NeurIPS 2024 Papers — Page 8
Conference on Neural Information Processing Systems · 4035 papers
Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition
Zi-Hao Zhou (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationRecognitionRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A Continuous Contrastive Learning (CCL) framework is proposed, which utilizes reliable and continuous pseudo-labels to improve representation learning in long-tail semi-supervised learning, addressing the issues of pseudo-label confirmation bias and inconsistency in the label distribution of unlabeled data.
Continuous Heatmap Regression for Pose Estimation via Implicit Neural Representation
Shengxiang Hu (Nanjing University of Science and Technology), Jin Wang (Nantong University)
Pose EstimationImage
🎯 What it does: This paper studies a continuous heatmap regression model called NerPE based on implicit neural representations for 2D human pose estimation.
Continuous Partitioning for Graph-Based Semi-Supervised Learning
Chester Holtz (University of California San Diego), Gal Mishne (University of California San Diego)
ClassificationOptimizationGraph Neural NetworkGraph
🎯 What it does: A CutSSL framework based on continuous non-convex quadratic programming is proposed to directly obtain integer label assignments in graph-structured semi-supervised learning, addressing the degradation problem of Laplacian learning under low label rates and class imbalance.
Continuous Product Graph Neural Networks
Aref Einizade (Telecom Paris Institute Polytechnique de Paris), Jhony H. Giraldo (Telecom Paris Institute Polytechnique de Paris)
Graph Neural NetworkGraphTime SeriesStochastic Differential Equation
🎯 What it does: A continuous product graph neural network called CITRUS is proposed, based on tensor partial differential equations, for joint learning and spatiotemporal prediction of multi-domain graph data.
Continuous Spatiotemporal Events Decoupling through Spike-based Bayesian Computation
Yajing Zheng (Peking University), Tiejun Huang (Peking University)
SegmentationAnomaly DetectionSpiking Neural NetworkTime Series
🎯 What it does: A Bayesian computing framework based on SNN is constructed, using WTA for the E-step and STDP for the M-step, to achieve motion segmentation of event streams.
Continuous Temporal Domain Generalization
Zekun Cai (University of Tokyo), Liang Zhao (Emory University)
Domain AdaptationTabularTime SeriesOrdinary Differential Equation
🎯 What it does: This work proposes the Continuous Time Domain Generalization (CTDG) framework Koodos, which addresses the limitation of traditional TDG that can only handle discrete time points, allowing for generalization of the model at any continuous time point.
Continuously Learning, Adapting, and Improving: A Dual-Process Approach to Autonomous Driving
Jianbiao Mei (Zhejiang University), Yu Qiao (Shanghai Artificial Intelligence Laboratory)
Autonomous DrivingTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageVideo
🎯 What it does: A closed-loop autonomous driving system named LeapAD is proposed, which combines human attention mechanisms and dual-process cognitive models to achieve selection of key objects, scene understanding, logical reasoning, and experiential memory, forming a closed-loop decision-making framework for continuous learning, adaptation, and improvement.
Contracting with a Learning Agent
Guru Guruganesh (Google Research), S. Matthew Weinberg (Princeton University)
OptimizationReinforcement Learning
🎯 What it does: This paper studies how the principal designs dynamic contracts to counteract learning agents in a repeated principal-agent setting, proving that the optimal dynamic contract under linear contracts and one-dimensional scaling takes the 'free-fall' form;
CONTRAST: Continual Multi-source Adaptation to Dynamic Distributions
Sk Miraj Ahmed (Brookhaven National Laboratory), Amit Roy-Chowdhury
Domain AdaptationImageBenchmark
🎯 What it does: A framework for dynamically adapting multi-source models during testing (CONTRAST) is proposed, which can perform weighted fusion of multi-source models and gradually update the most relevant models without source data, relying only on a small batch of test samples, thereby maintaining robustness to dynamic distributions.
Contrasting with Symile: Simple Model-Agnostic Representation Learning for Unlimited Modalities
Adriel Saporta (New York University), Rajesh Ranganath (New York University)
RetrievalRepresentation LearningContrastive LearningImageTextMultimodalityBiomedical DataAudio
🎯 What it does: This paper presents Symile, a contrastive learning framework for an arbitrary number of modalities that captures higher-order joint information while maintaining the simplicity of CLIP, generating modality-specific representations that can be directly used for zero-shot retrieval and downstream tasks.
Contrastive dimension reduction: when and how?
Sam Hawke (University of North Carolina), Didong Li (University of North Carolina)
Anomaly DetectionContrastive LearningImageBiomedical Data
🎯 What it does: This paper proposes a hypothesis testing method to determine whether there is unique information between the foreground group and the background group, and provides an estimator for quantifying the unique dimensions of the foreground;
Contrastive losses as generalized models of global epistasis
David H Brookes, Sam Sinai (Dyno Therapeutics)
Drug DiscoveryContrastive LearningBiomedical Data
🎯 What it does: The study uses contrastive loss (such as Bradley-Terry loss) to recover sparse latent ranking functions in protein fitness data with global interactions, demonstrating its effectiveness over traditional MSE loss under limited samples.
Contrastive-Equivariant Self-Supervised Learning Improves Alignment with Primate Visual Area IT
Thomas Edward Yerxa (New York University), SueYeon Chung (Flatiron Institute)
Representation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A contrastive-equivariant self-supervised learning framework is proposed that does not require transformation parameters, encouraging the network representation to maintain transformation-related structures.
Controlled maximal variability along with reliable performance in recurrent neural networks
Chiara Mastrogiuseppe (Universitat Pompeu Fabra), Rubén Moreno-Bote (Universitat Pompeu Fabra)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: Introducing the 'NeuroMOP' framework in random recurrent neural networks (RNNs) to maximize future action entropy for generating controllable neural variations and achieving long-term survival by avoiding terminal states;
Controlling Continuous Relaxation for Combinatorial Optimization
Yuma Ichikawa (Fujitsu Limited)
OptimizationGraph Neural NetworkGraph
🎯 What it does: Proposes a Continuous Relaxation Annealing (CRA) method to improve unsupervised learning-based combinatorial optimization solvers, eliminating manual rounding.
Controlling Counterfactual Harm in Decision Support Systems Based on Prediction Sets
Eleni Straitouri (Max Planck Institute for Software Systems), Manuel Gomez Rodriguez (Max Planck Institute for Software Systems)
ClassificationImage
🎯 What it does: This study investigates causal counterfactual harm in decision support systems based on predictive sets and proposes a theoretical and algorithmic framework to control its occurrence frequency.
Controlling Multiple Errors Simultaneously with a PAC-Bayes Bound
Reuben Adams (University College London), Benjamin Guedj (University College London)
OptimizationImageBiomedical Data
🎯 What it does: A PAC-Bayes generalization bound that can simultaneously control the distribution of multiple types of errors is proposed, and it is transformed into a differentiable training objective.
ControlMLLM: Training-Free Visual Prompt Learning for Multimodal Large Language Models
Mingrui Wu (Xiamen University), Rongrong Ji (National University of Singapore)
RecognitionObject DetectionDomain AdaptationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: A training-agnostic visual prompt injection method is proposed, which optimizes visual tokens using learnable latent variables during the inference phase, thereby controlling the attention of multimodal large language models to achieve fine-grained region description and reasoning.
ControlSynth Neural ODEs: Modeling Dynamical Systems with Guaranteed Convergence
Wenjie Mei (Southeast University), Shihua Li (Southeast University)
Time SeriesBiomedical DataOrdinary Differential Equation
🎯 What it does: A novel continuous-time neural network called ControlSynth Neural ODEs (CSODEs) is proposed for modeling highly nonlinear and scalable dynamical systems, ensuring convergence through solvable linear inequalities.
Convergence Analysis of Split Federated Learning on Heterogeneous Data
Pengchao Han (Guangdong University of Technology), Xin Liu (University of California)
OptimizationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: Proposes and analyzes the convergence of Split Federated Learning (SFL), providing convergence upper bounds for strongly convex, generally convex, and non-convex objectives under heterogeneous data and partial participation scenarios.
Convergence of $\text{log}(1/\epsilon)$ for Gradient-Based Algorithms in Zero-Sum Games without the Condition Number: A Smoothed Analysis
Ioannis Anagnostides (Carnegie Mellon University), Tuomas Sandholm (Carnegie Mellon University)
OptimizationGaussian Splatting
🎯 What it does: This paper studies the iterative complexity of gradient methods (OGDA, EGDA, IterSmooth) and OMWU in two-player zero-sum games through smoothed analysis, proving that the first three have polynomial smoothed complexity under high precision targets, while the complexity of OMWU remains non-polynomial.
Convergence of No-Swap-Regret Dynamics in Self-Play
Renato Paes Leme (Google Research), Jon Schneider (Google Research)
OptimizationReinforcement Learning from Human Feedback
🎯 What it does: The study investigates the convergence properties of self-play in symmetric zero-sum games when using a no-internal (exchange) regret learning algorithm.
Convolutional Differentiable Logic Gate Networks
Felix Petersen (Stanford University), Stefano Ermon (Stanford University)
ClassificationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A convolutional extension of a differentiable logic gate network (LGN) is proposed, utilizing techniques such as logic gate tree convolution, logical OR pooling, and residual initialization to train efficient logic gate networks on CIFAR-10 and MNIST, achieving extremely low gate counts with high accuracy.
Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods
Felix Dangel (Vector Institute)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: Convolutions and their related operations (self-differentiation, Jacobian, KFAC, etc.) are represented as tensor networks (TN) and implemented in a unified and scalable manner through einsum.
CooHOI: Learning Cooperative Human-Object Interaction with Manipulated Object Dynamics
Jiawei Gao (Shanghai AI Laboratory), Jiangmiao Pang (Shanghai AI Laboratory)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: Proposes the CooHOI framework, which first learns single-agent manipulation skills using AMP, and then transfers this skill to a multi-agent collaborative task of transporting long objects through CTDE+MAPPO;
Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of LLM Agents
Giorgio Piatti (ETH Zürich), Rada Mihalcea (University of Michigan)
TransformerLarge Language ModelAgentic AIText
🎯 What it does: Created the GOVSIM platform to simulate cooperation and governance among multiple agents in public resource sharing using large language models, and to assess its sustainability.
Cooperative Hardware-Prompt Learning for Snapshot Compressive Imaging
Jiamian Wang (Rochester Institute of Technology), ZHIQIANG TAO
RestorationCompressionFederated LearningTransformerPrompt EngineeringImage
🎯 What it does: A Federated Hardware Prompt Learning (FedHP) framework is proposed and implemented for collaborative training of multi-hardware configuration snapshot compression imaging (SCI) systems without sharing hardware configurations and measurement data, significantly improving reconstruction performance under different optical encoders.
CorDA: Context-Oriented Decomposition Adaptation of Large Language Models for Task-Aware Parameter-Efficient Fine-tuning
Yibo Yang (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A parameter-efficient fine-tuning method called CorDA based on context-oriented singular value decomposition is proposed, which allows for flexible switching between maintaining world knowledge and enhancing downstream task performance.
Corruption-Robust Linear Bandits: Minimax Optimality and Gap-Dependent Misspecification
Haolin Liu (University of Virginia), Chen-Yu Wei (University of Virginia)
Optimization
🎯 What it does: This paper studies the optimal strategies and lower bounds of linear bandits in online learning under the scenarios of contaminated rewards and model misspecification (gap-dependent misspecification), providing a unified analysis of strong and weak contamination and proposing feasible algorithms.
CosAE: Learnable Fourier Series for Image Restoration
Sifei Liu (NVIDIA), Jan Kautz (NVIDIA)
RestorationSuper ResolutionCompressionAuto EncoderImage
🎯 What it does: A self-encoder CosAE utilizing a learnable two-dimensional cosine frequency domain representation is proposed, achieving extreme compression and reconstruction of image details.
COSMIC: Compress Satellite Image Efficiently via Diffusion Compensation
Ziyuan Zhang (Tsinghua University), Hewu Li (Tsinghua University)
GenerationCompressionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A compression scheme for satellite images called COSMIC is proposed, which uses a lightweight encoder to compress images at the satellite end and employs a conditional diffusion model at the ground end to compensate for the details missing from the encoder, achieving efficient and high-quality image compression and reconstruction.
Cost-aware Bayesian Optimization via the Pandora's Box Gittins Index
Qian Xie (Cornell University), Alexander Terenin (Cornell University)
OptimizationTabular
🎯 What it does: This paper proposes a novel acquisition function PBGI based on the Pandora's Box Gittins index for making sampling decisions in Bayesian optimization considering evaluation costs, and validates its effectiveness on multi-dimensional, multi-modal problems.
Cost-efficient Knowledge-based Question Answering with Large Language Models
Junnan Dong (Hong Kong Polytechnic University), Xiao Huang (Hong Kong Polytechnic University)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: For the knowledge graph question answering task, a framework called Coke is proposed, which dynamically selects either LLM or a small KG model to balance reasoning accuracy and invocation cost under a limited budget.
CoSW: Conditional Sample Weighting for Smoke Segmentation with Label Noise
Lujian Yao (East China University of Science and Technology), Jingchao Peng (East China University of Science and Technology)
SegmentationImage
🎯 What it does: A conditional sample weighting method for label noise in smoke segmentation, CoSW, is proposed, which utilizes a multi-prototype framework and regularized internal entropy to dynamically identify and suppress noisy pixels.
CoSy: Evaluating Textual Explanations of Neurons
Laura Kopf (TU Berlin), Kirill Bykov (UMI Lab)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningImage
🎯 What it does: An automated evaluation framework named COSY is proposed to quantify the quality of neuron text explanations.
Counter-Current Learning: A Biologically Plausible Dual Network Approach for Deep Learning
Chia Hsiang Kao, Bharath Hariharan (Cornell University)
ClassificationExplainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a biologically interpretable Counter-Current Learning (CCL) framework, which consists of a dual network structure formed by a feedforward network and a corresponding feedback network. Learning is achieved by backpropagating information, local loss, and gradient separation between the two networks.
Counterfactual Fairness by Combining Factual and Counterfactual Predictions
Zeyu Zhou (Purdue University), David I. Inouye (Purdue University)
Tabular
🎯 What it does: Theoretical analysis of the trade-off between adversarial bias and predictive performance is conducted, proposing the optimal counterfactual fairness (CF) predictor, and providing error bounds and a practical plug-in algorithm PCF in the absence of complete causal knowledge.
CountGD: Multi-Modal Open-World Counting
Niki Amini-Naieni (Visual Geometry Group University of Oxford), Andrew Zisserman (Visual Geometry Group University of Oxford)
Object DetectionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This study proposes COUNTGD, a multimodal open-world counting model that can accept text descriptions, visual samples, or a combination of both to specify target objects and perform counting.
Coupled Mamba: Enhanced Multimodal Fusion with Coupled State Space Model
Wenbing Li (Huazhong University of Science and Technology), Wei Yang (Huazhong University of Science and Technology)
ClassificationTransformerMultimodality
🎯 What it does: The Coupled Mamba model is proposed, utilizing the state chain in state space models to achieve the coupling and fusion of multimodal information, enhancing the performance of multimodal sentiment analysis and classification tasks.
Covariate Shift Corrected Conditional Randomization Test
Bowen Xu (Harvard University), Molei Liu (Columbia University)
TabularBiomedical DataElectronic Health Records
🎯 What it does: A conditional independence testing method for covariate shift scenarios, called csPCR, is proposed, which can utilize source samples to infer the target population.
COVE: Unleashing the Diffusion Feature Correspondence for Consistent Video Editing
Jiangshan Wang (Tsinghua University), Xiu Li (Tsinghua University)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: Utilizing the inherent feature similarity in diffusion models, a cross-frame token correspondence is established to achieve temporal consistency and high-quality generation during video editing.
CoVoMix: Advancing Zero-Shot Speech Generation for Human-like Multi-talker Conversations
leying zhang, Michael Zeng (Microsoft)
GenerationData SynthesisTransformerFlow-based ModelAudio
🎯 What it does: This paper presents the CoVoMix system, which achieves zero-shot, natural multi-speaker, multi-turn dialogue speech synthesis.
Cracking the Code of Juxtaposition: Can AI Models Understand the Humorous Contradictions
Zhe Hu (Hong Kong Polytechnic University), Yu Yin (Case Western Reserve University)
RecognitionGenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: A task for understanding comparative contradictory narratives in comics is proposed, and a new YESBUT dataset is created.
Crafting Interpretable Embeddings for Language Neuroscience by Asking LLMs Questions
Vinamra Benara (University of California Berkeley), Jianfeng Gao (Microsoft Research)
ClassificationRetrievalExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBiomedical DataMagnetic Resonance Imaging
🎯 What it does: By having large language models answer a series of yes/no questions, interpretable text embeddings QA-Emb are constructed and used to predict human fMRI brain responses.
CRAYM: Neural Field Optimization via Camera RAY Matching
Liqiang Lin (Shenzhen University), Hui Huang (Shenzhen University)
Pose EstimationOptimizationNeural Radiance FieldImage
🎯 What it does: This paper proposes a method for joint optimization of camera pose and neural field (feature volume) through Camera Ray Matching (CRAYM);
Credal Deep Ensembles for Uncertainty Quantification
Kaizheng Wang (KU Leuven), Hans Hallez (KU Leuven)
ClassificationAnomaly DetectionImage
🎯 What it does: A new trustworthy deep ensemble model called Credal Deep Ensembles (CreDEs) is proposed, consisting of Credal-Set Neural Networks (CreNets), which can output probability bounds for each class, thereby forming a credal set that enhances the quantification of model uncertainty, especially epistemic uncertainty.
Credal Learning Theory
Michele Caprio (University of Manchester), Fabio Cuzzolin (Oxford Brookes University)
🎯 What it does: This paper proposes Credal Learning Theory (CLT), which extends traditional statistical learning theory to consider a framework that accounts for data distribution drift and uncertainty, deriving upper bounds on generalization error in various scenarios using credal sets.
Credit Attribution and Stable Compression
Roi Livni (Tel Aviv University), Chirag Pabbaraju (Stanford University)
CompressionSafty and Privacy
🎯 What it does: This paper proposes a theoretical framework for implementing 'academic attribution' and 'copyright ownership' in machine learning, defining new stability constraints such as **Counterfactual Attribution** (CCA) and **Sample Differential Privacy Compression** (Sample-DP-Compression), and provides a characterization of learnability under the PAC learning model.
CriticEval: Evaluating Large-scale Language Model as Critic
Tian Lan (Beijing Institute of Technology), Xian-Ling Mao (Beijing Institute of Technology)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: The CRITICEVAL benchmark has been constructed and released, systematically evaluating the criticism ability of large models across four dimensions (feedback, comparison, correction, meta-feedback) covering nine task scenarios.
CRONOS: Enhancing Deep Learning with Scalable GPU Accelerated Convex Neural Networks
Miria Feng (Stanford University), Mert Pilanci (Stanford University)
OptimizationConvolutional Neural NetworkImageText
🎯 What it does: The CRONOS algorithm is proposed, which utilizes convex optimization to train two-layer ReLU networks and is extended to arbitrary depth networks through CRONOS-AM, achieving efficient training on large-scale datasets.
Cross-Device Collaborative Test-Time Adaptation
Guohao Chen (South China University of Technology), Mingkui Tan (South China University of Technology)
Domain AdaptationFederated LearningTransformerImage
🎯 What it does: A multi-device collaborative terminal adaptation (CoLA) framework is proposed, achieving collaborative learning and knowledge sharing across devices and terminals with online adaptation.
Cross-modal Representation Flattening for Multi-modal Domain Generalization
Yunfeng FAN, Song Guo (Huazhong University of Science and Technology)
Domain AdaptationKnowledge DistillationRepresentation LearningContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes a Cross-Modal Representation Flattening (CMRF) method aimed at addressing the issues of modality competition and single-modal flattening differences in multi-modal domain generalization.
Cross-Modality Perturbation Synergy Attack for Person Re-identification
Yunpeng Gong (Xiamen University), Min Jiang (Xiamen University)
RecognitionRetrievalAdversarial AttackImageMultimodality
🎯 What it does: A universal perturbation attack method called CMPS is proposed for cross-modal person re-identification (RGB and infrared/thermal imaging) systems, which can jointly optimize perturbations across different modalities to confuse retrieval results.
Cross-model Control: Improving Multiple Large Language Models in One-time Training
Jiayi Wu (East China Normal University), Ming Gao (East China Normal University)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a cross-model control method (Cross-model Control, CMC) that enhances various large language models (LLMs) with a single training session. It achieves this by training a portable small 'Delta model' to adjust the output logits of different LLMs, enabling multi-model instruction following and forgetting tasks.
Cross-Scale Self-Supervised Blind Image Deblurring via Implicit Neural Representation
Tianjing Zhang (National University of Singapore), Hui Ji (South China University of Technology)
RestorationContrastive LearningImage
🎯 What it does: A self-supervised blind image deblurring method that does not require real images is proposed, utilizing cross-scale consistency constraints to achieve simultaneous estimation of the image and the blur kernel.
Cross-video Identity Correlating for Person Re-identification Pre-training
Jialong Zuo (Huazhong University of Science and Technology), Changxin Gao (Huazhong University of Science and Technology)
RecognitionRetrievalDomain AdaptationKnowledge DistillationConvolutional Neural NetworkTransformerContrastive LearningImageVideo
🎯 What it does: A cross-video identity association pre-training framework CION is proposed, which learns cross-video identity invariance from a large number of unlabeled video portraits using progressive multi-level denoising and identity-guided self-distillation.
CRT-Fusion: Camera, Radar, Temporal Fusion Using Motion Information for 3D Object Detection
Jisong Kim (Hanyang University), Jun Won Choi (Seoul National University)
Object DetectionAutonomous DrivingMultimodalityPoint Cloud
🎯 What it does: This paper proposes a radar-camera fusion framework named CRT-Fusion for 3D object detection, which compensates for the motion of dynamic targets through temporal information.
CryoGEM: Physics-Informed Generative Cryo-Electron Microscopy
Jiakai Zhang (ShanghaiTech University), Jingyi Yu (ShanghaiTech University)
GenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImagePhysics Related
🎯 What it does: CryoGEM is designed, combining physics-based cryo-EM simulation with unsupervised noise contrastive learning to generate realistic, annotated synthetic micrograph datasets.
CryoSPIN: Improving Ab-Initio Cryo-EM Reconstruction with Semi-Amortized Pose Inference
Shayan Shekarforoush (University of Toronto), David J. Fleet (University of Toronto)
Pose EstimationOptimizationConvolutional Neural NetworkAuto EncoderImageBiomedical Data
🎯 What it does: A semi-autonomous atomic-level cryo-electron microscopy reconstruction method called cryoSPIN is proposed, which combines a multi-head pose predictor and subsequent pixel-wise pose optimization.
Cryptographic Hardness of Score Estimation
Min Jae Song (Paul G. Allen School of Computer Science and Engineering University of Washington)
🎯 What it does: It is proven that without strong assumptions on data distribution, achieving an L₂ error of 1/√log d for score estimation is computationally difficult, and a corresponding statistical-computational gap is provided.
CSPG: Crossing Sparse Proximity Graphs for Approximate Nearest Neighbor Search
Ming Yang (Fudan University), Weiguo Zheng (Fudan University)
RetrievalComputational EfficiencyGraph Neural NetworkGraphBenchmark
🎯 What it does: By randomly partitioning the dataset and sharing routing vectors, multiple sparse adjacency graphs (CSPG) are constructed. During queries, a two-stage (fast approach + cross-graph expansion) greedy Beam Search is employed, significantly reducing distance calculations and search steps, thereby accelerating nearest neighbor retrieval.
Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance
Kuan Heng Lin (University of California), Bolei Zhou (University of California)
Image TranslationGenerationDiffusion modelImagePoint CloudMesh
🎯 What it does: This paper proposes Ctrl-X, a zero-shot control framework without training or guidance, which can achieve structural alignment and appearance transfer simultaneously in pre-trained text-to-image diffusion models.
CultureLLM: Incorporating Cultural Differences into Large Language Models
CHENG LI, Xing Xie (Microsoft Research)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes CultureLLM, a method for fine-tuning large language models (LLMs) with cultural awareness based on a small sample from the World Values Survey (WVS) and generating training data through semantic data augmentation.
CulturePark: Boosting Cross-cultural Understanding in Large Language Models
CHENG LI, Jindong Wang (William and Mary)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Developed the CulturePark multi-agent platform, using LLM to generate cross-cultural dialogue data, and fine-tuning culture-specific LLMs with this data for tasks such as content moderation, cultural alignment, and cultural education.
CuMo: Scaling Multimodal LLM with Co-Upcycled Mixture-of-Experts
Jiachen Li (Georgia Tech and UIUC), Longyin Wen (ByteDance Inc)
RecognitionData-Centric LearningTransformerLarge Language ModelMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: Incorporating a sparse mixture of experts (MoE) module into multimodal large language models, expanding parameters from both the visual encoder and the visual-language connector, and employing a co-upcycling strategy for expert initialization, along with a three-stage training process (pre-training MLP, pre-fine-tuning, visual instruction tuning) and auxiliary load balancing loss, enhances the model's performance on visual tasks.
Curriculum Fine-tuning of Vision Foundation Model for Medical Image Classification Under Label Noise
Yeonguk Yu (Gwangju Institute of Science and Technology), Kyoobin Lee (Gwangju Institute of Science and Technology)
ClassificationSupervised Fine-TuningImageBiomedical Data
🎯 What it does: This paper proposes CUFIT—a curriculum fine-tuning paradigm for label noise in medical image classification, which gradually filters clean samples and fine-tunes them by combining linear probing, internal adapters, and end adapters.
Curvature Clues: Decoding Deep Learning Privacy with Input Loss Curvature
Deepak Ravikumar (Purdue University), Kaushik Roy (Purdue University)
Safty and PrivacyAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a curvature-based theoretical framework and black-box membership inference attack by studying the distinguishability of training and testing samples based on input loss curvature, verifying its superiority.
Customized Multiple Clustering via Multi-Modal Subspace Proxy Learning
Jiawei Yao (University of Washington), Juhua Hu (University of Washington)
Representation LearningTransformerLarge Language ModelContrastive LearningImageMultimodality
🎯 What it does: Multi-Sub is proposed, a customizable multi-clustering method based on multi-modal subspace proxy learning, which can automatically generate corresponding clustering results based on user interests (such as color, variety, etc.);
Customized Subgraph Selection and Encoding for Drug-drug Interaction Prediction
Haotong Du (Northwestern Polytechnical University), Zhen Wang (Northwestern Polytechnical University)
Drug DiscoveryNeural Architecture SearchGraph Neural NetworkGraphBiomedical Data
🎯 What it does: A searchable subgraph selection and encoding framework is designed for drug-drug interaction prediction.
Customizing Language Models with Instance-wise LoRA for Sequential Recommendation
Xiaoyu Kong (University of Science and Technology of China), Xiangnan He (University of Science and Technology of China)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsSequential
🎯 What it does: For the sequence recommendation task, an Instance-wise LoRA (iLoRA) framework is proposed to dynamically generate LoRA parameters for each user sequence during the fine-tuning process of LLM, addressing the negative transfer problem caused by uniform LoRA.
CV-VAE: A Compatible Video VAE for Latent Generative Video Models
Sijie Zhao (Tencent AI Lab), Ying Shan (Tencent AI Lab)
GenerationData SynthesisCompressionComputational EfficiencyConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkImageVideo
🎯 What it does: Designed and trained a 3D video VAE (CV-VAE) compatible with existing image VAEs (such as Stable Diffusion), achieving a continuous latent space for spatiotemporal compression that can be directly used for video generation models.
CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns
Shengsheng Lin (South China University of Technology), Haocheng Zhong (Jinan University)
TransformerTime Series
🎯 What it does: The CycleNet model is proposed by explicitly learning cyclical patterns and making predictions on residuals;
CYCLO: Cyclic Graph Transformer Approach to Multi-Object Relationship Modeling in Aerial Videos
Trong-Thuan Nguyen (University of Arkansas), Khoa Luu (University of Arkansas)
Object DetectionObject TrackingTransformerVideo
🎯 What it does: A cyclic graph Transformer named CYCLO is proposed to generate video scene graphs from aerial videos captured by drones, achieving temporal modeling of multi-object relationships through a cyclic attention mechanism.
D-CPT Law: Domain-specific Continual Pre-Training Scaling Law for Large Language Models
Haoran Que (Taobao and Tmall Group of Alibaba), Bo Zheng (Taobao and Tmall Group of Alibaba)
TransformerLarge Language ModelText
🎯 What it does: The D-CPT rule is proposed, aiming to determine the optimal mixing ratio for large language models (LLMs) of different scales with limited training costs, in order to improve performance in specific downstream domains.
D-LLM: A Token Adaptive Computing Resource Allocation Strategy for Large Language Models
yikun jiang, John C.S. Lui (Chinese University of Hong Kong)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes D-LLMs, a dynamic inference framework for large language models, which achieves on-demand execution or skipping by inserting decision modules at each layer;
D-MiSo: Editing Dynamic 3D Scenes using Multi-Gaussians Soup
Joanna Waczynska, Przemysław Spurek (IDEAS National Centre for Research and Development)
Gaussian SplattingPoint Cloud
🎯 What it does: A dynamic 3D scene editing framework D-MiSo based on Gaussian Splatting is proposed, utilizing multi-Gaussian (core + sub-Gaussian) and triangle soup structures to achieve local editability and dynamic control of objects at any moment.
D2R2: Diffusion-based Representation with Random Distance Matching for Tabular Few-shot Learning
Ruoxue Liu (Hong Kong University of Science and Technology), Bingyi Jing
ClassificationRepresentation LearningDiffusion modelTabularBenchmark
🎯 What it does: This paper proposes an unsupervised representation learning framework D2R2 based on diffusion models to address the few-shot classification problem in tabular data.
DA-Ada: Learning Domain-Aware Adapter for Domain Adaptive Object Detection
Haochen Li (Intelligent Software Research Center, Institute of Software, Chinese Academy of Sciences), Ling Li (Intelligent Software Research Center, Institute of Software, Chinese Academy of Sciences)
Object DetectionDomain AdaptationTransformerVision Language ModelImage
🎯 What it does: Based on the Visual Language Model (VLM), a Domain-Aware Adapter (DA-Ada) is proposed to address the Domain Adaptation Object Detection (DAOD) problem.
DAGER: Exact Gradient Inversion for Large Language Models
Ivo Petrov (INSAIT Sofia University St Kliment Ohridski), Martin Vechev (ETH Zurich)
Federated LearningAdversarial AttackTransformerLarge Language ModelText
🎯 What it does: This paper studies a gradient inversion attack for large language models called DAGER, which can accurately recover complete batches of text from shared gradients in federated learning.
DALD: Improving Logits-based Detector without Logits from Black-box LLMs
Cong Zeng (MBZUAI), Dongkuan Xu (North Carolina State University)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The DALD framework is proposed, which achieves more accurate text source identification by aligning the distribution of the proxy model in black-box LLM detection.
DAPE: Data-Adaptive Positional Encoding for Length Extrapolation
Chuanyang Zheng (Chinese University of Hong Kong), Yu Li (Chinese University of Hong Kong)
TransformerText
🎯 What it does: A data-adaptive position encoding (DAPE) is proposed to enhance the long-sequence generalization ability of Transformers by dynamically adjusting attention values and original position biases.
DapperFL: Domain Adaptive Federated Learning with Model Fusion Pruning for Edge Devices
Yongzhe Jia (Nanjing University), Wanchun Dou (China University of Petroleum)
Domain AdaptationFederated LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Designed and implemented DapperFL, a federated learning framework for heterogeneous edge devices and domain shift, utilizing model fusion pruning and domain adaptive regularization to achieve personalized compression and domain generalization.
DARG: Dynamic Evaluation of Large Language Models via Adaptive Reasoning Graph
Zhehao Zhang (Dartmouth College), Diyi Yang (Stanford University)
Large Language ModelPrompt EngineeringMixture of ExpertsTextChain-of-Thought
🎯 What it does: Proposes the DARG framework, which utilizes the construction and perturbation of inference graphs to dynamically expand existing evaluation benchmarks, generating test samples with controllable complexity while maintaining language diversity and label correctness.
DarkSAM: Fooling Segment Anything Model to Segment Nothing
Ziqi Zhou (Huazhong University of Science and Technology), Hai Jin (Huazhong University of Science and Technology)
SegmentationAdversarial AttackImage
🎯 What it does: We propose DarkSAM, which achieves the first universal adversarial perturbation attack on the Segment Anything Model (SAM) and its variants, rendering the model unable to correctly segment under any prompt.
DARNet: Dual Attention Refinement Network with Spatiotemporal Construction for Auditory Attention Detection
Sheng Yan (Anhui University), Zhao Lv (Anhui University)
ClassificationRecognitionConvolutional Neural NetworkTime SeriesBiomedical Data
🎯 What it does: A dual-attention refinement-based spatiotemporal construction network (DARNet) is proposed for EEG signal auditory attention detection.
DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving
Yuxuan Tong (Tsinghua University), Junxian He (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper studies the performance of large language models on mathematical reasoning tasks and proposes a difficulty-aware rejection tuning method called DART, which is used to construct more challenging synthetic training data and enhance the model's reasoning capabilities.
DASH: Warm-Starting Neural Network Training in Stationary Settings without Loss of Plasticity
Baekrok Shin (Korea Advanced Institute of Science and Technology), Chulhee Yun (Korea Advanced Institute of Science and Technology)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This study investigates how using learned weights for warm starts when continuously adding new data can lead to a loss of plasticity in neural networks, resulting in decreased generalization performance. It proposes a Direction-Aware Weight Shrinkage (DASH) method aimed at forgetting noise and retaining features by proportionally shrinking weights in the direction of the gradient, thereby restoring plasticity and enhancing performance during warm starts.
DAT: Improving Adversarial Robustness via Generative Amplitude Mix-up in Frequency Domain
Fengpeng Li (University of Macau), Jiantao Zhou (University of Macau)
Adversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A bidirectional adversarial training (DAT) framework is proposed, which forces the model to focus on phase features unaffected by attacks by mixing amplitude spectra in the frequency domain and generating adversarial amplitudes, thereby enhancing robustness against various adversarial attacks.
Data Acquisition via Experimental Design for Data Markets
Charles Lu (Massachusetts Institute of Technology), Ramesh Raskar (Massachusetts Institute of Technology)
Federated LearningData-Centric LearningLarge Language ModelImageTextBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A verification-free data acquisition method for data markets, DAVED, is proposed, which directly selects the most valuable training samples from unlabeled test queries in a federated environment using experimental design principles.
Data Attribution for Text-to-Image Models by Unlearning Synthesized Images
Sheng-Yu Wang (Carnegie Mellon University), Richard Zhang (Adobe Research)
GenerationData-Centric LearningDiffusion modelImage
🎯 What it does: This paper proposes a method for data attribution in text-to-image models using machine 'unlearning' to synthesize images, which identifies the training samples that contribute the most to a generated image by forgetting the generated images on a trained model.
Data Augmentation with Diffusion for Open-Set Semi-Supervised Learning
Seonghyun Ban (KAIST), Kee-Eung Kim (KAIST)
ClassificationData SynthesisDiffusion modelImage
🎯 What it does: This study proposes a Discriminative Weighted Diffusion (DWD) framework based on diffusion models and discriminators for data augmentation in open semi-supervised learning with mismatched class distributions.
Data Distribution Valuation
Xinyi Xu (National University of Singapore), Giulia Fanti (Carnegie Mellon University)
Anomaly DetectionOptimizationData-Centric LearningImageTabular
🎯 What it does: A distribution-level data valuation method based on Maximum Mean Discrepancy (MMD) suitable for the Huber model is proposed, which can directly assess and compare the data distribution value of different suppliers from limited samples.
Data Free Backdoor Attacks
Bochuan Cao (Pennsylvania State University), Dawn Song (University of California Berkeley)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A data-independent, retraining-free, and architecture-agnostic backdoor attack called DFBA is proposed.
Data Mixture Inference Attack: BPE Tokenizers Reveal Training Data Compositions
Jonathan Hayase (University of Washington), Noah A. Smith (University of Washington)
Large Language ModelText
🎯 What it does: By analyzing the merge list of the BPE tokenizer, the mixed proportions of its training data in terms of language, code, and domain are inferred.
Data subsampling for Poisson regression with pth-root-link
Han Cheng Lie (University Potsdam), Alexander Munteanu (TU Dortmund University)
🎯 What it does: This paper proposes and analyzes a subsampling technique for Poisson regression, constructing a subset (coreset) that can approximate the negative log-likelihood within a (1±ε) error and provides an upper bound on its size.
Data-Driven Discovery of Dynamical Systems in Pharmacology using Large Language Models
Samuel Holt (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
Drug DiscoveryTransformerLarge Language ModelBiomedical DataOrdinary Differential Equation
🎯 What it does: A framework named Data-Driven Discovery (D3) is proposed, which utilizes large language models (LLM) to iteratively discover and optimize interpretable dynamic system models through a model generation, feature acquisition, and evaluation loop, primarily used for pharmacokinetic (PK) modeling in pharmacology.
Data-Efficient Learning with Neural Programs
Alaia Solko-Breslin (University of Pennsylvania), Eric Wong (University of Pennsylvania)
Data-Centric LearningReinforcement Learning from Human FeedbackReinforcement LearningTextBenchmark
🎯 What it does: Proposes the ISED algorithm, which supports end-to-end learning of a 'neural program' composed of any black-box program (Python code or GPT-4 API calls) and neural networks;
Data-Efficient Operator Learning via Unsupervised Pretraining and In-Context Learning
Wuyang Chen (Simon Fraser University), Michael W. Mahoney (International Computer Science Institute)
TransformerAuto EncoderTabularTime SeriesPhysics Related
🎯 What it does: Enhance the performance of PDE neural operators in data-scarce and OOD scenarios through unsupervised pre-training and context learning methods.
Data-faithful Feature Attribution: Mitigating Unobservable Confounders via Instrumental Variables
Qiheng Sun (Zhejiang University), Jinfei Liu (Zhejiang University)
Explainability and InterpretabilityTabular
🎯 What it does: A method is proposed to achieve feature attribution of the data generation process using instrumental variables to train unconfounded models.
Dataset Decomposition: Faster LLM Training with Variable Sequence Length Curriculum
Hadi Pouransari (Apple), Oncel Tuzel (Apple)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes a Dataset Decomposition method that splits the original text collection into fixed-length subsequence buckets and uses variable sequence length (VSL) and length-based learning curves for sampling during training;