NeurIPS 2023 Papers — Page 19
Conference on Neural Information Processing Systems · 3218 papers
MGDD: A Meta Generator for Fast Dataset Distillation
Songhua Liu (National University of Singapore), Xinchao Wang (National University of Singapore)
Knowledge DistillationData-Centric LearningMeta LearningImage
🎯 What it does: A meta-learning-based conditional generator is proposed, which quickly adapts to the target dataset for one-shot forward generation of dataset distillation without the need for backpropagation.
Michelangelo: Conditional 3D Shape Generation based on Shape-Image-Text Aligned Latent Representation
Zibo Zhao (Tencent PCG), Shenghua Gao (ShanghaiTech University)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningMultimodalityMesh
🎯 What it does: A framework is proposed that first aligns and then generates, utilizing a latent space of shape-image-text alignment for 3D shape generation.
MIM4DD: Mutual Information Maximization for Dataset Distillation
Yuzhang Shang (Illinois Institute of Technology), Yan Yan (Illinois Institute of Technology)
Data SynthesisKnowledge DistillationContrastive LearningImage
🎯 What it does: A framework called MIM4DD is proposed, which utilizes mutual information (MI) maximization to guide dataset distillation (DD). It constructs positive and negative sample pairs through contrastive learning and indirectly maximizes the MI between the real dataset and the synthetic dataset using NCE loss, thereby generating synthetic data with better information retention capabilities.
MIMEx: Intrinsic Rewards from Masked Input Modeling
Toru Lin (University of California Berkeley), Allan Jabri (University of California Berkeley)
Robotic IntelligenceTransformerReinforcement LearningAuto EncoderSequential
🎯 What it does: A general exploration framework based on masked input modeling, MIMEx, is proposed, which utilizes sequence-level masked autoencoders to generate intrinsic rewards that encourage agents to actively explore in sparse reward visual control environments.
MIMONets: Multiple-Input-Multiple-Output Neural Networks Exploiting Computation in Superposition
Nicolas Menet (ETH Zurich), Abbas Rahimi (IBM Research)
Computational EfficiencyConvolutional Neural NetworkTransformerImageSequential
🎯 What it does: This paper proposes the MIMONets framework, which can simultaneously handle multiple sets of inputs in a single forward pass, significantly improving inference throughput.
Mind the spikes: Benign overfitting of kernels and neural networks in fixed dimension
Moritz Haas (University of Tübingen), Ingo Steinwart (University of Stuttgart)
Tabular
🎯 What it does: The paper demonstrates that by designing a 'spiky-smooth' kernel and activation function with sharp high-frequency components, benign overfitting can still be achieved in fixed dimensions, and provides corresponding theoretical consistency and inconsistency analyses.
Minimax Forward and Backward Learning of Evolving Tasks with Performance Guarantees
Veronica Alvarez, Jose A. Lozano (University of the Basque Country)
ClassificationOptimizationImageTabular
🎯 What it does: This paper proposes a minimax risk classifier (IMRC) for gradually learning evolving tasks, which dynamically constructs uncertainty sets through forward and backward learning, thereby improving performance in multi-task sequences.
Minimax Optimal Rate for Parameter Estimation in Multivariate Deviated Models
Dat Do (University of Michigan), Nhat Ho (University of Texas at Austin)
Tabular
🎯 What it does: This paper studies the convergence rate of maximum likelihood estimation (MLE) for parameters λ, μ, and Σ in a mixture model with known distribution h0 and unknown component f, providing corresponding lower bounds and proving that optimal minimax rates can be achieved under different identifiability assumptions.
Minimax Risks and Optimal Procedures for Estimation under Functional Local Differential Privacy
Bonwoo Lee (Korea Advanced Institute of Science and Technology), Cheolwoo Park (Korea Advanced Institute of Science and Technology)
Safty and Privacy
🎯 What it does: This paper derives the optimal minimax risk for univariate mean estimation and nonparametric density estimation under the framework of Functional Local Differential Privacy (FLDP), and provides the corresponding closed-form optimal mechanisms.
Minimax-Optimal Location Estimation
Shivam Gupta (University of Texas at Austin), Paul Valiant (University of Texas at Austin)
OptimizationTabular
🎯 What it does: Two position estimators are proposed and implemented: one is the optimal point estimator, and the other is the optimal confidence interval estimator; both are designed for a translation model with a known distribution shape.
Minimum Description Length and Generalization Guarantees for Representation Learning
Milad Sefidgaran (Huawei Technologies France), Piotr Krasnowski (Huawei Technologies France)
CompressionRepresentation LearningConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a compression framework based on Minimum Description Length (MDL) and symmetric priors to provide an upper bound on the generalization error of representation learning, particularly for information bottleneck-style encoders.
Minimum norm interpolation by perceptra: Explicit regularization and implicit bias
Jiyoung Park (Texas A&M University), Stephan Wojtowytsch (University of Pittsburgh)
OptimizationTabular
🎯 What it does: This paper studies the minimum norm interpolation of shallow ReLU networks in an over-parameterized setting and proves that under appropriate regularization, the empirical risk minimizer converges to the minimum norm interpolator.
Minimum-Risk Recalibration of Classifiers
Zeyu Sun (University of Michigan), Alfred Hero (University of Michigan)
ClassificationOptimizationTabular
🎯 What it does: A minimum risk recalibration theory based on MSE decomposition is proposed, along with a unified risk upper bound and optimal bin number.
Mip-Grid: Anti-aliased Grid Representations for Neural Radiance Fields
Seungtae Nam (AI2XL KT), Eunbyung Park (Sungkyunkwan University)
GenerationData SynthesisOptimizationComputational EfficiencyNeural Radiance FieldMesh
🎯 What it does: A method called mip-Grid is proposed for anti-aliasing grid representation, aimed at improving the multi-scale rendering of NeRF and avoiding aliasing or blurriness at different camera distances.
Mirror Diffusion Models for Constrained and Watermarked Generation
Guan-Horng Liu (Georgia Institute of Technology), Molei Tao (Georgia Institute of Technology)
GenerationData SynthesisOptimizationDiffusion modelImage
🎯 What it does: A mirror diffusion model (MDM) is proposed, achieving trainable and simulation-free diffusion generation on convex constraint sets.
Mitigating Over-smoothing in Transformers via Regularized Nonlocal Functionals
Tam Minh Nguyen, Richard Baraniuk
ClassificationSegmentationTransformerImageText
🎯 What it does: The NeuTRENO method is proposed, which alleviates the over-smoothing problem of Transformers by incorporating a regularized non-local functional into self-attention.
Mitigating Source Bias for Fairer Weak Supervision
Changho Shin (University of Wisconsin Madison), Frederic Sala (University of Wisconsin Madison)
TabularBenchmark
🎯 What it does: This study investigates the fairness bias introduced by label functions in weakly supervised training and proposes a source bias mitigation method based on adversarial fairness.
Mitigating Test-Time Bias for Fair Image Retrieval
Fanjie Kong (Duke University), Ricardo Henao (Duke University)
RetrievalTransformerSupervised Fine-TuningVision Language ModelImage
🎯 What it does: This paper studies gender/race bias in image retrieval under neutral text queries and proposes a post-processing method called PBM to mitigate bias during testing.
Mitigating the Effect of Incidental Correlations on Part-based Learning
Gaurav Bhatt (University of British Columbia), Vineeth N. Balasubramanian
ClassificationRecognitionExplainability and InterpretabilityKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes DPViT, which learns decoupled components of foreground and background through a mixture-of-parts approach, alleviating the decline in interpretability and generalization caused by incidental correlations.
Mitigating the Popularity Bias of Graph Collaborative Filtering: A Dimensional Collapse Perspective
Yifei Zhang (Chinese University of Hong Kong), Irwin King (Chinese University of Hong Kong)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A decorrelation objective based on LogDet divergence is proposed, applied to Graph Collaborative Filtering (GCF), to alleviate popularity bias caused by dimensional collapse and improve the recommendation performance for unpopular items.
Mix-of-Show: Decentralized Low-Rank Adaptation for Multi-Concept Customization of Diffusion Models
Yuchao Gu (Show Lab), Mike Zheng Shou
GenerationDiffusion modelImage
🎯 What it does: The Mix-of-Show framework is proposed, achieving decentralized multi-concept customization, namely client-side single-concept LoRA tuning and central node gradient fusion, as well as region-controllable sampling for multi-concept image generation.
Mixed Samples as Probes for Unsupervised Model Selection in Domain Adaptation
Dapeng Hu (National University of Singapore), Xinchao Wang (National University of Singapore)
SegmentationDomain AdaptationHyperparameter SearchImage
🎯 What it does: A method for unsupervised domain adaptation model selection based on mixed samples, called MixVal, is proposed, which can evaluate and select the best UDA model using only unlabeled target domain data during the inference phase.
Mixed-Initiative Multiagent Apprenticeship Learning for Human Training of Robot Teams
Esmaeil Seraj (Georgia Institute of Technology), Matthew Gombolay (Georgia Institute of Technology)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningGenerative Adversarial NetworkSequential
🎯 What it does: Proposes the MixTURE framework, which utilizes demonstration data from a single human expert to train multi-robot teams to achieve collaborative tasks in complex, partially observable, and heterogeneous environments, while automatically learning efficient internal communication protocols.
MixFormerV2: Efficient Fully Transformer Tracking
Yutao Cui (Nanjing University), Limin Wang (Nanjing University)
Object TrackingComputational EfficiencyKnowledge DistillationTransformerVideo
🎯 What it does: MixFormerV2 is proposed, a vision object tracking framework completely based on Transformer, which removes convolutional heads and complex scoring modules;
Mixture Weight Estimation and Model Prediction in Multi-source Multi-target Domain Adaptation
Yuyang Deng (Pennsylvania State University), Mehrdad Mahdavi (Pennsylvania State University)
Domain AdaptationImageStochastic Differential Equation
🎯 What it does: A multi-source multi-target domain adaptation framework based on mixed weight estimation and model prediction is proposed to solve the efficient computation problem of optimal mixed weight selection and multi-target co-component minimization.
MKOR: Momentum-Enabled Kronecker-Factor-Based Optimizer Using Rank-1 Updates
Mohammad Mozaffari (University of Toronto), Maryam Mehri Dehnavi (University of Toronto)
OptimizationConvolutional Neural NetworkTransformerSupervised Fine-TuningAuto EncoderImageText
🎯 What it does: A second-order optimizer named MKOR is proposed, which combines the Kronecker factor and rank-1 updates with momentum, significantly reducing training latency and convergence error.
MMD-Fuse: Learning and Combining Kernels for Two-Sample Testing Without Data Splitting
Felix Biggs (University College London), Arthur Gretton (University College London)
Anomaly DetectionOptimizationAuto EncoderContrastive LearningImage
🎯 What it does: Proposes the MMD-FUSE statistic, which achieves adaptive kernel selection and fusion of multiple kernel MMD two-sample tests using soft maximization and unsupervised feature learning without splitting the data.
MMGP: a Mesh Morphing Gaussian Process-based machine learning method for regression of physical problems under nonparametrized geometrical variability
Fabien Casenave (Safran Tech), Xavier Roynard (Safran Tech)
Graph Neural NetworkMeshPhysics Related
🎯 What it does: A Gaussian Process regression method based on mesh deformation, finite element interpolation, and PCA dimensionality reduction is proposed to learn the physical simulation results under non-parametric geometric deformations.
Mnemosyne: Learning to Train Transformers with Transformers
Deepali Jain (Google DeepMind), Jie Tan (Google DeepMind)
OptimizationTransformerSupervised Fine-TuningImage
🎯 What it does: A new learnable optimizer Mnemosyne is proposed, which utilizes spatio-temporal low-rank implicit attention Transformers to train large models like Transformers without the need for task-specific optimizer tuning.
Mobilizing Personalized Federated Learning in Infrastructure-Less and Heterogeneous Environments via Random Walk Stochastic ADMM
Ziba Parsons (University of Michigan), Jin Lu (University of Georgia)
Federated LearningImageBenchmarkStochastic Differential Equation
🎯 What it does: A mobile personalized federated learning framework RWSADMM is proposed in an environment lacking infrastructure and with uneven data distribution, using server random walks and hard constraints to achieve local neighbor model consistency.
MoCa: Measuring Human-Language Model Alignment on Causal and Moral Judgment Tasks
Allen Nie (Stanford University), Tobias Gerstenberg (Stanford University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: By integrating causal and moral judgment factors from 24 cognitive science papers, a structured annotated causal and moral judgment challenge set was constructed, and various large language models' alignment performance was evaluated on this set.
Modality-Agnostic Self-Supervised Learning with Meta-Learned Masked Auto-Encoder
Huiwon Jang (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
Representation LearningMeta LearningTransformerAuto EncoderContrastive LearningImageMultimodalityTabularTime SeriesBenchmarkAudio
🎯 What it does: This paper proposes a modality-agnostic self-supervised learning framework called MetaMAE based on Masked Autoencoders (MAE), treating MAE as a meta-learning task to further enhance cross-modal representation learning effectiveness.
Modality-Independent Teachers Meet Weakly-Supervised Audio-Visual Event Parser
Yung-Hsuan Lai (National Taiwan University), Yu-Chiang Frank Wang (NVIDIA)
RecognitionData-Centric LearningTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes the VALOR method, which significantly improves model performance in weakly labeled audio-visual event parsing tasks by generating fine-grained, modality-aware pseudo-labels using large contrastive pre-trained models CLIP and CLAP, and extends this framework to weakly supervised audio-visual event localization tasks.
Mode Connectivity in Auction Design
Christoph Hertrich (London School of Economics and Political Science), László A. Végh (London School of Economics and Political Science)
OptimizationTabular
🎯 What it does: The paper discusses the pattern connectivity in auction design, particularly through neural networks (such as RochetNet) to learn and discover optimal auction mechanisms.
Model and Feature Diversity for Bayesian Neural Networks in Mutual Learning
Cuong Pham (Monash University), Thanh-Toan Do (Monash University)
ClassificationOptimizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes to enhance model performance by introducing diversity constraints in the parameter space and feature space within the framework of mutual learning in Bayesian neural networks.
Model Shapley: Equitable Model Valuation with Black-box Access
Xinyi Xu (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
ClassificationImageBiomedical Data
🎯 What it does: A fair model valuation method based on black-box access, called Model Shapley, is proposed, which abstracts model predictions using the Dirichlet distribution and measures model value through Shapley values.
Model Sparsity Can Simplify Machine Unlearning
Jinghan Jia (Michigan State University), Sijia Liu (IBM Research)
ClassificationOptimizationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: This paper explores and verifies that model sparsification achieved through weight pruning can significantly enhance the performance of approximate machine unlearning (MU). It proposes two new paradigms: 'prune first, then unlearn' and sparse-aware unlearning, and applies them to backdoor defense and transfer learning.
Model Spider: Learning to Rank Pre-Trained Models Efficiently
Yi-Kai Zhang (Nanjing University), Han-Jia Ye (Nanjing University)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelImageTextMultimodality
🎯 What it does: This paper proposes the MODEL SPIDER method, which achieves efficient model selection by mapping pre-trained models and tasks to vector spaces and learning the similarity between the two.
Model-Based Control with Sparse Neural Dynamics
Ziang Liu (Cornell University), Yunzhu Li (University of Illinois Urbana-Champaign)
OptimizationRobotic IntelligenceNeural Architecture SearchGraph Neural NetworkReinforcement LearningTabularTime Series
🎯 What it does: By gradually sparsifying the ReLU units in neural networks and solving the sparse dynamic model using Mixed Integer Programming (MIP), efficient model-based closed-loop control is achieved.
Model-Based Reparameterization Policy Gradient Methods: Theory and Practical Algorithms
Shenao Zhang (Northwestern University), Tuo Zhao (Georgia Tech)
Reinforcement LearningSequential
🎯 What it does: Analyzed and proved the convergence of the model-based reparameterization policy gradient method (RP-PGM) and the sources of gradient variance/bias, and proposed the Spectral Normalization technique to suppress gradient explosion and improve learning efficiency.
Model-enhanced Vector Index
Hailin Zhang (Peking University), Bin CUI
RetrievalTransformerTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a retrieval framework called Model-Enhanced Vector Index (MEVI), which integrates autoregressive sequence generation models and dual-tower dense retrieval models to achieve high recall and low-latency retrieval for large-scale corpora.
Model-Free Active Exploration in Reinforcement Learning
Alessio Russo (KTH Royal Institute of Technology), Alexandre Proutiere (KTH Royal Institute of Technology)
Reinforcement Learning
🎯 What it does: A model-free active exploration algorithm MF-BPI (and its deep learning version DBMF-BPI) is proposed, which dynamically allocates sampling resources by estimating the variance of the Q-function and value function, thereby quickly finding near-optimal policies in unknown MDPs.
Model-free Posterior Sampling via Learning Rate Randomization
Daniil Tiapkin (École Polytechnique), Pierre MENARD
OptimizationReinforcement LearningTabular
🎯 What it does: A computable model-agnostic posterior sampling algorithm RandQL based on learning rate randomization is proposed to achieve optimistic exploration without reward terms.
Model-Free Reinforcement Learning with the Decision-Estimation Coefficient
Dylan J Foster, Ayush Sekhari (Massachusetts Institute of Technology)
Meta LearningReinforcement Learning
🎯 What it does: A novel meta-algorithm (E2D.Opt) is proposed, which combines Optimistic Estimation with the Decision-Estimation Coefficient (DEC) framework, achieving sample-efficient learning in model-free reinforcement learning (model-free RL) and providing corresponding upper and lower bounds.
Modeling Dynamics over Meshes with Gauge Equivariant Nonlinear Message Passing
Jung Yeon Park (Northeastern University), Robin Walters (Northeastern University)
Graph Neural NetworkMesh
🎯 What it does: A new Gauge-equivariant nonlinear message passing model, Hermes, is proposed on three-dimensional surface meshes to learn and predict complex surface dynamics and other mesh-related tasks.
Modeling Human Visual Motion Processing with Trainable Motion Energy Sensing and a Self-attention Network
Zitang Sun (Kyoto University), Shin'ya Nishida (Nippon Telegraph and Telephone Corporation)
Convolutional Neural NetworkRecurrent Neural NetworkOptical FlowVideo
🎯 What it does: A two-stage visual motion processing model was designed and trained. The first stage uses trainable space-time Gabor filters to extract local motion energy, while the second stage implements global motion integration and separation through self-attention and recurrent networks, generating optical flow that aligns with human perception.
Modelling Cellular Perturbations with the Sparse Additive Mechanism Shift Variational Autoencoder
Michael Bereket (insitro), Theofanis Karaletsos (insitro)
GenerationRepresentation LearningDrug DiscoveryAuto EncoderBiomedical Data
🎯 What it does: This paper proposes a Sparse Additive Mechanism Shift Variational Autoencoder (SAMS-VAE) for learning the latent representations of cells under different interventions and predicting gene expression.
Modulated Neural ODEs
Ilze Amanda Auzina (University of Amsterdam), Efstratios Gavves (University of Amsterdam)
Time SeriesSequentialOrdinary Differential Equation
🎯 What it does: Proposes the Modulated Neural ODE (MoNODE) framework, which adds time-invariant modulation factors (static and dynamic modulators) to the traditional NODE, achieving separation of dynamic states and inherent factors;
Module-wise Adaptive Distillation for Multimodality Foundation Models
Chen Liang (Georgia Institute of Technology), Tianyi Zhou (University of Maryland)
CompressionOptimizationKnowledge DistillationTransformerVision Language ModelMultimodality
🎯 What it does: This paper proposes a module-level adaptive distillation method called OPTIMA, which enhances the compression effect of multimodal foundational models by dynamically selecting different modules for distillation.
Module-wise Training of Neural Networks via the Minimizing Movement Scheme
Skander Karkar (Criteo), patrick gallinari
OptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: A modular training method based on transport regularization (TRGL) is proposed to address the issues of overfitting and stagnation in modular training.
Molecule Joint Auto-Encoding: Trajectory Pretraining with 2D and 3D Diffusion
weitao Du, Shengchao Liu (Université de Montréal)
GenerationDrug DiscoveryGraph Neural NetworkDiffusion modelContrastive LearningGraph
🎯 What it does: This paper studies a joint self-supervised pre-training framework called MoleculeJAE, which learns to capture both 2D topology and 3D conformation information of molecules through diffusion trajectories.
Moment Matching Denoising Gibbs Sampling
Mingtian Zhang (University College London), David Barber (University College London)
RestorationGenerationData SynthesisDiffusion modelScore-based ModelImage
🎯 What it does: This paper proposes a sampling framework based on 'pseudo Gibbs sampling' and 'moment matching', utilizing the noise EBM trained through Denoising Score Matching (DSM) to sample directly from its implicit clean model.
MomentDiff: Generative Video Moment Retrieval from Random to Real
Pandeng Li (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
GenerationRetrievalDiffusion modelVideoText
🎯 What it does: A generative video moment retrieval framework called MomentDiff is proposed, simulating the retrieval process from random browsing to gradual localization, similar to human behavior.
Momentum Provably Improves Error Feedback!
Ilyas Fatkhullin (ETH Zurich), Peter Richtárik (King Abdullah University of Science and Technology)
OptimizationReinforcement LearningImage
🎯 What it does: In distributed uncompressed gradient descent, Polyak momentum is added to improve the EF21 error feedback algorithm, eliminating the need for large batch sizes and enhancing sample and communication efficiency;
Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture
Daniel Y Fu, Christopher Re
TransformerImageText
🎯 What it does: Proposed MONARCH MIXER (M2), which uses the Monarch matrix to achieve sub-quadratic mixing in both sequence and model dimensions.
Monitor-Guided Decoding of Code LMs with Static Analysis of Repository Context
Lakshya Agrawal, Sriram Rajamani
GenerationAI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposes Monitoring-Driven Decoding (MGD), which enhances code quality by applying static analysis from the IDE to constrain code in real-time during the generation process.
MonoUNI: A Unified Vehicle and Infrastructure-side Monocular 3D Object Detection Network with Sufficient Depth Clues
Jinrang Jia (Baidu Inc), Yifeng Shi (Baidu Inc)
Object DetectionAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a unified normalized depth objective and 3D cubic depth supervision to address the differences in focal length and pitch angle between monocular 3D detection on the vehicle side and the infrastructure side.
Monte Carlo Tree Search with Boltzmann Exploration
Michael Painter (Oxford Robotics Institute University of Oxford), Bruno Lacerda (Oxford Robotics Institute University of Oxford)
OptimizationReinforcement LearningSequential
🎯 What it does: This paper proposes two Monte Carlo tree search algorithms based on Boltzmann sampling, BTS and DENTS, to address the inconsistency issue of the original MENTS in maximizing rewards.
Moral Responsibility for AI Systems
Sander Beckers (University of Amsterdam)
🎯 What it does: A definition of moral responsibility for AI systems is proposed, based on causal models' CNESS causal conditions and comprehensive cognitive conditions.
MosaicBERT: A Bidirectional Encoder Optimized for Fast Pretraining
Jacob Portes (Databricks), Jonathan Frankle (Databricks)
TransformerLarge Language ModelText
🎯 What it does: This paper presents MosaicBERT, a BERT-style encoder optimized for fast pre-training.
Most Neural Networks Are Almost Learnable
Amit Daniely (Hebrew University and Google), Gal Vardi (TTI-Chicago and Hebrew University)
🎯 What it does: The paper proposes a Polynomial Time Approximation Scheme (PTAS) that can learn constant-depth neural networks initialized randomly with Xavier under any given error ϵ.
MotionGPT: Human Motion as a Foreign Language
Biao Jiang (Fudan University), Tao Chen (Fudan University)
GenerationPose EstimationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVideoText
🎯 What it does: Treating human motion as a foreign language, a unified MotionGPT model is constructed to accomplish multiple tasks (text-driven motion generation, motion description, motion prediction, etc.).
MoVie: Visual Model-Based Policy Adaptation for View Generalization
Sizhe Yang (Shanghai Qi Zhi Institute), Huazhe Xu (Tsinghua University)
Robotic IntelligenceReinforcement LearningImage
🎯 What it does: Proposes a perspective adaptation (MoVie) for visual model-based policies using a frozen dynamic model and spatial adaptive encoder during testing.
Multi Time Scale World Models
Vaisakh Shaj (Karlsruhe Institute Of Technology), Gerhard Neumann (Karlsruhe Institute Of Technology)
Robotic IntelligenceReinforcement Learning from Human FeedbackRecurrent Neural NetworkTransformerWorld ModelTime SeriesSequential
🎯 What it does: A multi-time scale world model (MTS3) is proposed, which can learn environmental dynamics at different time granularities and make long-horizon predictions.
Multi-Agent First Order Constrained Optimization in Policy Space
Youpeng Zhao (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
OptimizationReinforcement Learning
🎯 What it does: Proposes MAFOCOPS, a safety multi-agent reinforcement learning algorithm based on first-order constraint optimization.
Multi-Agent Learning with Heterogeneous Linear Contextual Bandits
Anh Do (Johns Hopkins University), Raman Arora (Johns Hopkins University)
Reinforcement Learning
🎯 What it does: This paper studies the heterogeneous multi-agent linear contextual bandit problem and proposes a distributed learning algorithm H-LINUCB based on UCB, providing theoretical upper and lower bounds when the heterogeneity ε is known.
Multi-Agent Meta-Reinforcement Learning: Sharper Convergence Rates with Task Similarity
Weichao Mao (University of Illinois Urbana-Champaign), Tamer Basar (University of Illinois Urbana-Champaign)
Meta LearningReinforcement LearningSequential
🎯 What it does: A multi-task meta-learning framework is proposed, providing optimization-based algorithms for zero-sum Markov games, Markov potential games, and general Markov games, and it is proven that under task similarity, the convergence speed can be significantly improved.
Multi-body SE(3) Equivariance for Unsupervised Rigid Segmentation and Motion Estimation
Jia-Xing Zhong (Oxford), Niki Trigoni (Oxford)
SegmentationPose EstimationAutonomous DrivingPoint Cloud
🎯 What it does: A lightweight two-headed network based on SE(3) equivariance is proposed, which simultaneously performs rigid body segmentation and motion estimation under unsupervised conditions.
Multi-Fidelity Multi-Armed Bandits Revisited
Xuchuang Wang (Chinese University of Hong Kong), John C.S. Lui (Chinese University of Hong Kong)
OptimizationReinforcement Learning
🎯 What it does: Research on the optimal arm identification (fixed confidence) and cumulative loss minimization problem under the multi-fidelity multi-armed bandit (MF-MAB) model.
Multi-Head Adapter Routing for Cross-Task Generalization
Lucas Caccia, Alessandro Sordoni
TransformerSupervised Fine-TuningText
🎯 What it does: After multi-task pre-training, few-shot fine-tuning across tasks is achieved using a multi-head routing-based adapter (MHR).
Multi-Modal Inverse Constrained Reinforcement Learning from a Mixture of Demonstrations
Guanren Qiao (Chinese University of Hong Kong), zhiqiang xu
Robotic IntelligenceReinforcement LearningFlow-based ModelContrastive LearningMultimodality
🎯 What it does: A multi-modal inverse constraint reinforcement learning (MMICRL) algorithm is proposed to unsupervisedly identify different types of experts in mixed expert demonstration data and estimate the corresponding constraint functions, thereby achieving multi-constraint imitation learning.
Multi-modal Queried Object Detection in the Wild
Yifan Xu (Institute of Automation, Chinese Academy of Sciences), Changsheng Xu (Institute of Automation, Chinese Academy of Sciences)
Object DetectionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a multimodal query object detection framework, MQ-Det, which achieves unsupervised open-set detection under the condition of providing only text descriptions and visual instances, and efficiently performs incremental tuning on baseline language query detection models (such as GLIP, GroundingDINO).
Multi-Object Representation Learning via Feature Connectivity and Object-Centric Regularization
Alex Foo (National University of Singapore), Mong-Li Lee
SegmentationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an unsupervised multi-object representation learning framework called OC-Net, which utilizes pixel feature connectivity clustering and further optimizes object representations through two types of regularization.
Multi-Objective Intrinsic Reward Learning for Conversational Recommender Systems
Zhendong Chu (University of Virginia), Hongning Wang (University of Virginia)
Recommendation SystemReinforcement LearningContrastive LearningText
🎯 What it does: To address the challenge of reward function design in dialog-based recommendation systems, this paper proposes a method for intrinsic reward learning based on multi-objective bi-level optimization (CRSIRL).
Multi-Player Zero-Sum Markov Games with Networked Separable Interactions
Chanwoo Park (Massachusetts Institute of Technology), Asuman E. Ozdaglar
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningGraph
🎯 What it does: This paper studies the structure, computational difficulty, and learning dynamics of networked separable interactions in multi-player zero-sum Markov games (Zero-Sum NMG). It presents a convergence proof for fictitious play under star networks and provides a value iteration algorithm for computing non-stationary Nash Equilibria (NE).
Multi-Prompt Alignment for Multi-Source Unsupervised Domain Adaptation
Haoran Chen (Fudan University), Yu-Gang Jiang (Fudan University)
Domain AdaptationPrompt EngineeringAuto EncoderContrastive LearningImage
🎯 What it does: This paper proposes a multi-source unsupervised domain adaptation framework called MPA based on prompt learning. It first learns low-dimensional prompts for each source-target domain pair, then uses an autoencoder to denoise and align the prompts, and finally makes predictions in the target domain. Additionally, it introduces the Latent Subspace Tuning (LST) strategy, which utilizes the learned latent space to quickly adapt to multiple target domains.
Multi-resolution Spectral Coherence for Graph Generation with Score-based Diffusion
Hyuna Cho (POSTECH), Won Hwa Kim (POSTECH)
GenerationData SynthesisGraph Neural NetworkDiffusion modelScore-based ModelGraph
🎯 What it does: This paper proposes Wave-GD, which constructs a score-based diffusion model for graph generation by capturing the spectral synergy of nodes and edges at multiple scales in the spectral wavelet transform space.
Multi-scale Diffusion Denoised Smoothing
Jongheon Jeong (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
RestorationOptimizationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a multi-scale diffusion denoised smoothing scheme, which utilizes the same diffusion model for random smoothing at different noise scales and selects the most reliable scale through cascading, balancing high accuracy with provable robustness.
Multi-Step Generalized Policy Improvement by Leveraging Approximate Models
Lucas Nunes Alegre, Bruno Castro da Silva
Reinforcement LearningTabularSequential
🎯 What it does: This paper studies a multi-step generalization strategy improvement method h-GPI for zero-shot transfer;
Multi-Swap k-Means++
Lorenzo Beretta (University of Copenhagen), Nikos Parotsidis (Google Research)
OptimizationTabularBiomedical Data
🎯 What it does: A local search k-means++ algorithm based on multiple exchanges is proposed, achieving a theoretical approximation of 9+ε, which is experimentally validated.
Multi-task Graph Neural Architecture Search with Task-aware Collaboration and Curriculum
Yijian Qin (Tsinghua University), Wenwu Zhu (Tsinghua University)
Neural Architecture SearchGraph Neural NetworkGraph
🎯 What it does: A method is proposed that can automatically search for the best graph neural network architecture in multi-task graph learning (MTGC 3), while also learning the collaborative relationships between tasks.
Multi-task learning with summary statistics
Parker Knight (Harvard University), Rui Duan (Harvard University)
TabularBiomedical Data
🎯 What it does: This paper proposes a framework for training multi-task learning models using only summary statistics from each task, and presents an adaptive parameter tuning strategy based on the Lepski method.
Multi-task Representation Learning for Pure Exploration in Bilinear Bandits
Subhojyoti Mukherjee (University of Wisconsin Madison), Robert D Nowak
OptimizationRepresentation Learning
🎯 What it does: This paper studies the pure exploration problem under multi-task representation learning in bilinear bandits, proposing the GOBLIN algorithm and providing a theoretical sample complexity analysis.
Multiclass Boosting: Simple and Intuitive Weak Learning Criteria
Nataly Brukhim (Princeton University), Shay Moran (Technion)
ClassificationOptimization
🎯 What it does: Proposed weak learning conditions and efficient algorithms for multi-class Boosting, addressing the limitations of traditional weak learning conditions for multi-class problems.
MultiFusion: Fusing Pre-Trained Models for Multi-Lingual, Multi-Modal Image Generation
Marco Bellagente (Stability AI), Samuel Weinbach (Aleph Alpha)
GenerationData SynthesisTransformerSupervised Fine-TuningDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes MULTIFUSION, a diffusion generation model capable of receiving multimodal inputs (text + images) in any order and supporting multiple languages (English, German, French, Spanish, Italian). It utilizes pre-trained language models, image prefixes, and adapters for modular fusion, ultimately achieving multilingual and multimodal inference for diffusion models trained on a single language.
Multimodal Deep Learning Model Unveils Behavioral Dynamics of V1 Activity in Freely Moving Mice
Aiwen Xu (University of California Santa Barbara), Michael Beyeler (University of California Santa Barbara)
Convolutional Neural NetworkRecurrent Neural NetworkVideoMultimodality
🎯 What it does: Establish a multimodal recurrent neural network that combines behavioral states such as eye movement, head movement, and movement speed with visual input to predict the neural activity of V1 during free mouse movement.
MultiMoDN—Multimodal, Multi-Task, Interpretable Modular Networks
Vinitra Swamy (École Polytechnique Fédérale de Lausanne), Mary-Anne Hartley (Yale University)
Explainability and InterpretabilityRecurrent Neural NetworkTransformerMultimodalityTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: We propose MultiModN, a sequentially composable multimodal multitask modular network that maintains predictive performance even when different modalities are missing.
Multinomial Logistic Regression: Asymptotic Normality on Null Covariates in High-Dimensions
Kai Tan (Rutgers University), Pierre C Bellec
ClassificationOptimizationTabularBiomedical Data
🎯 What it does: This paper studies the asymptotic distribution of the maximum likelihood estimation (MLE) of multi-class logistic regression under high dimensions (where the number of samples and features are of the same order).
Multiplication-Free Transformer Training via Piecewise Affine Operations
Atli Kosson (École Polytechnique Fédérale de Lausanne), Martin Jaggi (École Polytechnique Fédérale de Lausanne)
TransformerImageText
🎯 What it does: Replace all multiplications in Transformer training with piecewise affine operations (PAM), achieving a completely multiplication-free training process.
Multiply Robust Federated Estimation of Targeted Average Treatment Effects
Larry Han (Northeastern University), Jose R Zubizarreta
Federated LearningSafty and PrivacyTabularBiomedical Data
🎯 What it does: This paper proposes a privacy-preserving federated multi-site estimation method for estimating treatment effects in target populations, capable of handling both covariate shift and covariate mismatch simultaneously.
Multitask Learning with No Regret: from Improved Confidence Bounds to Active Learning
Pier Giuseppe Sessa (ETH Zurich), Andreas Krause (ETH Zurich)
Drug DiscoveryTabularBiomedical Data
🎯 What it does: This study proposes a new multi-task learning framework that provides confidence intervals for multi-task regression without prior knowledge and designs an adaptive regret-free algorithm to optimize the learning of multiple tasks.
MuSe-GNN: Learning Unified Gene Representation From Multimodal Biological Graph Data
Tianyu Liu (Yale University), Hongyu Zhao (Yale University)
Representation LearningDrug DiscoveryGraph Neural NetworkTransformerContrastive LearningMultimodalityBiomedical Data
🎯 What it does: The MuSe-GNN model is proposed, which integrates single-cell and spatial transcriptomics multimodal data in a gene-centric manner to generate unified gene embeddings.
Mutual Information Regularized Offline Reinforcement Learning
Xiao Ma (Sea AI Lab), Shuicheng YAN
Reinforcement LearningTabularBenchmark
🎯 What it does: By constructing a lower bound of state-action mutual information as regularization, we directly constrain the policy improvement direction in offline RL to address the distribution shift problem.
Mutual-Information Regularized Multi-Agent Policy Iteration
Jiangxing Wang (Peking University), Zongqing Lu (Peking University)
Reinforcement Learning
🎯 What it does: A multi-agent policy iteration algorithm based on mutual information regularization, MIPI, is proposed to enhance generalization capabilities under dynamic team compositions.
MVDiffusion: Enabling Holistic Multi-view Image Generation with Correspondence-Aware Diffusion
Shitao Tang (Simon Fraser University), Yasutaka Furukawa (Simon Fraser University)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImageVideo
🎯 What it does: This study proposes the MVDiffusion framework, which can generate multi-view images at once while ensuring the consistency of pixel correspondence between views.
NAP: Neural 3D Articulated Object Prior
Jiahui Lei (University of Pennsylvania), Kostas Daniilidis (University of Pennsylvania)
GenerationData SynthesisGraph Neural NetworkDiffusion modelPoint CloudMesh
🎯 What it does: A neural 3D articulated object prior based on diffusion models (NAP) is proposed, which can generate complete joint structures and component geometric models from scratch.
NAR-Former V2: Rethinking Transformer for Universal Neural Network Representation Learning
Yun Yi (Intellifusion), Xiaoyu Wang (Intellifusion)
Representation LearningNeural Architecture SearchGraph Neural NetworkTransformerGraph
🎯 What it does: This paper proposes NAR-Former V2, a universal neural network representation learning framework that integrates Transformer and GNN, capable of handling both cell structure models and complete deep networks.
NAS-X: Neural Adaptive Smoothing via Twisting
Dieterich Lawson (Google Research), Scott Linderman
Recurrent Neural NetworkTime SeriesSequentialBiomedical Data
🎯 What it does: A new method called NAS-X is proposed, which combines the Reweighted Sleep (RWS) framework with Smoothing Sequential Monte Carlo (SMC) for inference and learning in nonlinear latent variable models.
Nash Regret Guarantees for Linear Bandits
Ayush Sawarni (Indian Institute of Science), Siddharth Barman (Indian Institute of Science)
OptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper obtains a tight upper bound on a strengthened concept of regret (Nash regret) within the framework of stochastic linear bandits. Nash regret is defined as the difference between the unknown optimal value and the geometric mean of the expected rewards accumulated by the linear bandit algorithm.
Natural Actor-Critic for Robust Reinforcement Learning with Function Approximation
Ruida Zhou (Texas A&M University), Chao Tian (Texas A&M University)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: A robust natural actor-critic (RNAC) algorithm based on double sampling (DS) and integral probability metric (IPM) uncertainty sets is proposed for achieving robust reinforcement learning using function approximation in large state spaces.