning for Visual Object Tracking
Xinyu Zhou (Fudan University), Wenqiang Zhang (Fudan University)
Object TrackingTransformerDiffusion modelImageVideo
🎯 What it does: Introducing noise box training in visual object tracking, achieving a tracker with single forward inference by denoising block by block within the model.
DEX: Data Channel Extension for Efficient CNN Inference on Tiny AI Accelerators
Taesik Gong (UNIST), Chulhong Min (Nokia Bell Labs)
Computational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: On small AI accelerators, the spatial information of the original image is expanded into additional channels through patch-wise even sampling and channel-wise stacking, rewriting the input of the first layer CNN, thereby making full use of unused processors and memory, improving inference accuracy without increasing latency.
DFA-GNN: Forward Learning of Graph Neural Networks by Direct Feedback Alignment
Gongpei Zhao (Beijing Jiaotong University), Haibin Ling (Stony Brook University)
Graph Neural NetworkGraph
🎯 What it does: A forward learning framework for graph neural networks (GNN) based on Direct Feedback Alignment (DFA), called DFA-GNN, is proposed to address the unsuitability of traditional backpropagation (BP) for graph data.
DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization
Yueming Xu (Fudan University), Li Zhang (Fudan University)
Pose EstimationOptimizationGaussian SplattingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper presents DG-SLAM, a real-time SLAM system for dynamic environments based on 3D Gaussian splatting, achieving high-precision camera pose estimation and high-quality map reconstruction.
DHA: Learning Decoupled-Head Attention from Transformer Checkpoints via Adaptive Heads Fusion
Yilong Chen (Institute of Information Engineering, Chinese Academy of Sciences), Yu Sun (Baidu Inc.)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: The multi-head attention (MHA) of large models is transformed into decoupled head attention (DHA) through an adaptive head fusion method, significantly reducing KV cache usage and inference costs while achieving efficient inference without compromising performance.
DI-MaskDINO: A Joint Object Detection and Instance Segmentation Model
Zhixiong Nan (Chongqing University), Jifeng Dai (Tsinghua University)
Object DetectionSegmentationAutonomous DrivingTransformerImage
🎯 What it does: This paper analyzes the performance imbalance phenomenon of detection and instance segmentation tasks in the first layer of the Transformer decoder in MaskDINO, and proposes the DI-MaskDINO model to alleviate this imbalance and enhance the overall performance of joint detection and instance segmentation.
Diff-eRank: A Novel Rank-Based Metric for Evaluating Large Language Models
Lai Wei (Shanghai Jiao Tong University), Weiran Huang (Shanghai Jiao Tong University)
TransformerLarge Language ModelTextMultimodality
🎯 What it does: This paper proposes an intrinsic evaluation metric based on effective rank difference (Diff-eRank) to quantify the denoising capability of LLMs in training.
DiffAug: A Diffuse-and-Denoise Augmentation for Training Robust Classifiers
Chandramouli Shama Sastry (Dalhousie University), Sageev Oore (Dalhousie University)
ClassificationDomain AdaptationConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: A single-step diffusion-denoising data augmentation method based on diffusion models (DiffAug) is proposed and applied to train image classifiers to enhance robustness.
DiffCut: Catalyzing Zero-Shot Semantic Segmentation with Diffusion Features and Recursive Normalized Cut
Paul Couairon (Sorbonne Université), Nicolas THOME
SegmentationDiffusion modelImage
🎯 What it does: Utilizing the final self-attention features of the diffusion UNet encoder, combined with Recursive Normalized Cut to achieve unsupervised zero-shot semantic segmentation.
Diffeomorphic interpolation for efficient persistence-based topological optimization
Mathieu Carrière (Inria d'Université Côte d'Azur), Théo Lacombe (Université Gustave Eiffel)
OptimizationAuto EncoderPoint Cloud
🎯 What it does: This paper proposes a method to extend sparse topological gradients into a globally smooth vector field through differentiable homeomorphic interpolation. Based on this, a gradient descent algorithm that can be combined with subsampling is designed for topological optimization of large-scale point clouds and topological regularization of the latent space in black-box autoencoders.
Differentiable Modal Synthesis for Physical Modeling of Planar String Sound and Motion Simulation
Jin Woo Lee (Seoul National University), Kyogu Lee (Seoul National University)
GenerationData SynthesisNeural Radiance FieldPhysics RelatedAudio
🎯 What it does: A differentiable modal synthesis model is studied to simulate the spatiotemporal motion and sound of planar nonlinear string bodies.
Differentiable Quantum Computing for Large-scale Linear Control
Connor Clayton (University of Maryland), Xiaodi Wu (University of Maryland)
OptimizationReinforcement LearningPhysics Related
🎯 What it does: An end-to-end quantum algorithm is proposed for solving large linear quadratic regulator (LQR) problems.
Differentiable Structure Learning with Partial Orders
Taiyu Ban (University of Science and Technology of China), Huanhuan Chen (University of Science and Technology of China)
GraphBiomedical Data
🎯 What it does: Incorporate partial order constraints into the differentiable structure learning framework and propose a pluggable module for continuous optimization.
Differentiable Task Graph Learning: Procedural Activity Representation and Online Mistake Detection from Egocentric Videos
Luigi Seminara (University of Catania), Antonino Furnari (University of Catania)
Anomaly DetectionOptimizationRepresentation LearningTransformerSupervised Fine-TuningVideo
🎯 What it does: This paper directly optimizes the edge weights of the task graph through maximum likelihood estimation for differentiable optimization, proposing the TGML loss function, which can learn interpretable task graphs in neural networks through gradient descent.
Differential Privacy in Scalable General Kernel Learning via $K$-means Nystr{\"o}m Random Features
Bonwoo Lee (KAIST), Cheolwoo Park (KAIST)
OptimizationSafty and PrivacyGaussian SplattingTabular
🎯 What it does: A scalable differential privacy kernel learning framework based on K-means Nyström approximation is proposed, compatible with any kernel and achieving DP ERM, KME, and general data publishing.
Differentially Private Equivalence Testing for Continuous Distributions and Applications
Or Sheffet (Bar-Ilan University), Daniel Omer (Bar-Ilan University)
Safty and PrivacyComputational Efficiency
🎯 What it does: This paper proposes the first differential privacy equivalence (similarity) detection algorithm for continuous distributions, which can determine whether two continuous distributions are equal or differ by at least α in the A_k-norm.
Differentially Private Graph Diffusion with Applications in Personalized PageRanks
Rongzhe Wei (Georgia Institute of Technology), Pan Li (Georgia Institute of Technology)
Recommendation SystemSafty and PrivacyGraph Neural NetworkGraph
🎯 What it does: A framework for edge-level differential privacy that injects Laplace noise during the graph diffusion process is proposed, combined with a degree-based threshold function and ∞-Wasserstein distance tracking to achieve privacy protection for personalized PageRank.
Differentially Private Optimization with Sparse Gradients
Badih Ghazi (Google Research), Pasin Manurangsi (Google Research)
OptimizationSafty and Privacy
🎯 What it does: This paper studies the differential privacy (DP) optimization problem under gradient sparsity (where each sample has at most s non-zero components), providing mean estimation, approximate optimal error rates for convex and non-convex objectives, and proposing a stochastic gradient descent (SGD) algorithm with bias reduction.
Differentially Private Reinforcement Learning with Self-Play
Dan Qiao (University of California San Diego), Yu-Xiang Wang (University of California San Diego)
Autonomous DrivingReinforcement LearningFinance Related
🎯 What it does: This paper studies the multi-agent reinforcement learning problem under differential privacy constraints and proposes a new algorithm, DP-Nash-VI, which can effectively perform self-play while satisfying differential privacy.
Differentially Private Set Representations
Sarvar Patel (Google), Kevin Yeo (Columbia University)
Safty and Privacy
🎯 What it does: This paper studies the problem of representing a set of size k from the large universe in a differentially private manner, proposing two construction methods that achieve (ϵ, δ)-DP and pure ϵ-DP representations.
Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement
Jeremiah Birrell (Texas State University), Jason Pacheco (University of Arizona)
OptimizationSafty and PrivacyImage
🎯 What it does: A new differential privacy stochastic gradient descent (DP-SGD) accounting method is proposed, with a comprehensive R' enyi differential privacy (RDP) analysis for fixed-size subsampling (FSwoR and FSwR).