AAAI 2025 Papers — Page 18
AAAI Conference on Artificial Intelligence · 3028 papers
Look Inside for More: Internal Spatial Modality Perception for 3D Anomaly Detection
Hanzhe Liang (Shenzhen University), Jinbao Wang (Shenzhen University)
Anomaly DetectionPoint Cloud
🎯 What it does: Proposes an Internal Space Modal Perception (ISMP) framework that utilizes an internal perspective to extract 3D anomaly detection features.
Loosely Synchronized Rule-Based Planning for Multi-Agent Path Finding with Asynchronous Actions
Shuai Zhou (South China University of Technology), Zhongqiang Ren (Shanghai Jiao Tong University)
OptimizationGraphBenchmark
🎯 What it does: This paper proposes a rule-based loosely coupled planning algorithm (LSRP and its SWAP variant) to address the issue of asynchronous actions in multi-agent path planning, capable of quickly obtaining acceptable approximate solutions under a large number of agents (up to 1000).
LoRID: Low-Rank Iterative Diffusion for Adversarial Purification
Geigh Zollicoffer (Georgia Institute of Technology), Manish Bhattarai (Los Alamos National Laboratory)
RestorationAdversarial AttackConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes LoRID, a method for purifying adversarial samples using low-rank tensor decomposition and multi-round early-cycle diffusion denoising.
Low-Light Image Enhancement via Generative Perceptual Priors
Han Zhou (McMaster University), Jun Chen (McMaster University)
RestorationTransformerVision Language ModelDiffusion modelImage
🎯 What it does: This paper proposes a low-light image enhancement framework based on Generative Perceptual Priors (GPP), called GPP-LLIE. It first utilizes a pre-trained vision-language model (LLaVA) to extract global and local perceptual priors from the image, and then guides the enhancement through GPP-LN and LPP-Attn in a diffusion Transformer, achieving more realistic and detail-rich enhancements.
LPCG: A Self-conditional Architecture for Labeled Point Cloud Generation
Dongshuo Huang (Northwestern Polytechnical University), Yilei Shi (Shanghai Jiao Tong University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkPoint Cloud
🎯 What it does: A self-conditioning architecture LPCG is designed, utilizing geometric and view features from unlabeled point clouds to train a generator that produces high-quality labeled 3D point clouds during inference.
LRM-LLaVA: Overcoming the Modality Gap of Multilingual Large Language-Vision Model for Low-Resource Languages
Junchen Li (Du Xiaoman Finance), Qingxuan Sun (Du Xiaoman Finance)
TransformerLarge Language ModelVision Language ModelTextMultimodality
🎯 What it does: A multilingual large language-vision model LRM-LLaVA has been constructed to bridge the modality gap between vision and text for low-resource languages.
LS-TGNN: Long and Short-Term Temporal Graph Neural Network for Session-Based Recommendation
Zhonghong Ou (Beijing University of Posts and Telecommunications), Tu Ao (Beijing University of Posts and Telecommunications)
Recommendation SystemGraph Neural NetworkTransformerContrastive LearningGraphSequential
🎯 What it does: A Long Short-Term Temporal Graph Neural Network (LS-TGNN) is proposed for session-based recommendation, combining long and short-term interest decoupling with temporal session graphs.
LTLf Synthesis on First-Order Agent Programs in Nondeterministic Environments
Till Hofmann (RWTH Aachen University), Jens Claßen (Roskilde University)
OptimizationRobotic Intelligence
🎯 What it does: In a non-deterministic environment, this study investigates how to perform LTLf synthesis for Golog programs based on scenario arithmetic to achieve specified temporal goals.
LTLf Synthesis Under Unreliable Input
Christian Hagemeier (University of Oxford), Moshe Y. Vardi (Rice University)
Reinforcement Learning
🎯 What it does: A method for LTLf synthesis that simultaneously satisfies the main objective and backup objective when the input variables are unreliable is proposed, addressing the need to consider both fully observable and partially observable scenarios during synthesis.
LVPTrack: High Performance Domain Adaptive UAV Tracking with Label Aligned Visual Prompt Tuning
Hongjing Wu (Sun Yat-sen University), Wenqi Ren (Sun Yat-sen University)
Object TrackingDomain AdaptationKnowledge DistillationTransformerPrompt EngineeringVideo
🎯 What it does: A domain-adaptive drone tracker LVPTrack based on visual prompt tuning is proposed, capable of achieving robust tracking in foggy and nighttime environments.
M^3EL: A Multi-task Multi-topic Dataset for Multi-modal Entity Linking
Fang Wang (Peking University), Yi Liang (Xinjiang University)
TransformerLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A multi-task, multi-topic, multi-modal entity linking dataset M EL 3 has been constructed, consisting of 79K instances and 318.5K images, and a training strategy that incorporates entity descriptions into the text for modal enhancement has been proposed.
M2Flow: A Motion Information Fusion Framework for Enhanced Unsupervised Optical Flow Estimation in Autonomous Driving
Xunpei Sun (Sun Yat-sen University), Zuoxun Hou (Beijing Institute of Space Mechanics and Electricity)
Autonomous DrivingConvolutional Neural NetworkOptical FlowImageVideo
🎯 What it does: The paper presents M2Flow, a multi-frame unsupervised optical flow estimation framework that addresses occlusion issues by utilizing four-frame input combined with motion information fusion.
M²N: A Progressive Macro-to-Micro 3D Modeling Scheme for Unveiling Drug-Target Affinity
Tianxu Lv (Jiangnan University), Xiang Pan (Jiangnan University)
Drug DiscoveryGraph Neural NetworkTransformerGraphBiomedical Data
🎯 What it does: An evolutionary macro-to-micro 3D modeling network M²N is proposed for drug-target affinity (DTA) prediction.
M2OST: Many-to-one Regression for Predicting Spatial Transcriptomics from Digital Pathology Images
Hongyi Wang (Zhejiang University), Lanfen Lin (Zhejiang University)
TransformerImageBiomedical Data
🎯 What it does: A multi-scale to one regression Transformer (M2OST) is proposed, which can utilize pathological images of different resolutions to predict spatial transcriptomics (ST) gene expression.
M²RL-Net: Multi-View and Multi-Level Relation Learning Network for Weakly-Supervised Image Forgery Detection
Jiafeng Li (East China Normal University), Lianghua He (Tongji University)
Anomaly DetectionTransformerContrastive LearningImage
🎯 What it does: The paper proposes a weakly supervised image forgery detection network M RL‑Net based on multi-view and multi-layer relationship learning, which can achieve forgery detection and pixel-level localization using only image-level labels.
M3Net: Efficient Time-Frequency Integration Network with Mirror Attention for Audio Classification on Edge
Xuanming Jiang (Xi'an Jiaotong University), Xueming Qian (Xi'an Jiaotong University)
ClassificationComputational EfficiencyConvolutional Neural NetworkAudio
🎯 What it does: A lightweight audio classification network M3Net is proposed, suitable for edge devices.
M3Net: Multimodal Multi-task Learning for 3D Detection, Segmentation, and Occupancy Prediction in Autonomous Driving
Xuesong Chen (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
Object DetectionSegmentationAutonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: A multi-modal multi-task network M3Net is proposed, capable of simultaneously performing 3D detection, BEV semantic segmentation, and 3D occupancy prediction.
MABR: Multilayer Adversarial Bias Removal Without Prior Bias Knowledge
Maxwell J. Yin (University of Western Ontario), Charles Ling (University of Western Ontario)
ClassificationTransformerSupervised Fine-TuningGenerative Adversarial NetworkText
🎯 What it does: Proposes the MABR framework, which dynamically identifies biased samples at multiple levels of the Transformer encoder through an auxiliary bias detector, and eliminates these biases using adversarial training without any prior bias knowledge or protected attribute labels.
MaFeRw: Query Rewriting with Multi-Aspect Feedbacks for Retrieval-Augmented Large Language Models
Yujing Wang (Beihang University), Zhiming Zheng (Beijing Academy of Blockchain and Edge Computing)
GenerationRetrievalTransformerLarge Language ModelReinforcement LearningTextRetrieval-Augmented Generation
🎯 What it does: Developed a multi-dimensional feedback query rewriting method called MaFeRw, which enhances the retrieval and generation quality of multi-turn dialogue RAG systems.
MAGIC: Generating Self-Correction Guideline for In-Context Text-to-SQL
Arian Askari (Leiden University), Xinye Tang (Microsoft)
GenerationOptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes the MAGIC multi-agent framework, which automatically generates text-to-SQL self-correction criteria and uses them during the inference phase;
MagicMan: Generative Novel View Synthesis of Humans with 3D-Aware Diffusion and Iterative Refinement
Xu He (Shenzhen International Graduate School, Tsinghua University), Haolin Zhuang
GenerationData SynthesisPose EstimationDiffusion modelImage
🎯 What it does: MagicMan is designed, a multi-view diffusion model based on the pre-trained Stable Diffusion 1.5 and the SMPL-X 3D body shape prior, capable of generating dense and high-quality consistent views from a single portrait, supporting 3D human reconstruction.
MagicNaming: Consistent Identity Generation by Finding a “Name Space” in T2I Diffusion Models
Jing Zhao (National University of Defense Technology), Yuhua Tang (National University of Defense Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The concept of 'Name Space' is proposed, where a trained image encoder maps any face to this space, and achieves consistent generation of ordinary identities by embedding predicted names into text prompts, maintaining the generative capabilities of the original diffusion model.
MaintaAvatar: A Maintainable Avatar Based on Neural Radiance Fields by Continual Learning
Shengbo Gu (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
GenerationPose EstimationNeural Radiance FieldImage
🎯 What it does: A sustainable and updatable virtual character model based on NeRF has been constructed, supporting rapid learning of new appearances from a small number of samples while retaining the rendering capability of old appearances.
Maintaining Fairness in Logit-based Knowledge Distillation for Class-Incremental Learning
Zijian Gao (National University of Defense Technology), Huaimin Wang (National University of Defense Technology)
ClassificationKnowledge DistillationImage
🎯 What it does: This paper proposes a preprocessing method for Z-score normalization of the logit of student and teacher models, eliminating the conflict between KD and CE objectives, thereby maintaining fairness between old and new classes in class-incremental learning, and further enhancing internal fairness within old classes through internal class relationship distillation.
Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental Learning
Wei Chen (Southeast University), Yi Zhou (Southeast University)
Domain AdaptationRepresentation LearningContrastive LearningImage
🎯 What it does: A method for incremental learning using domain shift and contrastive learning, called DisCo, is proposed to reduce catastrophic forgetting in class-incremental learning.
Making Large Vision Language Models to Be Good Few-Shot Learners
Fan Liu (Hohai University), Jun Zhou (Hong Kong University of Science and Technology)
ClassificationMeta LearningTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: By performing meta-learning-based instruction fine-tuning on large-scale visual language models, and incorporating character perturbation label enhancement and attribute description-based candidate filtering, the performance in few-shot classification tasks is significantly improved.
MalCL: Leveraging GAN-Based Generative Replay to Combat Catastrophic Forgetting in Malware Classification
Jimin Park (Ewha Womans University), Se Eun Oh (Ewha Womans University)
ClassificationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkTabular
🎯 What it does: Developed the MalCL system, which uses GAN to generate replay samples for continual learning to mitigate catastrophic forgetting in malware classification.
MalDetectFormer: Leveraging Sparse SpatioTemporal Information for Effective Malicious Traffic Detection
Shuai Zhang (Zhongguancun Laboratory), Bo Li (Beihang University)
Anomaly DetectionTransformerTime Series
🎯 What it does: Proposes the MalDetectFormer model, which enhances malicious traffic detection and identification using sparse spatiotemporal attention and subgraph convolutional networks;
Mamba YOLO: A Simple Baseline for Object Detection with State Space Model
Zeyu Wang (Zhejiang Normal University), Hongbo Li (Beijing Geekplus Technology)
Object DetectionImage
🎯 What it does: A real-time object detection framework Mamba YOLO based on a linear complexity state space model (SSM) is proposed, achieving high-performance detection models trained from scratch without large-scale pre-training.
Mamba-CAD: State Space Model for 3D Computer-Aided Design Generative Modeling
Xueyang Li (Fudan University), Xiangdong Zhou (Fudan University)
GenerationData SynthesisTransformerGenerative Adversarial NetworkSequential
🎯 What it does: A self-supervised generative framework called Mamba-CAD based on the Mamba state space model is proposed, capable of handling longer parameterized CAD sequences and generating complex 3D CAD models.
MambaLCT: Boosting Tracking via Long-term Context State Space Model
Xiaohai Li (Guangxi Normal University), Shuxiang Song (Guangxi Normal University)
Object TrackingTransformerVideo
🎯 What it does: Proposes MambaLCT, a framework that enhances target tracking accuracy through long-term contextual information.
MambaPro: Multi-Modal Object Re-identification with Mamba Aggregation and Synergistic Prompt
Yuhao Wang (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
RecognitionRetrievalPrompt EngineeringContrastive LearningImageMultimodality
🎯 What it does: The MambaPro framework is proposed for the task of multi-modal person re-identification, combining the CLIP pre-trained model, parallel feedforward adapters, synergistic residual prompts, and the Mamba aggregation module to achieve efficient feature learning.
MAMS: Model-Agnostic Module Selection Framework for Video Captioning
Sangho Lee (Sungkyunkwan University), Hogun Park (Sungkyunkwan University)
GenerationTransformerVideoTextMultimodality
🎯 What it does: This paper proposes the MAMS (Model-Agnostic Module Selection) framework, which can dynamically select appropriately sized generation modules based on the visual information of each video segment and filter important visual tokens at both the frame and word levels.
Manhattan Self-Attention Diffusion Residual Networks with Dynamic Bias Rectification for BCI-based Few-Shot Learning
Hao Wang (Harbin Engineering University), Yiming Xu (Tokyo Institute of Technology)
ClassificationMeta LearningConvolutional Neural NetworkDiffusion modelMultimodalityBiomedical Data
🎯 What it does: Proposes the MSADiff-Resnet model and a complete multi-source brain-computer interface few-shot learning framework to address distribution bias and sample insufficiency issues.
Manta: Enhancing Mamba for Few-Shot Action Recognition of Long Sub-Sequence
Wenbo Huang (Southeast University), Miki Haseyama (Hokkaido University)
RecognitionContrastive LearningVideo
🎯 What it does: Proposes the Manta framework, adapting Mamba for few-shot action recognition of long subsequences, and incorporates Matryoshka Mamba and mixed contrastive learning;
MapExpert: Online HD Map Construction with Simple and Efficient Sparse Map Element Expert
Dapeng Zhang (Lanzhou University), Qingguo Zhou (Lanzhou University)
Object DetectionAutonomous DrivingTransformerMixture of ExpertsSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes MapExpert, an online HD map construction method based on sparse experts and learnable weighted moving downsampling.
MAPF-GPT: Imitation Learning for Multi-Agent Pathfinding at Scale
Anton Andreychuk (AIRI), Alexey Skrynnik (Federal Research Center 'Computer Science and Control' of the Russian Academy of Sciences)
TransformerSupervised Fine-TuningTabular
🎯 What it does: Designed and trained a Transformer-based multi-agent pathfinding model MAPF-GPT, which learns from expert solutions through supervised imitation learning and generates safe paths in a distributed environment without communication.
Mapping from Meaning: Addressing the Miscalibration of Prompt-Sensitive Language Models
Kyle Cox (University of Texas at Austin), Ying Ding (University of Texas at Austin)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Using semantic-preserving paraphrase perturbation sampling and embedding variance uncertainty measurement, the uncertainty of LLM in question-answering tasks is evaluated and calibrated for improvement.
Marginal Benefit Driven RL Teacher for Unsupervised Environment Design
Dexun Li (Singapore Management University), Pradeep Varakantham (Singapore Management University)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: Proposes an unsupervised environment design method based on marginal returns, using a teacher-student cycle to generate training environments suitable for RL learners, and enhancing generalization ability through a representative state diversity mechanism.
Markov Balance Satisfaction Improves Performance in Strictly Batch Offline Imitation Learning
Rishabh Agrawal (University of Southern California), Ashutosh Nayyar (University of Southern California)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningFlow-based ModelSequential
🎯 What it does: In a strictly offline batch imitation learning environment, an algorithm based on the Markov balance equation is proposed, which estimates the state-action transition density using conditional normal flows and combines it with behavior cloning to achieve learning without rewards and interactions.
MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesis
Wanggui He (Alibaba Group), Hao Jiang (Alibaba Group)
GenerationData SynthesisTransformerLarge Language ModelMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: The MARS framework is proposed, utilizing the pre-trained LLM (Qwen-7B) combined with the trainable visual expert in the SemVIE component to achieve efficient text-to-image (T2I) synthesis, supporting both Chinese and English bilingual and text-image joint generation.
Masked Language Modeling Becomes Conditional Density Estimation for Tabular Data Synthesis
Seunghwan An (University of Seoul), Jong-June Jeon (University of Seoul)
Data SynthesisTransformerTabular
🎯 What it does: A conditional density estimation method based on a masked language model, MaCoDE, is proposed to generate synthetic data that retains the statistical features of tabular data while achieving high machine learning performance.
MaskPrompt: Open-Vocabulary Affordance Segmentation with Object Shape Mask Prompts
Dongpan Chen (Beijing University of Technology), Baocai Yin (Beijing University of Technology)
Object DetectionSegmentationTransformerPrompt EngineeringImageBenchmark
🎯 What it does: This paper proposes the open vocabulary segmentation task and the corresponding benchmark dataset, along with the MaskPrompt method.
MaskViM: Domain Generalized Semantic Segmentation with State Space Models
Jiahao Li (Xiamen University), Yanyun Qu (Xiamen University)
SegmentationDomain AdaptationSupervised Fine-TuningImage
🎯 What it does: This paper proposes a semantic segmentation network called MaskViM based on a state space model, which enhances generalization to unknown domains by learning distinguishable binary masks to filter out non-causal labels in images.
MASS: Overcoming Language Bias in Image-Text Matching
Jiwan Chung (Yonsei University), Youngjae Yu (Yonsei University)
RetrievalTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A point mutual information-based image-text matching scoring framework called MASS is proposed, which requires no additional training and aims to reduce the language bias of visual language models.
Massively Parallel Continuous Local Search for Hybrid SAT Solving on GPUs
Yunuo Cen (National University of Singapore), Xuanyao Fong (National University of Singapore)
OptimizationBenchmark
🎯 What it does: A parallel algorithm based on FFT heuristics, FastFourierSAT, is proposed for efficiently executing continuous local search to solve mixed SAT formulas on GPUs.
Matching While Perceiving: Enhance Image Feature Matching with Applicable Semantic Amalgamation
Shihua Zhang (Wuhan University), Jiayi Ma (Wuhan Institute of Technology)
RecognitionRetrievalDomain AdaptationGraph Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes the SemaGlue framework, which injects semantic features from a pre-trained semantic segmentation model into the image feature matching process to improve matching accuracy in sparse textures and occluded scenes.
Math-PUMA: Progressive Upward Multimodal Alignment to Enhance Mathematical Reasoning
Wenwen Zhuang (University of Chinese Academy of Sciences), Jin Zeng (University of Chinese Academy of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodalityBenchmark
🎯 What it does: A three-stage training framework called Math-PUMA is proposed and implemented to enhance the performance of multimodal large language models in mathematical reasoning tasks.
MathSpeech: Leveraging Small LMs for Accurate Conversion in Mathematical Speech-to-Formula
Sieun Hyeon (Seoul National University), Jaeyoung Do (Seoul National University)
RecognitionGenerationTransformerLarge Language ModelSupervised Fine-TuningTextAudio
🎯 What it does: Proposes the MathSpeech pipeline, which first corrects ASR output using a small LLM, and then directly generates LaTeX formulas, supporting the generation of subtitles for math lectures.
Max-Mahalanobis Anchors Guidance for Multi-View Clustering
Pei Zhang (National University of Defense Technology), Ivor Tsang (Agency for Science Technology and Research)
OptimizationRepresentation LearningMultimodality
🎯 What it does: Proposes a maximum Mahalanobis distance-based Anchor design (MMA) as a guide to enhance representation learning and clustering performance in multi-view clustering.
Maximizing the Position Embedding for Vision Transformers with Global Average Pooling
Wonjun Lee (Yonsei University), Suhyun Kim (Korea Institute of Science and Technology)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: Proposes the MPVG method, maximizing the expression of positional information when using global average pooling in Vision Transformers.
MCGAN: Enhancing GAN Training with Regression-Based Generator Loss
Baoren Xiao (University College London), Weixin Yang (University of Oxford)
GenerationData SynthesisGenerative Adversarial NetworkImageVideoTime SeriesFinance Related
🎯 What it does: An algorithm named MCGAN is proposed, which enhances the training of Generative Adversarial Networks (GANs) through a regression loss function, addressing the lack of supervision in generator training.
McHirc: A Multimodal Benchmark for Chinese Idiom Reading Comprehension
Tongguan Wang (Huazhong Agricultural University), Ying Sha (Huazhong Agricultural University)
Graph Neural NetworkTransformerContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: This paper first constructs the first multimodal Chinese idiom reading comprehension dataset MChIRC (44,433 image-text pairs, 2,926 idioms), and then proposes a multimodal idiom reading comprehension model DCIGN based on dual contrastive learning and graph convolutional networks, using images to assist text in enhancing idiom understanding and selection.
MDD-5k: A New Diagnostic Conversation Dataset for Mental Disorders Synthesized via Neuro-Symbolic LLM Agents
Congchi Yin (Theta Health Inc.), Xun Jiang (Theta Health Inc.)
GenerationData SynthesisTransformerLarge Language ModelTextElectronic Health Records
🎯 What it does: A neural-symbolic multi-agent framework was constructed to synthesize diverse psychological disorder diagnostic dialogues using anonymized real medical records, and the largest annotated diagnostic dialogue dataset in Chinese, MDD-5k, was released.
MDFG: Multi-Dimensional Fine-Grained Modeling for Fatigue Detection
Mei Wang (Wuhan University), Mang Ye (Wuhan University)
ClassificationRecognitionConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: This paper studies the problem of multi-dimensional fine-grained fatigue detection and proposes the MDFG method, which automatically generates multi-dimensional fine-grained labels (time, type, level) using eye movement wavelet features, and constructs intermediate fatigue states through base class extraction and synthesis to achieve accurate recognition of real fatigue videos.
Measuring Cross-Modal Interactions in Multimodal Models
Laura Wenderoth (University of Cambridge), Mateja Jamnik (University of Cambridge)
Explainability and InterpretabilityMultimodalityTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes and implements an interaction measurement for multimodal models called InterSHAP, which quantifies the interaction effects between different modalities and provides explanations at both global and local levels.
Measuring Human and AI Values Based on Generative Psychometrics with Large Language Models
Haoran Ye (Peking University), Guojie Song (Peking University)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: A generative psychological measurement framework based on large language models (GPV) is proposed and implemented, which can automatically extract 'perceptions' from unstructured text and convert them into value vectors for both humans and LLMs.
MEATRD: Multimodal Anomalous Tissue Region Detection Enhanced with Spatial Transcriptomics
Kaichen Xu (Zhongnan University of Economics and Law), Xiaobo Sun (Zhongnan University of Economics and Law)
Anomaly DetectionGraph Neural NetworkTransformerMultimodalityBiomedical Data
🎯 What it does: This paper proposes a multi-modal abnormal tissue region detection method called MEATRD, which combines histological images with spatial transcriptomic data to achieve precise localization and identification of abnormal tissue regions (ATR).
Mechanism Design for Connecting Regions Under Disruptions
Hau Chan (University of Nebraska-Lincoln), Chenhao Wang (Beijing Normal University)
Optimization
🎯 What it does: The study constructs optimal paths (bridges/channels) to minimize social costs and maximize costs through mechanism design when two areas are disconnected due to obstacles, and designs corresponding strategy-independent mechanisms.
Mediation Analysis for Probabilities of Causation
Yuta Kawakami (Mohamed bin Zayed University of Artificial Intelligence), Jin Tian (Mohamed bin Zayed University of Artificial Intelligence)
Tabular
🎯 What it does: A novel probabilistic causal (PoC) metric based on causal mediation analysis is proposed—controlled direct PNS, natural direct PNS, and natural indirect PNS, along with their identifiability theorems and estimation methods.
Medical Manifestation-Aware De-Identification
Yuan Tian (Shanghai AI Laboratory), Guangtao Zhai (Shanghai Jiao Tong University)
ClassificationSegmentationGenerationData SynthesisSafty and PrivacyTransformerSupervised Fine-TuningDiffusion modelGenerative Adversarial NetworkImageBiomedical Data
🎯 What it does: A MeMa dataset (42,307 synthetic patient facial images) was constructed using a Stable Diffusion generative model based on real patient photos, and MedSem-DeID was proposed on this dataset, achieving facial identity removal while retaining medical diagnostic information with reversible decryption.
Medical MLLM Is Vulnerable: Cross-Modality Jailbreak and Mismatched Attacks on Medical Multimodal Large Language Models
Xijie Huang (Beihang University), Chengwei Pan (Beihang University)
Adversarial AttackTransformerLarge Language ModelMultimodalityBiomedical Data
🎯 What it does: This paper studies the security vulnerabilities of medical multimodal large language models (MedMLLM), proposing cross-modal jailbreaking and mismatch attacks, and constructing the 3MAD dataset.
Medical Multimodal Model Stealing Attacks via Adversarial Domain Alignment
Yaling Shen (Bosch Center for Artificial Intelligence), Mario Fritz (CISPA Helmholtz Center for Information Security)
Domain AdaptationAdversarial AttackLarge Language ModelGenerative Adversarial NetworkMultimodalityBiomedical Data
🎯 What it does: This study investigates the theft attacks on medical multimodal large models and proposes the ADA-STEAL method based on adversarial domain alignment.
MEDSAGE: Enhancing Robustness of Medical Dialogue Summarization to ASR Errors with LLM-generated Synthetic Dialogues
Kuluhan Binici (ASUS Intelligent Cloud Services), Stefan Winkler (ASUS Intelligent Cloud Services)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataAudio
🎯 What it does: Using large language models (LLM) to generate synthetic medical dialogue texts similar to real ASR errors through in-context learning and error-tagged grammar, serving as data augmentation to enhance the robustness of medical dialogue summarization models against ASR errors.
MegActor-Sigma: Unlocking Flexible Mixed-Modal Control in Portrait Animation with Diffusion Transformer
Shurong Yang (MEGVII Technology), Jin Wang (University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodalityAudio
🎯 What it does: Developed MegActor-Sigma, a hybrid modal conditional diffusion transformer (DiT) that allows for flexible control of portrait animation using both visual and audio signals.
Meme Trojan: Backdoor Attacks Against Hateful Meme Detection via Cross-Modal Triggers
Ruofei Wang (Hong Kong Baptist University), Renjie Wan (Hong Kong Baptist University)
ClassificationAdversarial AttackTransformerImageTextMultimodality
🎯 What it does: This paper proposes the Meme Trojan framework to implement backdoor attacks on hate meme detection models.
Memorize and Rank: Elevating Large Language Models for Clinical Diagnosis Prediction
Mingyu Derek Ma (University of California), Wei Wang (University of California)
ClassificationGenerationRecommendation SystemTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningTextBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes and implements MERA, a model that utilizes large language models for clinical diagnosis prediction, capable of directly generating a set of diagnoses for the next visit from the patient's historical diagnosis codes.
Memory Efficient Matting with Adaptive Token Routing
Yiheng Lin (Beijing Jiaotong University), Yunchao Wei (Beijing Jiaotong University)
SegmentationComputational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: The MEMatte framework is proposed, introducing a router and a lightweight token refinement module before the global attention in the Transformer, significantly reducing memory usage and inference speed for high-resolution image matting.
Memory-Augmented Re-Completion for 3D Semantic Scene Completion
Yu-Wen Tseng (National Taiwan University), Wen-Huang Cheng (National Taiwan University)
SegmentationAutonomous DrivingTransformerPoint CloudBenchmark
🎯 What it does: A memory-enhanced re-completion framework MARE is proposed to address the issue of insufficient visibility in 3D semantic scene completion from camera inputs.
Memory-Reduced Meta-Learning with Guaranteed Convergence
Honglin Yang (Xiamen University), Xiao Yu (Xiamen University)
OptimizationMeta LearningImage
🎯 What it does: A meta-learning algorithm is designed that does not use historical lower-level parameters/gradients, significantly reducing memory consumption while ensuring sublinear convergence.
Mental-Perceiver: Audio-Textual Multi-Modal Learning for Estimating Mental Disorders
Jinghui Qin (Guangdong University of Technology), Rumin Zhang (Guangdong Shuye Intelligent Technology Co., Ltd.)
ClassificationTransformerTextMultimodalityAudio
🎯 What it does: A large-scale multimodal dataset of Chinese adolescent anxiety and depression, MMPsy, has been constructed, and a Mental-Perceiver model based on Transformer full attention has been proposed to estimate psychological disorders from both speech and text simultaneously.
MENTOR: Multi-level Self-supervised Learning for Multimodal Recommendation
Jinfeng Xu (University of Hong Kong), Edith C. H. Ngai (Hong Kong Polytechnic University)
Recommendation SystemGraph Neural NetworkContrastive LearningMultimodality
🎯 What it does: The MENTOR framework is proposed in multimodal recommendation, utilizing multi-level self-supervised learning for cross-modal alignment while preserving interaction information.
MEPNet: Medical Entity-Balanced Prompting Network for Brain CT Report Generation
Xiaodan Zhang (Beijing University of Technology), Liangqiong Qu (University of Hong Kong)
GenerationConvolutional Neural NetworkLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBiomedical DataComputed Tomography
🎯 What it does: A Medical Entity Balance Prompt Network (MEPNet) is proposed, utilizing large language models to generate brain CT reports.
MergeNet: Knowledge Migration Across Heterogeneous Models, Tasks, and Modalities
Kunxi Li (Zhejiang University), Fei Wu (Shanghai Jiao Tong University)
Domain AdaptationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkTransformerImageTextMultimodality
🎯 What it does: The MergeNet framework achieves cross-model, cross-task, and cross-modal knowledge transfer by treating model parameters as carriers of knowledge, directly bridging networks with different structures and functions through parameter adapters in the parameter space.
Merging Mechanisms for Ads and Organic Items in E-commerce Platforms
Nan An (Renmin University of China), Liang Zhang (Renmin University of China)
Recommendation SystemOptimization
🎯 What it does: This paper proposes a merging mechanism that simultaneously handles advertisements and organic products on multi-slot e-commerce search result pages, requiring compliance with incentive compatibility (IC), individual rationality (IR), feasibility, and avoiding the repetition of the same product appearing as both an advertisement and an organic listing.
MeRino: Entropy-Driven Design for Generative Language Models on IoT Devices
Youpeng Zhao (University of Central Florida), Jun Wang (University of Technology Sydney)
GenerationOptimizationComputational EfficiencyNeural Architecture SearchTransformerLarge Language ModelText
🎯 What it does: A lightweight generative language model MeRino for IoT devices is designed, utilizing information entropy-driven mathematical programming for rapid model generation.
Mesh Watermark Removal Attack and Mitigation: A Novel Perspective of Function Space
Xingyu Zhu (Southern University of Science and Technology), Xuetao Wei (Southern University of Science and Technology)
RestorationAdversarial AttackMesh
🎯 What it does: This paper proposes a function space-based attack method called FUNCEVADE and a corresponding defense method called FUNCMARK, aimed at achieving robustness against topological transformations in 3D mesh watermarking.
Mesoscopic Insights: Orchestrating Multi-Scale & Hybrid Architecture for Image Manipulation Localization
Xuekang Zhu (Sichuan University), Ji-Zhe Zhou (Sichuan University)
Image TranslationObject DetectionAnomaly DetectionComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes the Mesorch architecture, which uses a parallel CNN and Transformer dual-path encoder to simultaneously extract micro details and macro semantics, achieving precise localization of image tampering areas.
MetaNeRV: Meta Neural Representations for Videos with Spatial-Temporal Guidance
Jialong Guo (Zhejiang University), Haishuai Wang (Shanghai Jiaotong University)
CompressionOptimizationMeta LearningVideo
🎯 What it does: This paper proposes MetaNeRV, a meta-learning based NeRV video implicit representation framework that can quickly adapt to new videos and supports multi-scale spatial supervision and advanced temporal guidance.
MetaSymNet: A Tree-like Symbol Network with Adaptive Architecture and Activation Functions
Yanjie Li (Institute of Semiconductor, Chinese Academy of Sciences), Meilan Hao (Institute of Semiconductor, Chinese Academy of Sciences)
OptimizationTabularBenchmarkPhysics Related
🎯 What it does: This paper proposes MetaSymNet, a tree-structured adaptive symbolic network that transforms symbolic regression from combinatorial optimization to numerical optimization, and achieves automatic symbolic selection through an evolvable PANGU meta-function.
Metric Distortion of Line-up Elections: The Right Person for the Right Job
Christopher Jerrett (Rensselaer Polytechnic Institute), Elliot Anshelevich (Rensselaer Polytechnic Institute)
🎯 What it does: A mechanism for multi-position ranking elections is proposed under the spatial preference model, and a constant-level metric distortion upper bound is provided.
Metric-Agnostic Continual Learning for Sustainable Group Fairness
Heng Lian (William and Mary), Yi He (William and Mary)
Domain AdaptationRepresentation LearningAdversarial AttackGraph Neural NetworkTabular
🎯 What it does: The MacFRL method is proposed in the context of continual learning, utilizing task reordering and flexible representation learning to achieve group fairness in continual learning with single-label initial tasks and unlabeled subsequent tasks.
MFL-Owner: Ownership Protection for Multi-modal Federated Learning via Orthogonal Transform Watermark
Keke Gai (Beijing Institute of Technology), Qi Wu (Australian Institute of Machine Learning)
Federated LearningSafty and PrivacyImageTextMultimodality
🎯 What it does: The MFL-Owner framework is proposed, which uses orthogonal transformations to embed watermarks, protecting the ownership of multimodal federated learning models and enabling traceability.
MGDA: Model-based Goal Data Augmentation for Offline Goal-conditioned Weighted Supervised Learning
Xing Lei (Xi'an Jiaotong University), Donglin Wang (Westlake University)
Reinforcement LearningSequential
🎯 What it does: To address the lack of trajectory stitching capability in the GCWSL method for offline goal-conditioned reinforcement learning, this paper proposes a model-based goal data augmentation method called MGDA.
MHBench: Demystifying Motion Hallucination in VideoLLMs
Ming Kong (Zhejiang University), Qiang Zhu (Zhejiang University)
ClassificationRecognitionTransformerVision Language ModelContrastive LearningVideoBenchmark
🎯 What it does: This paper first introduces the concept of 'Motion Hallucination' in video LLMs and systematically evaluates it by constructing the MHBench benchmark.
MI-CAPTCHA: Enhance the Security of CAPTCHA Using Mooney Images
Jingmeng Li (Fudan University), Hui Wei (Fudan University)
RecognitionObject DetectionSegmentationSafty and PrivacyConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A framework named HiMI was designed and implemented to generate high-quality Mooney images (MI) from natural images, with adjustable parameters to control their perceived difficulty. Based on these MIs, two new CAPTCHA schemes (based on object detection and instance segmentation) were designed and their security was validated through experiments.
Micro-macro Wavelet-based Gaussian Splatting for 3D Reconstruction from Unconstrained Images
Yihui Li (Beihang University), Di Huang (Beihang University)
RestorationDepth EstimationGaussian SplattingImage
🎯 What it does: Achieving high-quality 3D reconstruction from unconstrained image collections using Gaussian spot technology.
MIDI-GPT: A Controllable Generative Model for Computer-Assisted Multitrack Music Composition
Philippe Pasquier (Simon Fraser University), Maryam Safi (Steinberg Media Technologies GmbH)
GenerationTransformerLarge Language ModelAudio
🎯 What it does: Designed and released a controllable multi-track MIDI generation model based on Transformer, MIDI-GPT, for computer-assisted composition, supporting track/bar filling and attribute control.
MIETT: Multi-Instance Encrypted Traffic Transformer for Encrypted Traffic Classification
Xu-Yang Chen (Nanjing University), Han-Jia Ye (Nanjing University)
ClassificationSafty and PrivacyTransformerContrastive LearningTime Series
🎯 What it does: A Multi-Instance Encrypted Traffic Transformer (MIETT) is proposed for encrypted network traffic classification, combining a two-level attention mechanism and specialized pre-training tasks.
MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity
Kanghyun Choi (Seoul National University), Jinho Lee (Seoul National University)
ClassificationObject DetectionSegmentationData SynthesisKnowledge DistillationTransformerImage
🎯 What it does: MimiQ is proposed, a data-free quantization method for Vision Transformers that generates synthetic samples and performs distillation by aligning multi-head attention similarities to achieve low-bit quantization.
MIMTrack: In-Context Tracking via Masked Image Modeling
Xingmei Wang (Harbin Engineering University), Zining Yan (National University of Singapore)
Object TrackingGenerationTransformerImageVideo
🎯 What it does: A single-stream generative tracker MIMTrack based on Masked Image Modeling (MIM) and context learning is designed, transforming object tracking into a pixel-level image generation task.
Mind Individual Information! Principal Graph Learning for Multimedia Recommendation
Penghang Yu (Nanjing University of Posts and Telecommunications), Bing-Kun Bao (Nanjing University of Posts and Telecommunications)
Recommendation SystemGraph Neural NetworkContrastive LearningMultimodalityGraph
🎯 What it does: This paper proposes a primary graph learning framework (PGL) that extracts primary subgraphs from the user-item interaction graph and performs message passing on them to better utilize local structural information to enhance multimodal recommendation performance.
Mind the Uncertainty in Human Disagreement: Evaluating Discrepancies Between Model Predictions and Human Responses in VQA
Jian Lan (Ludwig Maximilian University of Munich), Barbara Plank (Ludwig Maximilian University of Munich)
ClassificationRecognitionTransformerVision Language ModelImageMultimodality
🎯 What it does: This paper studies human uncertainty in answers (HUD) in visual question answering (VQA), evaluates the differences between model prediction distributions and human response distributions, and explores whether temperature scaling calibration can improve alignment with humans.
MindPainter: Efficient Brain-Conditioned Painting of Natural Images via Cross-Modal Self-Supervised Learning
Muzhou Yu (Xi'an Jiaotong University), Kaisheng Ma (Tsinghua University)
Image TranslationRestorationGenerationTransformerDiffusion modelContrastive LearningImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: Developed MindPainter, a cross-modal self-supervised learning framework that directly edits natural images using visual fMRI brain signals;
MindTuner: Cross-Subject Visual Decoding with Visual Fingerprint and Semantic Correction
Zixuan Gong (Tongji University), Duoqian Miao (Tongji University)
GenerationRetrievalDiffusion modelContrastive LearningImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: MindTuner is proposed, a cross-subject visual decoding framework that utilizes visual fingerprints and text assistance to encode fMRI, ultimately achieving high-quality image reconstruction.
Minimal Change in Modal Logic S5
Carlos Aguilera-Ventura (National Centre for Scientific Research), Andreas Herzig (National Centre for Scientific Research)
🎯 What it does: This paper extends the AGM framework of belief revision to modal logic S5 and proposes three new modal revision axioms (M1–M3). It also defines a class of pseudo-distances between sets and three axioms (A3–A5) to ensure that the revision operation meets the requirement of minimal change.
Mining In-distribution Attributes in Outliers for Out-of-distribution Detection
Yutian Lei (University of Electronic Science and Technology of China), Pei Liu (University of Electronic Science and Technology of China)
Anomaly DetectionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a multi-view learning framework MVOL, which utilizes the few in-distribution (ID) features contained in auxiliary outlier samples during training to improve OOD detection performance.
Mitigating Feature Gap for Adversarial Robustness by Feature Disentanglement
Nuoyan Zhou (Xidian University), Xinbo Gao (Chongqing University of Posts and Telecommunications)
Representation LearningAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Adversarial fine-tuning of pre-trained models is performed, using feature decoupling to eliminate confused latent features that lead to feature gaps, thereby enhancing adversarial robustness.
Mitigating Hallucinations in Large Vision-Language Models by Adaptively Constraining Information Flow
Jiaqi Bai (Guangzhou University), Zhihong Tian (Guangzhou University)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: The ADAVIB method is proposed, which injects noise into the visual-language projector of LVLM using the Variational Information Bottleneck (VIB) and adaptively adjusts the noise intensity through entropy to suppress the overconfidence of visual tokens in the LLM embedding space, thereby reducing object hallucinations in generated text.
Mitigating Pervasive Modality Absence Through Multimodal Generalization and Refinement
Wuliang Huang (Institute of Computing Technology, Chinese Academy of Sciences), Yifan Wang (Tsinghua University)
ClassificationRecognitionGraph Neural NetworkTransformerPrompt EngineeringImageTextMultimodalityAudio
🎯 What it does: A multi-modal generalization and refinement (MGR) framework is proposed to address the bias and performance degradation issues caused by the common absence of modalities during training and inference stages.