ICLR 2025 Papers — Page 22
International Conference on Learning Representations · 3704 papers
Mixture Compressor for Mixture-of-Experts LLMs Gains More
Wei Huang (Beihang University), XIAOJUAN QI
CompressionOptimizationTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: This paper proposes an untrained mixed compression framework MC, which significantly reduces the storage and activation costs of expert parameters in MoE LLMs through pre-loaded mixed precision quantization and online dynamic pruning.
Mixture of Attentions For Speculative Decoding
Matthieu Zimmer (Huawei Noah's Ark Lab), Jun Wang (UCL Centre for Artificial Intelligence)
GenerationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a Mixture of Attentions architecture to enhance the speed and accuracy of Speculative Decoding (SD) in both single-machine and client-server scenarios, addressing issues of partial observability, off-policy nature, and target layer selection present in existing methods.
Mixture of Experts Made Personalized: Federated Prompt Learning for Vision-Language Models
Jun Luo (University of Pittsburgh), Shandong Wu (University of Pittsburgh)
Federated LearningPrompt EngineeringMixture of ExpertsVision Language ModelImageMultimodality
🎯 What it does: The pFedMoAP framework is proposed, which implements personalized prompt learning for CLIP-like VLM in federated learning through a Mixture of Experts approach.
Mixture of In-Context Prompters for Tabular PFNs
Derek Qiang Xu (University of California), Wei Wang (University of California)
ClassificationComputational EfficiencyPrompt EngineeringMixture of ExpertsTabularBenchmark
🎯 What it does: Proposes the MIXTUREPFN framework, which utilizes Sparse Mixture Context Prompts (MICP) and Context-Aware Fine-Tuning (CAPFN) to enhance the efficiency and effectiveness of Prior-Fitted Networks on large-scale tabular data.
Mixture of Parrots: Experts improve memorization more than reasoning
Samy Jelassi (Harvard University), eran malach
TransformerMixture of ExpertsTextGraph
🎯 What it does: This study investigates the performance differences between Mixture-of-Experts (MoE) and dense Transformers in memory tasks (such as phone book and world knowledge question answering) and reasoning tasks (such as graph path and mathematical reasoning).
Mixture-of-Agents Enhances Large Language Model Capabilities
Junlin Wang (Duke University), James Zou (Stanford University)
TransformerLarge Language ModelPrompt EngineeringMixture of ExpertsTextBenchmark
🎯 What it does: Proposes the Mixture-of-Agents (MoA) framework, which achieves iterative improvement of multiple models on the same task through collaborative reasoning and aggregation of multi-layer LLMs.
MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering
Jun Shern Chan (OpenAI), Lilian Weng (OpenAI)
TransformerLarge Language ModelAgentic AITabularBenchmark
🎯 What it does: Establish the MLE-bench benchmark, utilizing 75 Kaggle competition datasets to evaluate the capabilities of AI agents in machine learning engineering tasks (such as data preprocessing, model training, and experiment management) and directly compare them with human performance on the Kaggle leaderboard.
MLLM as Retriever: Interactively Learning Multimodal Retrieval for Embodied Agents
Junpeng Yue (Peking University), Zongqing Lu (Peking University)
RetrievalRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningAgentic AIMultimodality
🎯 What it does: This paper proposes a trajectory retriever MART based on a multi-modal large language model (MLLM), which utilizes interactive feedback to generate preference pairs, fine-tunes the MLLM to evaluate trajectory effectiveness, and compresses trajectories through trajectory abstraction to reduce context window usage.
MLLM can see? Dynamic Correction Decoding for Hallucination Mitigation
Chenxi Wang (Zhejiang University), Huajun Chen (Zhejiang University)
RecognitionGenerationTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: This paper analyzes the internal mechanism of hallucination generation in MLLM through experimental analysis and proposes a training-free dynamic correction decoding method, DeCo, to reduce hallucinations.
MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMs
Jiarui Zhang (University of Southern California), Filip Ilievski (Vrije Universiteit Amsterdam)
RecognitionObject DetectionComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: This study investigates the performance bottlenecks of multimodal large language models in recognizing small visual details and proposes a training-free visual cropping method to enhance accuracy.
MLPs Learn In-Context on Regression and Classification Tasks
William Lingxiao Tong, Cengiz Pehlevan (Harvard University)
ClassificationAnomaly DetectionComputational EfficiencyTransformerTabular
🎯 What it does: This study investigates the performance of Multi-Layer Perceptrons (MLP) in In-Context Learning (ICL) tasks, comparing it with Transformer and MLP-Mixer in synthetic regression, classification, and psychological relational reasoning tasks.
MM-EMBED: UNIVERSAL MULTIMODAL RETRIEVAL WITH MULTIMODAL LLMS
Sheng-Chieh Lin (University of Waterloo), Wei Ping (NVIDIA)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningTextMultimodality
🎯 What it does: This study investigates the construction of a general multimodal retriever using a multimodal large language model (MLLM), enhancing retrieval performance through modality-aware hard negative sample mining, continuous fine-tuning, and zero-shot prompt-based re-ranking.
MM1.5: Methods, Analysis & Insights from Multimodal LLM Fine-tuning
Haotian Zhang (Apple), Yinfei Yang (Apple)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: MM1.5 is proposed, a new multimodal large language model that enhances text-rich image understanding, visual reference and localization, and multi-image reasoning capabilities through a three-stage training process (large-scale pre-training, continuous pre-training, and supervised fine-tuning).
MMAD: A Comprehensive Benchmark for Multimodal Large Language Models in Industrial Anomaly Detection
Xi Jiang (Southern University of Science and Technology), Feng Zheng (Tencent)
Anomaly DetectionTransformerLarge Language ModelImageMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed the MMAD benchmark for evaluating the performance of multimodal large language models in industrial anomaly detection, and systematically generated nearly 40,000 multiple-choice questions;
MMAU: A Massive Multi-Task Audio Understanding and Reasoning Benchmark
S Sakshi (University of Maryland), Dinesh Manocha (University of Maryland)
ClassificationRecognitionRetrievalTransformerLarge Language ModelMultimodalityBenchmarkAudio
🎯 What it does: The MMAU benchmark is proposed to evaluate the capabilities of multimodal audio understanding and reasoning models.
MMDisCo: Multi-Modal Discriminator-Guided Cooperative Diffusion for Joint Audio and Video Generation
Akio Hayakawa (Sony AI), Yuki Mitsufuji (Sony Group Corporation)
GenerationData SynthesisDiffusion modelVideoMultimodalityAudio
🎯 What it does: This paper proposes a lightweight joint guidance module that utilizes pre-trained unimodal diffusion models (audio and video) to collaboratively generate aligned audio-video pairs without the need for backpropagation on the base model, thus maintaining the model's transferability.
MMDT: Decoding the Trustworthiness and Safety of Multimodal Foundation Models
Chejian Xu (University of Illinois), Dawn Song (University of California)
Safty and PrivacyLarge Language ModelImageTextMultimodality
🎯 What it does: The MMDT platform is proposed to conduct a unified evaluation of multimodal foundational models (MMFMs) across six dimensions: security, hallucination, bias, privacy, adversarial robustness, and out-of-distribution (OOD) robustness.
MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans?
YiFan Zhang, Rong Jin (Meta AI)
RecognitionAutonomous DrivingTransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Two fully manually annotated high-resolution multimodal large model evaluation benchmarks, MME-RealWorld and MME-RealWorld-CN, have been proposed, covering 5 real-world scenarios and including 43 sub-tasks, with a total of 29,429 question-answer pairs.
MMed-RAG: Versatile Multimodal RAG System for Medical Vision Language Models
Peng Xia (University of North Carolina at Chapel Hill), Huaxiu Yao (Stanford University)
GenerationRetrievalConvolutional Neural NetworkVision Language ModelMultimodalityBiomedical DataRetrieval-Augmented Generation
🎯 What it does: This paper proposes MMed-RAG, a multimodal retrieval-augmented generation system aimed at reducing factual hallucinations in medical text generation.
MMEgo: Towards Building Egocentric Multimodal LLMs for Video QA
Hanrong Ye (Hong Kong University of Science and Technology), Yinfei Yang (Apple)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVideoTextMultimodalityBenchmark
🎯 What it does: This paper develops MM-Ego, a multimodal large language model specifically designed for first-person long videos, and constructs a large-scale egocentric QA dataset, the EgoMemoria evaluation benchmark, and a bias evaluation method.
MMFakeBench: A Mixed-Source Multimodal Misinformation Detection Benchmark for LVLMs
Xuannan Liu (Beijing University of Posts and Telecommunications), Zhaofeng He (Beijing University of Posts and Telecommunications)
TransformerLarge Language ModelVision Language ModelVision-Language-Action ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposes the MMFakeBench mixed-source multimodal information forgery detection benchmark and evaluates the performance of various detection methods and LVLMs;
MMIE: Massive Multimodal Interleaved Comprehension Benchmark for Large Vision-Language Models
Peng Xia (University of North Carolina at Chapel Hill), Huaxiu Yao (University of Chicago)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: The MMIE benchmark is proposed to evaluate the capabilities of large visual-language models in interleaved text-image understanding and generation, providing automated evaluation metrics.
MMIU: Multimodal Multi-image Understanding for Evaluating Large Vision-Language Models
Fanqing Meng (Shanghai Jiao Tong University), Wenqi Shao (Shanghai AI Laboratory)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageVideoMultimodalityPoint CloudBenchmark
🎯 What it does: This paper proposes the MMIU benchmark, which systematically evaluates the performance of large visual language models in multi-image understanding tasks.
MMKE-Bench: A Multimodal Editing Benchmark for Diverse Visual Knowledge
Yuntao Du., Qing Li
TransformerLarge Language ModelImageTextMultimodalityBenchmark
🎯 What it does: This paper presents MMKE-Bench, a multimodal knowledge editing benchmark that can evaluate the editing capabilities of large multimodal models on visual entities, visual semantics, and user-specific knowledge in real-world scenarios.
MMQA: Evaluating LLMs with Multi-Table Multi-Hop Complex Questions
Jian Wu (Institute of Science), Yue Zhang (Westlake University)
RetrievalTransformerLarge Language ModelTabularBenchmark
🎯 What it does: This paper proposes and constructs the MMQA multi-table multi-hop question answering benchmark to evaluate the understanding and reasoning capabilities of large models in tasks such as multi-table retrieval, SQL generation, question answering, and primary-foreign key selection.
MMR: A Large-scale Benchmark Dataset for Multi-target and Multi-granularity Reasoning Segmentation
Donggon Jang (KAIST), Daeshik Kim (KAIST)
Object DetectionSegmentationLarge Language ModelSupervised Fine-TuningImageBenchmark
🎯 What it does: A multi-target, multi-granularity (object-level and part-level) reasoning segmentation dataset MMR with 194K entries was constructed, and a multi-target, multi-granularity reasoning segmentation auxiliary model M-SA was proposed;
MMRole: A Comprehensive Framework for Developing and Evaluating Multimodal Role-Playing Agents
Yanqi Dai (Renmin University of China), Zhiwu Lu (Renmin University of China)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Proposed and implemented the concept and framework of Multi-modal Role-playing Agents (MRPA) with MMRole, constructed the MMRole-Data dataset, and trained the first dedicated MRPA model, MMRole-Agent.
MMSearch: Unveiling the Potential of Large Models as Multi-modal Search Engines
Dongzhi Jiang (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
RetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposes the MMSEARCH-ENGINE pipeline, enabling any large-scale multimodal model to perform multimodal AI search.
MMTEB: Massive Multilingual Text Embedding Benchmark
Kenneth Enevoldsen (Aarhus University), Niklas Muennighoff (Stanford University)
RetrievalRepresentation LearningTextBenchmark
🎯 What it does: This paper created the MMTEB benchmark through large-scale community collaboration, covering over 500 tasks and more than 250 languages, achieving a comprehensive and multilingual evaluation of text embedding models.
MMWorld: Towards Multi-discipline Multi-faceted World Model Evaluation in Videos
Xuehai He (University of California, Santa Cruz), Xin Eric Wang (University of California, Santa Cruz)
RetrievalReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: The MMWorld benchmark is proposed to evaluate the world modeling capabilities of multimodal large language models (MLLMs) in multidisciplinary and multi-faceted video understanding.
Modality-Specialized Synergizers for Interleaved Vision-Language Generalists
Zhiyang Xu (Virginia Tech), Lifu Huang (University of California)
GenerationData SynthesisTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Proposed the Modal Specialization Coordinator (MOSS) and the large-scale interactive instruction dataset LEAFINSTRUCT, improving the interleaved generation of multimodal text and images;
MoDeGPT: Modular Decomposition for Large Language Model Compression
Chi-Heng Lin (Samsung Research America), Yen-Chang Hsu (Samsung Research America)
CompressionTransformerLarge Language ModelText
🎯 What it does: A gradient-free module-level matrix decomposition compression method called MoDeGPT is proposed, which can jointly compress multiple weight matrices in the Transformer layers, significantly reducing parameters and computational load.
Model Editing as a Robust and Denoised variant of DPO: A Case Study on Toxicity
Rheeya Uppaal (University of Wisconsin Madison), Junjie Hu (University of Wisconsin Madison)
GenerationOptimizationTransformerLarge Language ModelText
🎯 What it does: A gradient-free model editing method based on projection subspace, ProFS, is proposed to reduce the toxicity of large language models.
Model Equality Testing: Which Model is this API Serving?
Irena Gao (Stanford University), Carlos Guestrin (Stanford University)
Large Language ModelPrompt EngineeringText
🎯 What it does: A Model Equality Testing framework is proposed, utilizing two-sample testing methods to evaluate whether the output distribution of a black-box LLM API is consistent with that of a reference model.
Model merging with SVD to tie the Knots
George Stoica (Georgia Tech), Judy Hoffman (Georgia Tech)
ClassificationRecognitionTransformerSupervised Fine-TuningImageTextBenchmark
🎯 What it does: The KnOTS method is proposed, which improves multi-task model merging by aligning LoRA task updates using SVD, and provides a new benchmark for evaluating general models.
Model Risk-sensitive Offline Reinforcement Learning
Gwangpyo Yoo (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
Autonomous DrivingReinforcement LearningTabularTime SeriesFinance Related
🎯 What it does: This paper proposes a model risk-based offline reinforcement learning framework (MR-IQN) that minimizes worst-case risk under a given set of feasible scenarios, thereby enhancing robustness in risk-sensitive tasks.
Model-Agnostic Knowledge Guided Correction for Improved Neural Surrogate Rollout
Bharat Srikishan (Stevens Institute of Technology), Nikhil Muralidhar (Stevens Institute of Technology)
OptimizationConvolutional Neural NetworkReinforcement LearningSequentialPhysics Related
🎯 What it does: A model-agnostic and simulator-agnostic hybrid method called HyPER is proposed, which intelligently invokes high-cost physical simulators using reinforcement learning to correct autoregressive predictions when errors occur in neural approximation models, significantly reducing cyclical errors.
Model-agnostic meta-learners for estimating heterogeneous treatment effects over time
Dennis Frauen (LMU Munich), Stefan Feuerriegel (LMU Munich)
Meta LearningTransformerTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes various model-agnostic meta-learners for estimating heterogeneous treatment effects (HTE) that change over time, implemented on any machine learning model (such as transformers).
Model-based Offline Reinforcement Learning with Lower Expectile Q-Learning
Kwanyoung Park (Yonsei University), Youngwoon Lee (Yonsei University)
Reinforcement LearningWorld ModelTabular
🎯 What it does: A model-based offline reinforcement learning method called LEQ is proposed, which utilizes next expectation regression and λ-returns to achieve low-bias value estimation on trajectories generated by the model and directly optimizes the policy.
Model-based RL as a Minimalist Approach to Horizon-Free and Second-Order Bounds
Zhiyong Wang (Chinese University of Hong Kong), Wen Sun (Cornell University)
Reinforcement Learning
🎯 What it does: This paper proves that the simplest model-based reinforcement learning (MLE + planning) can achieve almost no periodic dependence and second-order instance dependence convergence/performance guarantees in both online and offline scenarios.
Model-Free Offline Reinforcement Learning with Enhanced Robustness
Chi Zhang (University of Central Florida), Yue Wang (University of Central Florida)
Reinforcement LearningSequential
🎯 What it does: A model-free offline reinforcement learning algorithm is proposed, aiming to improve both robustness and scalability, addressing the uncertainty issues caused by model mismatch and insufficient datasets.
Modeling Complex System Dynamics with Flow Matching Across Time and Conditions
Martin Rohbeck (Genentech), Romain Lopez (Stanford University)
Flow-based ModelTime SeriesBiomedical DataOrdinary Differential Equation
🎯 What it does: Using the Multi-Boundary Flow Matching (MMFM) method, we learn the dynamics of complex systems based on time snapshots and unpaired data under multiple conditions, and perform interpolation for missing time points.
Modeling dynamic social vision highlights gaps between deep learning and humans
Kathy Garcia (Johns Hopkins University), Leyla Isik (Johns Hopkins University)
Large Language ModelVideoTextMultimodalityMagnetic Resonance Imaging
🎯 What it does: In dynamic social visual contexts, over 350 images, videos, and language models were evaluated and compared to predict human behavior ratings and fMRI neural responses.
Modeling Fine-Grained Hand-Object Dynamics for Egocentric Video Representation Learning
Baoqi Pei (Zhejiang University), Limin Wang (Zhejiang University)
Object DetectionRepresentation LearningRobotic IntelligenceTransformerLarge Language ModelContrastive LearningVideo
🎯 What it does: Proposed the HOD data generation pipeline and EgoVideo model, combining fine-grained dynamic learning of hand-object interactions to enhance egocentric video representation.
Modeling Future Conversation Turns to Teach LLMs to Ask Clarifying Questions
Michael JQ Zhang, Eunsol Choi (University of Texas at Austin)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes a method for training large language models using a dual-round preference annotation approach, enabling them to actively ask clarifying questions when faced with ambiguous queries to improve answer accuracy.
Modeling Unseen Environments with Language-guided Composable Causal Components in Reinforcement Learning
Xinyue Wang (University of California San Diego), Biwei Huang (University of California San Diego)
Robotic IntelligenceReinforcement LearningWorld ModelText
🎯 What it does: This paper proposes the World Modeling with Compositional Causal Components (WM3C) framework, which utilizes language as a composable causal control signal to learn decomposable causal components and build reversible world models, enhancing the generalization ability of reinforcement learning in unseen environments.
MoDGS: Dynamic Gaussian Splatting from Casually-captured Monocular Videos with Depth Priors
Qingming LIU, Junhui Hou (City University of Hong Kong)
GenerationData SynthesisDepth EstimationGaussian SplattingOptical FlowVideo
🎯 What it does: This paper studies a dynamic view synthesis method called MoDGS based on monocular video captured by a nearly stationary or slowly moving camera, aimed at generating high-quality images from arbitrary viewpoints.
MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts
Peng Jin (Peking University), Shuicheng YAN
TransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Proposes the MoE++ framework, which integrates traditional FFN experts with three types of zero-computation experts (Zero, Copy, Constant) to achieve adaptive routing;
MOFFlow: Flow Matching for Structure Prediction of Metal-Organic Frameworks
Nayoung Kim (Korea Advanced Institute of Science and Technology), Sungsoo Ahn (Korea Advanced Institute of Science and Technology)
GenerationOptimizationGraph Neural NetworkFlow-based ModelGraph
🎯 What it does: This paper presents MOFFLOW, a deep generative model that utilizes Riemannian flow matching specifically for predicting the crystal structures of metal-organic frameworks (MOFs);
MoLEx: Mixture of Layer Experts for Fine-tuning with Sparse Upcycling
Rachel Teo, Tan Minh Nguyen
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsTextBenchmark
🎯 What it does: A Mixture of Layer Experts (MoLEx) method is proposed, which achieves the reuse and exchange of hierarchical information by treating model layers as experts and implementing sparse routing during the fine-tuning of large-scale language models.
MolSpectra: Pre-training 3D Molecular Representation with Multi-modal Energy Spectra
Liang Wang (Chinese Academy of Sciences)
Representation LearningDrug DiscoveryTransformerContrastive LearningMultimodalityGraphPhysics Related
🎯 What it does: Using quantum mechanical energy spectrum information for pre-training 3D molecular representations, the MolSpectra framework is proposed, which includes the SpecFormer multi-spectrum encoder, a mask completion objective, and a contrastive learning strategy.
Moner: Motion Correction in Undersampled Radial MRI with Unsupervised Neural Representation
Qing Wu (ShanghaiTech University), Hongjiang Wei (Shanghai Jiao Tong University)
RestorationOptimizationContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: An unsupervised method named Moner is proposed, which utilizes implicit neural representations to reconstruct high-quality MR images without motion artifacts from undersampled radial MRI k-space, without relying on external training data.
Monet: Mixture of Monosemantic Experts for Transformers
Jungwoo Park (Korea University), Jaewoo Kang (Korea University)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: We propose MONET, a sparse mixture of experts Transformer architecture that can scale to 262,144 experts at each layer, aimed at enhancing the interpretability and controllability of LLMs.
Monitoring Latent World States in Language Models with Propositional Probes
Jiahai Feng (University of California Berkeley), Jacob Steinhardt (University of California Berkeley)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: The researchers proposed a proposition probe that can decode logical propositions from the internal activations of language models and achieve semantic combinations through binding subspaces.
MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion
Junyi Zhang (University of California Berkeley), Ming-Hsuan Yang (University of California Merced)
Pose EstimationDepth EstimationAutonomous DrivingTransformerSimultaneous Localization and MappingOptical FlowVideoPoint Cloud
🎯 What it does: Directly estimate the geometry of dynamic scenes on each frame using point maps, and further extract camera poses and video depth;
Monte Carlo Planning with Large Language Model for Text-Based Game Agents
Zijing Shi (University of Technology Sydney), Ling Chen (University of Technology Sydney)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: The MC-DML algorithm is proposed, using large language models (LLM) as a prior strategy for Monte Carlo Tree Search (MCTS), and dynamically adjusting action value assessments through intra-trial and cross-trial memory to enhance planning efficiency in text games.
Montessori-Instruct: Generate Influential Training Data Tailored for Student Learning
Xiaochuan Li (Tsinghua University), Chenyan Xiong (Carnegie Mellon University)
Data SynthesisOptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A data synthesis framework named MONTESSORI-INSTRUCT is designed to generate synthetic data that aligns with students' learning preferences by optimizing the parameters of a teacher model, thereby enhancing the performance of the student model.
MOOSE-Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses
Zonglin Yang (Nanyang Technological University), Dongzhan Zhou (Shanghai Artificial Intelligence Laboratory)
Drug DiscoveryTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: A multi-agent framework MOOSE-Chem based on large language models has been developed, capable of automatically retrieving inspirations, generating chemical hypotheses, and ranking them, given only the context of chemical research. Additionally, a benchmark TOMATO-Chem consisting of 51 high-impact chemistry papers has been constructed.
Moral Alignment for LLM Agents
Elizaveta Tennant (University College London), Mirco Musolesi (University College London)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper trains LLM agents in the Iterated Prisoner's Dilemma (IPD) using PPO for RL fine-tuning on Gemma2-2b-it, leveraging predefined intrinsic moral rewards (deontological ethics, utilitarianism, and game rewards), and further evaluates their performance in other matrix games as well as their ability to 'forget' existing selfish strategies.
More Experts Than Galaxies: Conditionally-Overlapping Experts with Biologically-Inspired Fixed Routing
Sagi Shaier (University of Colorado Boulder), Matt Jones
ClassificationOptimizationTransformerMixture of ExpertsImageText
🎯 What it does: A sparse neural network method named COMET is proposed, which generates input-related sparse masks using fixed random projections and k-winner-take-all, thereby forming exponentially overlapping experts within the network.
More RLHF, More Trust? On The Impact of Preference Alignment On Trustworthiness
Aaron Jiaxun Li, Himabindu Lakkaraju
Reinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: The system evaluated the effects of RLHF on five trust indicators and proposed a data attribution method based on influence functions to explain the results.
Morphing Tokens Draw Strong Masked Image Models
Taekyung Kim (NAVER AI Lab), Dongyoon Han (NAVER AI Lab)
SegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper proposes a Dynamic Token Morphing (DTM) method that addresses spatial inconsistency in MIM by aggregating context while maintaining the number of tokens, thereby enhancing the pre-training performance of ViT.
MorphoDiff: Cellular Morphology Painting with Diffusion Models
Zeinab Navidi (University of Toronto), BO WANG
GenerationData SynthesisDrug DiscoveryDiffusion modelImageBiomedical Data
🎯 What it does: MorphoDiff, a cell morphology prediction framework based on diffusion models, has been developed to generate high-resolution cell morphology images based on chemical or genetic perturbations.
MOS: Model Synergy for Test-Time Adaptation on LiDAR-Based 3D Object Detection
Zhuoxiao Chen (University of Queensland), Yadan Luo (University of Queensland)
Object DetectionDomain AdaptationAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes the MOS (Model Synergy) framework for online test-time adaptation (TTA) of LiDAR-based 3D object detection, constructing a super model by dynamically selecting and weighting combinations of historical checkpoints, and performing single-step fine-tuning on the current batch using self-supervised pseudo-labels.
MoS: Unleashing Parameter Efficiency of Low-Rank Adaptation with Mixture of Shards
Sheng Wang (University of Hong Kong), Chuan Wu (University of Hong Kong)
TransformerSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: This paper proposes a Mixture of Shards (MOS) method, which achieves significant parameter-efficient fine-tuning by combining global sharing with various nearly cost-free differentiation strategies in the LoRA model.
MotherNet: Fast Training and Inference via Hyper-Network Transformers
Andreas C Mueller, Raghu Ramakrishnan (Microsoft)
ClassificationKnowledge DistillationTransformerTabular
🎯 What it does: A Transformer-based super network called MotherNet is proposed, which can directly generate a small MLP for tabular classification in a single forward pass based on the training set.
Motion Control of High-Dimensional Musculoskeletal Systems with Hierarchical Model-Based Planning
Yunyue Wei (Tsinghua University), Yanan Sui (Tsinghua University)
OptimizationRobotic IntelligenceReinforcement LearningMultimodality
🎯 What it does: MPC 2 is proposed—a hierarchical model predictive control method that achieves zero-training, near-real-time control of high-dimensional musculoskeletal systems (such as a human model with 700 tendon units) under various motion tasks (standing, walking, motion imitation, etc.).
Motion-Agent: A Conversational Framework for Human Motion Generation with LLMs
Qi Wu (Hong Kong University of Science and Technology), Chi-Keung Tang (Hong Kong University of Science and Technology)
GenerationData SynthesisPose EstimationTransformerLarge Language ModelSupervised Fine-TuningAgentic AIVideoText
🎯 What it does: Proposes the Motion-Agent framework, utilizing a pre-trained LLM to achieve the generation, editing, and understanding of 3D human actions;
MotionAura: Generating High-Quality and Motion Consistent Videos using Discrete Diffusion
Onkar Kishor Susladkar (Northwestern University), Rekha Singhal (TCS Research)
RestorationGenerationTransformerDiffusion modelVideoText
🎯 What it does: The MotionAura framework is proposed to achieve high-quality video generation and video restoration based on text and sketches.
MotionClone: Training-Free Motion Cloning for Controllable Video Generation
Pengyang Ling (University of Science and Technology of China), Yi Jin (University of Science and Technology of China)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: We propose MotionClone, a training-free video motion cloning framework that can transfer motion from a reference video to a new scene, supporting text-to-video and image-to-video.
MotionDreamer: One-to-Many Motion Synthesis with Localized Generative Masked Transformer
Yilin Wang (University of Alberta), Li cheng
GenerationData SynthesisPose EstimationTransformerGenerative Adversarial NetworkVideoSequential
🎯 What it does: The MotionDreamer framework is proposed to generate diverse new motions from a single reference action.
MovieDreamer: Hierarchical Generation for Coherent Long Visual Sequences
Canyu Zhao (Zhejiang University), Chunhua Shen (Zhejiang University)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelVideoTextMultimodality
🎯 What it does: A hierarchical generative framework called MovieDreamer is proposed, which first uses an autoregressive model to predict visual tokens of sparse keyframes, and then generates high-quality long videos through diffusion rendering;
MP-Mat: A 3D-and-Instance-Aware Human Matting and Editing Framework with Multiplane Representation
Siyi Jiao (Huazhong University of Science and Technology), Mike Zheng Shou (National University of Singapore)
Image TranslationSegmentationTransformerImage
🎯 What it does: A 3D and instance-aware human segmentation and editing framework based on multi-plane representation, called MP-Mat, is proposed.
mPLUG-Owl3: Towards Long Image-Sequence Understanding in Multi-Modal Large Language Models
Jiabo Ye (Alibaba Group), Jingren Zhou (Alibaba Group)
RecognitionGenerationComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodality
🎯 What it does: Proposes the mPLUG-Owl3 multimodal large language model, which supports long image sequences and video understanding, achieving efficient inference on multimodal tasks.
MQuAKE-Remastered: Multi-Hop Knowledge Editing Can Only Be Advanced with Reliable Evaluations
Shaochen Zhong, Xia Hu (Rice University)
Large Language ModelTextBenchmark
🎯 What it does: This paper conducts a systematic audit of the MQUAKE knowledge editing dataset, discovering and fixing 33%–76% of errors, releasing a bug-free version called MQUAKE-REMASTERED, and providing a dynamic masking tool.
MR-GSM8K: A Meta-Reasoning Benchmark for Large Language Model Evaluation
Zhongshen Zeng (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
Meta LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: A meta-reasoning-based mathematical reasoning evaluation benchmark MR-GSM8K has been constructed, and the MR-Score evaluation metric has been proposed.
MRAG-Bench: Vision-Centric Evaluation for Retrieval-Augmented Multimodal Models
Wenbo Hu (University of California Los Angeles), Nanyun Peng (Stanford University)
RetrievalTransformerVision Language ModelImageMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper presents MRAG-BENCH, a multimodal evaluation benchmark focused on enhancing visual retrieval, designed to assess the capabilities of large visual-language models (LVLM) in visual knowledge retrieval and utilization.
MrSteve: Instruction-Following Agents in Minecraft with What-Where-When Memory
Junyeong Park (KAIST), Sungjin Ahn (New York University)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningVision Language ModelMultimodality
🎯 What it does: A novel low-level controller MrSteve is proposed, which integrates the 'What-Where-When' recall mechanism to enhance the efficiency of long-sequence task execution in Minecraft.
MrT5: Dynamic Token Merging for Efficient Byte-level Language Models
Julie Kallini (Stanford University), Róbert Csordás (Stanford University)
CompressionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A byte-level language model named MrT5 is proposed, which compresses sequence length by dynamically removing unimportant bytes through the insertion of learnable deletion gates in fixed layers of the encoder.
MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance
Xierui Wang (Zhejiang University), Hao Jiang (Alibaba Group)
GenerationData SynthesisDiffusion modelImageVideoBenchmark
🎯 What it does: The MS-Diffusion framework is proposed to achieve zero-shot multi-subject personalized image generation, with precise correspondence between subject positions and text descriptions achieved through layout guidance.
MTSAM: Multi-Task Fine-Tuning for Segment Anything Model
Xuehao Wang (Southern University of Science and Technology), Yu Zhang (Southern University of Science and Technology)
SegmentationDepth EstimationTransformerSupervised Fine-TuningImage
🎯 What it does: In MTSAM, the authors restructured the SAM architecture by removing the prompt encoder and adding task embeddings, allowing the model to output multiple tasks at once with variable channel numbers, and employed the low-rank tensor decomposition method ToRA for efficient multi-task fine-tuning;
MTU-Bench: A Multi-granularity Tool-Use Benchmark for Large Language Models
Pei Wang (Alibaba Group), Bo Zheng (University of Chinese Academy of Sciences)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes the MTU-Bench multi-granularity tool usage benchmark, which includes the training set MTU-Instruct and the evaluation set MTU-Eval, covering single/multiple rounds, single/multiple tools, and OOD scenarios, and provides automatic evaluation metrics that do not require GPT.
Mufu: Multilingual Fused Learning for Low-Resource Translation with LLM
Zheng Wei Lim (University of Melbourne), Trevor Cohn (Google)
Knowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Using large-scale language models (LLM) to generate multilingual auxiliary translation candidates and integrating them with post-editing tasks to improve the translation quality of low-resource languages.
MuHBoost: Multi-Label Boosting For Practical Longitudinal Human Behavior Modeling
Nguyen T Thach, Hau Chan (University of Nebraska Lincoln)
ClassificationRecommendation SystemLarge Language ModelPrompt EngineeringTabularTime SeriesElectronic Health Records
🎯 What it does: Proposes MuHBoost and its two variants, using a multi-label boosting framework to predict universal health data, addressing issues of heterogeneous data and resource consumption.
MuirBench: A Comprehensive Benchmark for Robust Multi-image Understanding
Fei Wang (University of Southern California), Muhao Chen
Large Language ModelImageMultimodalityBenchmark
🎯 What it does: A multi-image understanding benchmark MUIRBENCH has been constructed, containing 11,264 images and 2,600 multiple-choice questions, covering 12 types of tasks and 10 kinds of image relationships.
Multi-agent cooperation through learning-aware policy gradients
Alexander Meulemans (Google), Joao Sacramento (Google)
Meta LearningRecurrent Neural NetworkReinforcement LearningTabularSequential
🎯 What it does: This paper proposes a bias-free, higher-order derivative-free learning perception strategy gradient algorithm (COALA-PG) aimed at multi-agent social dilemmas, and uses it to achieve cooperation among self-interested agents and shape weaker learners in both general and sequential social dilemmas.
Multi-Dimensional Conformal Prediction
Yam Tawachi (Tel Aviv University), Bracha Laufer-Goldshtein (Tel Aviv University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This study investigates a conformal prediction method that utilizes multi-dimensional nonconformity measures, generating multi-dimensional scores through self-ensemble and partitioning cells in a high-dimensional score space to form deformable coverage regions.
Multi-domain Distribution Learning for De Novo Drug Design
Arne Schneuing (Ecole Polytechnique Federale de Lausanne), Bruno Correia (Ecole Polytechnique Federale de Lausanne)
Drug DiscoveryFlow-based ModelGraph
🎯 What it does: A new structured drug design generation model DRUGFLOW (and its extension FLEXFLOW) has been developed, which can simultaneously learn ligand atomic coordinates, atomic types, bond types, and protein side chain conformations, and can adaptively adjust ligand size during the generation process.
Multi-Draft Speculative Sampling: Canonical Decomposition and Theoretical Limits
Ashish J Khisti, Christos Louizos (Qualcomm AI Research)
OptimizationComputational EfficiencyLarge Language ModelText
🎯 What it does: This paper proposes a token-level optimal selection rule in multi-draft inference and provides an analytical decomposition of two-step importance-weighted sampling.
Multi-Field Adaptive Retrieval
Millicent Li (Northeastern University), Patrick Xia (Microsoft)
RetrievalContrastive LearningText
🎯 What it does: A multi-field adaptive retrieval framework (MFAR) is proposed, which can split semi-structured documents into fields and use lexical retrieval (BM25) and dense retrieval (Contriever) separately, and then adaptively weight and fuse the scores of each field based on query conditions.
Multi-Label Node Classification with Label Influence Propagation
Yifei Sun (Zhejiang University), Bingsheng He (Capital One)
ClassificationGraph Neural NetworkGraphBiomedical Data
🎯 What it does: This paper proposes a multi-label node classification method based on label influence propagation, LIP, which enhances multi-label node classification performance by analyzing and utilizing the positive and negative influences between labels in the graph.
Multi-Label Test-Time Adaptation with Bound Entropy Minimization
Xiangyu Wu (Nanjing University of Science and Technology), Jianfeng Lu (Nanjing University of Science and Technology)
Object DetectionDomain AdaptationPrompt EngineeringVision Language ModelImage
🎯 What it does: A multi-label test-time adaptation framework ML-TTA is proposed, which utilizes boundary entropy minimization (BEM) and text titles as pseudo-images to simultaneously enhance the confidence of multi-label predictions.
Multi-level Certified Defense Against Poisoning Attacks in Offline Reinforcement Learning
Shijie Liu (University of Melbourne), Benjamin I. P. Rubinstein (Defence Science and Technology Group)
Reinforcement LearningTabular
🎯 What it does: A multi-level authentication defense mechanism is proposed to address data poisoning attacks in offline reinforcement learning, ensuring robustness under different poisoning levels.
Multi-modal Agent Tuning: Building a VLM-Driven Agent for Efficient Tool Usage
Zhi Gao (Peking University), Qing Li (State Key Laboratory of General Artificial Intelligence)
Robotic IntelligenceAI Code AssistantTransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: A multi-modal agent tuning method is proposed, which fine-tunes the Visual Language Model (VLM) by automatically generating multi-modal tool usage data, and constructs the T3-Agent;
Multi-Modal and Multi-Attribute Generation of Single Cells with CFGen
Alessandro Palma (Helmholtz Munich), Fabian J Theis
GenerationData SynthesisFlow-based ModelMultimodalityBiomedical Data
🎯 What it does: CFGen is a conditional flow-based generative model that can generate multimodal data such as single-cell RNA sequencing and DNA accessibility while preserving the characteristics of discrete data, and it supports multi-attribute conditional control.
Multi-modal brain encoding models for multi-modal stimuli
SUBBA REDDY OOTA, Bapi Raju Surampudi (Indian Institute of Information Technology Hyderabad)
TransformerVideoMultimodalityMagnetic Resonance ImagingAudio
🎯 What it does: Using cross-modal and joint pre-trained Transformer models for brain encoding under multimodal stimuli (film + audio), comparing their alignment performance with unimodal models;
Multi-objective antibody design with constrained preference optimization
Milong Ren (Chinese Academy of Sciences), Haicang Zhang (Shanghai Jiao Tong University)
OptimizationDrug DiscoveryDiffusion modelBiomedical Data
🎯 What it does: Proposes the AbNovo framework, which utilizes constraint preference optimization to achieve multi-objective antibody design, balancing binding affinity with self-aggregation, specificity, stability, and other developability metrics.
Multi-objective Differentiable Neural Architecture Search
Rhea Sanjay Sukthanker (University of Freiburg), Frank Hutter (University of Technology Nuremberg)
OptimizationNeural Architecture SearchImage
🎯 What it does: A differentiable neural architecture search algorithm is proposed, capable of generating a complete Pareto front for multiple devices and multiple objectives (such as accuracy, latency, and energy consumption) in a single search process;
Multi-Perspective Data Augmentation for Few-shot Object Detection
Anh Khoa Nguyen Vu, Tam V. Nguyen (University of Dayton)
Object DetectionTransformerLarge Language ModelPrompt EngineeringDiffusion modelImage
🎯 What it does: For few-shot object detection, a multi-view data augmentation framework (MPAD) is proposed, which uses a controllable diffusion model to synthesize typical and difficult samples, thereby alleviating overfitting and improving detection performance.
Multi-Resolution Decomposable Diffusion Model for Non-Stationary Time Series Anomaly Detection
Guojin Zhong (Hunan University), Long Chen (Hong Kong University of Science and Technology)
Anomaly DetectionTransformerDiffusion modelTime Series
🎯 What it does: A multi-resolution decomposable diffusion model (MODEM) is proposed for anomaly detection in non-stationary time series.