ICLR 2024 Papers — Page 6
International Conference on Learning Representations · 2260 papers
DiffEnc: Variational Diffusion with a Learned Encoder
Beatrix Miranda Ginn Nielsen (Technical University of Denmark), Ole Winther (Technical University of Denmark)
GenerationDiffusion modelImage
🎯 What it does: This paper proposes DiffEnc—a variational diffusion model that incorporates a time-dependent encoder into the diffusion process, maintaining the sampling time while enhancing the model's flexibility.
Diffeomorphic Mesh Deformation via Efficient Optimal Transport for Cortical Surface Reconstruction
Thanh Tung Le, Xiaohui Xie (University of California)
RestorationSegmentationOptimizationMeshBiomedical DataMagnetic Resonance ImagingAlzheimer's DiseaseOrdinary Differential Equation
🎯 What it does: A mesh reconstruction network based on differential homeomorphic deformation is proposed, utilizing probability measures and sliced Wasserstein distance for high-precision, non-self-intersecting reconstruction of the cortical surface;
Differentiable Euler Characteristic Transforms for Shape Classification
Ernst Röell (Helmholtz Munich), Bastian Rieck (Helmholtz Munich)
ClassificationOptimizationRepresentation LearningGraph Neural NetworkPoint CloudMeshGraph
🎯 What it does: A differentiable Euler Characteristic Transform (DECT) is proposed, enabling end-to-end learning and optimization of shape features.
Differentiable Learning of Generalized Structured Matrices for Efficient Deep Neural Networks
Changwoo Lee (University of Michigan), Hun-Seok Kim (University of Michigan)
OptimizationComputational EfficiencyTransformerSupervised Fine-TuningImageText
🎯 What it does: This paper proposes a differentiable general structured matrix format—Generalized Block-Low-Rank (GBLR), and learns its width and position parameters through gradient descent, aiming to achieve efficient deep neural networks with lower multiplication costs.
Differentially Private SGD Without Clipping Bias: An Error-Feedback Approach
Xinwei Zhang (University of Minnesota), Mingyi Hong (University of Minnesota)
OptimizationSafty and PrivacyTransformerSupervised Fine-TuningImageText
🎯 What it does: A differential privacy SGD algorithm named DiceSGD is proposed, which eliminates the constant bias introduced by gradient clipping by incorporating an error feedback mechanism into the gradient clipping process while maintaining privacy guarantees.
Differentially Private Synthetic Data via Foundation Model APIs 1: Images
Zinan Lin (Microsoft Research), Sergey Yekhanin (Microsoft Research)
Data SynthesisSafty and PrivacyDiffusion modelImageBiomedical Data
🎯 What it does: Using the inference API of large pre-trained models, a training-free differential privacy synthetic data generation framework called Private Evolution (PE) is proposed, which generates synthetic samples that closely resemble the distribution of private data through gradual evolution.
DIFFTACTILE: A Physics-based Differentiable Tactile Simulator for Contact-rich Robotic Manipulation
Zilin Si (Carnegie Mellon University), Chuang Gan (Massachusetts Institute of Technology)
Robotic IntelligenceReinforcement LearningImagePhysics Related
🎯 What it does: A differentiable physics-based haptic simulator DIFFTACTILE has been designed and implemented, supporting soft sensors, objects made of different materials, and differentiable contact dynamics, while also providing optical response simulation.
Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization
Dinghuai Zhang (Mila - Quebec AI Institute), Yoshua Bengio (Mila - Quebec AI Institute)
GenerationOptimizationDiffusion modelFlow-based ModelTabularBenchmarkStochastic Differential Equation
🎯 What it does: This paper proposes a new framework for sampling using diffusion models under known unnormalized target densities—Diffusion Generative Flow Sampler (DGFS). By introducing a 'flow function' at each step of the diffusion process to approximate the marginal distribution of that step, and utilizing Subtrajectory Balance or local learning signals, the model can update parameters based solely on incomplete trajectories, significantly reducing gradient variance and improving credit assignment.
Diffusion in Diffusion: Cyclic One-Way Diffusion for Text-Vision-Conditioned Generation
Ruoyu Wang (Wuhan University), Yu Wu (Wuhan University)
GenerationData SynthesisDiffusion modelImageTextMultimodalityOrdinary Differential Equation
🎯 What it does: A training-independent Circular One-Way Diffusion (COW) method is proposed, achieving high-fidelity generation of visual and textual conditions by injecting visual conditions into a pre-trained diffusion model and cyclically perturbing the reconstruction.
Diffusion Model for Dense Matching
Jisu Nam (Korea University), Seungryong Kim (KT)
Image TranslationSuper ResolutionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a dense correspondence framework called DiffMatch based on a conditional diffusion model, aimed at estimating pixel-level correspondences between paired images.
Diffusion Models for Multi-Task Generative Modeling
Changyou Chen (University at Buffalo), Belinda Zeng (Amazon)
RestorationSegmentationGenerationData SynthesisDiffusion modelImageMultimodality
🎯 What it does: This paper proposes a multimodal diffusion model (MT-Diffusion) that can simultaneously learn and generate various modalities of data (such as images, labels, semantic segmentation masks, etc.) within a unified diffusion space, and achieves cross-modal information sharing through multi-task loss.
Diffusion Posterior Sampling for Linear Inverse Problem Solving: A Filtering Perspective
Zehao Dou (Yale University), Yang Song (OpenAI)
RestorationSuper ResolutionDiffusion modelScore-based ModelImage
🎯 What it does: This paper proposes a method called Filtering Posterior Sampling (FPS) for zero-shot solutions to linear inverse problems using prior information from diffusion models without retraining the diffusion model.
Diffusion Sampling with Momentum for Mitigating Divergence Artifacts
Suttisak Wizadwongsa (Vistec), Supasorn Suwajanakorn (Vistec)
GenerationData SynthesisOptimizationDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes embedding Heavy Ball momentum into numerical solvers and extending it to a higher-order Generalized Heavy Ball (GHVB) method to alleviate divergence artifacts in low-step diffusion sampling.
Diffusion-TS: Interpretable Diffusion for General Time Series Generation
Xinyu Yuan (Hefei University of Technology), Yan Qiao (Hefei University of Technology)
GenerationData SynthesisExplainability and InterpretabilityTransformerDiffusion modelTime SeriesFinance Related
🎯 What it does: A time series generation framework based on diffusion models, Diffusion-TS, is proposed, utilizing a transformer architecture and decoupled seasonal-trend-error decomposition to achieve high-quality synthesis of multivariate long-term time series.
DiffusionNAG: Predictor-guided Neural Architecture Generation with Diffusion Models
Sohyun An (Korea Advanced Institute of Science and Technology), Sung Ju Hwang (Korea Advanced Institute of Science and Technology)
OptimizationNeural Architecture SearchTransformerDiffusion modelScore-based ModelImageStochastic Differential Equation
🎯 What it does: This paper presents DiffusionNAG—a conditional neural architecture generation framework based on diffusion models, which can directly guide the generation of neural network architectures that meet target attributes (such as high accuracy and low latency) using a predictor.
DiffusionSat: A Generative Foundation Model for Satellite Imagery
Samar Khanna (Stanford University), Stefano Ermon (Stanford University)
RestorationGenerationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes DiffusionSat, a foundational model for satellite image generation and inverse problem solving based on Stable Diffusion.
DiLu: A Knowledge-Driven Approach to Autonomous Driving with Large Language Models
Licheng Wen (Shanghai Artificial Intelligence Laboratory), Yu Qiao (East China Normal University)
Autonomous DrivingTransformerLarge Language ModelReinforcement LearningSequentialChain-of-Thought
🎯 What it does: Proposed and implemented a knowledge-driven autonomous driving framework DiLu, utilizing large language models (LLM) for reasoning and reflection, and accumulating experience through a memory module;
Directly Fine-Tuning Diffusion Models on Differentiable Rewards
Kevin Clark (Google DeepMind), David J. Fleet (Google DeepMind)
GenerationOptimizationReinforcement LearningDiffusion modelImage
🎯 What it does: This paper proposes Direct Reward Fine-Tuning (DRaFT) and its variants (DRaFTK, DRaFT-LV), which directly maximize a differentiable reward function by performing complete or truncated backpropagation on the diffusion sampling chain, with model parameters using LoRA;
Dirichlet-based Per-Sample Weighting by Transition Matrix for Noisy Label Learning
HeeSun Bae (KAIST), Il-chul Moon
ClassificationImage
🎯 What it does: This paper proposes a resampling method called RENT based on the noise transfer matrix to improve the learning effectiveness with noisy labels.
Discovering Failure Modes of Text-guided Diffusion Models via Adversarial Search
Qihao Liu (Johns Hopkins University), Alan Yuille (Johns Hopkins University)
GenerationData SynthesisOptimizationExplainability and InterpretabilityAdversarial AttackTransformerLarge Language ModelDiffusion modelGenerative Adversarial NetworkImageText
🎯 What it does: This paper proposes the SAGE method, which uses adversarial search to automatically discover failure modes of text-guided diffusion models in text, token, and latent spaces.
Discovering modular solutions that generalize compositionally
Simon Schug (ETH Zurich), Angelika Steger (ETH Zurich)
Knowledge DistillationMeta Learning
🎯 What it does: This paper studies the method of achieving compositional generalization through modular hypernetworks in multi-task learning, using a teacher-student framework to validate theory and practice.
Discovering Temporally-Aware Reinforcement Learning Algorithms
Matthew Thomas Jackson (University of Oxford), Jakob Nicolaus Foerster (University of Oxford)
OptimizationMeta LearningReinforcement LearningSequential
🎯 What it does: A time-aware RL objective function is proposed, which adapts to different training durations by incorporating training cycle information into the learning algorithm.
DisenBooth: Identity-Preserving Disentangled Tuning for Subject-Driven Text-to-Image Generation
Hong Chen (Tsinghua University), Wenwu Zhu (Tsinghua University)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningImageText
🎯 What it does: Proposes the DisenBooth framework, which utilizes identity-preserving embeddings and identity-agnostic embeddings for decoupled fine-tuning, achieving theme-driven text-to-image generation.
Disentangling Time Series Representations via Contrastive Independence-of-Support on l-Variational Inference
Khalid Oublal (Institute Polytechnique de Paris), François Roueff (OneTech TotalEnergies)
Representation LearningRecurrent Neural NetworkAuto EncoderContrastive LearningTime Series
🎯 What it does: This paper proposes a recognizable decoupled representation learning framework for time series, capable of separating the power consumption contributions of different appliances from device power consumption data.
Dissecting learning and forgetting in language model finetuning
Xiao Zhang (Tsinghua University), Ji Wu (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningTextBiomedical Data
🎯 What it does: Decomposes the learning and forgetting processes of domain fine-tuned language models, assessing the variations in biases related to themes, styles, and factual knowledge.
Dissecting Sample Hardness: A Fine-Grained Analysis of Hardness Characterization Methods for Data-Centric AI
Nabeel Seedat (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
Data-Centric LearningConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: A fine-grained hardness type classification system is proposed, and a unified evaluation framework H-CAT is developed to quantitatively assess the performance of 13 hardness characterization methods across 8 hardness types.
DistillSpec: Improving Speculative Decoding via Knowledge Distillation
Yongchao Zhou (University of Toronto), Rishabh Agarwal (Google DeepMind)
GenerationComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Improving the alignment between the small model (draft) and the large model (target) in Speculative Decoding through knowledge distillation, significantly enhancing inference speed while maintaining generation quality.
Distinguished In Uniform: Self-Attention Vs. Virtual Nodes
Eran Rosenbluth (RWTH Aachen University), Martin Grohe (RWTH Aachen University)
Graph Neural NetworkTransformerGraph
🎯 What it does: This paper theoretically proves and experimentally compares the unified expressive power of Graph Transformer (GT) and Message Passing Graph Neural Network with Virtual Nodes (MPGNN+VN), finding that neither is a global universal approximator, and each cannot fully simulate the functions of the other.
Distributional Preference Learning: Understanding and Accounting for Hidden Context in RLHF
Anand Siththaranjan (University of California), Dylan Hadfield-Menell (Massachusetts Institute of Technology)
Adversarial AttackReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningText
🎯 What it does: The study investigates how preference learning aggregates information in the presence of hidden contexts (such as diverse preferences, competing objectives, and unconscious factors) and proposes Distributional Preference Learning (DPL) to explicitly model the effects of hidden contexts; it also validates the mitigation effect of DPL on 'jailbreak' attacks in the RLHF tasks of large language models.
Distributionally Robust Optimization with Bias and Variance Reduction
Ronak Mehta (University of Washington), Zaid Harchaoui (University of Washington)
OptimizationTabular
🎯 What it does: A stochastic gradient algorithm named Prospect is proposed for solving distributionally robust optimization problems with spectral risk measures (such as CVaR, extremile, ESRM);
DittoGym: Learning to Control Soft Shape-Shifting Robots
Suning Huang (Tsinghua University), Vincent Sitzmann (Massachusetts Institute of Technology)
Robotic IntelligenceReinforcement LearningMeshBenchmark
🎯 What it does: The study focuses on soft robots that can dynamically change their shape during their lifespan and proposes a control framework based on reinforcement learning.
Diverse Projection Ensembles for Distributional Reinforcement Learning
Moritz Akiya Zanger, Matthijs T. J. Spaan (Delft University of Technology)
Reinforcement LearningTabular
🎯 What it does: This paper proposes the construction of a distributed reinforcement learning projection ensemble using multiple distribution projections (classification and quantiles) to enhance exploration and uncertainty estimation.
Divide and not forget: Ensemble of selectively trained experts in Continual Learning
Grzegorz Rypeść (Warsaw University of Technology), Bartłomiej Twardowski (Universitat Autònoma de Barcelona)
ClassificationKnowledge DistillationConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: This paper proposes a sample-free continuous learning method called SEED, which utilizes a limited number of expert networks and fine-tunes a single expert for each new task, thereby reducing forgetting and promoting expert diversity.
Diving Segmentation Model into Pixels
Chen Gan (Nanjing University), Junfeng Zhang (Nanjing University)
SegmentationDomain AdaptationContrastive LearningImage
🎯 What it does: A pixel learning-based semantic segmentation framework called PiXL is proposed, which utilizes pixel-level subdomain partitioning, prototype generation, and asymmetric contrastive learning to achieve fine-grained alignment for pixel-level variance.
DMBP: Diffusion model-based predictor for robust offline reinforcement learning against state observation perturbations
Zhihe YANG, Yunjian Xu (Chinese University of Hong Kong)
Reinforcement LearningDiffusion modelTabular
🎯 What it does: A diffusion model-based predictor (DMBP) is proposed to denoise state observation disturbances in offline reinforcement learning, thereby enhancing policy robustness.
DMV3D: Denoising Multi-view Diffusion Using 3D Large Reconstruction Model
Yinghao Xu (Adobe Research), Kai Zhang (Adobe Research)
RestorationGenerationData SynthesisTransformerDiffusion modelNeural Radiance FieldImageText
🎯 What it does: A single-stage 3D generative model DMV3D has been developed, achieving rapid 3D generation through multi-view image denoising and NeRF reconstruction.
DNA-GPT: Divergent N-Gram Analysis for Training-Free Detection of GPT-Generated Text
Xianjun Yang (University of California), Haifeng Chen (NEC Laboratories America)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This paper proposes a zero-training, interpretable GPT text detection method called DNA-GPT, which utilizes text truncation and regeneration, and then judges whether the text is generated by LLM based on n-gram similarity or probability differences.
DNABERT-2: Efficient Foundation Model and Benchmark For Multi-Species Genomes
Zhihan Zhou (Northwestern University), Han Liu (Northwestern University)
TransformerSupervised Fine-TuningBiomedical DataBenchmark
🎯 What it does: An efficient foundational model for multi-species genome pre-training, DNABERT-2, has been constructed, with improvements such as BPE tokenization and ALiBi positional encoding.
Do Generated Data Always Help Contrastive Learning?
Yifei Wang (Peking University), Yisen Wang (Peking University)
Data SynthesisRepresentation LearningDiffusion modelContrastive LearningImage
🎯 What it does: The study investigates the role of generated data in contrastive learning, finding that generated data can sometimes harm performance, and proposes an adaptive data augmentation strategy called AdaInf.
Does CLIP’s generalization performance mainly stem from high train-test similarity?
Prasanna Mayilvahanan (University of Tübingen), Wieland Brendel (Tübingen AI Center)
ClassificationDomain AdaptationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: The paper quantifies the perceptual similarity between the CLIP training set and the OOD benchmark set, and based on this, conducts data pruning experiments to test whether high training-testing similarity is the main reason for CLIP's outstanding performance on OOD.
Does Progress On Object Recognition Benchmarks Improve Generalization on Crowdsourced, Global Data?
Megan Richards (Meta AI), Mark Ibrahim (Meta AI)
RecognitionDomain AdaptationImageBenchmark
🎯 What it does: A systematic evaluation of nearly 100 visual models on global crowdsourced datasets reveals a significant gap in progress between standard ImageNet advancements and global distribution transfer performance, as well as an expansion of regional disparities.
Does Writing with Language Models Reduce Content Diversity?
Vishakh Padmakumar (New York University), He He (New York University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Evaluate the impact of different LLMs (GPT-3, InstructGPT) on the diversity and homogenization of argumentative writing in controlled experiments, and quantify the contributions of model text and user text to diversity.
DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models
Yung-Sung Chuang (Massachusetts Institute of Technology), Pengcheng He (Microsoft)
GenerationTransformerLarge Language ModelContrastive LearningText
🎯 What it does: A decoding strategy called DoLa is proposed, which does not require retrieval or fine-tuning, to enhance the factuality of large language models by comparing the logits of different Transformer layers.
Domain constraints improve risk prediction when outcome data is missing
Sidhika Balachandar (Cornell University), Emma Pierson (Cornell University)
Biomedical DataElectronic Health Records
🎯 What it does: A Bayesian model is proposed to estimate disease risk and evaluate human testing decisions in selective labeling scenarios (only observing the results of tested subjects).
Domain Randomization via Entropy Maximization
Gabriele Tiboni (Politecnico di Torino), Georgia Chalvatzaki (University of Wrzburg)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: The DORAEMON method is proposed, which automatically enhances the cross-domain generalization ability of reinforcement learning strategies through entropy maximization of dynamic parameter distributions in simulations.
Domain-Agnostic Molecular Generation with Chemical Feedback
Yin Fang (Zhejiang University), Huajun Chen (Zhejiang University)
GenerationDrug DiscoveryTransformerAuto EncoderGraph
🎯 What it does: A molecular pre-trained language model called MOLGEN based on SELFIES was constructed and trained, guided by a chemical feedback mechanism for molecular generation.
Domain-Inspired Sharpness-Aware Minimization Under Domain Shifts
Ruipeng Zhang (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Domain AdaptationOptimizationConvolutional Neural NetworkTransformerContrastive LearningImageBenchmark
🎯 What it does: This paper proposes Domain-Inspired Sharpness-Aware Minimization (DISAM), which enhances convergence speed and generalization performance in multi-source domain learning with domain shift by introducing domain-level convergence consistency constraints in the perturbation generation of SAM.
Don't Judge by the Look: Towards Motion Coherent Video Representation
Yitian Zhang (Northeastern University), Yun Fu (Northeastern University)
ClassificationRecognitionRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningVideo
🎯 What it does: This paper proposes a Motion Coherent Augmentation (MCA) for video understanding, which achieves hue perturbation in the RGB space through an efficient SwapMix operation and incorporates Variation Alignment (VA) to align predictions of different appearances of the same video, encouraging the model to focus on motion information rather than static appearance.
Don't Play Favorites: Minority Guidance for Diffusion Models
Soobin Um (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This study investigates the problem of generating low-density (minority) samples using diffusion models and proposes a controllable sampling framework.
Don't Trust: Verify -- Grounding LLM Quantitative Reasoning with Autoformalization
Jin Peng Zhou (Cornell University), Yuhuai Wu (xAI)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Utilizing large language models to automatically translate natural language math problems and their solutions into Isabelle form, and using automated theorem provers to verify and filter out correct answers.
DORSal: Diffusion for Object-centric Representations of Scenes $\textit{et al.}$
Allan Jabri (University of California Berkeley), Thomas Kipf (Google DeepMind)
GenerationData SynthesisDiffusion modelImageVideo
🎯 What it does: Combining frozen object center slots with video diffusion networks to generate editable new perspectives of 3D scenes;
DOS: Diverse Outlier Sampling for Out-of-Distribution Detection
Wenyu Jiang (Southern University of Science and Technology), Hongxin Wei (Southern University of Science and Technology)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A diversity external sample sampling strategy named DOS is designed and implemented to select external samples that can construct a global compact discriminative boundary during the training process, thereby enhancing the model's ability to detect out-of-distribution (OOD) data.
Doubly Robust Instance-Reweighted Adversarial Training
Daouda Sow, Yingbin Liang (Ohio State University)
OptimizationAdversarial AttackImage
🎯 What it does: Proposed a dual robust instance-weighted adversarial training framework and implemented the corresponding algorithm.
Doubly Robust Proximal Causal Learning for Continuous Treatments
Yong Wu (Fudan University), Xinwei Sun (Fudan University)
Generative Adversarial NetworkTabular
🎯 What it does: A kernel-based double robust estimator is proposed to estimate the causal effects of continuous treatments within the proximal causal learning framework.
DP-OPT: Make Large Language Model Your Privacy-Preserving Prompt Engineer
Junyuan Hong (University of Texas at Austin), Zhangyang Wang (University of Chicago)
ClassificationSafty and PrivacyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This work proposes Differentially-Private Offsite Prompt Tuning (DP-OPT), an end-to-end privacy-preserving framework that trains discrete prompts using private data locally and infers on cloud LLMs;
DP-SGD Without Clipping: The Lipschitz Neural Network Way
Louis Béthune (IRIT Université Paul Sabatier), David Vigouroux (IRT Saint Exupéry)
OptimizationSafty and PrivacyConvolutional Neural NetworkImageTabular
🎯 What it does: Proposes Clipless DP-SGD, which utilizes Lipschitz constraints on parameters to provide an analytically upper bound on gradient norms, thereby eliminating the clipping process for each sample gradient in traditional DP-SGD.
DQ-LoRe: Dual Queries with Low Rank Approximation Re-ranking for In-Context Learning
Jing Xiong (Sun Yat-Sen University), Xiaodan Liang (Shenzhen MSU-BIT University)
RetrievalTransformerLarge Language ModelTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: A framework named DQ-LoRe is proposed, which first allows the LLM to generate Chain-of-Thought (CoT), and then uses the CoT along with the question to query the retriever, automatically selecting examples used in In-Context Learning (ICL).
DragonDiffusion: Enabling Drag-style Manipulation on Diffusion Models
Chong Mou (Peking University), Jian Zhang (Peking University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: We propose DragonDiffusion, a method for drag-and-drop image editing that utilizes the correspondence of image features in pre-trained diffusion models, supporting tasks such as object movement, scaling, replacement, and pasting without the need for additional training.
DREAM: Dual Structured Exploration with Mixup for Open-set Graph Domain Adaption
Nan Yin (Mohamed bin Zayed University of Artificial Intelligence), Xiao Luo (University of California)
Domain AdaptationGraph Neural NetworkGraph
🎯 What it does: This paper proposes the task of Open-Set Graph Domain Adaptation for graph data and presents a DREAM framework based on a dual-branch approach (graph-level representation and subgraph enhanced representation) combined with Mixup.
DreamClean: Restoring Clean Image Using Deep Diffusion Prior
Jie Xiao (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationDiffusion modelImageOrdinary Differential Equation
🎯 What it does: This paper proposes an unsupervised, training-free image restoration framework called DreamClean, which utilizes the prior of diffusion models to directly restore clear images from degraded images.
DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior
Jingxiang Sun (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationData SynthesisDiffusion modelScore-based ModelImageMesh
🎯 What it does: Introducing DreamCraft3D: It first generates high-quality 2D images, then hierarchically sculpts geometry and enhances textures, ultimately producing globally consistent and detail-rich 3D objects.
DreamFlow: High-quality text-to-3D generation by Approximating Probability Flow
Kyungmin Lee (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImageTextOrdinary Differential Equation
🎯 What it does: A text-to-3D generation method called DreamFlow based on probabilistic flow ODE is proposed, employing a coarse-to-fine three-stage optimization framework to achieve high-quality, high-resolution (1024×1024) 3D content generation.
DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation
Jiaxiang Tang (Peking University), Gang Zeng (Baidu Inc.)
GenerationData SynthesisComputational EfficiencyDiffusion modelScore-based ModelGaussian SplattingImageTextMesh
🎯 What it does: A generative model based on 3D Gaussian splatting has been implemented, capable of quickly generating high-quality textured meshes from a single image or text.
DreamLLM: Synergistic Multimodal Comprehension and Creation
Runpei Dong (Xi'an Jiaotong University), Li Yi
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelScore-based ModelImageTextMultimodality
🎯 What it does: The DREAMLLM framework is proposed, enabling multimodal large language models to possess both understanding and creative capabilities.
DreamSmooth: Improving Model-based Reinforcement Learning via Reward Smoothing
Vint Lee (University of California), Youngwoon Lee (Yonsei University)
Reinforcement LearningWorld ModelSequential
🎯 What it does: This study investigates the challenge of reward prediction in model-based reinforcement learning and proposes improving the prediction performance of the reward model through temporal smoothing of the reward signal (DreamSmooth).
DreamTime: An Improved Optimization Strategy for Diffusion-Guided 3D Generation
Yukun Huang (International Digital Economy Academy), Lei Zhang (International Digital Economy Academy)
GenerationOptimizationDiffusion modelScore-based ModelNeural Radiance FieldImage
🎯 What it does: This paper proposes an improved optimization strategy—DreamTime, which significantly enhances the convergence speed, quality, and diversity of 3D models by employing time-prioritized score distillation sampling (TP-SDS) during the 3D generation process, replacing the traditional uniform time step sampling.
DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization
Guowei Xu (Tsinghua University), Huazhe Xu (Tsinghua University)
Robotic IntelligenceReinforcement LearningImage
🎯 What it does: An algorithm called DrM is proposed to enhance sample efficiency and performance by minimizing the dormant ratio in visual reinforcement learning models.
Dropout Enhanced Bilevel Training
Peiran Yu (University of Maryland), Heng Huang (University of Maryland)
OptimizationMeta LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: The paper constructs a random bi-level optimization model that incorporates a random dropout mask in a bi-level optimization task, and designs a dropout bi-level optimization algorithm based on the existing single-loop algorithm (FSLA). It then conducts theoretical analysis on its convergence and complexity, and performs experimental validation on data cleaning and meta-learning tasks.
Dropout-Based Rashomon Set Exploration for Efficient Predictive Multiplicity Estimation
Hsiang Hsu (JPMorgan Chase Bank), Chun-Fu Chen
ClassificationObject DetectionComputational EfficiencyConvolutional Neural NetworkTabular
🎯 What it does: By using Dropout technology in neural networks to quickly explore the Rashomon set, we can efficiently estimate predictive multiplicity.
DrS: Learning Reusable Dense Rewards for Multi-Stage Tasks
Tongzhou Mu (University of California San Diego), Hao Su (University of California San Diego)
Robotic IntelligenceReinforcement LearningTabularBenchmark
🎯 What it does: The DrS method is proposed, which learns reusable dense rewards from sparse rewards and optional demonstrations by utilizing phase indicators of multi-stage tasks, alleviating the burden of manual reward engineering.
DRSM: De-Randomized Smoothing on Malware Classifier Providing Certified Robustness
Shoumik Saha (University of Maryland), Tudor Dumitras (University of Maryland)
Anomaly DetectionAdversarial AttackConvolutional Neural NetworkTabular
🎯 What it does: This paper proposes DRSM (De-Randomized Smoothed MalConv), which achieves provable robustness detection for malicious binary files through window ablation and de-randomized smoothing.
DSPy: Compiling Declarative Language Model Calls into State-of-the-Art Pipelines
Omar Khattab (Stanford University), Christopher Potts (Stanford University)
TransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the DSPy programming model, defines signatures, modules, and teleprompters, and implements a compiler for automatic optimization of the LM pipeline; experiments are conducted on mathematical word problems (GSM8K) and multi-hop question answering (HotPotQA).
Dual Associated Encoder for Face Restoration
YU-JU TSAI, Ming-Hsuan Yang (Google Research)
RestorationTransformerGenerative Adversarial NetworkImage
🎯 What it does: Proposes a dual-branch encoder combined with code cluster priors to achieve blind face restoration.
Dual RL: Unification and New Methods for Reinforcement and Imitation Learning
Harshit Sikchi (University of Texas at Austin), Scott Niekum (University of Massachusetts Amherst)
Robotic IntelligenceReinforcement LearningTabularBenchmark
🎯 What it does: This paper unifies various existing methods through linear programming and dual transformations of RL and IL, and proposes two new algorithms: ReCOIL (offline IL, no discriminator, relaxed coverage assumption) and f-DVL (offline RL, improved implicit maximizer, enhanced training stability).
Dual-Encoders for Extreme Multi-label Classification
Nilesh Gupta (University of Texas at Austin), Inderjit S Dhillon
ClassificationRetrievalContrastive LearningText
🎯 What it does: This study investigates how to achieve state-of-the-art retrieval performance in extreme multi-label classification (XMC) tasks using a dual encoder (DE) model, proposing an improved contrastive loss function.
Duolando: Follower GPT with Off-Policy Reinforcement Learning for Dance Accompaniment
Li Siyao (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
GenerationTransformerReinforcement LearningVideo
🎯 What it does: This study investigates the dance accompaniment task, constructs the DD100 dual dance dataset, and proposes the Duolando accompaniment generation framework based on GPT and off-policy reinforcement learning.
DV-3DLane: End-to-end Multi-modal 3D Lane Detection with Dual-view Representation
Yueru Luo (Chinese University of Hong Kong Shenzhen), Zhen Li (Chinese University of Hong Kong Shenzhen)
Object DetectionAutonomous DrivingTransformerContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: This study proposes an end-to-end multimodal 3D lane detection framework, DV-3DLane, which can simultaneously utilize camera images and LiDAR point clouds, and learn and fuse features in both perspective view (PV) and bird's eye view (BEV) spaces.
Dynamic Discounted Counterfactual Regret Minimization
Hang Xu (Institute of Automation, Chinese Academy of Sciences), Jian Cheng (Institute of Automation, Chinese Academy of Sciences)
OptimizationReinforcement LearningSequential
🎯 What it does: A dynamic discounting CFR framework DDCFR is proposed, which quickly approaches Nash equilibrium in uncertain information games using a learnable discount scheme.
Dynamic Layer Tying for Parameter-Efficient Transformers
Tamir David Hay (Tel Aviv University), Lior Wolf (Tel Aviv University)
OptimizationTransformerReinforcement LearningText
🎯 What it does: By dynamically deciding which Transformer layers share weights during training through reinforcement learning, the number of trainable parameters is significantly reduced.
Dynamic Neighborhood Construction for Structured Large Discrete Action Spaces
Fabian Akkerman (University of Twente), Maximilian Schiffer (Technical University of Munich)
OptimizationReinforcement Learning
🎯 What it does: A dynamic neighborhood construction (DNC) based actor-critic algorithm is proposed for directly learning continuous actions and mapping them to discrete actions in a structured large discrete action space (SLDAS).
Dynamic Neural Response Tuning
Tian Qiu (Zhejiang University), Mingli Song (Zhejiang University)
ClassificationRecognitionConvolutional Neural NetworkTransformerImageGraph
🎯 What it does: Proposed and implemented a Dynamic Neural Response Tuning (DNRT) mechanism, which includes Response Adaptive Activation (RAA) and Aggregated Response Regularization (ARR), and embedded it into various ANN architectures, significantly improving model performance.
Dynamic Sparse No Training: Training-Free Fine-tuning for Sparse LLMs
Yuxin Zhang (Xiamen University), Rongrong Ji (Xiamen University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A training-free sparse LLM fine-tuning method (DS/ban_circle T) is proposed, which enhances the performance of sparse models by iteratively adding and removing weights (prune-and-grow) on a pre-sparsified model, using only matrix multiplication to compute reconstruction error, without the need for backpropagation or weight updates.
Dynamic Sparse Training with Structured Sparsity
Mike Lasby (University of Calgary), Yani Ioannou
OptimizationComputational EfficiencyConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: A sparse training method called SRigL is proposed, which dynamically learns constant fan-in structured sparsity during the training process while maintaining generalization performance comparable to unstructured DST.
Dynamics-Informed Protein Design with Structure Conditioning
Urszula Julia Komorowska (University of Cambridge), Mateja Jamnik (University of Cambridge)
Protein Structure PredictionDiffusion modelScore-based ModelBiomedical Data
🎯 What it does: Incorporating kinetic constraints (based on the lowest non-trivial modes) into the protein generation process using diffusion models, and further achieving joint conditional generation of structure and dynamics.
DynaVol: Unsupervised Learning for Dynamic Scenes through Object-Centric Voxelization
Yanpeng Zhao (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
GenerationData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkNeural Radiance FieldVideo
🎯 What it does: Learn object-centric decomposition of 3D dynamic scenes under unsupervised conditions, and model the spatial and temporal features of each object through voxelization.
DyST: Towards Dynamic Neural Scene Representations on Real-World Videos
Maximilian Seitzer (Max Planck Institute for Intelligent Systems), Mehdi S. M. Sajjadi (Google DeepMind)
GenerationData SynthesisPose EstimationTransformerVideo
🎯 What it does: A Transformer-based DyST model is proposed, which learns an implicit representation of separable camera poses and scene dynamics from monocular real-time video, enabling new view synthesis and video manipulation.
DyVal: Dynamic Evaluation of Large Language Models for Reasoning Tasks
Kaijie Zhu (Microsoft Research), Xing Xie (Microsoft Research)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposed and implemented the DYVAL dynamic evaluation protocol, which dynamically generates inference task samples using directed acyclic graphs.
Early Neuron Alignment in Two-layer ReLU Networks with Small Initialization
Hancheng Min (University of Pennsylvania), Rene Vidal
ClassificationImage
🎯 What it does: This paper studies the problem of training a two-layer ReLU network for binary classification under small initialization conditions using gradient flow. The analysis shows that in the early stages of training, the neurons in the first layer align with the positive and negative data according to the weights of the second layer.
Early Stopping Against Label Noise Without Validation Data
Suqin Yuan (University of Sydney), Tongliang Liu (Nanyang Technological University)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: A novel early stopping method called Label Wave is proposed, which locates the optimal stopping point of the model in the presence of label noise by utilizing prediction changes from the training set, thereby avoiding overfitting to incorrect labels.
EasyTPP: Towards Open Benchmarking Temporal Point Processes
Siqiao Xue (Ant Group), Hongyuan Mei (TTIC)
Recurrent Neural NetworkTransformerTime SeriesSequentialBenchmarkOrdinary Differential Equation
🎯 What it does: This paper presents EasyTPP, a unified open-source benchmark library for standardizing and transparently training, evaluating, and comparing temporal point process models.
EBMDock: Neural Probabilistic Protein-Protein Docking via a Differentiable Energy Model
Huaijin Wu (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
Protein Structure PredictionGraph Neural NetworkContrastive LearningTabularBiomedical DataStochastic Differential Equation
🎯 What it does: A geometric deep learning framework called EBMDock based on an energy model is proposed, which describes the probability distribution of protein-protein docking using statistical potential energy and finds the lowest energy conformation through Langevin dynamics sampling.
ECoFLaP: Efficient Coarse-to-Fine Layer-Wise Pruning for Vision-Language Models
Yi-Lin Sung (University of North Carolina Chapel Hill), Mohit Bansal (University of North Carolina Chapel Hill)
OptimizationComputational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: An efficient coarse-to-fine layer pruning method ECoFLaP is proposed for single pruning of large-scale vision-language models.
EControl: Fast Distributed Optimization with Compression and Error Control
Yuan Gao (Universität des Saarlandes), Sebastian U Stich
OptimizationConvolutional Neural NetworkImage
🎯 What it does: A fault compensation mechanism named EControl is proposed, which is compatible with any shrinkage compressor and suitable for distributed optimization of heterogeneous data.
ED-NeRF: Efficient Text-Guided Editing of 3D Scene With Latent Space NeRF
JangHo Park, Jong Chul Ye (KAIST)
GenerationComputational EfficiencyDiffusion modelNeural Radiance FieldImage
🎯 What it does: An efficient method called ED-NeRF is proposed for training and editing NeRF in the latent space of the Latent Diffusion model, which can quickly modify real 3D scenes based on text prompts.
Effective and Efficient Federated Tree Learning on Hybrid Data
Qinbin Li (University of California Berkeley), Dawn Song (University of California Berkeley)
Anomaly DetectionFederated LearningComputational EfficiencyTabular
🎯 What it does: A HybridTree algorithm for federated gradient boosting trees in a mixed data environment is proposed, which supports multiple parties to jointly train decision trees without sharing raw data.
Effective Data Augmentation With Diffusion Models
Brandon Trabucco (Carnegie Mellon University), Ruslan Salakhutdinov (Carnegie Mellon University)
ClassificationGenerationData SynthesisDiffusion modelImageStochastic Differential Equation
🎯 What it does: This paper proposes DA-Fusion, a semantic enhancement method for image-to-image based on Stable Diffusion, which uses a pre-trained diffusion model to semantically edit real images and generate diverse data augmentation samples.
Effective pruning of web-scale datasets based on complexity of concept clusters
Amro Kamal Mohamed Abbas (Meta AI), Ari S. Morcos (University of California San Diego)
Computational EfficiencyData-Centric LearningTransformerContrastive LearningImageText
🎯 What it does: Refining large-scale web datasets through deduplication, CLIP-score filtering, and density-based pruning (DBP) based on concept complexity, significantly reducing computational costs and improving model performance in large-scale CLIP training.
Effective Structural Encodings via Local Curvature Profiles
Lukas Fesser (Harvard University), Melanie Weber (Harvard University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: Proposed and implemented a local curvature distribution (LCP) structure encoding based on discrete Ricci curvature, and used it as node features input into various graph neural networks.
Efficient and Scalable Graph Generation through Iterative Local Expansion
Andreas Bergmeister (ETH Zurich), Roger Wattenhofer (ETH Zurich)
GenerationData SynthesisGraph Neural NetworkDiffusion modelPoint CloudGraph
🎯 What it does: A method for generating graphs through gradual local expansion (from a single node to a complete graph) is proposed, utilizing a denoising diffusion model to recover the edge and node features at each step, and introducing a local PPGN layer to maintain high expressive power.
Efficient Backdoor Attacks for Deep Neural Networks in Real-world Scenarios
Ziqiang Li (University of Science and Technology of China), Bin Li (University of Science and Technology of China)
Adversarial AttackConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A backdoor attack framework in 'data-constrained' real-world scenarios is proposed, achieving efficient backdoor injection through improved trigger design and feature suppression.