NeurIPS 2024 Papers — Page 23
Conference on Neural Information Processing Systems · 4035 papers
MaVEn: An Effective Multi-granularity Hybrid Visual Encoding Framework for Multimodal Large Language Model
Chaoya Jiang (National Engineering Research Center for Software Engineering Peking University), Shikun Zhang (National Engineering Research Center for Software Engineering Peking University)
RecognitionObject DetectionSegmentationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: This paper proposes MaVEn—a multi-granularity hybrid visual encoding framework that combines discrete visual symbol sequences (coarse-grained semantics) with continuous visual vector sequences (fine-grained features). It also designs a dynamic visual tag compression mechanism based on discrete symbols to enhance the performance of multimodal large language models in multi-image reasoning and single-image tasks.
Maximizing utility in multi-agent environments by anticipating the behavior of other learners
Angelos Assos (Massachusetts Institute of Technology), Constantinos Costis Daskalakis
Optimization
🎯 What it does: This paper studies how an optimizer can maximize its own payoff by predicting the opponent's behavior in a two-player repeated game, given that the opponent uses mean-based learning algorithms (such as MWU and Replicator Dynamics).
Maximum Entropy Inverse Reinforcement Learning of Diffusion Models with Energy-Based Models
Sangwoong Yoon (Korea Institute for Advanced Study), Frank C. Park (Seoul National University)
GenerationAnomaly DetectionReinforcement LearningDiffusion modelImage
🎯 What it does: A framework based on maximum entropy inverse reinforcement learning (IRL) called DxMI is proposed to improve sample quality in few-step diffusion model generation, and energy learning is achieved without MCMC through joint training of the energy-based model (EBM).
Maximum Entropy Reinforcement Learning via Energy-Based Normalizing Flow
Chen-Hao Chao (National Tsing Hua University), Chun-Yi Lee (National Tsing Hua University)
Reinforcement LearningFlow-based ModelSequential
🎯 What it does: A maximum entropy reinforcement learning framework MEow based on energy flow (EBFlow) is proposed, achieving integrated training for policy evaluation and improvement.
MC-DiT: Contextual Enhancement via Clean-to-Clean Reconstruction for Masked Diffusion Models
Guanghao Zheng (Shanghai Jiao Tong University), Hongkai Xiong (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: This paper proposes MC-DiT, which enhances the context information extraction capability of the Diffusion Transformer through mask-to-clean reconstruction of clear images, and prevents model collapse with two EMA decoder branches.
MDAgents: An Adaptive Collaboration of LLMs for Medical Decision-Making
Yubin Kim (Massachusetts Institute of Technology), Hae Won Park (Massachusetts Institute of Technology)
TransformerLarge Language ModelImageVideoTextMultimodalityBiomedical DataRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: An adaptive multi-agent framework called MDAgents is proposed, which utilizes large language models (LLMs) to dynamically construct collaborative structures for individuals, multidisciplinary teams (MDTs), and integrated teams (ICTs) in the medical decision-making process.
Mean-Field Analysis for Learning Subspace-Sparse Polynomials with Gaussian Input
Ziang Chen (Massachusetts Institute of Technology), Rong Ge (Duke University)
Gaussian Splatting
🎯 What it does: This paper studies the learning of subspace sparse polynomials using a two-layer neural network and stochastic gradient descent under high-dimensional Gaussian inputs, and provides necessary and approximately sufficient conditions for learnability through mean field analysis.
Mean-Field Langevin Dynamics for Signed Measures via a Bilevel Approach
Guillaume Wang (École polytechnique fédérale de Lausanne), Lénaïc Chizat (École polytechnique fédérale de Lausanne)
OptimizationTabularStochastic Differential Equation
🎯 What it does: This paper proposes the application of Mean Field Langevin Dynamics (MFLD) to convex optimization problems with signed measures, achieved through a bilevel dimensionality reduction approach.
Meaningful Learning: Enhancing Abstract Reasoning in Large Language Models via Generic Fact Guidance
Kai Xiong (Harbin Institute of Technology), Yixin Cao (Fudan University)
TransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: This study investigates the abstract reasoning capabilities of large language models (LLMs), proposes a new evaluation metric called AbsAcc, constructs an abstract reasoning dataset based on general facts named AbsR, and designs a 'Mean Learning' (MeanLearn) paradigm that enables LLMs to implicitly learn and apply general facts through both knowledge and reasoning, thereby improving their performance on multiple reasoning and language understanding benchmarks.
Measuring Dejavu Memorization Efficiently
Narine Kokhlikyan (Meta), Kamalika Chaudhuri (Meta)
RecognitionRetrievalComputational EfficiencyConvolutional Neural NetworkVision Language ModelImageText
🎯 What it does: A single-model déjà vu memory measurement method is proposed, which can evaluate the memory level of image and visual language models without the need to train additional models.
Measuring Goal-Directedness
Matt MacDermott (Imperial College London), Tom Everitt (Google DeepMind)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes and implements the Maximum Entropy Goal Orientation (MEG) as a formal metric to measure the degree of goal orientation in causal models (CBN, CID) and Markov Decision Processes (MDP), and provides corresponding computational algorithms.
Measuring Mutual Policy Divergence for Multi-Agent Sequential Exploration
Haowen Dou (Xi'an Jiaotong University), Badong Chen (Xi'an Jiaotong University)
Robotic IntelligenceReinforcement LearningAgentic AISequential
🎯 What it does: This paper proposes MADPO, a multi-agent reinforcement learning framework that enhances heterogeneous task exploration by maximizing mutual policy differences under sequential updates.
Measuring Per-Unit Interpretability at Scale Without Humans
Roland S. Zimmermann (Max Planck Institute for Intelligent Systems), Wieland Brendel (Max Planck Institute for Intelligent Systems)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: This paper proposes the Machine Interpretability Score (MIS), a fully automated method for evaluating the interpretability of visual model units, and conducts large-scale interpretability analysis on over 700,000 units across 835 models.
Measuring Progress in Dictionary Learning for Language Model Interpretability with Board Game Models
Adam Karvonen (Independent), Samuel Marks (Northeastern University)
Explainability and InterpretabilitySupervised Fine-TuningAuto EncoderText
🎯 What it does: This paper proposes two new evaluation metrics—coverage and board reconstruction—by training a sparse autoencoder (SAE) on language models of board games (chess and Go). It also introduces the p-annealing training technique, aiming to objectively measure the interpretability and information capture capability of the SAE.
MECD: Unlocking Multi-Event Causal Discovery in Video Reasoning
Tieyuan Chen (Shanghai Jiao Tong University), Weiyao Lin (Shanghai Jiao Tong University)
TransformerVideoMultimodalityChain-of-Thought
🎯 What it does: This paper proposes the Multi-Event Causal Discovery (MECD) task and dataset, and designs a Granger-causal based multimodal model VGCM to construct a complete event-level causal graph from long videos.
Mechanism design augmented with output advice
George Christodoulou (Aristotle University of Thessaloniki), Ioannis Vlachos (Athens University of Economics and Business)
Recommendation SystemOptimizationReinforcement Learning
🎯 What it does: A new framework for mechanism design is proposed: improving the approximate performance of mechanisms by providing only output suggestions (recommended results), and based on this, designing provably strategic mechanisms for various problems;
Med-Real2Sim: Non-Invasive Medical Digital Twins using Physics-Informed Self-Supervised Learning
Keying Kuang (University of California Berkeley), Ahmed Alaa (University of California Berkeley)
Convolutional Neural NetworkSupervised Fine-TuningImageBiomedical DataUltrasound
🎯 What it does: A self-supervised learning framework called Med-Real2Sim is proposed, which predicts personalized digital twin models from non-invasive medical imaging (such as echocardiograms) and simulates the cardiac pressure-volume loop.
Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification
Yihe Wang (University of North Carolina), Xiang Zhang (University of North Carolina)
ClassificationTransformerTime SeriesBiomedical DataElectrocardiogram
🎯 What it does: This study proposes a multi-granularity block Transformer named Medformer, specifically designed for the classification of medical time series data, combining cross-channel blocking, granularity embedding, and a two-stage self-attention mechanism;
MediQ: Question-Asking LLMs and a Benchmark for Reliable Interactive Clinical Reasoning
Shuyue Stella Li (University of Washington), Yulia Tsvetkov (University of Washington)
TransformerLarge Language ModelPrompt EngineeringTextBiomedical DataBenchmark
🎯 What it does: An interactive medical diagnosis evaluation framework, MEDIQ, is proposed, simulating the question-and-answer process between doctors and patients, focusing on the active questioning and information gathering capabilities of LLMs.
Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length
Xuezhe Ma (AI at Meta), Chunting Zhou (AI at Meta)
TransformerLarge Language ModelTextMultimodality
🎯 What it does: MEGALODON is proposed, an efficient LLM pre-training and inference architecture that supports infinite context length.
MeLLoC: Lossless Compression with High-order Mechanism Learning
Xinyue Luo (Fudan University), Yu Chen (Shanghai University of Finance and Economics)
CompressionTime SeriesPhysics Related
🎯 What it does: By learning the underlying partial differential equation mechanisms of scientific data, the data is transformed into sparse source terms and lossless compression is achieved using entropy coding.
Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibration
Wenjie Fu (Huazhong University of Science and Technology), Tao Jiang (Huazhong University of Science and Technology)
Adversarial AttackTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes a membership inference attack (MIA) for fine-tuning large language models (LLMs), which generates reference data through self-prompting and utilizes probability variance metrics to identify training records.
Membership Inference Attacks against Large Vision-Language Models
Zhan Li (EPFL), Volkan Cevher (EPFL)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A benchmark specifically for membership inference attacks (MIA) on large-scale vision-language models (VLLMs) has been constructed (VL-MIA), and a cross-modal attack pipeline along with a novel MaxRényi-K% evaluation metric has been proposed.
Membership Inference on Text-to-Image Diffusion Models via Conditional Likelihood Discrepancy
Shengfang Zhai (Peking University), Yang Liu (Nanyang Technological University)
GenerationAdversarial AttackDiffusion modelImageText
🎯 What it does: This paper proposes a member inference attack called CLiD based on conditional likelihood differences, targeting text-to-image diffusion models.
MeMo: Meaningful, Modular Controllers via Noise Injection
Megan Tjandrasuwita (Massachusetts Institute of Technology), Wojciech Matusik (Massachusetts Institute of Technology)
Robotic IntelligenceGraph Neural NetworkTransformerReinforcement LearningTabular
🎯 What it does: Proposes the MeMo framework, which utilizes a single robot training module to create a modular controller that can be quickly transferred to new robots composed of the same assembly parts;
Memorize What Matters: Emergent Scene Decomposition from Multitraverse
Yiming Li (New York University), Jose M. Alvarez (NVIDIA)
SegmentationAutonomous DrivingKnowledge DistillationRobotic IntelligenceGaussian SplattingSimultaneous Localization and MappingVideoBenchmark
🎯 What it does: This paper proposes a 3D Gaussian Mapping (3DGM) framework that utilizes RGB videos from multiple driving perspectives for self-supervised 3D environment mapping (EnvGS) and 2D instantaneous object segmentation (EmerSeg), achieving robot memory and localization capabilities without LiDAR and manual annotations.
Memory-Efficient Gradient Unrolling for Large-Scale Bi-level Optimization
Qianli Shen (National University of Singapore), Kenji Kawaguchi (National University of Singapore)
OptimizationKnowledge DistillationLarge Language ModelImageText
🎯 What it does: A novel algorithm named (FG)U² is proposed, which combines forward gradient expansion and forward gradient methods to address memory and approximation issues in large-scale bilevel optimization.
Memory-Efficient LLM Training with Online Subspace Descent
Kaizhao Liang (University of Texas at Austin), qiang liu
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: An Online Subspace Descent algorithm is proposed, which updates the projection matrix using online PCA, avoiding the SVD computation in traditional low-rank training methods, thus achieving memory-efficient LLM pre-training.
MemoryFormer : Minimize Transformer Computation by Removing Fully-Connected Layers
Ning Ding (Peking University), Yunhe Wang (Huawei)
Computational EfficiencyTransformerText
🎯 What it does: Proposes MemoryFormer, a Transformer architecture that uses a learnable local sensitive hash table instead of fully connected layers, significantly reducing FLOPs.
MemVLT: Vision-Language Tracking with Adaptive Memory-based Prompts
Xiaokun Feng (University of Chinese Academy of Sciences), Kaiqi Huang (University of Chinese Academy of Sciences)
Object TrackingTransformerVision Language ModelMultimodality
🎯 What it does: A MemVLT visual-language tracker is proposed, which utilizes a memory mechanism to adaptively generate multimodal prompts for tracking targets that change dynamically over time.
Mesa-Extrapolation: A Weave Position Encoding Method for Enhanced Extrapolation in LLMs
Xin Ma (Digital Research Institute Enn Group), Xiaoxu Ma (Digital Research Institute Enn Group)
TransformerLarge Language ModelText
🎯 What it does: A method called Mesa-Extrapolation is proposed, which is a cost-free woven positional encoding method that enables LLMs to achieve scalable reasoning in contexts beyond the maximum training length.
MeshFormer : High-Quality Mesh Generation with 3D-Guided Reconstruction Model
Minghua Liu (UC San Diego), Hao Su (UC San Diego)
GenerationTransformerMesh
🎯 What it does: This paper proposes MeshFormer, a model capable of generating high-quality 3D meshes in a single forward inference under sparse views.
MeshXL: Neural Coordinate Field for Generative 3D Foundation Models
Sijin Chen (Tencent PCG), Tao Chen (ShanghaiTech University)
GenerationData SynthesisTransformerLarge Language ModelMultimodalityMesh
🎯 What it does: The study proposes MeshXL, which utilizes Neural Coordinate Field (NeurCF) to map explicit coordinates of 3D meshes to implicit neural embeddings, and directly generates high-quality meshes through an autoregressive Transformer (an OPT variant);
Meta 3D AssetGen: Text-to-Mesh Generation with High-Quality Geometry, Texture, and PBR Materials
Yawar Siddiqui, David Novotny
GenerationData SynthesisTransformerDiffusion modelMesh
🎯 What it does: Meta 3D AssetGen is proposed, which can quickly generate high-quality geometric, textured, and physically-based rendering (PBR) material 3D meshes from text or image prompts.
Meta-Controller: Few-Shot Imitation of Unseen Embodiments and Tasks in Continuous Control
Seongwoong Cho (Korea Advanced Institute of Science and Technology), Seunghoon Hong (Korea Advanced Institute of Science and Technology)
Robotic IntelligenceMeta LearningTransformerSequential
🎯 What it does: This paper proposes a Transformer-based few-shot behavior cloning framework that can generalize to unseen robot shapes and tasks with only five reward-free demonstrations provided.
Meta-Diffu$B$: A Contextualized Sequence-to-Sequence Text Diffusion Model with Meta-Exploration
Yunyen Chuang, Hung-yi Lee (National Taiwan University)
GenerationData SynthesisMeta LearningTransformerReinforcement LearningDiffusion modelText
🎯 What it does: This paper proposes Meta-DiffuB, a scheduling-utilization framework based on Meta-Exploration, to achieve context-aware noise scheduling in sequence-to-sequence (Seq2Seq) text diffusion models, thereby enhancing the quality and diversity of text generation.
Meta-DT: Offline Meta-RL as Conditional Sequence Modeling with World Model Disentanglement
Zhi Wang (Nanjing University), Chunlin Chen (Nanjing University)
Meta LearningTransformerReinforcement LearningWorld ModelSequential
🎯 What it does: This paper proposes Meta-DT, a Transformer-based offline Meta-RL framework that utilizes a world model to decompose task information and achieve task generalization through self-guided complementary prompts.
Meta-Exploiting Frequency Prior for Cross-Domain Few-Shot Learning
Fei Zhou (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
Domain AdaptationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: A framework for meta-learning using frequency priors in cross-domain few-shot learning is proposed, called Meta-Exploiting Frequency Prior (MEF).
Meta-Learning Universal Priors Using Non-Injective Change of Variables
Yilang Zhang (University of Minnesota), Georgios B. Giannakis (University of Minnesota)
Meta LearningImage
🎯 What it does: This paper proposes a method that utilizes non-injective transformations (NCoV) to learn a more expressive prior distribution, thereby enhancing the performance of meta-learning in few-shot tasks.
Meta-Reinforcement Learning with Universal Policy Adaptation: Provable Near-Optimality under All-task Optimum Comparator
Siyuan Xu (Pennsylvania State University), Minghui Zhu (Pennsylvania State University)
Meta LearningReinforcement Learning
🎯 What it does: This paper proposes a dual-layer optimization-based Meta-RL framework (BO-MRL), which collects data in a single instance during the task adaptation phase and updates the policy through multi-step general policy optimization (covering PPO, NPG, policy gradient, etc.), and provides an upper bound on the expected optimality gap compared to the optimal policy for all tasks.
MetaAligner: Towards Generalizable Multi-Objective Alignment of Language Models
Kailai Yang (University of Manchester), Sophia Ananiadou (University of Manchester)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposes MetaAligner, a policy model-agnostic multi-objective preference alignment method that achieves multi-objective alignment without tuning the target model;
Metacognitive Capabilities of LLMs: An Exploration in Mathematical Problem Solving
Aniket Rajiv Didolkar (Mila), Sanjeev Arora (Princeton University)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Automatically label skills for mathematical problems using LLM, build a skill example library, and enhance the mathematical reasoning performance of LLM through few-shot prompting based on skill matching.
MetaCURL: Non-stationary Concave Utility Reinforcement Learning
Bianca Marin Moreno (Inria), Nadia Oudjane (EDF Research and Development)
OptimizationReinforcement Learning
🎯 What it does: The MetaCURL algorithm is proposed to address the online reinforcement learning problem of nonlinear objective functions (CURL) in non-stationary environments.
MetaLA: Unified Optimal Linear Approximation to Softmax Attention Map
Yuhong Chou (Hong Kong Polytechnic University), Guoqi Li (Institute of Automation Chinese Academy of Sciences)
RetrievalOptimizationTransformerImageText
🎯 What it does: This paper studies and proposes MetaLA, unifying LinFormer/SSM/LinRNN into a general linear attention mechanism, and constructs a model that achieves the optimal linear approximation of softmax attention, validating its effectiveness across various tasks.
MetaUAS: Universal Anomaly Segmentation with One-Prompt Meta-Learning
Bin-Bin Gao (Tencent YouTu Lab)
SegmentationAnomaly DetectionMeta LearningConvolutional Neural NetworkImage
🎯 What it does: A meta-learning framework called MetaUAS is proposed, which can achieve general anomaly segmentation based on a single normal image prompt without relying on language prompts or target anomaly datasets.
Meteor: Mamba-based Traversal of Rationale for Large Language and Vision Models
Byung-Kwan Lee (KAIST), Yong Man Ro (KAIST)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: An efficient LLVM model called Meteor is proposed, which integrates long-form multifaceted reasoning (rationale) through the Mamba architecture and a 'traversal reasoning' mechanism, achieving image understanding and multi-domain reasoning capabilities.
Metric Flow Matching for Smooth Interpolations on the Data Manifold
Kacper Kapusniak (University of Oxford), Francesco Di Giovanni (University of Oxford)
Data SynthesisOptimizationFlow-based ModelImagePoint CloudBiomedical Data
🎯 What it does: A data-driven Riemannian metric-based method for simulation-free conditional flow matching (Metric Flow Matching, MFM) is proposed, which generates more natural probability paths by minimizing the kinetic energy of the interpolation path in the metric, keeping the interpolation on the data manifold.
Metric from Human: Zero-shot Monocular Metric Depth Estimation via Test-time Adaptation
Yizhou Zhao (Carnegie Mellon University), Min Xu (Carnegie Mellon University)
Depth EstimationDomain AdaptationDiffusion modelImage
🎯 What it does: By generating human figures from input images during testing and using the real size of the human figures as a scale prior, unsupervised zero-shot monocular metric depth estimation is achieved.
Metric Space Magnitude for Evaluating the Diversity of Latent Representations
Katharina Limbeck (Helmholtz Munich), Bastian Rieck (Helmholtz Munich)
GenerationRepresentation LearningImageTextGraph
🎯 What it does: This paper proposes a multi-scale diversity assessment method based on the size of metric spaces (magnitude) to measure the intrinsic diversity of potential representations and detect mode collapse and loss in generative models.
Metric Transforms and Low Rank Representations of Kernels for Fast Attention
Timothy Zer-An Chu (Independent Researcher), Zhao Song (Simons Institute for the Theory of Computing)
🎯 What it does: A new tool called 'the representation theory of hyperrectangles' is proposed, and it is used to achieve three theoretical results: ① It proves that low-degree polynomials are the only functions that can maintain low rank; ② It completely characterizes the positive definite kernels of Manhattan distance, showing that they are equivalent to completely monotonic functions; ③ It fully classifies the metric transformations from Manhattan to Manhattan or Euclidean distances, indicating that they are exactly Bernstein functions.
MG-Net: Learn to Customize QAOA with Circuit Depth Awareness
Yang Qian (University of Sydney), Dacheng Tao (Nanyang Technological University)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper proposes MG-Net, an end-to-end deep learning framework for dynamically generating optimal mixed Hamiltonians for a given problem and circuit depth, thereby improving the approximation ratio of QAOA.
MGF: Mixed Gaussian Flow for Diverse Trajectory Prediction
Jiahe Chen (Zhejiang University), Jiangmiao Pang (Shanghai AI Laboratory)
GenerationData SynthesisAutonomous DrivingFlow-based ModelTime SeriesSequential
🎯 What it does: This paper studies a regularized flow model based on a mixed Gaussian prior for multimodal trajectory prediction.
MicroAdam: Accurate Adaptive Optimization with Low Space Overhead and Provable Convergence
Ionut-Vlad Modoranu (Institute of Science and Technology Austria), Dan Alistarh (Institute of Science and Technology Austria)
OptimizationTransformerSupervised Fine-TuningImageText
🎯 What it does: This paper proposes an adaptive optimizer named MICROADAM, which significantly reduces the memory footprint of the optimizer's state by compressing before the gradient enters the Adam state and using an error feedback mechanism, while maintaining theoretical convergence guarantees.
Microstructures and Accuracy of Graph Recall by Large Language Models
Yanbang Wang (Cornell University), Jon Kleinberg (Cornell University)
TransformerLarge Language ModelTextGraph
🎯 What it does: A systematic evaluation of the ability of large language models (LLMs) to recall graph structures from text was conducted and compared with human memory patterns.
MIDGArD: Modular Interpretable Diffusion over Graphs for Articulated Designs
Quentin Leboutet (Intel Labs), Kai Yuan (Intel Labs)
GenerationExplainability and InterpretabilityGraph Neural NetworkDiffusion modelMultimodalityGraph
🎯 What it does: The MIDGArD framework is proposed, which uses diffusion models to achieve interpretable, controllable, and simulative 3D joint object generation, divided into two steps: structure generation and shape generation.
MILP-StuDio: MILP Instance Generation via Block Structure Decomposition
Haoyang Liu (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
OptimizationGraphBenchmark
🎯 What it does: A framework for generating MILP instances based on block structure decomposition, called MILP-StuDio, is proposed. It utilizes the block structure splitting of CCM and generates high-quality instances through three block operations (reduction, replacement, expansion).
Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games
The Viet Bui (Singapore Management University), Thanh Hong Nguyen (University of Oregon)
Reinforcement Learning
🎯 What it does: A multi-agent reinforcement learning framework called IMAX-PPO, which integrates imitation learning, is proposed to predict and utilize the opponent's next state in multi-agent competitive environments.
MimicTalk: Mimicking a personalized and expressive 3D talking face in minutes
Zhenhui Ye (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationData SynthesisFlow-based ModelNeural Radiance FieldVideoAudio
🎯 What it does: This paper proposes the MimicTalk framework, which utilizes a pre-trained 3D crowd-agnostic NeRF model for personalized talking face generation. It achieves high-quality, fast, and expressive video generation through static-dynamic hybrid adaptation and context-stylized audio to motion model.
Mind the Gap Between Prototypes and Images in Cross-domain Finetuning
Hongduan Tian (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
ClassificationDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: For cross-domain few-shot classification, we propose the Contrastive Prototype-Image Adaptation (CoPA) method, which learns independent representation transformations for prototypes and image instances on a frozen pre-trained backbone.
Mind the Gap: A Causal Perspective on Bias Amplification in Prediction & Decision-Making
Drago Plecko (Columbia University), Elias Bareinboim (Columbia University)
TabularBiomedical DataElectronic Health Records
🎯 What it does: This paper studies the fairness issues that may arise when converting continuous predictive probabilities into binary decisions, introduces the concept of 'margin complement', and provides a bias amplification decomposition based on causal pathways. It then defines weak/strong business necessity and offers evaluation algorithms.
Mind the Graph When Balancing Data for Fairness or Robustness
Jessica Schrouff (Google DeepMind), Silvia Chiappa (Google DeepMind)
ImageText
🎯 What it does: This paper studies the use of joint data balancing to eliminate undesirable dependencies among variables, labels, and auxiliary factors in fairness and robustness tasks, and explores its success and failure patterns under different causal graph structures.
Mind's Eye of LLMs: Visualization-of-Thought Elicits Spatial Reasoning in Large Language Models
Wenshan Wu (Microsoft Research), Furu Wei (Microsoft Research)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This study investigates the spatial reasoning capabilities of LLMs and proposes the Visualization-of-Thought (VoT) prompt, which allows the model to generate a two-dimensional 'mental image' in text form at each step, improving performance on multi-hop spatial reasoning tasks.
MindMerger: Efficiently Boosting LLM Reasoning in non-English Languages
Zixian Huang (Nanjing University), Fei Yuan (Shanghai Artificial Intelligence Laboratory)
TransformerLarge Language ModelText
🎯 What it does: By mapping the representations of multilingual pre-trained models to the representation space of large language models (LLMs) and incrementally fusing the input based on this, MindMerger is proposed to enhance the performance of LLMs in non-English multilingual reasoning and language understanding tasks.
MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attention
Huiqiang Jiang (Microsoft Corporation), Lili Qiu (Microsoft Corporation)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Accelerate the pre-filling stage of long-context LLMs by using dynamic sparse attention to significantly reduce computational load.
Mini-Sequence Transformers: Optimizing Intermediate Memory for Long Sequences Training
Cheng Luo (California Institute of Technology), Anima Anandkumar (California Institute of Technology)
TransformerLarge Language ModelText
🎯 What it does: By splitting the MLP and LM-Head modules of the Transformer into mini-sequences, the memory usage of intermediate activations is significantly reduced, enabling the training of very long sequences of LLM on a single GPU.
MiniCache: KV Cache Compression in Depth Dimension for Large Language Models
Akide Liu (Monash University), Bohan Zhuang (Zhejiang University)
CompressionTransformerLarge Language ModelText
🎯 What it does: In the inference of large language models, to address the memory bottleneck caused by the linear growth of KV cache with sequence length, the MiniCache scheme is proposed. It achieves significant memory usage compression by performing cross-layer compression of KV caches between adjacent layers in the deep layers of the model.
Minimax Optimal and Computationally Efficient Algorithms for Distributionally Robust Offline Reinforcement Learning
Zhishuai Liu (Duke University), Pan Xu (Duke University)
Computational EfficiencyReinforcement LearningTabular
🎯 What it does: An optimal and computable algorithm based on linear function approximation in distributed robust offline reinforcement learning is proposed, introducing two value iteration methods: DRPVI and VA-DRPVI.
Minimizing UCB: a Better Local Search Strategy in Local Bayesian Optimization
Zheyi Fan (Academy of Mathematics and Systems Science, Chinese Academy of Sciences), Qingpei Hu (Academy of Mathematics and Systems Science, Chinese Academy of Sciences)
OptimizationTabular
🎯 What it does: This paper proposes two local Bayesian optimization algorithms based on Gaussian processes: MinUCB and LA-MinUCB, which achieve more efficient local search by minimizing UCB.
Minimum Entropy Coupling with Bottleneck
MohammadReza Ebrahimi, Ashish J Khisti
CompressionReinforcement LearningSequential
🎯 What it does: This paper proposes distortion compression under log-loss and introduces output distribution constraints, forming a Minimum Entropy Coupling with Bottleneck (MEC-B) framework, which is divided into Entropy-Bounded Information Maximization (EBIM) and Minimum Entropy Coupling (MEC) parts;
Mining and Transferring Feature-Geometry Coherence for Unsupervised Point Cloud Registration
KeZheng Xiong, Cheng Wang (Xiamen University)
Autonomous DrivingOptimizationRepresentation LearningContrastive LearningPoint Cloud
🎯 What it does: An unsupervised point cloud registration method called INTEGER is proposed, which utilizes feature-geometry consistency to mine pseudo-labels and learns density-invariant features through mixed density student learning.
Mirror and Preconditioned Gradient Descent in Wasserstein Space
Clément Bonet (CREST ENSAE IP Paris), Anna Korba (CREST ENSAE IP Paris)
OptimizationBiomedical Data
🎯 What it does: This paper studies the extension of mirror descent and preconditioned gradient descent in Wasserstein space, provides discrete-time convergence theory, and verifies convergence on various objective functions. It subsequently demonstrates significant acceleration effects through preconditioned gradient descent on single-cell biology datasets.
MiSO: Optimizing brain stimulation to create neural population activity states
Yuki Minai (Carnegie Mellon University), Byron M. Yu (Carnegie Mellon University)
OptimizationConvolutional Neural NetworkBiomedical Data
🎯 What it does: This study proposes the MiSO framework, which utilizes closed-loop optimization and convolutional neural network predictions to quickly guide neural population activity to a specified state among thousands of stimulus parameter configurations.
Mission Impossible: A Statistical Perspective on Jailbreaking LLMs
Jingtong Su (New York University), Karen Ullrich (Meta AI)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: A statistical theoretical framework is proposed to analyze the phenomenon that LLMs are still vulnerable to jailbreaks after pre-training alignment, and based on this, the E-RLHF method is designed to enhance security.
Mitigating Backdoor Attack by Injecting Proactive Defensive Backdoor
Shaokui Wei (Chinese University of Hong Kong), Baoyuan Wu (Chinese University of Hong Kong)
ClassificationAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: A method for actively defending against backdoor attacks is proposed, which involves injecting a defensive backdoor (PDB) during the training phase.
Mitigating Biases in Blackbox Feature Extractors for Image Classification Tasks
Abhipsa Basu (Indian Institute of Science), Venkatesh Babu Radhakrishnan
ClassificationTransformerContrastive LearningImage
🎯 What it does: The study proposes a clustering adaptive margin loss to alleviate model bias in image classification tasks with frozen pre-trained model feature extractors and no biased labels.
Mitigating Covariate Shift in Behavioral Cloning via Robust Stationary Distribution Correction
Seokin Seo (Korea Advanced Institute of Science and Technology), Kee-Eung Kim (Korea Advanced Institute of Science and Technology)
Robotic IntelligenceReinforcement LearningTabular
🎯 What it does: A method called DrilDICE is proposed for offline imitation learning, designed to address the covariate shift problem where expert demonstration data deviates from the stationary distribution in state distribution, by creating a distributionally robust behavior cloning objective.
Mitigating Object Hallucination via Concentric Causal Attention
Yun Xing (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
RecognitionObject DetectionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes the Concentric Causal Attention (CCA) method, which reduces the relative distance between visual tokens and instruction tokens by reordering visual tokens into concentric hierarchies and rewriting the attention mask, thereby suppressing object hallucinations in large visual language models (LVLMs);
Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy
Cameron Allen (University of California Berkeley), George Konidaris (Brown University)
OptimizationRecurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: Proposes the λ-difference metric to detect and mitigate partial observability in sequential decision-making, incorporating it as an auxiliary loss in Recurrent PPO.
Mitigating Reward Overoptimization via Lightweight Uncertainty Estimation
Xiaoying Zhang (ByteDance Research), Yang Liu (University of California Santa Cruz)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningText
🎯 What it does: This paper proposes a lightweight uncertainty estimation based on the last layer embedding of the reward model, and designs a distributionally robust adversarial policy optimization (ADVPO) based on this estimation to mitigate the issue of reward over-optimization in RLHF.
Mitigating Spurious Correlations via Disagreement Probability
Hyeonggeun Han (Seoul National University), Jungwoo Lee (Hodoo AI Labs)
ClassificationData-Centric LearningGenerative Adversarial NetworkImageText
🎯 What it does: This paper proposes a debiasing method in the absence of bias labels, utilizing the disagreement probability of adversarial bias models for sample weighting/resampling, thereby enhancing the model's consistency and performance on samples with or without spurious correlations.
Mixed Dynamics In Linear Networks: Unifying the Lazy and Active Regimes
Zhenfeng Tu (New York University), Arthur Jacot (New York University)
🎯 What it does: A self-consistent dynamical formula is presented in shallow linear networks, unifying the lazy and balanced training dynamics, and revealing an intermediate mixed dynamic;
MixEval: Deriving Wisdom of the Crowd from LLM Benchmark Mixtures
Jinjie Ni (National University of Singapore), Yang You (National University of Singapore)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: MixEval is proposed, which constructs an unbiased and efficient LLM evaluation set by mining network user queries and matching them with existing benchmarks, and further introduces MixEval-Hard and a dynamic updating mechanism.
Mixture of Adversarial LoRAs: Boosting Robust Generalization in Meta-Tuning
Xu Yang (City University of Hong Kong), Ying Wei (Zhejiang University)
Domain AdaptationAdversarial AttackMeta LearningTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes a method based on Adversarial Meta-Tuning (AMT), which enhances robust generalization on out-of-distribution (OOD) tasks by introducing low-rank adapters (LoRA) on a pre-trained Vision Transformer and applying adversarial perturbations on query samples and their singular values/vectors.
Mixture of Demonstrations for In-Context Learning
Song Wang (University of Virginia), Jundong Li (University of Virginia)
ClassificationGenerationData SynthesisOptimizationTransformerLarge Language ModelMixture of ExpertsContrastive LearningText
🎯 What it does: This paper proposes a framework called MoD (Mixture of Demonstrations) for automatically selecting high-quality examples in the context of In-Context Learning (ICL) with large language models, significantly improving model performance.
Mixture of Experts Meets Prompt-Based Continual Learning
Minh Le (VinAI Research), Nhat Ho (University of Texas at Austin)
TransformerPrompt EngineeringMixture of ExpertsImage
🎯 What it does: This paper studies the relationship between prefix tuning and self-attention, viewing it as a special type of expert model, and based on this, proposes a Nonlinear Residual Gate (NoRGa) to enhance the performance of prompt-based continual learning.
Mixture of In-Context Experts Enhance LLMs' Long Context Awareness
Hongzhan Lin (Renmin University of China), Rui Yan (Renmin University of China)
TransformerLarge Language ModelMixture of ExpertsTextBenchmark
🎯 What it does: This paper proposes the Mixture of In-Context Experts (MoICE) plugin, which enhances the LLM's ability to perceive long text contexts by dynamically selecting multiple RoPE angles (considered as experts) through routers added to each attention head.
Mixture of Link Predictors on Graphs
Li Ma (Shanghai Jiao Tong University), Jiliang Tang (Michigan State University)
Graph Neural NetworkMixture of ExpertsGraph
🎯 What it does: This paper proposes a hybrid expert model for graph link prediction called Link-MoE, which can dynamically select the most suitable GNN predictor based on various heuristic features of node pairs.
Mixture of Nested Experts: Adaptive Processing of Visual Tokens
Gagan Jain (Google DeepMind), Sujoy Paul (Google DeepMind)
ClassificationOptimizationComputational EfficiencyTransformerMixture of ExpertsImageVideo
🎯 What it does: Proposes Mixture of Nested Experts (MoNE), a dynamic routing mechanism for visual Transformers based on nested models, which allocates different computational resources according to token importance;
Mixture of neural fields for heterogeneous reconstruction in cryo-EM
Axel Levy (Stanford University), Ellen D Zhong
Image TranslationSegmentationOptimizationProtein Structure PredictionMixture of ExpertsAuto EncoderImage
🎯 What it does: The Hydra method is proposed, utilizing a mixture model of K neural fields for fully automated heterogeneous reconstruction in single-particle cryo-EM, capable of simultaneously inferring the combination category, pose, and continuous conformational states of the particles.
Mixture of Scales: Memory-Efficient Token-Adaptive Binarization for Large Language Models
Dongwon Jo (Seoul National University), Jae-Joon Kim (Seoul National University)
CompressionComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: This paper proposes a binaryization technique called BinaryMoS, which achieves efficient binaryization of LLMs by using multiple learnable scaling experts on binary weights and dynamically generating scaling factors for each token.
Mixture of Tokens: Continuous MoE through Cross-Example Aggregation
Szymon Antoniak (University of Warsaw), Sebastian Jaszczur (University of Warsaw)
TransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Designed and evaluated a continuous Mixture of Experts architecture called Mixture of Tokens (MoT) for training and inference of autoregressive language models.
Mixtures of Experts for Audio-Visual Learning
Ying Cheng (Fudan University), Rui Feng (Fudan University)
RecognitionSegmentationTransformerMixture of ExpertsMultimodalityAudio
🎯 What it does: Proposes AVMoE - an audio-visual learning framework that injects unimodal and cross-modal Adapters into a frozen pre-trained model and dynamically allocates weights through Mixture of Experts (MoE).
MKGL: Mastery of a Three-Word Language
Lingbing Guo (Zhejiang University), Huajun Chen (Zhejiang University)
TransformerLarge Language ModelContrastive LearningTextGraphRetrieval-Augmented Generation
🎯 What it does: The research combines large language models with knowledge graphs, introducing a specialized three-word knowledge graph language (KGL) to enhance the capabilities of LLMs in completing and generating tasks related to knowledge graphs through dictionaries, context retrieval, and KGL embeddings.
MMSite: A Multi-modal Framework for the Identification of Active Sites in Proteins
Song Ouyang (Wuhan University), Bo Du (Wuhan University)
RecognitionProtein Structure PredictionTransformerPrompt EngineeringMultimodalityBiomedical Data
🎯 What it does: A multi-modal framework called MMSite is proposed, which jointly identifies protein active sites using protein sequences and multi-attribute text descriptions.
MO-DDN: A Coarse-to-Fine Attribute-based Exploration Agent for Multi-Object Demand-driven Navigation
Hongcheng Wang (Peking University), Hao Dong (Peking University)
Object DetectionRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningAgentic AITextPoint CloudBenchmark
🎯 What it does: This paper proposes a Multi-Object Demand-Driven Navigation (MO-DDN) benchmark and designs a Coarse-to-Fine Attribute Exploration Agent (C2FAgent) to find combinations of objects that satisfy basic and preferred demands under natural language requirement instructions.
Mobile-Agent-v2: Mobile Device Operation Assistant with Effective Navigation via Multi-Agent Collaboration
Junyang Wang (Beijing Jiaotong University), Jitao Sang (Beijing Jiaotong University)
Large Language ModelAgentic AIVision Language ModelTextMultimodality
🎯 What it does: A multi-agent architecture called Mobile-Agent-v2 has been designed and implemented to perform complex UI operation tasks on mobile devices, with three main agents—planning, decision-making, and reflection—collaborating to navigate task progress and focal content.
Mobility-LLM: Learning Visiting Intentions and Travel Preference from Human Mobility Data with Large Language Models
Letian Gong (Beijing Jiaotong University), Huaiyu Wan (Beijing Jiaotong University)
TransformerLarge Language ModelSupervised Fine-TuningTime SeriesSequential
🎯 What it does: A unified framework called Mobility-LLM is proposed, which utilizes large language models (LLM) to learn from check-in sequences, extracting access intentions and travel preferences to accomplish three types of tasks: location prediction, trajectory user association, and time prediction.
Model Based Inference of Synaptic Plasticity Rules
Yash Mehta (Howard Hughes Medical Institute), Jan Funke (Howard Hughes Medical Institute)
Time Series
🎯 What it does: A gradient descent-based parameterized plasticity rule inference method has been developed, which can automatically learn synaptic plasticity functions from neural or behavioral trajectories and has been validated on simulations and fruit fly behavioral data.
Model Collapse Demystified: The Case of Regression
Elvis Dohmatob (Meta), Julia Kempe (New York University)
Image
🎯 What it does: This paper provides a theoretical analysis of the phenomenon of 'Model Collapse' that occurs when a model is trained repeatedly on data it generates itself, within the framework of high-dimensional linear regression (equivalent to kernel regression), and presents precise error expressions.
Model Decides How to Tokenize: Adaptive DNA Sequence Tokenization with MxDNA
Lifeng Qiao (Shanghai Artificial Intelligence Laboratory), Wanli Ouyang (Shanghai Artificial Intelligence Laboratory)
TransformerMixture of ExpertsBiomedical DataBenchmark
🎯 What it does: This paper proposes a framework MxDNA that can automatically learn DNA sequence segmentation strategies through gradient descent;