AAAI 2026 Papers — Page 23
AAAI Conference on Artificial Intelligence · 4149 papers
Machine Pareidolia: Protecting Facial Image with Emotional Editing
Binh M. Le (Sungkyunkwan University), Simon S. Woo (Sungkyunkwan University)
Image TranslationSafty and PrivacySupervised Fine-TuningDiffusion modelScore-based ModelImage
🎯 What it does: Disguising the original face identity as a target identity through emotion editing to achieve facial privacy protection.
MACoT: Synthesizing Chains of Thought for Small Models via Multi-Agent Collaboration
Guokai Tang (Huazhong University of Science and Technology), Feng Zhao (Huazhong University of Science and Technology)
Data SynthesisExplainability and InterpretabilityComputational EfficiencyTransformerSupervised Fine-TuningAgentic AIPrompt EngineeringTextChain-of-Thought
🎯 What it does: Designed and implemented the MACoT multi-agent framework to synthesize chain-of-thought (CoT) specifically tailored for small language models (≤7B parameters), generated an 1879-example mathematics reasoning CoT dataset based on this framework, and subsequently fine-tuned Qwen2.5-7B-Instruct using LoRA.
MacPrompt: Maraconic-Guided Jailbreak Against Text-to-Image Models
Xi Ye (Wuhan University), Jiayi Yu (Tianjin University)
GenerationAdversarial AttackPrompt EngineeringMultimodality
🎯 What it does: This paper proposes MacPrompt, a cross-lingual black-box attack method targeting text-to-image models, capable of bypassing safety filters and generating NSFW or restricted images by constructing hybrid-language 'macaronic words.'
MACRec: A Multi-View Subspace Alignment Framework for Contrastive Sampling Calibration in Recommendation
Junping Liu (Wuhan Textile University), Yi Guo (Western Sydney University)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposed the MACRec framework, which utilizes multi-perspective subspace alignment and contrastive learning sampling calibration to address the problem of fake negative samples in GCL recommendations.
MACS: Multi-source Audio-to-image Generation with Contextual Significance and Semantic Alignment
Hao Zhou (Nanyang Technological University), Adams Wai-Kin Kong (Nanyang Technological University)
GenerationConvolutional Neural NetworkDiffusion modelContrastive LearningImageMultimodalityAudio
🎯 What it does: Propose the MACS framework, implementing a two-stage audio-image conversion method that first separates multi-source audio and then generates images.
MacVQA: Adaptive Memory Allocation and Global Noise Filtering for Continual Visual Question Answering
Zhifei Li (Hubei University), Bing Yang (Hubei University)
TransformerVision Language ModelAuto EncoderMultimodality
🎯 What it does: Proposed the MacVQA framework for continual visual question answering tasks, aiming to simultaneously address knowledge retention, adaptation to new tasks, and visual noise interference issues.
MAGIC: Mastering Physical Adversarial Generation in Context Through Collaborative LLM Agents
Yun Xing (University of Alberta), Qing Guo (Nankai University)
Adversarial AttackTransformerLarge Language ModelAgentic AIPrompt EngineeringDiffusion modelImageMultimodality
🎯 What it does: This paper proposes a framework named MAGIC, which utilizes a multi-modal large language model to collaboratively generate and deploy physical adversarial patches, automatically adapting to the testing scenario context;
MagicPaint: Operate Anything for Image Inpainting with Diffusion Model
Qinhong Yang (University of Science and Technology of China), Nenghai Yu (Beijing Electronic Science and Technology Institute)
RestorationVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Propose a unified diffusion model called MagicPaint, supporting three restoration tasks: object addition, removal, and unconditional restoration under text and image conditions;
Magnitude-Modulated Equivariant Adapter for Parameter-Efficient Fine-Tuning of Equivariant Graph Neural Networks
Dian Jin (Hong Kong Polytechnic University), Xiaoming Tao (Hong Kong Polytechnic University)
Computational EfficiencyRepresentation LearningDrug DiscoveryGraph Neural NetworkSupervised Fine-TuningGraphBenchmark
🎯 What it does: Designed and implemented a new equivariant parameter-efficient fine-tuning method called Magnitude-Modulated Equivariant Adapter (MMEA), which fine-tunes equivariant graph neural networks based on spherical harmonics by using a lightweight scalar gate to modulate the amplitude of each multiplicity channel while preserving rotational equivariance.
MAISI-v2: Accelerated 3D High-Resolution Medical Image Synthesis with Rectified Flow and Region-specific Contrastive Loss
Can Zhao (NVIDIA), Daguang Xu (NVIDIA)
GenerationData SynthesisDiffusion modelRectified FlowAuto EncoderContrastive LearningBiomedical DataComputed Tomography
🎯 What it does: Develop the MAISI-v2 3D medical image synthesis framework, employing rectified flow to achieve 33× sampling acceleration, and introduce region-specific contrast loss to enhance conditional consistency and detail quality.
MAJIC: Markovian Adaptive Jailbreaking via Iterative Composition of Diverse Innovative Strategies
Weiwei Qi (Zhejiang University), Kui Ren (Zhejiang University)
Adversarial AttackLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: This paper proposes MAJIC, a Markov chain-based black-box LLM jailbreaking framework that dynamically combines multiple disguise strategies to generate jailbreaking prompts, significantly improving attack success rates and query efficiency.
Make Foundation Models Trustworthy Again: Causal Fine-Adaptation for Medical Image Segmentation
Hongpeng Yang (University of South Carolina), Fei Guo (Central South University)
SegmentationDomain AdaptationConvolutional Neural NetworkTransformerMixture of ExpertsBiomedical Data
🎯 What it does: Developed the CausalBridgeNet framework, leveraging frozen visual foundation models and achieving domain adaptation in medical image segmentation through a causal correction module.
Make LVLMs Focus: Context-Aware Attention Modulation for Better Multimodal In-Context Learning
Yanshu Li (Brown University), Ruixiang Tang (Rutgers University)
Prompt EngineeringVision Language ModelMultimodality
🎯 What it does: This study proposes a training-free, plug-and-play attention modulation method called CAMA to enhance the performance of large vision-language models (LVLM) in multimodal context learning.
Make Model Transparent: Brain Network Analysis via Causal and Knowledge Graph Learning
Lingyuan Meng, Xinwang Liu (College Of Computer Science And Technology National University Of Defense Technology)
Explainability and InterpretabilityGraph Neural NetworkTransformerLarge Language ModelBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose the BrainCKT framework, leveraging causal knowledge graphs driven by large language models and knowledge graph enhancement, combined with graph Transformers to improve the interpretability and performance of brain network analysis.
Makespan Investigations of Sequential, Parallel, PO, and POCL Plans
Harrison Oates (Australian National University), Pascal Bercher (Australian National University)
Optimization
🎯 What it does: This paper reveals through formal proof that the convertibility of four plan representations—sequential, parallel, partial order (PO), and partial order causal link (POCL)—in terms of makespan is one-way and asymmetric; it also proves that converting POCL plans into optimal parallel plans is an NP-complete problem. Furthermore, it provides the computational complexity of the existence of parallel, PO, and POCL plans under makespan constraints (binary encoding is PSPACE-complete, decimal encoding is NP-complete), and refutes the previous assertion that planning graphs (Graphplan) can maximize parallelism.
Making Every Head Count: Sparse Attention Without the Speed-Performance Trade-off
Mingkuan Zhao (Xi'an Jiaotong University), Xiaoyan Zhu (Tsinghua University)
Computational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Propose a new self-attention mechanism called SPAttention, which divides the attention distance spectrum into non-overlapping segments, allowing each attention head to focus on different distance ranges and thus eliminating redundant computations in multi-head attention.
Making Sense of LLM Decisions: A Prototype-based Framework for Explainable Classification
Bowen Wei (George Mason University), Ziwei Zhu (George Mason University)
ClassificationExplainability and InterpretabilityKnowledge DistillationText
🎯 What it does: This paper proposes ProtoSurE, a post-hoc explanation framework based on sentence-level prototypes, to provide interpretable and faithful reasoning processes for text classification in large language models.
Making Visual Dialogue More Engaging: A New Task, Method, and Metric
Guanghui Ye (Hunan University), Zhihua Jiang (Jinan University)
TransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodalityBenchmarkAudio
🎯 What it does: Propose the audio-enhanced visual dialogue task (AVD) and design the VITA-DM model along with the MME evaluation metric, aiming to enhance the emotional engagement and participation of the conversation.
MAMA-Memeia! Multi-Aspect Multi-Agent Collaboration for Depressive Symptoms Identification in Memes
Siddhant Agarwal (University of Illinois at Chicago), Shweta Yadav (University of Illinois at Chicago)
ClassificationTransformerLarge Language ModelAgentic AIPrompt EngineeringMultimodalityBenchmark
🎯 What it does: Investigated how to identify depressive symptoms in online memes and constructed a multimodal dataset and a multi-agent reasoning framework.
Mamba-Driven Multi-View Discriminative Clustering via Global-Local Cross-View Sequence Modeling
Yuanyang Zhang, Yijie Lin (Hong Kong University of Science and Technology)
Representation LearningAuto EncoderContrastive LearningMultimodality
🎯 What it does: Proposed an end-to-end multi-view clustering framework called MGLC, which jointly models cross-view, global, and local structures via the Mamba sequence model, designs Cross-Mamba Fusion for dynamic fusion of different representations, and introduces dual-calibrated contrastive learning to leverage high-confidence pseudo labels for improving clustering quality.
MambaOVSR: Multiscale Fusion with Global Motion Modeling for Chinese Opera Video Super-Resolution
Hua Chang (Wuhan University of Science and Technology), Kui Jiang (Harbin Institute of Technology)
Super ResolutionConvolutional Neural NetworkTransformerVideo
🎯 What it does: Proposed a Mamba-based multi-scale fusion network called MambaOVSR to enhance the spatial and temporal resolution of Chinese opera videos, and constructed the first large-scale Chinese opera video dataset named COVC.
MambaSeg: Harnessing Mamba for Accurate and Efficient Image-Event Semantic Segmentation
Fuqiang Gu (Chongqing University), Zhenliang Ni (Chongqing University)
SegmentationTransformerMultimodalityBenchmark
🎯 What it does: Propose the MambaSeg dual-branch framework for multimodal semantic segmentation of RGB and event data, and design the dual-dimensional interaction module (DDIM) to finely fuse spatial and temporal information.
Managing Infinite Abstractions in Numeric Pattern Database Heuristics
Markus Fritzsche, Alexander Shleyfman (Technion Israel Institute Of Technology)
OptimizationBenchmark
🎯 What it does: Proposed a numerical planning pattern database (PDB) heuristic method for infinite abstractions, with improvements in A* guided goal-oriented abstract exploration, fragmented abstract heuristic enhancements, and failure query backup strategies.
Maniflat3D: Learning 3D Geometry Through Planar Representations from Multi-Layer Unwrapping
Zijian Cao (Chinese University of Hong Kong), Fangxin Wang (Chinese University of Hong Kong)
ClassificationCompressionConvolutional Neural NetworkAuto EncoderGaussian SplattingPoint CloudMesh
🎯 What it does: This paper proposes the Maniflat3D framework, which converts irregular 3D geometries such as point clouds and 3D Gaussian renderings into 2D UV grids through multi-layer Ball-Pivoting reconstruction and SLIM parameterization, and then utilizes mature 2D convolutional networks for 3D shape classification and compression.
ManiLong-Shot: Interaction-Aware One-Shot Imitation Learning for Long-Horizon Manipulation
Zixuan Chen (Nanjing University), Yang Gao (YiLi Normal University)
Robotic IntelligenceTransformerTextPoint Cloud
🎯 What it does: Proposed the ManiLong-Shot framework, enabling learning and completing long-horizon pre-grasping tasks (such as grasping and placing) with a single demonstration
ManipDreamer3D: Synthesizing Plausible Robotic Manipulation Video with Occupancy-aware 3D Trajectory
Ying Li (Peking University), Sirui Han (Hong Kong University of Science and Technology)
GenerationData SynthesisOptimizationRobotic IntelligenceDiffusion modelAuto EncoderImageVideoText
🎯 What it does: Propose the ManipDreamer3D framework, which first reconstructs a 3D occupancy map and plans an optimized 3D grasping trajectory using single-view images and text instructions, then generates realistic and physics-compliant robot grasping videos through a trajectory-to-video diffusion model.
ManipLVM-R1: Reinforcement Learning for Reasoning in Embodied Manipulation with Large Vision-Language Models
Zirui Song (Mohamed bin Zayed University of Artificial Intelligence), Xiuying Chen (Mohamed bin Zayed University of Artificial Intelligence)
Robotic IntelligenceReinforcement LearningVision Language ModelVision-Language-Action ModelTextMultimodalitySequential
🎯 What it does: Propose ManipLVM-R1, a reinforcement learning framework based on verifiable rewards, enabling large vision-language models (LVLMs) to accomplish robotic grasping and trajectory planning tasks without expensive human annotation;
Manipulating the Mind’s Eye: A-SAGE, the Attention-Based Attack on ViT Explainability
Boshi Zheng, Jiabin Liu (Beijing Institute of Technology)
Explainability and InterpretabilityAdversarial AttackTransformerImage
🎯 What it does: A dual-objective adversarial attack framework named A-SAGE is proposed for Vision Transformer (ViT), which can not only cause model misclassification but also generate credible yet erroneous attention explanations.
Manipulation Intention Understanding for Zero-Shot Composed Image Retrieval
Yuanmin Tang (Chinese Academy of Sciences), Qi Wu (Augusta University)
RetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodalityChain-of-Thought
🎯 What it does: Proposes an intent-centric image-text dataset and the De-MINDS framework to enhance user intent understanding in zero-shot compositional image retrieval.
Many Minds, One Path: LLM-Augmented Consensus Decision for Distributed Control in Multi-Agent Collaborative Stable Scenarios
Zhuohao Yu (University of Chinese Academy of Sciences), Qing Wang (University of Chinese Academy of Sciences)
Explainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningPrompt Engineering
🎯 What it does: Propose the LLMASC framework, utilizing a semantic-aware encoder, an LLM-driven consensus decision module, and a policy execution controller to achieve long-term stable control in distributed multi-agent systems.
MAPI-GNN: Multi-Activation Plane Interaction Graph Neural Network for Multimodal Medical Diagnosis
Ziwei Qin (Southwest Jiaotong University), Jun Li (Southwest Jiaotong University)
ClassificationGraph Neural NetworkMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a Multi-Activation Plane Interaction Graph Neural Network (MAPI-GNN), which dynamically constructs multi-graphs for each patient using a multi-dimensional feature discriminator and fuses them in two hierarchical levels to achieve multi-modal medical diagnosis.
MAPS: Multi-Agent Personality Shaping for Collaborative Reasoning
Jian Zhang (Xi'an Jiaotong University), Jun Liu (Nanyang Technological University)
TransformerLarge Language ModelAgentic AIMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This study proposes a multi-agent collaboration framework called MAPS based on the Big Five personality theory for multimodal complex reasoning.
MARE: Multimodal Analogical Reasoning for Disease Evolution-Aware Radiology Report Generation
Qingqing Gao (Beijing University of Technology), Zhaohui Liu (Beijing University of Technology)
GenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBiomedical DataChain-of-Thought
🎯 What it does: Propose an end-to-end multimodal analogy reasoning framework called MARE for generating radiology reports based on the longitudinal evolution of medical imaging.
Margin-Aware Preference Optimization for Aligning Diffusion Models Without Reference
Jiwoo Hong (KAIST AI), Jongheon Jeong (Theia Insights)
GenerationReinforcement Learning from Human FeedbackSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: This paper proposes a reference-free text-to-image diffusion model alignment method (MaPO), which directly learns model generation preferences by performing marginal optimization based on the Bradley-Terry model on the likelihood gap between selected and rejected samples.
Marginalized Generalized IoU (MGIoU): A Unified Objective Function for Optimizing Convex Parametric Shapes
Duy-Tho Le (Monash University), Hamid Rezatofighi (Monash University)
Object DetectionPose EstimationOptimizationRobotic IntelligenceImagePoint CloudMesh
🎯 What it does: Proposes a unified shape optimization loss MGIoU and its extensions MGIoU+ and MGIoU-, which optimize shape similarity by projecting 2D/3D rotation, 6-DoF, unordered polygons, trajectory prediction, and grasp detection tasks into a one-dimensional GIoU along the shape normal.
MARLIN: Multi-Agent Reinforcement Learning for Incremental DAG Discovery
Dong Li (Baylor University), Chen Zhao (College of William and Mary)
Recurrent Neural NetworkGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Propose an efficient multi-agent reinforcement learning framework MARLIN for online incremental learning of directed acyclic graphs (DAGs);
MARPO: A Reflective Policy Optimization for Multi-Agent Reinforcement Learning
Cuiling Wu (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
OptimizationReinforcement Learning
🎯 What it does: Proposed a Multi-Agent Reflective Policy Optimization (MARPO) framework that combines trajectory reflection mechanisms and KL-driven asymmetric clipping to improve sample efficiency and training stability.
MARS: A Meta-Adaptive Reinforcement Learning Framework for Risk-Aware Multi-Agent Portfolio Management
Jiayi Chen (New Jersey Institute of Technology), Guiling Wang (New Jersey Institute of Technology)
Reinforcement LearningTime SeriesFinance Related
🎯 What it does: Proposed the MARS framework, utilizing multi-agent, risk-aware reinforcement learning methods for portfolio management.
MaRS: A Multi-modality Very-high-resolution Remote Sensing Foundation Model with Cross-Granularity Meta-Modality Learning
Ruoyu Yang (Wuhan University), Yanfei Zhong (Wuhan University)
ClassificationRecognitionObject DetectionSegmentationTransformerContrastive LearningImageMultimodality
🎯 What it does: This paper constructs a large-scale MaRS-16M VHR SAR-optical paired dataset and trains a cross-modal VHR remote sensing foundation model named MaRS based on this dataset, for multi-modal fine-grained tasks.
MARS: Multi-Agent Adaptive Reasoning with Socratic Guidance for Automated Prompt Optimization
Jian Zhang (Xi'an Jiaotong University), Erik Cambria (Nanyang Technological University)
OptimizationComputational EfficiencyAgentic AIPrompt EngineeringText
🎯 What it does: Proposes a multi-agent, Socratic-style guided automated prompt optimization framework called MARS, which uses a Planner to generate task-specific optimization paths and refines prompts through a Teacher-Critic-Student three-party dialogue loop, with the final performance evaluated by the Target agent;
MARS: Multimodal Adaptive Reasoning Model for Avoiding Overthinking
Tan Yue (Wangxuan Institute of Computer Technology, Peking University), Dongyan Zhao (State Key Laboratory of General Artificial Intelligence)
Computational EfficiencyReinforcement Learning from Human FeedbackTransformerReinforcement LearningVision Language ModelTextMultimodalityChain-of-Thought
🎯 What it does: Proposed the MARS model, which employs three-stage training to achieve adaptive reasoning based on problem difficulty, significantly shortening the CoT length and improving accuracy.
MartDE: A Privacy-Preserving and Cost-Efficient Evaluation Framework for Data Marketplaces
Xinyuan Qian (University of Electronic Science and Technology of China), Guowen Xu (University of Electronic Science and Technology of China)
Safty and PrivacyComputational EfficiencyImage
🎯 What it does: This study proposes the MartDE framework, achieving efficient and fair model update transactions in data markets through privacy-preserving model evaluation.
Mask the Redundancy: Evolving Masking Representation Learning for Multivariate Time-Series Clustering
Zexi Tan (Guangdong University of Technology), Yiqun Zhang (Guangdong University of Technology)
Representation LearningConvolutional Neural NetworkAuto EncoderContrastive LearningTime Series
🎯 What it does: Proposes the EMTC method, integrating dynamic masking with multi-view learning for unsupervised clustering of multivariate time series data.
Mask2IV: Interaction-Centric Video Generation via Mask Trajectories
Gen Li (Nanyang Technological University), Laura Sevilla-Lara (University of Edinburgh)
GenerationData SynthesisRobotic IntelligenceTransformerPrompt EngineeringVision Language ModelDiffusion modelAuto EncoderVideoTextBenchmark
🎯 What it does: Propose the Mask2IV two-stage framework, which first predicts the interaction trajectories (mask sequences) between humans/robots and objects, and then generates high-quality interaction videos conditioned on these trajectories; achieves controllable video synthesis without requiring dense hand or robot mask inputs; simultaneously supports text or position control, allowing users to specify interaction objects and precisely locate the final positions.
MaskAD: Parallel Masked Autoencoder for Multi-class Unsupervised Anomaly Detection
Ruiying Lu (Xidian University), Junwei Zhang (Xidian University)
Anomaly DetectionConvolutional Neural NetworkTransformerAuto EncoderImage
🎯 What it does: Propose MaskAD, a parallel multi-branch masked autoencoder for multi-class unsupervised anomaly detection, which locates anomalies by leveraging reconstruction differences under different masks.
MaskAnyNet: Rethinking Masked Image Regions as Valuable Information in Supervised Learning
Jingshan Hong (Zhejiang University of Technology), Li Zhao (Zhejiang Normal University)
ClassificationObject DetectionSegmentationImage
🎯 What it does: Propose MaskAnyNet in supervised learning, which employs a dual-branch structure to reuse occluded image regions as auxiliary information to enhance feature representation capability.
Masked Clustering Prediction for Unsupervised Point Cloud Pre-training
Bin Ren (University of Trento), Guofeng Mei (Shanghai Jiao Tong University)
ClassificationObject DetectionSegmentationGraph Neural NetworkTransformerAuto EncoderContrastive LearningPoint Cloud
🎯 What it does: Propose MaskClu, an unsupervised ViT pre-training framework combining masked point modeling with clustering prediction for learning dense semantic features of 3D point clouds.
MASP: Multi-Aspect Guided Emotion Reasoning with Soft Prompt Tuning In Vision-Language Models
SangEun Lee (Electronics and Telecommunications Research Institute), Wonseok Chae (Electronics and Telecommunications Research Institute)
ClassificationRecognitionTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Propose the MASP framework, which extracts emotion-related visual features through a multi-perspective cross-attention module and achieves visual emotion recognition by optimizing the language model with soft prompts.
Mass Concept Erasure in Diffusion Models with Concept Hierarchy
Jiahang Tu (Zhejiang University), Hui Qian (Zhejiang University)
GenerationKnowledge DistillationDiffusion modelImageVideo
🎯 What it does: Designed a batch concept elimination method based on concept hierarchy, utilizing parent-child structures to perform group-level suppression on related sub-concepts, and proposed the SuPLoRA mechanism to preserve the generation capability of parent concepts during sub-concept elimination.
Massively Parallel Proof-Number Search for Impartial Games and Beyond
Tomáš Čížek (Charles University), Martin Schmid (Charles University)
Optimization
🎯 What it does: Proposed a two-level parallel proof-number search algorithm (PNS-PDFPN), integrating Grundy numbers to reduce game trees, and implemented solutions for impartial games like Sprouts on a distributed cluster.
Matching Policy Design for Gig Platforms with “Priority” Features
Evan Yifan Xu (Southeast University), Pan Xu (New Jersey Institute of Technology)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: Designed a novel matching strategy that integrates online matching with queueing theory, proposing an LP-based sampling framework to maximize platform revenue while ensuring waiting time guarantees for users of different priorities.
MathSE: Improving Multimodal Mathematical Reasoning via Self-Evolving Iterative Reflection and Reward-Guided Fine-Tuning
Jinhao Chen (Beihang University), Jie Tang (Beihang University)
Reinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Propose a mathematical self-evolution framework named MathSE, which continuously enhances the mathematical reasoning ability of multi-modal large language models through multi-round supervised fine-tuning, reward-guided feedback, and self-reflection loops.
MathSmith: Towards Extremely Hard Mathematical Reasoning by Forging Synthetic Problems with a Reinforced Policy
Shaoxiong Zhan (Tsinghua University), Fei Tan (East China Normal University)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: Proposes the MathSmith framework, which can randomly sample concept-explanation pairs from PlanetMath and autonomously generate high-difficulty math problems;
Matrix Editing Meets Fair Clustering: Parameterized Algorithms and Complexity
Robert Ganian (TU Wien), Simon Wietheger (TU Wien)
Optimization
🎯 What it does: The paper studies the discrete mean clustering problem under fairness constraints (equivalent to editing a matrix into a finite number of fair rows), conducting a comprehensive analysis of its parameterized complexity, and presents W[1]-hardness, FPT algorithms under certain conditions, fixed-parameter approximation algorithms, and tree-width based FPT algorithms.
Matrix-Free Two-to-Infinity and One-to-Two Norms Estimation
Askar Tsyganov (Hse University), Maxim Rakhuba (Hse University)
Recommendation SystemOptimizationImageTabular
🎯 What it does: Studied the estimation of matrix two→infinity (∥·∥₂→∞) and one→two (∥·∥₁→₂) norms in a matrix-free setting, proposing TwINEst and TwINEst++ algorithms using random Rademacher vectors and Hutchinson's diagonal estimation.
MAU-GPT: Enhancing Multi-type Industrial Anomaly Understanding via Anomaly-aware and Generalist Experts Adaptation
Zhuonan Wang (Zhejiang University), Jun Xiao (Zhejiang University)
Anomaly DetectionLarge Language ModelSupervised Fine-TuningMixture of ExpertsVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposed the MAU-Set industrial multi-type anomaly understanding dataset and MAU-GPT multimodal large model, aiming to achieve hierarchical QA evaluation from binary classification to complex reasoning;
MAUGen: A Unified Diffusion Approach for Multi-Identity Facial Expression and AU Label Generation
Xiangdong Li (Zhejiang University), Siyang Song (University of Exeter)
GenerationData SynthesisTransformerDiffusion modelImageMultimodality
🎯 What it does: Proposes MAUGen, a diffusion-based multimodal framework capable of synthesizing realistic facial images with precise Action Unit (AU) labels from a single text prompt, across multiple identities.
MAVERIX: Multimodal Audio-Visual Evaluation and Recognition IndeX
Liuyue Xie (Carnegie Mellon University), Laszlo A. Jeni
Large Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: This paper introduces MAVERIX, a unified benchmark for audio-visual multimodal understanding, containing 700 videos and 2,556 multiple-choice and open-ended questions.
MAVIS: A Benchmark for Multimodal Source Attribution in Long-form Visual Question Answering
Seokwon Song (Seoul National University), Gunhee Kim (Seoul National University)
RetrievalTransformerVision Language ModelMultimodalityBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Design and release the MAVIS benchmark for evaluating multimodal source attribution in long-text visual question answering;
Maximizing Schatten-p Norm Regularization Toward Balance
Fangfang Li (Xidian University), Qin Li (Xidian University)
OptimizationRepresentation LearningImageText
🎯 What it does: Propose and verify that maximizing Schatten-p norm regularization can effectively guide clustering results toward balance, combined with anchor graph to achieve efficient clustering.
MCGS: Markov Chain Gaussian Splatting for Dynamic Scenes Reconstruction
Yuzhong Wang (Macau University of Science and Technology), Zhongheng Chen (Macau University of Science and Technology)
GenerationComputational EfficiencyTransformerGaussian SplattingVideoPoint Cloud
🎯 What it does: Proposed a dynamic scene reconstruction method combining Markov chains with 3D Gaussian Splatting (MCGS).
MCI-Net: A Robust Multi-Domain Context Integration Network for Point Cloud Registration
Shuyuan Lin (Jinan University), Jian Weng (Jinan University)
Pose EstimationAutonomous DrivingGraph Neural NetworkPoint Cloud
🎯 What it does: Propose a multi-domain context integration network, MCI-Net, for achieving high-quality point cloud registration.
MCIE: Multimodal LLM-Driven Complex Instruction Image Editing with Spatial Guidance
Xuehai Bai, Jack Ma (Monash University)
GenerationLarge Language ModelDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: Propose a multimodal large language model-driven complex instruction image editing method called MCIE-E1
MCMoE: Completing Missing Modalities with Mixture of Experts for Incomplete Multimodal Action Quality Assessment
Huangbiao Xu (Fuzhou University), Jinglin Xu (University of Science and Technology Beijing)
RestorationTransformerMixture of ExpertsMultimodality
🎯 What it does: This paper proposes the MCMoE framework, which utilizes a hybrid expert and an adaptive gating modality generator to simultaneously learn unimodal and cross-modal features in a single-stage training, addressing the modality missing problem in action quality assessment.
MCP-AgentBench: Evaluating Real-World Language Agent Performance with MCP-Mediated Tools
Zikang Guo (University of Science and Technology of China), Zhendong Mao (University of Science and Technology of China)
Large Language ModelAgentic AITextBenchmarkChain-of-Thought
🎯 What it does: This paper constructs MCP-AgentBench, a language agent evaluation benchmark based on the Model Context Protocol (MCP), which includes 33 executable, stateless, text-based interactive MCP servers (with 188 tools in total) and 600 well-designed queries (categorized into six types based on single-machine/multi-machine and parallel/sequential invocation), and proposes MCP-Eval—a terminal task success rate evaluation framework based on LLM-as-a-judge.
MCPTox: A Benchmark for Tool Poisoning on Real-World MCP Servers
Zhiqiang Wang (University of Science and Technology of China), Xiangyang Li
Adversarial AttackLarge Language ModelBenchmark
🎯 What it does: Constructed the MCPTox benchmark to systematically evaluate the robustness of LLM agents against tool poisoning attacks on real MCP servers, generating 1,348 malicious test cases;
MCTS-SQL: Light-Weight LLMs Can Master the Text-to-SQL Through Monte Carlo Tree Search
Shuozhi Yuan (China Telecom Digital Intelligence), Zhao Jin (China Telecom Digital Intelligence)
Computational EfficiencyAI Code AssistantTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Use a lightweight large language model to iteratively optimize SQL generation through Monte Carlo Tree Search (MCTS), achieving natural language to SQL conversion.
MCTSr-Zero: Self-Reflective Psychological Counseling Dialogues Generation via Principles and Adaptive Exploration
Hao Lu (JianChengXingYun Technology Co., Ltd.), Chen Li (JianChengXingYun Technology Co., Ltd.)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Introducing the self-reflective MCTS+LLM framework MCTSr-Zero for psychological counseling dialogue generation, which generates high-quality dialogues through mechanisms such as domain alignment, regeneration, and meta-prompt adaptation.
MCW-KD: Multi-Cost Wasserstein Knowledge Distillation for Large Language Models
Hoang Tran Vuong (Hanoi University of Science and Technology), Trung Le (Monash University)
Knowledge DistillationLarge Language ModelText
🎯 What it does: Propose a multi-cost Wasserstein knowledge distillation framework, MCW-KD, to achieve more robust knowledge transfer when there are tokenizer and architectural differences between teacher and student models.
MdaIF: Robust One-Stop Multi-Degradation-Aware Image Fusion with Language-Driven Semantics
Jing Li (East China Normal University), Bin Yang (Central University of Finance and Economics)
RestorationConvolutional Neural NetworkTransformerLarge Language ModelMixture of ExpertsVision Language ModelImageMultimodality
🎯 What it does: Propose a multi-degradation-aware single-site image fusion framework (MdaIF) based on the semantic prior of a vision-language model (VLM), utilizing a Mixture of Experts (MoE) and a degradation-aware channel attention module (DCAM) to achieve infrared-visible image fusion under different weather degradation conditions such as fog, rain, and snow.
MDBench: Benchmarking Data-Driven Methods for Model Discovery
Amirmohammad Ziaei Bideh (CUNY), Jonathan Gryak (CUNY)
TransformerBenchmarkPhysics Related
🎯 What it does: This paper proposes MDBench, an open-source model discovery benchmark framework for ODE and PDE systems.
MDF: A Modality-Aware Disentanglement and Fusion Framework for Multimodal Sentiment Analysis
Zhongquan Jian (Minjiang University), Qingqiang Wu (Xiamen University)
ClassificationRecurrent Neural NetworkTransformerLarge Language ModelTextMultimodalityBenchmarkAudio
🎯 What it does: Designed and implemented the MDF framework, which first decomposes audio into text and acoustic components to generate cross-modal heterogeneity, then enhances heterogeneous features using the CHE module, and completes sentiment analysis by adaptively fusing multimodal information through MAW.
MDiff4STR: Mask Diffusion Model for Scene Text Recognition
Yongkun Du (Fudan University), Yu-Gang Jiang (Beijing Jiaotong University)
RecognitionTransformerDiffusion modelImageText
🎯 What it does: Proposed and implemented a scene text recognition framework based on the Mask Diffusion Model (MDiff4STR), which balances recognition accuracy and inference speed.
MDK12-Bench: A Multi-Discipline Benchmark for Evaluating Reasoning in Multimodal Large Language Models
Pengfei Zhou (National University of Singapore), Kaipeng Zhang (Shanghai Innovation Institute)
TransformerLarge Language ModelTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose MDK12-Bench, a large-scale multimodal evaluation benchmark based on real K-12 exams, comprising 141K instances across six subjects and a six-level knowledge hierarchy.
MDMLP-EIA: Multi-domain Dynamic MLPs with Energy Invariant Attention for Time Series Forecasting
Hu Zhang (Changsha University), Yongfang Xie (Central South University)
Time SeriesBenchmark
🎯 What it does: A time series forecasting framework named MDMLP-EIA based on multi-domain dynamic MLP and energy-invariant attention is constructed.
MDND: Unsupervised Learning Guided by Non-Differentiable Refinement for Shape Correspondence
Qinsong Li (Central South University), Shengjun Liu (Central South University)
OptimizationRepresentation LearningConvolutional Neural NetworkDiffusion modelMeshBenchmark
🎯 What it does: Proposes the MDND framework, integrating non-differentiable iterative optimization with deep functional mapping to achieve unsupervised learning.
ME-SFDA: Marginal Exploration with Pyramidal Atkinson-Shiffrin Memory for Source-Free Domain Adaptation
Chunzhi Liu (South China Normal University), Yuwu Lu (South China Normal University)
Domain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Propose a source-agnostic domain adaptation method ME-SFDA based on a pyramidal Atkinson-Shiffrin memory model, achieving effective transfer to the target domain through two-step splitting, memory fusion, and adversarial clustering.
Measuring the Unmeasurable: Unveiling Latent Cognitive Capabilities of LLM
Cui Danxin (Fudan University), Yilun Liu (Huawei)
Large Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Constructed a hierarchical classification of cognitive abilities based on the ACT-R cognitive architecture, and introduced the multilingual CogProbe benchmark and corresponding CogEval dataset to fine-grained evaluate LLMs' performance on 16 cognitive operations.
Measuring What Matters: Scenario-Driven Evaluation for Trajectory Predictors in Autonomous Driving
Longchao Da (Arizona State University), Manish Saroya (Honda Research Institute)
Autonomous DrivingRecurrent Neural NetworkGraph Neural NetworkVideoGraph
🎯 What it does: This paper proposes an adaptive evaluation framework called ED-Eva based on scene criticality, dynamically balancing the accuracy of trajectory predictors and multimodal diversity, and introduces the GMM-Area Diversity (GAD) metric and a graph-based scene classification network named ScenarioNN;
MechaFormer: Sequence Learning for Kinematic Mechanism Design Automation
Diana Bolanos (Autodesk Research), Pradeep Kumar Jayaraman (Autodesk Research)
GenerationOptimizationRobotic IntelligenceTransformerSequential
🎯 What it does: Studies the automation of mechanical mechanism design, proposing the MechaFormer model, which transforms the mechanism synthesis problem into a conditional sequence generation task. The model generates DSL strings describing mechanism topology and geometric parameters using a Transformer, achieving one-time complete design;
Mechanistic Dissection of Cross-Attention Subspaces in Text-to-Image Diffusion Models
Jun-Hyun Bae (Kyungpook National University), Heechul Jung (Kyungpook National University)
GenerationExplainability and InterpretabilityComputational EfficiencyTransformerDiffusion modelImageText
🎯 What it does: This paper performs singular value decomposition (SVD) on the cross-attention output-value (OV) circuit in text-to-image diffusion models, revealing that semantic concepts are encoded in low-dimensional spectral subspaces, and verifies their functionality by intervening in these subspaces during the generation process.
MedEyes: Learning Dynamic Visual Focus for Medical Progressive Diagnosis
Chunzheng Zhu (Hunan University), Jianxin Lin (Hunan University)
TransformerReinforcement LearningVision Language ModelImageTextMultimodalityBiomedical Data
🎯 What it does: Propose the MedEyes framework, combining scan-and-drill dual-mode visual exploration with dual-stream GRPO reinforcement learning, to achieve progressive visual attention and chain-of-thought reasoning in medical visual question answering.
MedGR2: Breaking the Data Barrier for Medical Reasoning via Generative Reward Learning
Weihai Zhi (Guangdong Institute of Intelligence Science and Technology), Shangyang Li (Guangdong Institute of Intelligence Science and Technology)
Data SynthesisData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelMultimodalityBiomedical Data
🎯 What it does: MedGR 2 automatically generates high-quality medical VLM training data through a self-improving generative-reward learning framework, and then uses reinforcement learning to enhance cross-modal and cross-task reasoning based on this data.
Medical Image Segmentation with Minimal Labeling Effort: How Far Can We Push the Limits?
Yizhe Zhang (Nanjing University of Science and Technology)
SegmentationTransformerContrastive LearningImageBiomedical DataMagnetic Resonance ImagingUltrasound
🎯 What it does: In medical image segmentation, a cyclic self-training framework is proposed to achieve near fully supervised performance using only one labeled image and a large amount of unlabeled data.
Medical Vision–Language Pretraining with LLM-Guided Temporal Supervision
Liang Bai (Shanxi University), Xian Yang (University of Manchester)
ClassificationRetrievalRepresentation LearningTransformerLarge Language ModelContrastive LearningMultimodalityBiomedical DataElectronic Health Records
🎯 What it does: This study proposes the TAMM framework, which leverages large language models (LLMs) to generate temporal trend labels and reasoning explanations for medical image-text pairs, achieving temporal alignment and semantic consistency pretraining for electronic health record sequences.
MedLA: A Logic-Driven Multi-Agent Framework for Complex Medical Reasoning with Large Language Models
Siqi Ma (Westlake University), Zelin Zang (CAIR, Hong Kong Institute of Science and Innovation, Chinese Academy of Sciences)
TransformerLarge Language ModelAgentic AIPrompt EngineeringTextBiomedical DataBenchmark
🎯 What it does: Proposed a multi-agent framework called MEDLA based on a logical tree, which can decompose medical question-answering into three-step reasoning (major premise, minor premise, conclusion), and let agents iteratively refine the reasoning tree through multi-round graph-guided discussions, ultimately achieving consistent and traceable diagnostic conclusions.
MedMKEB: A Comprehensive Knowledge Editing Benchmark for Medical Multimodal Large Language Models
Dexuan Xu (Peking University), Yu Huang (China Academy of Chinese Medical Sciences)
Adversarial AttackTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityBiomedical DataBenchmark
🎯 What it does: This paper proposes and implements MedMKEB—a comprehensive benchmark for knowledge editing in medical multimodal large language models (MLLMs), covering five dimensions: reliability, locality, generalizability, transferability, and robustness.
MedOmni-45°: A Safety–Performance Benchmark for Reasoning-Oriented LLMs in Medicine
Kaiyuan Ji (Shanghai Artificial Intelligence Laboratory), Ning Liu (Shanghai Jiao Tong University)
TransformerPrompt EngineeringBiomedical DataBenchmarkChain-of-Thought
🎯 What it does: Constructed a specialized benchmark called MedOmni-45° to evaluate the safety of medical LLMs during reasoning, and tested the safety-performance trade-offs on seven LLMs.
MedReasoner: Reinforcement Learning Drives Reasoning Grounding from Clinical Thought to Pixel-Level Precision
Zhonghao Yan (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
SegmentationTransformerLarge Language ModelReinforcement LearningPrompt EngineeringBiomedical DataBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the Unified Medical Reasoning and Localization (UMRG) task, constructs the U-MRG-14K dataset containing implicit clinical queries, chain reasoning, and pixel-level masks, and designs the MedReasoner framework, which decouples medical reasoning and segmentation modules and optimizes the reasoning module using reinforcement learning.
MedS³: Towards Medical Slow Thinking with Self-Evolved Soft Dual-sided Process Supervision
Shuyang Jiang (Fudan University), Yu Wang (Shanghai Jiao Tong University)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBiomedical DataBenchmarkChain-of-Thought
🎯 What it does: Proposed MedS 3, a self-evolving slow thinking framework that enables small medical language models to generate reliable step-by-step reasoning paths through MCTS, thereby self-improving and ultimately achieving high-performance clinical reasoning.
MedSAMix: A Training-Free Model Merging Approach for Medical Image Segmentation
Yanwu Yang (University Hospital Tubingen), Thomas Wolfers (University Hospital Tubingen)
SegmentationHyperparameter SearchTransformerImageBiomedical Data
🎯 What it does: A training-agnostic model merging framework called MedSAMix is proposed for medical image segmentation, which can enhance performance by automatically searching for merging configurations to combine the general foundation model SAM with specialized models MedSAM and MedicoSAM;
MedSpaformer: A Transferable Transformer with Multi-Granularity Token Sparsification for Medical Time Series Classification
Jiexia Ye, Fugee Tsung (Hong Kong University Of Science And Technology)
ClassificationTransformerContrastive LearningTime SeriesBiomedical DataAlzheimer's DiseaseElectrocardiogram
🎯 What it does: Proposed a transferable Transformer called MedSpaformer for medical time series classification.
Medverse: A Universal Model for Full-Resolution 3D Medical Image Segmentation, Transformation and Enhancement
Jiesi Hu (Harbin Institute of Technology at Shenzhen), Ting Ma (Harbin Institute of Technology at Shenzhen)
SegmentationConvolutional Neural NetworkMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose Medverse, a general context learning (ICL) framework capable of performing 3D medical image segmentation, transformation, and enhancement tasks within the same model.
Melodia: Training-Free Music Editing Guided by Attention Probing in Diffusion Models
Yi Yang (South China University of Technology), Qi Liu (South China University of Technology)
GenerationTransformerDiffusion modelAudio
🎯 What it does: Proposed a source-music-agnostic audio editing method called Melodia, which modifies music attributes by selectively replacing attention information in the self-attention layer of a diffusion model while preserving the original rhythm and melody.
Mem-PAL: Towards Memory-based Personalized Dialogue Assistants for Long-term User-Agent Interaction
Zhaopei Huang (Renmin University of China), Qin Jin (Renmin University of China)
GenerationData SynthesisTransformerLarge Language ModelTextGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper constructs a Chinese long-term user-assistant interaction dataset PAL-Set and benchmark PAL-Bench, and proposes a hierarchical heterogeneous memory framework H Memory 2 to enhance personalized dialogue generation.
Mem4D: Decoupling Static and Dynamic Memory for Dynamic Scene Reconstruction
Xudong Cai (Renmin University of China), Deying Li (Shanghai Jiao Tong University)
Pose EstimationDepth EstimationTransformerVideoPoint Cloud
🎯 What it does: Propose the Mem4D framework, which employs a dual-memory system (Transient Dynamics Memory and Persistent Structure Memory) to achieve online dynamic scene reconstruction from monocular videos.
Membership Inference Attack Against Large Language Model-Based Recommendation Systems: A New Distillation-Based Paradigm
Cuihong Li, Jitao Sang (Beijing Jiaotong University)
Recommendation SystemKnowledge DistillationAdversarial AttackTransformerLarge Language ModelText
🎯 What it does: Studied a membership inference attack against LLM-based recommendation systems and proposed a novel attack paradigm based on knowledge distillation.
MemeBQ:Memory Efficient Binary Quantization of LLMs
Yuanhui Wang (Sanya Nanhai Innovation and Development Base of Harbin Engineering University), Qinghao Hu (Institute of Automation Chinese Academy of Sciences)
Computational EfficiencyTransformerText
🎯 What it does: Propose a binary post-training quantization framework called MemeBQ, which reduces additional bitmap memory and improves quantization quality by leveraging row similarity clustering to share bitmaps and performing fine-grained segmentation within each group via k-means.
MemGuide: Intent-Driven Memory Selection for Goal-Oriented Multi-Session LLM Agents
Yiming Du (Chinese University of Hong Kong), Kam-Fai Wong (Chinese University of Hong Kong)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the MemGuide framework, achieving efficient dialogue in multi-session task-oriented conversations through intent-driven memory selection.
MEML-GRPO: Heterogeneous Multi-Expert Mutual Learning for RLVR Advancement
Weitao Jia (ByteDance Inc), Can Huang (University College Dublin)
Reinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningMixture of ExpertsText
🎯 What it does: Propose the MEML-GRPO framework, integrating the reasoning styles of multiple heterogeneous experts into a single model through systematic prompts, and enhancing RLVR performance via reinforcement learning and mutual learning;