ICLR 2024 Papers — Page 14
International Conference on Learning Representations · 2260 papers
MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models
Deyao Zhu (King Abdullah University of Science and Technology), Mohamed Elhoseiny (King Abdullah University of Science and Technology)
GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: MiniGPT-4 is constructed, a multimodal model that aligns a pretrained visual encoder with the Vicuna LLM by training only a single linear projection layer.
MiniLLM: Knowledge Distillation of Large Language Models
Yuxian Gu (Tsinghua University), Minlie Huang (Tsinghua University)
Knowledge DistillationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes a knowledge distillation method based on reverse KL divergence, called MINILLM, to effectively transfer the generative distribution of large language models to smaller models.
Minimax optimality of convolutional neural networks for infinite dimensional input-output problems and separation from kernel methods
Yuto Nishimura (University of Tokyo), Taiji Suzuki (University of Tokyo)
Convolutional Neural Network
🎯 What it does: This study investigates the approximation and estimation error of Dilated Convolutional Neural Networks (Dilated CNN) for nonlinear operator learning in infinite-dimensional input and output spaces. It proves that it achieves a minimax optimal convergence rate and has a better error lower bound compared to linear estimators (such as kernel ridge regression and k-NN), highlighting the feature learning advantages of deep networks in high-dimensional input-output tasks.
Minimum width for universal approximation using ReLU networks on compact domain
Namjun Kim (Korea University), Sejun Park (Korea University)
Auto Encoder
🎯 What it does: This paper studies the minimum width required for RELU networks to achieve universal approximation of L_p functions on compact domains, deriving the exact value of the minimum width w_min as max{d, d_xy, 2}.
MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback
Xingyao Wang (University of Illinois Urbana-Champaign), Heng Ji (University of Illinois Urbana-Champaign)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes the MINT benchmark for evaluating the ability of large language models to utilize tools and natural language feedback to solve tasks in multi-turn interactions.
MIntRec2.0: A Large-scale Benchmark Dataset for Multimodal Intent Recognition and Out-of-scope Detection in Conversations
Hanlei Zhang (Tsinghua University), Yanting Chen (Tsinghua University)
ClassificationRecognitionTransformerLarge Language ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: A large-scale multimodal dialogue intent recognition and out-of-domain detection dataset MIntRec2.0 has been constructed, along with a unified evaluation framework.
Mirage: Model-agnostic Graph Distillation for Graph Classification
Mridul Gupta (Indian Institute of Technology), Sayan Ranu (Indian Institute of Technology)
Computational EfficiencyKnowledge DistillationGraph Neural NetworkGraph
🎯 What it does: A model-independent graph data distillation method called MIRAGE is designed, which compresses data by decomposing the graph into computational trees and utilizing their frequent patterns.
Mitigating Emergent Robustness Degradation while Scaling Graph Learning
Xiangchi Yuan (Brandeis University), Chuxu Zhang (Brandeis University)
OptimizationAdversarial AttackGraph Neural NetworkMixture of ExpertsAuto EncoderGraphBenchmark
🎯 What it does: To address the issue of 'destructive degradation' in graph neural networks under high-intensity node injection attacks, the DRAGON framework is proposed, which enhances the model's robustness and scalability against attacks through a preprocessed denoising autoencoder and a hierarchical differential privacy mixture of experts mechanism.
Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuning
Fuxiao Liu (University of Maryland), Lijuan Wang (Microsoft Corporation)
Object DetectionAnomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: A large-scale multimodal instruction tuning dataset (LRV-Instruction) and evaluation method (GAVIE) are proposed to reduce the hallucination phenomenon of large-scale multimodal models (LMM) under image + text instructions. Fine-tuning MiniGPT4 and mPLUG-Owl on this dataset significantly improves the model's authenticity and instruction-following ability.
Mitigating the Curse of Dimensionality for Certified Robustness via Dual Randomized Smoothing
Song Xia (Nanyang Technological University), Henghui Ding (Nanyang Technological University)
ClassificationOptimizationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: In response to the dimensionality curse faced by Randomized Smoothing for high-dimensional inputs when providing ℓ2 certified robustness, this paper proposes the Dual Randomized Smoothing (DRS) mechanism, which first divides the input image into two low-dimensional sub-images, then smooths them separately and aggregates the results to obtain the certified robustness of the original input.
Mixed-Type Tabular Data Synthesis with Score-based Diffusion in Latent Space
Hengrui Zhang (University of Illinois at Chicago), George Karypis (Amazon Web Services)
GenerationData SynthesisTransformerDiffusion modelScore-based ModelAuto EncoderTabular
🎯 What it does: A hybrid table data synthesis framework named TABSYN is proposed, which maps table data to the VAE latent space and trains a score-based diffusion model in the latent space to achieve high-quality synthesis.
MixSATGEN: Learning Graph Mixing for SAT Instance Generation
Xinyan Chen (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
GenerationData SynthesisOptimizationGraph Neural NetworkGraphBenchmark
🎯 What it does: This paper proposes MixSATGEN, a method for generating instances with adjustable hardness through graph matching and substructure replacement on the literal-clause graph of SAT.
MixSup: Mixed-grained Supervision for Label-efficient LiDAR-based 3D Object Detection
Yuxue Yang (University of Chinese Academy of Sciences), Zhaoxiang Zhang (University of Chinese Academy of Sciences)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: The MixSup framework is proposed, which combines a large number of coarse-grained clustering labels with a small number of precise box labels for 3D object detection of LiDAR point clouds, significantly reducing labeling costs and achieving a general integration for various detectors.
Mixture of LoRA Experts
Xun Wu (Tsinghua University), Furu Wei (Microsoft Research)
TransformerLarge Language ModelMixture of ExpertsImageTextMultimodality
🎯 What it does: This paper studies and proposes a new method for combining multiple LoRA (Low-Rank Adaptation) models—MOLE (Mixture of LoRA Experts). It achieves hierarchical weighted combinations by using learnable gating functions at each layer, allowing for a more efficient and dynamic integration of multiple pre-trained LoRAs.
Mixture of Weak and Strong Experts on Graphs
Hanqing Zeng (Meta AI), Jiebo Luo (University of Rochester)
ClassificationOptimizationExplainability and InterpretabilityGraph Neural NetworkMixture of ExpertsGraph
🎯 What it does: A mixed weak-strong expert (Mowst) model is proposed, combining a lightweight MLP with a powerful GNN to adaptively separate node features and neighborhood structures, and achieve expert collaboration through confidence gating.
Mixture-of-Experts Meets Instruction Tuning: A Winning Combination for Large Language Models
Sheng Shen (Google), Denny Zhou (Google)
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: This paper proposes the FLAN-MOE model, which combines sparse Mixture-of-Experts (MoE) with instruction tuning to achieve efficient scaling of large-scale language models and better general performance.
MMD Graph Kernel: Effective Metric Learning for Graphs via Maximum Mean Discrepancy
Yan Sun (Chinese University of Hong Kong), Jicong Fan (Chinese University of Hong Kong)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A graph kernel based on Maximum Mean Discrepancy (MMD) (MMD-GK) and its deep learnable version (Deep MMD-GK) are proposed. After obtaining node representations through message passing, MMD comparison is directly performed on graph distributions to obtain inter-graph similarity.
MMICL: Empowering Vision-language Model with Multi-Modal In-Context Learning
Haozhe Zhao (Peking University), Baobao Chang (Peking University)
RecognitionGenerationData-Centric LearningTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: A novel visual-language model MMICL is proposed, which can efficiently handle interleaved multimodal prompts containing multiple images, achieving multimodal context learning (ICL).
Model Merging by Uncertainty-Based Gradient Matching
Nico Daheim (Technical University of Darmstadt), Mohammad Emtiyaz Khan (RIKEN Center for Advanced Intelligence Project)
ClassificationOptimizationTransformerLarge Language ModelSupervised Fine-TuningImageText
🎯 What it does: A model merging method based on uncertainty gradient matching is proposed, utilizing gradient mismatch error to improve traditional parameter weighted averaging, thereby merging pre-trained models trained on different datasets more accurately.
Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs
Suyu Ge (University of Illinois Urbana-Champaign), Jianfeng Gao (Microsoft)
GenerationCompressionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Designed and implemented an adaptive KV cache compression method called FastGen, which significantly reduces GPU memory usage during LLM inference while maintaining generation quality.
Modeling Boundedly Rational Agents with Latent Inference Budgets
Athul Paul Jacob (Massachusetts Institute of Technology), Jacob Andreas (Massachusetts Institute of Technology)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningAgentic AISequential
🎯 What it does: This paper proposes and validates a 'Latent Inference Budget Model' (L-IBM), which explicitly simulates the computational constraints of different agents by introducing latent budget variables into any inference algorithm. It learns their objective functions and budget distributions, and applies them to three types of tasks: maze navigation, semantic reasoning, and chess.
Modeling state-dependent communication between brain regions with switching nonlinear dynamical systems
Orren Karniol-Tambour (Princeton Neuroscience Institute), Jonathan W. Pillow (Princeton Neuroscience Institute)
TransformerTime Series
🎯 What it does: A multi-region switching nonlinear dynamical model (MR-SDS) is proposed to capture state-dependent communication between brain regions.
Modelling complex vector drawings with stroke-clouds
Alexander Ashcroft (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: A complex vector drawing generation model based on 'stroke cloud' has been proposed, which can view drawings as a collection of an arbitrary number of Bezier curves. It generates new high-complexity vector drawings through a Set Transformer for encoding, MLP-based conditional diffusion decoding, and a latent diffusion generator.
ModernTCN: A Modern Pure Convolution Structure for General Time Series Analysis
Luo donghao, wang xue
ClassificationAnomaly DetectionConvolutional Neural NetworkTime Series
🎯 What it does: A pure convolutional model, ModernTCN, is proposed, achieving state-of-the-art performance in time series analysis tasks.
Modulate Your Spectrum in Self-Supervised Learning
Xi Weng (Beihang University), Lei Huang (Mohamed bin Zayed University of Artificial Intelligence)
Object DetectionSegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes the Spectral Transformation framework, exploring spectral transformations beyond whitening to avoid dimensional collapse in self-supervised learning, and introduces the IterNorm with trace loss (INTL) method.
Modulated Phase Diffusor: Content-Oriented Feature Synthesis for Detecting Unknown Objects
Aming WU, Cheng Deng (Xidian University)
Object DetectionDiffusion modelImage
🎯 What it does: This paper proposes a phase-based diffusion generative model called Modulated Phase Diffusion (MPD), which enhances unsupervised out-of-distribution (OOD) object detection performance by performing Gaussian averaging on the phase of ID features during forward diffusion and designing two reverse processes to synthesize OOD features and augmented features.
MOFDiff: Coarse-grained Diffusion for Metal-Organic Framework Design
Xiang Fu (Massachusetts Institute of Technology), Jake Allen Smith
GenerationOptimizationGraph Neural NetworkDiffusion modelContrastive LearningGraph
🎯 What it does: A framework for generating and optimizing MOFs based on a coarse-grained diffusion model, named MOFDiff, is proposed, which can generate and optimize MOF structures without relying on predefined topologies.
MOFI: Learning Image Representations from Noisy Entity Annotated Images
Wentao Wu (Apple AI ML), Yinfei Yang (Apple AI ML)
ClassificationRetrievalRepresentation LearningTransformerContrastive LearningImageText
🎯 What it does: In this paper, the authors constructed a large-scale image-entity dataset (I2E) and trained a new visual foundation model MOFI using three pre-training methods: supervised, contrastive, and multi-task. The model demonstrated strong performance on multiple image retrieval and classification benchmarks.
MogaNet: Multi-order Gated Aggregation Network
Siyuan Li (Westlake University), Stan Z. Li (Westlake University)
ClassificationObject DetectionSegmentationPose EstimationConvolutional Neural NetworkSupervised Fine-TuningImageVideo
🎯 What it does: A fully convolutional network called MogaNet is proposed, based on multi-order game theory interactions, integrating multi-order gated aggregation and channel redistribution modules to achieve more efficient visual representation.
Mol-Instructions: A Large-Scale Biomolecular Instruction Dataset for Large Language Models
Yin Fang (Zhejiang University), Huajun Chen (Zhejiang University)
Drug DiscoveryTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBiomedical Data
🎯 What it does: Designed and released the Mol-Instructions dataset, covering three main categories of instructions: molecules, proteins, and biological texts, with approximately 2,043,587 entries, and subsequently used this data for instruction tuning of LLMs;
Momentum Benefits Non-iid Federated Learning Simply and Provably
Ziheng Cheng (Peking University), Kun Yuan (Peking University)
OptimizationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes FedAvg-M, SCAFFOLD-M, and their variance-reduced versions by incorporating momentum into the local SGD steps of two mainstream federated learning algorithms, Federated Averaging (FedAvg) and SCAFFOLD, to improve convergence performance under non-iid conditions.
Monte Carlo guided Denoising Diffusion models for Bayesian linear inverse problems.
Gabriel Cardoso (Ecole Polytechnique), Eric Moulines (Ecole Polytechnique)
RestorationGenerationSuper ResolutionDiffusion modelImage
🎯 What it does: A sampling method for Bayesian linear inverse problems, MCGdiff, is proposed, utilizing a diffusion generative model prior to reconstruct unknown quantities.
More is Better: when Infinite Overparameterization is Optimal and Overfitting is Obligatory
James B Simon, Mikhail Belkin (University of California San Diego)
Convolutional Neural NetworkImage
🎯 What it does: This paper proves that under the framework of random feature regression and kernel ridge regression, increasing both the feature dimension and the sample size can strictly reduce the test error under optimal ridge regularization conditions. It further shows that in tasks with power law spectral structures, overfitting (where training error is much lower than test error) is necessary to achieve approximately optimal performance.
Most discriminative stimuli for functional cell type clustering
Max F Burg, Alexander S Ecker
OptimizationBiomedical Data
🎯 What it does: A method combining deep prediction models with EM-style clustering is proposed, utilizing the most discriminative stimuli (MDS) to simultaneously optimize stimuli and cell type clustering.
Motif: Intrinsic Motivation from Artificial Intelligence Feedback
Martin Klissarov (Mila), Mikael Henaff (FAIR at Meta)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Proposes the Motif method, which utilizes large language models (LLM) to evaluate preferences for environmental event descriptions (captions), generating intrinsic rewards for training agents in reinforcement learning;
Motion Guidance: Diffusion-Based Image Editing with Differentiable Motion Estimators
Daniel Geng (University of Michigan), Andrew Owens (University of Michigan)
Image TranslationGenerationDiffusion modelOptical FlowImage
🎯 What it does: Using optical flow estimators as guiding signals, dense motion editing of any pixel in the image is achieved during the sampling process of the diffusion model, supporting complex operations such as translation, rotation, and deformation.
MOTOR: A Time-to-Event Foundation Model For Structured Medical Records
Ethan Steinberg (Stanford University), Nigam Shah
TransformerTime SeriesSequentialBiomedical DataElectronic Health Records
🎯 What it does: This study proposes and pre-trains a time-to-event (TTE) foundational model named MOTOR, utilizing self-supervised TTE objective learning on large-scale clinical event sequences and transferring it to various medical prediction tasks.
MovingParts: Motion-based 3D Part Discovery in Dynamic Radiance Field
Kaizhi Yang (University of Science and Technology of China), Hao Su (University of California, San Diego)
Object DetectionSegmentationNeural Radiance FieldPoint Cloud
🎯 What it does: Dynamic scene reconstruction and motion-driven 3D part discovery based on NeRF, utilizing Eulerian and Lagrangian dual perspectives to model object motion and achieve part segmentation and editing.
MT-Ranker: Reference-free machine translation evaluation by inter-system ranking
Ibraheem Muhammad Moosa (Pennsylvania State University), Wenpeng Yin (Pennsylvania State University)
TransformerSupervised Fine-TuningText
🎯 What it does: A reference-free MT evaluation method called MT-Ranker is proposed, which directly predicts which of the two translations is better by comparing translations of the same source sentence, rather than providing an absolute quality score.
MUFFIN: Curating Multi-Faceted Instructions for Improving Instruction Following
Renze Lou (Pennsylvania State University), Wenpeng Yin (Pennsylvania State University)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: A new dataset construction paradigm called 'Scaling Tasks per Input' is proposed, and the first corresponding dataset MUFFIN is constructed to enhance the instruction-following ability of large language models.
Multi-granularity Correspondence Learning from Long-term Noisy Videos
Yijie Lin (Sichuan University), Xi Peng (Sichuan University)
RetrievalRepresentation LearningContrastive LearningVideoText
🎯 What it does: This study proposes Norton, a framework based on noise-robust optimal transport for learning multi-granularity correspondences across clips and subtitles from long-duration, noisy videos;
Multi-modal Gaussian Process Variational Autoencoders for Neural and Behavioral Data
Rabia Gondur (Fordham University), Stephen L Keeley
GenerationData SynthesisRepresentation LearningAuto EncoderMultimodalityTime Series
🎯 What it does: This paper proposes a multimodal Gaussian Process Variational Autoencoder (MM-GPVAE) that can simultaneously learn shared and independent temporal latent variables of neural activity and behavioral/stimulus data within the same model, achieving unsupervised spatiotemporal structure extraction and reconstruction.
Multi-Resolution Diffusion Models for Time Series Forecasting
Lifeng Shen (Hong Kong University of Science and Technology), James Kwok
GenerationData SynthesisOptimizationConvolutional Neural NetworkDiffusion modelTime Series
🎯 What it does: A multi-resolution diffusion model (mr-Diff) is proposed, which gradually decomposes time series into trends from fine to coarse and performs conditional denoising at different resolutions.
Multi-resolution HuBERT: Multi-resolution Speech Self-Supervised Learning with Masked Unit Prediction
Jiatong Shi (Carnegie Mellon University), Anna Sun (Meta AI)
RecognitionRepresentation LearningTransformerSupervised Fine-TuningAudio
🎯 What it does: This paper proposes a multi-resolution HuBERT (MR-HuBERT) that achieves single-model multi-resolution speech representation learning through hierarchical Transformer and HuBERT-style masked unit prediction.
Multi-Scale Representations by Varying Window Attention for Semantic Segmentation
Haotian Yan (Beijing University of Posts and Telecommunications), Chuang Zhang (Beijing University of Posts and Telecommunications)
SegmentationTransformerImage
🎯 What it does: A variable window attention (VWA) and a multi-scale decoder (VWFormer) based on it are proposed for efficiently learning multi-scale feature representations in semantic segmentation tasks.
Multi-Source Diffusion Models for Simultaneous Music Generation and Separation
Giorgio Mariani (Sapienza University of Rome), Emanuele Rodolà (Sapienza University of Rome)
GenerationData SynthesisDiffusion modelScore-based ModelOrdinary Differential EquationAudio
🎯 What it does: Train a multi-source diffusion model that can perform music mixing generation, source interpolation (partial generation), and source separation within the same model.
Multi-task Learning with 3D-Aware Regularization
Wei-Hong Li (University of Edinburgh), Hakan Bilen (University of Edinburgh)
SegmentationDepth EstimationNeural Radiance FieldImage
🎯 What it does: To address the problem of multi-task dense prediction, a 3D-aware multi-task learning framework is proposed that projects shared features into a triplane three-dimensional space and regularizes through differentiable rendering.
Multi-Task Reinforcement Learning with Mixture of Orthogonal Experts
Ahmed Hendawy (TU Darmstadt), Carlo D'Eramo
Reinforcement LearningMixture of Experts
🎯 What it does: Designed and implemented the MOORE (Mixture of Orthogonal Experts) framework, which enforces orthogonality among expert representations on the Stiefel manifold using the Gram-Schmidt process, thereby providing diverse and interpolable shared representations for multi-task reinforcement learning, ultimately learning a unified policy.
Multi-View Causal Representation Learning with Partial Observability
Dingling Yao (Institute of Science and Technology Austria), Francesco Locatello (Institute of Science and Technology Austria)
Representation LearningContrastive LearningMultimodality
🎯 What it does: This study explores a partially observable representation learning framework for multi-view nonlinear mixtures, providing a unified identifiability theory.
Multi-View Representation is What You Need for Point-Cloud Pre-Training
Siming Yan (University of Texas at Austin), Qixing Huang (University of Texas at Austin)
Object DetectionSegmentationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: A multi-view projection-based pre-training framework for 3D point clouds is proposed by projecting the knowledge of a pre-trained 2D network onto 3D point clouds.
Multilinear Operator Networks
Yixin Cheng (École Polytechnique Fédérale de Lausanne), Volkan Cevher (University of Wisconsin-Madison)
ClassificationRecognitionImage
🎯 What it does: This paper proposes a Fully Multilinear Operator Network (MONet) that captures high-order interactions of input elements using only linear and multilinear operations, completely independent of activation functions.
Multilingual Jailbreak Challenges in Large Language Models
Yue Deng (Alibaba Group), Lidong Bing (Alibaba Group)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study addresses the jailbreak challenges of multilingual large language models (LLMs), constructs the MultiJail dataset, and proposes the SELF-DEFENSE framework to enhance multilingual security.
Multimarginal Generative Modeling with Stochastic Interpolants
Michael Samuel Albergo, Eric Vanden-Eijnden (New York University)
Image TranslationGenerationImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A multi-marginal stochastic interpolation framework is proposed to generate a joint distribution between given K+1 marginal distributions and to identify multi-way correspondences.
Multimodal Learning Without Labeled Multimodal Data: Guarantees and Applications
Paul Pu Liang (Carnegie Mellon University), Russ Salakhutdinov
OptimizationData-Centric LearningMultimodalityBenchmark
🎯 What it does: In a semi-supervised scenario with only single-modal labeled data and unlabeled multi-modal paired data, a method for estimating multi-modal interactions (redundant, unique, and collaborative) is proposed, utilizing these interaction quantities to predict multi-modal model performance, guide data collection, and model selection.
Multimodal Molecular Pretraining via Modality Blending
Qiying Yu (Institute for AI Industry Research Tsinghua University), Jingjing Liu (Institute for AI Industry Research Tsinghua University)
Drug DiscoveryTransformerContrastive LearningMultimodalityGraph
🎯 What it does: A relational hierarchical multimodal molecular pre-training method called MOLEBLEND is proposed, which first mixes the atomic relationship matrices from 2D molecular graphs and 3D spatial structures into a unified input, and then encodes them through a Transformer to predict 2D/3D relationships, completing self-supervised learning.
Multimodal Patient Representation Learning with Missing Modalities and Labels
Zhenbang Wu (University of Illinois Urbana-Champaign), Faraz Faghri (National Institutes of Health)
Representation LearningGraph Neural NetworkContrastive LearningMultimodalityBiomedical DataAlzheimer's DiseaseElectronic Health Records
🎯 What it does: The MUSE method is proposed, which utilizes a patient-modal bipartite graph combined with mutually consistent contrastive learning to handle missing modalities and labels in multimodal clinical representation learning.
Multimodal Web Navigation with Instruction-Finetuned Foundation Models
Hiroki Furuta (University of Tokyo), Izzeddin Gur (Google DeepMind)
TransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelImageTextMultimodality
🎯 What it does: WebGUM has been developed, a multimodal web navigation agent based on instruction fine-tuning, which utilizes webpage screenshots, HTML, and natural language instructions to complete complex interactive tasks.
Multiscale Positive-Unlabeled Detection of AI-Generated Texts
Yuchuan Tian (Peking University), Yunhe Wang (Huawei)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes a Multi-Scale Positive and Negative Labeling (MPU) framework, modeling AI text detection as a partial positive and negative labeling problem, addressing the bottleneck of detecting short texts.
Multisize Dataset Condensation
Yang He (Agency for Science Technology and Research), Ivor Tsang
Data SynthesisCompressionImage
🎯 What it does: A multi-size dataset compression method (MDC) is proposed, which can obtain a multi-size synthetic dataset with a single compression.
MuSc: Zero-Shot Industrial Anomaly Classification and Segmentation with Mutual Scoring of the Unlabeled Images
Xurui Li (Huazhong University of Science and Technology), Yu Zhou (University of Trento)
ClassificationSegmentationAnomaly DetectionTransformerImage
🎯 What it does: A completely zero-shot industrial anomaly classification and segmentation method called MuSc is proposed, which detects anomalies by mutual scoring between unlabeled test images.
MuSR: Testing the Limits of Chain-of-thought with Multistep Soft Reasoning
Zayne Rea Sprague (University of Texas), Greg Durrett (University of Texas)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: The MuSR dataset is proposed, constructing multi-step soft reasoning tasks through natural language narratives, covering three domains: murder reasoning, object placement, and team allocation.
MUSTARD: Mastering Uniform Synthesis of Theorem and Proof Data
Yinya Huang (City University of Hong Kong), Xiaodan Liang (Shenzhen Campus of Sun Yat-sen University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: By interacting with large language models and the Lean prover, automatically generate step-by-step mathematical problems, natural language answers, and formal proofs, while filtering data verified by the prover.
MVDream: Multi-view Diffusion for 3D Generation
Yichun Shi (ByteDance), Xiao Yang (ByteDance)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldImagePoint Cloud
🎯 What it does: We propose and train MVDream—a multi-view diffusion model that can generate multi-view consistent images based on text descriptions, and use it as a 3D prior to generate 3D NeRF models through Score Distillation Sampling, while also supporting multi-view DreamBooth refinement with a small number of 2D examples.
MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo
Chenjie Cao (Fudan University), Yanwei Fu (Fudan University)
Depth EstimationTransformerImage
🎯 What it does: The multi-view stereo reconstruction method MVSFormer++ based on Transformer introduces cross-view attention and position encoding optimization in two major modules: feature extraction and cost volume regularization.
NaturalSpeech 2: Latent Diffusion Models are Natural and Zero-Shot Speech and Singing Synthesizers
Kai Shen (Zhejiang University), Jiang Bian (Zhejiang University)
GenerationData SynthesisDiffusion modelAudio
🎯 What it does: Developed NaturalSpeech 2, which combines continuous vector audio codecs and latent diffusion models to achieve natural, zero-shot TTS and singing synthesis;
Navigating Dataset Documentations in AI: A Large-Scale Analysis of Dataset Cards on HuggingFace
Xinyu Yang (Cornell University), James Zou (Stanford University)
Text
🎯 What it does: This study examined the structure, completeness, and content of 7,433 dataset documents on Hugging Face, combining it with download statistics, usage details, and human evaluations;
Navigating Text-To-Image Customization: From LyCORIS Fine-Tuning to Model Evaluation
SHIH-YING YEH, Yanmin Gong (University of Texas at San Antonio)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper presents the LyCORIS open-source library for efficiently fine-tuning various low-rank adapters (LoRA, LoHa, LoKr, etc.) for Stable Diffusion, and provides a systematic evaluation framework.
Navigating the Design Space of Equivariant Diffusion-Based Generative Models for De Novo 3D Molecule Generation
Tuan Le (Pfizer Research and Development), Kristof T Schütt (Pfizer Research and Development)
GenerationDrug DiscoveryGraph Neural NetworkDiffusion modelGraph
🎯 What it does: A diffusion model EQGAT-diff based on E(3)-equivariant graph attention has been developed for 3D molecular generation.
Near-Optimal Quantum Algorithm for Minimizing the Maximal Loss
Hao Wang (Peking University), Tongyang Li (Peking University)
Optimization
🎯 What it does: A quantum algorithm is proposed for finding the minimization point of the maximum value in a set of N convex Lipschitz functions, which can solve the problem with nearly optimal query complexity under a quantum zero-order oracle.
Near-Optimal Solutions of Constrained Learning Problems
Juan Elenter (University of Pennsylvania), Alejandro Ribeiro (University of Pennsylvania)
OptimizationTabular
🎯 What it does: This paper conducts a theoretical analysis of the Lagrangian dual algorithm for non-convex constrained learning problems, proving that under sufficiently rich model capacity, simply using the last or best iteration of the primal solution can approximate the optimal solution while ensuring constraint feasibility, thus eliminating the traditional necessity for randomized sampling; this theory is validated on the COMPAS dataset, demonstrating a significant decrease in constraint violations as model capacity increases.
Nearly $d$-Linear Convergence Bounds for Diffusion Models via Stochastic Localization
Joe Benton (University of Oxford), George Deligiannidis (University of Oxford)
OptimizationDiffusion modelStochastic Differential Equation
🎯 What it does: This paper presents convergence bounds for denoising diffusion models, proving that under the assumption of data distribution having a finite second moment, the convergence rate of the diffusion model is linear in terms of data dimension (with an additional logarithmic factor).
NECO: NEural Collapse Based Out-of-distribution detection
Mouïn Ben Ammar (U2IS Lab ENSTA Paris), Gianni Franchi (SafranTech)
Anomaly DetectionTransformerImage
🎯 What it does: A post-hoc OOD detection method called NECO is proposed, based on the geometric structure generated by the convergence phenomenon (Neural Collapse) of neural networks at the end of training, to directly determine whether the input is an outlier sample on a trained model.
NEFTune: Noisy Embeddings Improve Instruction Finetuning
Neel Jain (University of Maryland), Tom Goldstein (University of Maryland)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes an enhancement method (NEFTune) that injects random noise into embedding vectors during instruction fine-tuning, and verifies that it can significantly improve the performance of large language models in dialogue quality.
Negative Label Guided OOD Detection with Pretrained Vision-Language Models
Xue Jiang (Southern University of Science and Technology), Bo Han (Hong Kong Baptist University)
Domain AdaptationAnomaly DetectionTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: A zero-shot OOD detection framework called NegLabel is proposed, which is based on a pre-trained vision-language model and utilizes negative labels mined from a large corpus to distinguish between ID and OOD samples.
Negatively Correlated Ensemble Reinforcement Learning for Online Diverse Game Level Generation
Ziqi Wang (Southern University of Science and Technology), Xin Yao (Lingnan University)
GenerationReinforcement LearningSequentialBenchmark
🎯 What it does: A method for online diversified game level generation based on negative correlation ensemble reinforcement learning is proposed.
Nemesis: Normalizing the Soft-prompt Vectors of Vision-Language Models
Shuai Fu (Southern University of Science and Technology), Yu Zhang (Southern University of Science and Technology)
ClassificationDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper conducts systematic experiments on the norm of learnable soft prompt vectors in visual-language models, discovering the 'Low-Norm Effect.' Based on this, a soft prompt normalization method named Nemesis is proposed, which can dynamically adjust the norm during the soft prompt tuning process to enhance model performance.
NeRM: Learning Neural Representations for High-Framerate Human Motion Synthesis
Dong Wei (Nanjing University of Science and Technology), Jianfeng Lu (Nanjing University of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodality
🎯 What it does: A high frame rate human motion synthesis framework named NeRM is designed, combining implicit neural representations with diffusion models to achieve motion generation at arbitrary frame rates.
NetInfoF Framework: Measuring and Exploiting Network Usable Information
Meng-Chieh Lee (Carnegie Mellon University), Christos Faloutsos (Carnegie Mellon University)
ClassificationRecommendation SystemComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: The NETINFOF framework is proposed to measure and utilize the available information in graph structures and node features (NUI) without training a model, thereby achieving efficient inference in link prediction and node classification tasks.
Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings
Edouard YVINEC, Kevin Bailly (Datakalab)
CompressionTransformerLarge Language ModelDiffusion modelImage
🎯 What it does: This paper proposes a method for compressing network memory usage (JLCM) through a learnable codebook and mapping, which can compress large-scale models (such as Llama-7B, Stable Diffusion, etc.) to a few GB while maintaining nearly original accuracy.
Neur2RO: Neural Two-Stage Robust Optimization
Justin Dumouchelle (University of Toronto), Elias Boutros Khalil
OptimizationTabular
🎯 What it does: This paper proposes Neur2RO, a two-stage robust optimization solving framework that embeds neural network predictions of the second-stage value function into Column-Constraint Generation (CCG).
Neural Active Learning Beyond Bandits
Yikun Ban (University of Illinois Urbana-Champaign), Jingrui He (University of Illinois Urbana-Champaign)
ClassificationComputational EfficiencyReinforcement LearningTabular
🎯 What it does: A neural network-based active learning framework is proposed, featuring a novel utilization network and exploration network, achieving active learning in both streaming and pooling settings.
Neural Architecture Retrieval
Xiaohuan Pei (University of Sydney), Chang Xu (University of Sydney)
RetrievalNeural Architecture SearchGraph Neural NetworkContrastive LearningGraph
🎯 What it does: The problem of Neural Architecture Retrieval is proposed, and an efficient retrieval framework is provided.
Neural Atoms: Propagating Long-range Interaction in Molecular Graphs through Efficient Communication Channel
Xuan Li (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
Drug DiscoveryGraph Neural NetworkTransformerGraph
🎯 What it does: Proposes the Neural Atoms method, which maps original atoms to a few learnable neural atoms, performs information exchange, and then projects back to atoms to enhance the GNN's ability to capture long-range interactions.
Neural Auto-designer for Enhanced Quantum Kernels
Cong Lei (University of Sydney), Tongliang Liu (University of Sydney)
ClassificationAnomaly DetectionOptimizationTabular
🎯 What it does: Developed the QuKerNet framework, which automates the design of quantum feature maps tailored for specific tasks to enhance the performance of quantum kernels on NISQ devices.
Neural Common Neighbor with Completion for Link Prediction
Xiyuan Wang (Peking University), Muhan Zhang (Peking University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: The Neural Common Neighbor (NCN) model is proposed, and based on this, a soft completion technique (NCNC) is introduced to address the issue of graph incompleteness, thereby achieving more accurate link prediction.
Neural Contractive Dynamical Systems
Hadi Beik Mohammadi, Leonel Rozo (Bosch Center for Artificial Intelligence)
Robotic IntelligenceAuto EncoderTime SeriesOrdinary Differential Equation
🎯 What it does: A neural network architecture called Neural-Collapsible Dynamical Systems (NCDS) is proposed, which can learn globally collapsible dynamics from demonstration data while ensuring stability.
Neural Field Classifiers via Target Encoding and Classification Loss
Xindi Yang (Beijing Jiaotong University), Mingming Sun (Baidu Research)
ClassificationRestorationNeural Radiance FieldImage
🎯 What it does: This paper proposes a framework that transforms neural fields from regression to classification - the Neural Field Classifier (NFC), which enhances the rendering and reconstruction performance of neural fields through target encoding and classification loss.
Neural Fine-Tuning Search for Few-Shot Learning
Panagiotis Eustratiadis (University of Edinburgh), Timothy Hospedales (University of Edinburgh)
Domain AdaptationMeta LearningNeural Architecture SearchConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: Proposes the NFTS (Neural Fine-Tuning Search) framework, which utilizes neural architecture search to automatically determine which layers to freeze and which layers to insert adapters in a pre-trained network, achieving adaptive fine-tuning for cross-domain few-shot learning.
Neural Fourier Transform: A General Approach to Equivariant Representation Learning
Masanori Koyama (Preferred Networks), Takeru Miyato (University of Tübingen)
CompressionRepresentation LearningConvolutional Neural NetworkTransformerAuto EncoderImageTime Series
🎯 What it does: Proposes the Neural Fourier Transform (NFT) framework, which learns equivariant representations of unknown nonlinear group actions in data.
Neural Language of Thought Models
Yi-Fu Wu (Rutgers University), Sungjin Ahn (KAIST)
GenerationData SynthesisTransformerAuto EncoderImage
🎯 What it does: The Neural Language of Thought Model (NLoTM) is proposed, achieving unsupervised learning of structured, composable discrete semantic representations and their generation in images.
Neural Neighborhood Search for Multi-agent Path Finding
Zhongxia Yan (Massachusetts Institute of Technology), Cathy Wu (Massachusetts Institute of Technology)
OptimizationConvolutional Neural NetworkTransformerContrastive LearningGraphBenchmark
🎯 What it does: A neural network architecture that integrates 3D convolution and Transformer attention is designed to evaluate and select a large neighborhood search (LNS) subset in multi-agent pathfinding (MAPF), thereby rapidly improving paths during the iterative process.
Neural Network-Based Score Estimation in Diffusion Models: Optimization and Generalization
Yinbin Han (University of Southern California), Renyuan Xu (University of Southern California)
OptimizationDiffusion modelScore-based Model
🎯 What it does: This paper analyzes a two-layer fully connected neural network trained by gradient descent, proving that it can learn the score function in diffusion models, and provides theoretical analysis of optimization and generalization.
Neural Optimal Transport with General Cost Functionals
Arip Asadulaev (Artificial Intelligence Research Institute), Evgeny Burnaev (Artificial Intelligence Research Institute)
Image TranslationOptimizationSupervised Fine-TuningImage
🎯 What it does: A neural network-based algorithm is proposed for calculating optimal transport plans with general cost functions.
Neural Polynomial Gabor Fields for Macro Motion Analysis
Chen Geng (Stanford University), Jiajun Wu (Zhejiang University)
GenerationExplainability and InterpretabilityVideo
🎯 What it does: Proposes Phase-PGF, which uses low-dimensional phase representation for macro motion to complete editing tasks such as motion cycle detection, separation, amplification, and smoothing.
Neural Processing of Tri-Plane Hybrid Neural Fields
Adriano Cardace (University of Bologna), Luigi di Stefano
ClassificationSegmentationConvolutional Neural NetworkTransformerPoint CloudBenchmark
🎯 What it does: This paper proposes a direct approach to processing tri-plane hybrid neural fields to accomplish 3D object classification and segmentation tasks without the need to first reconstruct the neural fields into explicit representations.
Neural Rate Control for Learned Video Compression
Yiwei Zhang (Shanghai Jiao Tong University), Li Song (Shanghai Jiao Tong University)
CompressionConvolutional Neural NetworkVideo
🎯 What it does: A learning-based video compression rate control system based on a fully neural network is proposed, which can accurately allocate the bit rate for each frame under a given target bit rate and achieve fine encoding parameter mapping.
Neural SDF Flow for 3D Reconstruction of Dynamic Scenes
Wei Mao (Australian National University), miaomiao Liu
Flow-based ModelNeural Radiance FieldPoint Cloud
🎯 What it does: This paper proposes a dynamic scene 3D reconstruction method based on SDF flow, which captures surface evolution by estimating the first derivative of the SDF over time and derives the scene flow from it.
Neural Snowflakes: Universal Latent Graph Inference via Trainable Latent Geometries
Haitz Sáez de Ocáriz Borde (University of Oxford), Anastasis Kratsios (McMaster University)
Graph Neural NetworkPoint CloudGraph
🎯 What it does: A differentiable Neural Snowflake architecture is proposed, utilizing MLP and trainable fractal metrics to achieve isometric embedding of arbitrary finite weighted graphs, with experimental validation conducted on hidden graph inference tasks.
Neural Spectral Methods: Self-supervised learning in the spectral domain
Yiheng Du (University of California), Aditi S. Krishnapriyan (University of California)
TabularPhysics Related
🎯 What it does: In the spectral domain, a neural operator is used to learn the mapping of parameterized PDE solutions, achieving self-supervised training solely by minimizing the residual, without the need for internal solution data.
Neural structure learning with stochastic differential equations
Benjie Wang (University of California Los Angeles), Wenbo Gong (Microsoft Research)
Time SeriesStochastic Differential Equation
🎯 What it does: A new structure learning method called SCOTCH is proposed, which combines neural stochastic differential equations (SDE) and variational inference to infer the posterior distribution of possible structures.