AAAI 2023 Papers — Page 10
AAAI Conference on Artificial Intelligence · 1578 papers
MetaZSCIL: A Meta-Learning Approach for Generalized Zero-Shot Class Incremental Learning
Yanan Wu (Beijing Jiaotong University), Yang Wang (Concordia University)
Meta LearningGenerative Adversarial NetworkImage
🎯 What it does: This paper studies a generalized zero-shot class incremental learning method for scenarios without training samples.
Metric Multi-View Graph Clustering
Yuze Tan (Sichuan University), Shudong Huang (Sichuan University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: A unified multi-view clustering framework that combines smooth representation learning, linear metric learning, and multi-view graph fusion is proposed.
Metric Nearness Made Practical
Wenye Li (Chinese University of Hong Kong), Zichen Ma (Chinese University of Hong Kong)
OptimizationComputational EfficiencyTabular
🎯 What it does: This paper proposes a two-stage method for solving the metric approximation problem of noisy distance matrices, first quickly obtaining an approximate solution within a subset of isometric embeddings, and then refining it to a global optimum through alternating projections.
Metric Residual Network for Sample Efficient Goal-Conditioned Reinforcement Learning
Bo Liu (University of Texas at Austin), Peter Stone (University of Texas at Austin)
Reinforcement Learning
🎯 What it does: This paper proposes the Metric Residual Network (MRN), a neural network architecture based on the triangle inequality, aimed at improving sample efficiency in goal-conditioned reinforcement learning.
MGFN: Magnitude-Contrastive Glance-and-Focus Network for Weakly-Supervised Video Anomaly Detection
Yingxian Chen (University of Hong Kong), Yik-Chung Wu (University of Hong Kong)
Anomaly DetectionTransformerContrastive LearningVideo
🎯 What it does: This paper proposes a weakly supervised video anomaly detection framework based on visual Transformers—MGFN, which utilizes a two-stage processing of global 'glance' and local 'focus', combined with feature magnitude amplification and contrast loss, to achieve efficient localization of anomalous frames.
MGTANet: Encoding Sequential LiDAR Points Using Long Short-Term Motion-Guided Temporal Attention for 3D Object Detection
Junho Koh (Hanyang University), Jun Won Choi (Hanyang University)
Object DetectionAutonomous DrivingConvolutional Neural NetworkPoint CloudSequential
🎯 What it does: This paper proposes a new 3D object detection network called MGTANet, which can encode continuous LiDAR point cloud sequences and utilize temporal information to improve detection accuracy.
MGTCF: Multi-Generator Tropical Cyclone Forecasting with Heterogeneous Meteorological Data
Cheng Huang (Zhejiang University of Technology), YuQuan Wu (Chinese Academy of Sciences)
GenerationData SynthesisRecurrent Neural NetworkGenerative Adversarial NetworkMultimodalityTime Series
🎯 What it does: A multi-generator multi-modal tropical cyclone prediction framework (MGTCF) is proposed, utilizing heterogeneous meteorological data and environmental information to predict trajectory and intensity.
MHCCL: Masked Hierarchical Cluster-Wise Contrastive Learning for Multivariate Time Series
Qianwen Meng (Shandong University), Zhiqi Shen (Nanyang Technological University)
Anomaly DetectionRepresentation LearningContrastive LearningTime Series
🎯 What it does: A mask-based hierarchical clustering contrastive learning framework (MHCCL) is proposed, which enhances unsupervised time series representation learning by removing outliers with an upward mask and filtering positive and negative samples with a downward mask.
MicroAST: Towards Super-fast Ultra-Resolution Arbitrary Style Transfer
Zhizhong Wang (Zhejiang University), Dongming Lu (Zhejiang University)
Image TranslationSuper ResolutionComputational EfficiencyConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: MicroAST is proposed, a lightweight and ultra-fast super-resolution arbitrary style transfer model that can complete style transfer on 4K images in about 0.5 seconds;
MIDMs: Matching Interleaved Diffusion Models for Exemplar-Based Image Translation
Junyoung Seo (Korea University), Seungryong Kim (Korea University)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: A Matching Interleaved Diffusion Model (MIDMs) is proposed, which alternately performs cross-domain matching and diffusion generation in the latent space to achieve more reliable example-driven image translation.
MIGA: A Unified Multi-Task Generation Framework for Conversational Text-to-SQL
Yingwen Fu (Guangdong University of Foreign Studies), Yue Lin (Columbia University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: The MIGA framework is proposed, utilizing multi-task pre-training and SQL perturbation to enhance the performance of generative pre-trained language models in conversational Text-to-SQL tasks.
MIMO Is All You Need：A Strong Multi-in-Multi-Out Baseline for Video Prediction
Shuliang Ning (Chinese University of Hong Kong, Shenzhen), Shuguang Cui (Chinese University of Hong Kong, Shenzhen)
GenerationData SynthesisAnomaly DetectionTransformerVideo
🎯 What it does: A multi-input multi-output (MIMO) video prediction framework called MIMO-VP based on Transformer is proposed to address the error accumulation problem of traditional single-input single-output models.
Min-Max Submodular Ranking for Multiple Agents
Qingyun Chen (University of California), Ruilong Zhang (City University of Hong Kong)
OptimizationBiomedical Data
🎯 What it does: This paper proposes and analyzes the minimum maximum coverage problem for multi-agent submodular ranking, providing an approximate algorithm and validating its effectiveness through experiments.
Mind the Gap: Polishing Pseudo Labels for Accurate Semi-supervised Object Detection
Lei Zhang (Northwestern Polytechnical University), Wei Wei (Northwestern Polytechnical University)
Object DetectionImage
🎯 What it does: This paper proposes a dual pseudo-label refinement framework that utilizes two structurally different refinement networks (classification refinement network and box regression refinement network) to correct the pseudo-labels generated by the teacher detector, achieving end-to-end joint training within a semi-supervised object detection framework.
Minimax AUC Fairness: Efficient Algorithm with Provable Convergence
Zhenhuan Yang (Etsy), Yiming Ying (University at Albany)
OptimizationTabular
🎯 What it does: A framework for minimizing and maximizing AUC fairness by considering both intra-group and inter-group AUC in AUC optimization is proposed, aiming to eliminate the model's differential ranking bias towards different sensitive attribute groups.
Mining and Applying Composition Knowledge of Dance Moves for Style-Concentrated Dance Generation
Xinjian Zhang (Fudan University), Longwen Gao (China University of Petroleum)
GenerationRecurrent Neural NetworkTransformerAuto EncoderContrastive LearningVideo
🎯 What it does: A music-to-dance generation framework based on style embeddings is proposed, which achieves precise control and diverse generation of dance styles by learning dance prototypes and forming controllable style embeddings through linear combinations.
Minority-Oriented Vicinity Expansion with Attentive Aggregation for Video Long-Tailed Recognition
WonJun Moon (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
RecognitionTransformerVideo
🎯 What it does: This paper proposes two core technologies for long-tail video recognition (VLTR): a learnable feature aggregator and minority-oriented neighborhood expansion (MOVE).
Mitigating Adversarial Norm Training with Moral Axioms
Taylor Olson (Northwestern University), Kenneth D. Forbus (Northwestern University)
Adversarial AttackTabular
🎯 What it does: This paper constructs an 'intuition and construction' method to resist adversarial training for ethical AI by incorporating moral axioms and Dempster-Shafer logical reasoning into a normative learning framework.
Mitigating Artifacts in Real-World Video Super-resolution Models
Liangbin Xie (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Ying Shan (Tencent)
RestorationSuper ResolutionRecurrent Neural NetworkGenerative Adversarial NetworkOptical FlowVideo
🎯 What it does: Proposes the Hidden State Attention (HSA) module, which directly corrects the hidden states in the recursive video super-resolution model to reduce artifacts in real scenes.
Mitigating Negative Style Transfer in Hybrid Dialogue System
Shimin Li (Fudan University), Xipeng Qiu (Fudan University)
GenerationDomain AdaptationTransformerPrompt EngineeringContrastive LearningText
🎯 What it does: A latent variable encoding-decoding model HiS-Dialog based on contrastive learning is designed and implemented to decouple and control the dialogue style in mixed dialogue systems.
Mixed-Variable Black-Box Optimisation Using Value Proposal Trees
Yan Zuo (Amazon), Iadine Chades (CSIRO)
OptimizationTabular
🎯 What it does: A hybrid variable black-box optimization method called Value Proposal Trees (VPT) is proposed, which generates 'value proposals' through tree regression and jointly optimizes categorical and continuous variables from a global perspective.
Mixture Manifold Networks: A Computationally Efficient Baseline for Inverse Modeling
Gregory P. Spell (Duke University), Jordan M. Malof (University of Montana)
OptimizationComputational EfficiencyTabularBenchmark
🎯 What it does: This paper proposes the Mixture Manifold Network (MMN) to solve inverse problems by mixing multiple backward models with a single forward model, using the forward model to generate data for training, thereby improving the accuracy and speed of inverse modeling.
Mixture Uniform Distribution Modeling and Asymmetric Mix Distillation for Class Incremental Learning
Sunyuan Qiang (Macau University of Science and Technology), Du Zhang (National Laboratory of Pattern Recognition)
Knowledge DistillationImage
🎯 What it does: Proposes a reverse mixed distillation method to alleviate the distribution mismatch problem in class-incremental learning through modeling with mixed uniform distribution and incremental sampling.
MMTN: Multi-Modal Memory Transformer Network for Image-Report Consistent Medical Report Generation
Yiming Cao (Shandong University), Yonghui Xu (Shandong University)
GenerationTransformerImageTextMultimodality
🎯 What it does: A multi-modal memory Transformer network (MMTN) is designed to simultaneously process medical images, terminology knowledge, and text reports, generating image-report consistent medical reports.
MNER-QG: An End-to-End MRC Framework for Multimodal Named Entity Recognition with Query Grounding
Meihuizi Jia (Beijing Institute of Technology), Jiaqi Li (Beijing Institute of Technology)
RecognitionObject DetectionTransformerVision Language ModelTextMultimodality
🎯 What it does: This paper proposes an end-to-end MRC framework MNER-QG for multimodal named entity recognition and query localization.
MobileTL: On-Device Transfer Learning with Inverted Residual Blocks
Hung-Yueh Chiang (University of Texas), Diana Marculescu (University of Texas)
ClassificationComputational EfficiencyImage
🎯 What it does: An efficient transfer learning method called MobileTL is proposed for edge devices, achieving low memory and low computational training for models based on Inverted Residual Block (IRB);
Model-Based Offline Reinforcement Learning with Local Misspecification
Kefan Dong (Stanford University), Emma Brunskill (Stanford University)
Reinforcement Learning
🎯 What it does: A model-based algorithm for offline reinforcement learning is proposed, which provides a lower bound on policy value through lazy estimation of local model errors for state-action pairs, and achieves safe policy improvement based on this.
Model-Based Reinforcement Learning with Multinomial Logistic Function Approximation
Taehyun Hwang (Seoul National University), Min-hwan Oh (Seoul National University)
Reinforcement LearningTabular
🎯 What it does: This paper studies model-based reinforcement learning and proposes the UCRL-MNL algorithm for approximating the MDP transition model using multinomial logistic regression functions, providing a regret upper bound that matches the lower bound.
Model-Checking for Ability-Based Logics with Constrained Plans
Stéphane Demri (Université Paris-Saclay), Raul Fervari (Universidad Nacional de Córdoba)
🎯 What it does: This paper studies the model checking problem of the logic of ability knowledge (knowing-how) and its variants with multi-agent, regular constraints, and budget constraints, providing corresponding complexity analysis.
Modeling Human Trust and Reliance in AI-Assisted Decision Making: A Markovian Approach
Zhuoyan Li (Purdue University), Ming Yin (Purdue University)
Recommendation SystemRecurrent Neural NetworkSupervised Fine-TuningTabularSequentialFinance Related
🎯 What it does: A framework for emotional processes based on hidden Markov models is proposed to characterize human trust and reliance behavior in AI-assisted decision-making, and the model is trained and inferred in experiments.
Models as Agents: Optimizing Multi-Step Predictions of Interactive Local Models in Model-Based Multi-Agent Reinforcement Learning
Zifan Wu (Sun Yat-sen University), Hankz Hankui Zhuo (Huawei Noah's Ark Lab)
OptimizationReinforcement LearningAgentic AISequential
🎯 What it does: In model-based multi-agent reinforcement learning, the MAG framework is proposed, treating local models as multi-step decision-makers and combining the current policy as the environment to optimize multi-step prediction errors, thereby reducing error propagation and improving model quality;
MoEC: Mixture of Expert Clusters
Yuan Xie (Microsoft Research), Furu Wei (Microsoft Research)
TransformerMixture of ExpertsText
🎯 What it does: The Mixture of Expert Clusters (MoEC) method is proposed, which addresses the issues of sparse data allocation and overfitting in large-scale MoE models by clustering experts and introducing variance constraints and cluster-level expert dropout.
Molformer: Motif-Based Transformer on 3D Heterogeneous Molecular Graphs
Fang Wu (Westlake University), Stan Z. Li (Westlake University)
Drug DiscoveryGraph Neural NetworkTransformerReinforcement LearningGraphBiomedical Data
🎯 What it does: A Motif-based Transformer model called Molformer is proposed, which represents molecules using heterogeneous molecular graphs (including atom layers and Motif layers) and learns their 3D geometric information.
Monitoring Arithmetic Temporal Properties on Finite Traces
Paolo Felli (University of Bologna), Sarah Winkler (Sapienza University of Rome)
🎯 What it does: A monitoring framework for finite-time linear temporal properties with arithmetic constraints (ALTL_f) is proposed, along with a theoretical foundation for determining monitorability.
Moral Machine or Tyranny of the Majority?
Michael Feffer (Carnegie Mellon University), Zachary C. Lipton (Carnegie Mellon University)
🎯 What it does: This paper explores the issues of fairness and strategic behavior when using the average preference vector aggregation mechanism in two groups (majority and minority) through theoretical analysis, revealing that the interests of minority groups are often underestimated, and proving that the majority group can completely monopolize decision-making in the presence of pure Nash equilibrium. It further provides the necessary and sufficient conditions for the existence of pure Nash equilibrium and its specific form, and compares it conceptually with other aggregation mechanisms such as the median and random dictatorship.
Moving-Landmark Assisted Distributed Learning Based Decentralized Cooperative Localization (DL-DCL) with Fault Tolerance
Shubhankar Gupta (Indian Institute of Science), Suresh Sundaram (Indian Institute of Science)
OptimizationRobotic IntelligenceSimultaneous Localization and Mapping
🎯 What it does: This study addresses the problem of collaborative localization of multiple robots in uncertain environments and proposes a decentralized collaborative localization (DL-DCL) algorithm based on distributed learning. It utilizes maneuverable milestone information to adaptively learn and fuse weights, achieving fault tolerance against sensor failures and dynamic biases. High-fidelity simulation validation was conducted on the Gazebo+ROS2 platform.
MPMQA: Multimodal Question Answering on Product Manuals
Liang Zhang (Renmin University of China), Qin Jin (Renmin University of China)
GenerationRetrievalTransformerTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Proposes the Multi-modal Product Manual Question Answering (MPMQA) task and constructs a large multi-modal dataset PM209;
MRCN: A Novel Modality Restitution and Compensation Network for Visible-Infrared Person Re-identification
Yukang Zhang (Xiamen University), Hanzi Wang (Xiamen University)
RecognitionRetrievalContrastive LearningImageMultimodality
🎯 What it does: A Modality Restitution and Compensation Network (MRCN) for cross-modal recognition of visible light and infrared portraits is proposed.
MSDC: Exploiting Multi-State Power Consumption in Non-intrusive Load Monitoring Based on a Dual-CNN Model
Jialing He (Chongqing University), Liehuang Zhu (Beijing Institute of Technology)
Convolutional Neural NetworkTime Series
🎯 What it does: This paper proposes a multi-state dual CNN model (MSDC) for non-intrusive load monitoring tasks, which decomposes aggregated power signals into the power consumption of individual devices.
MulGT: Multi-Task Graph-Transformer with Task-Aware Knowledge Injection and Domain Knowledge-Driven Pooling for Whole Slide Image Analysis
Weiqin Zhao (University of Hong Kong), Lequan Yu (University of Hong Kong)
ClassificationGraph Neural NetworkTransformerImageBiomedical Data
🎯 What it does: This paper proposes and implements MulGT, a multi-task graph-Transformer framework for simultaneously performing tumor typing and staging diagnostic tasks on whole slide images (WSI).
Multi-Action Dialog Policy Learning from Logged User Feedback
Shuo Zhang (Xi'an Jiaotong University), Junlan Feng (China Mobile Research)
Recommendation SystemReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningText
🎯 What it does: A multi-action dialogue strategy learning framework based on BanditMatch is proposed, which enhances multi-action dialogue strategies by utilizing bandit feedback from historical dialogues.
Multi-Aspect Explainable Inductive Relation Prediction by Sentence Transformer
Zhixiang Su (Nanyang Technological University), Lizhen Cui (Shandong University)
Explainability and InterpretabilityKnowledge DistillationTransformerGraph
🎯 What it does: This paper proposes a path reasoning model KRST based on sentence Transformers for interpretable incremental relation prediction in knowledge graphs.
Multi-Classifier Adversarial Optimization for Active Learning
Lin Geng (Nanjing University of Aeronautics and Astronautics), Jie Qin (Nanjing University of Aeronautics and Astronautics)
ClassificationObject DetectionOptimizationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A multi-classifier adversarial optimization active learning framework (MAOAL) is proposed, which aligns the distribution of labeled and unlabeled samples through adversarial training between a main classifier and two auxiliary classifiers, and selects the most informative samples for labeling.
Multi-Domain Generalized Graph Meta Learning
Mingkai Lin (Nanjing University), Sanglu Lu (Nanjing University)
Domain AdaptationMeta LearningGraph Neural NetworkGraph
🎯 What it does: The MD-Gram framework is proposed to address the issues of inconsistent domains, feature space discrepancies, and distribution heterogeneity in multi-domain graph learning.
Multi-Label Few-Shot ICD Coding as Autoregressive Generation with Prompt
Zhichao Yang (University of Massachusetts), Hong Yu (University of Massachusetts)
ClassificationGenerationTransformerSupervised Fine-TuningPrompt EngineeringTextBiomedical DataElectronic Health Records
🎯 What it does: For the ICD coding task, the authors propose a framework based on autoregressive generation and prompts: first, pre-training is conducted to generate assessment and plan texts using SOAP-structured clinical records, and then the model is guided by prompts to sequentially generate disease descriptions and map them back to ICD codes, achieving multi-label classification, especially with significant results in low-sample scenarios.
Multi-Level Compositional Reasoning for Interactive Instruction Following
Suvaansh Bhambri (Yonsei University), Jonghyun Choi (Yonsei University)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningAgentic AIMultimodalityBenchmark
🎯 What it does: A multi-level combinatorial reasoning robot, MCR-Agent, is proposed, which decomposes instructions into observable sub-goals and learns navigation and interaction sub-policies separately.
Multi-Level Confidence Learning for Trustworthy Multimodal Classification
Xiao Zheng (National University of Defense Technology), Wei Zhang (Shandong Computer Science Center)
ClassificationGraph Neural NetworkMultimodalityBiomedical DataAlzheimer's Disease
🎯 What it does: A trustworthy multimodal classification network MLCLNet is proposed, which enhances the credibility of features, labels, and decisions through multi-layer confidence learning.
Multi-Level Wavelet Mapping Correlation for Statistical Dependence Measurement: Methodology and Performance
Yixin Ren (Fudan University), Shuigeng Zhou (Fudan University)
Tabular
🎯 What it does: Multi-layer Wavelet Mapping Correlation (MWMC) is proposed as a new metric for measuring the nonlinear dependence between two continuous variables, and a permutation test method is designed based on this metric for independence testing.
Multi-Mask Label Mapping for Prompt-Based Learning
Jirui Qi (Beihang University), Yongyi Mao (University of Ottawa)
ClassificationTransformerPrompt EngineeringText
🎯 What it does: A Multi-Mask Label Mapping (MMLM) framework is proposed for few-shot classification in prompt-based learning.
Multi-Modal Knowledge Hypergraph for Diverse Image Retrieval
Yawen Zeng (ByteDance), Wenfeng Li (ByteDance)
RetrievalGraph Neural NetworkContrastive LearningImageMultimodality
🎯 What it does: A multi-modal knowledge hypergraph (MKHG) has been designed and implemented, achieving diversified image retrieval by explicitly extending sub-semantics in keyword search.
Multi-Modality Deep Network for Extreme Learned Image Compression
Xuhao Jiang (Fudan University), Liquan Shen (Shanghai University)
CompressionAuto EncoderGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: A text-guided multimodal image compression framework TGIC is proposed, which utilizes textual semantic information to enhance image compression quality at extremely low bit rates.
Multi-Relational Contrastive Learning Graph Neural Network for Drug-Drug Interaction Event Prediction
Zhankun Xiong (Huazhong Agricultural University), Wen Zhang (Binghamton University)
Drug DiscoveryGraph Neural NetworkContrastive LearningGraphBiomedical Data
🎯 What it does: This paper proposes a Multi-Relation Contrastive Learning Graph Neural Network (MRCGNN) for predicting drug-drug interaction events (DDI events).
Multi-Resolution Monocular Depth Map Fusion by Self-Supervised Gradient-Based Composition
Yaqiao Dai (National University of Defense Technology), Kai Xu (National University of Defense Technology)
Depth EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A multi-resolution gradient domain self-supervised fusion module is proposed to enhance the detail and global accuracy of monocular depth estimation.
Multi-Scale Control Signal-Aware Transformer for Motion Synthesis without Phase
Lintao Wang (Shanghai AI Laboratory), Zhiyong Wang (School of Computer Science, The University of Sydney)
GenerationData SynthesisRobotic IntelligenceGraph Neural NetworkTransformerTime SeriesSequential
🎯 What it does: A multi-scale control signal perception transformer (MCS-T) is proposed to achieve real-time controllable action generation without the need for phase assistance.
Multi-Source Survival Domain Adaptation
Ammar Shaker (NEC Laboratories Europe), Carolin Lawrence (NEC Laboratories Europe)
Domain AdaptationBiomedical Data
🎯 What it does: A method for multi-source survival domain adaptation is proposed, aimed at effectively adapting from multiple survival source domains to a new survival target domain, especially in cases of data scarcity and partial information censorship.
Multi-Stage Facility Location Problems with Transient Agents
Xuezhen Wang (Lingnan University), Minming Li (City University of Hong Kong)
Optimization
🎯 What it does: This paper studies the one-dimensional multi-stage facility location problem, considering temporary agents (who can stay for a limited number of stages after arrival) and the migration costs of facilities at different stages; it proposes optimal algorithms and a stochastic mechanism based on population strategy invariance under two models: one with no migration costs and one with migration costs.
Multi-Stream Representation Learning for Pedestrian Trajectory Prediction
Yuxuan Wu (Xi'an Jiaotong University), Wei Tang (University of Illinois)
Representation LearningTransformerGenerative Adversarial NetworkMultimodality
🎯 What it does: Designed and implemented a multi-stream representation learning module to learn the spatiotemporal and interaction features of pedestrian trajectories under the CVAE framework, achieving multimodal prediction.
Multi-Unit Auctions for Allocating Chance-Constrained Resources
Anna Gautier (University of Oxford), Michael Wooldridge (University of Oxford)
OptimizationReinforcement LearningTabular
🎯 What it does: A bidding-based 'ACCR' mechanism is proposed for the pre-allocation of multiple units of resources in multi-agent systems with uncertain resource usage, ensuring that the probability of violating resource constraints does not exceed a preset threshold by setting confidence constraints.
Multi-View Domain Adaptive Object Detection on Camera Networks
Yan Lu (New York University), Yuanchao Shu (Zhejiang University)
Object DetectionDomain AdaptationVideo
🎯 What it does: This paper studies a domain adaptation object detection method in multi-camera networks with passive domain data and only target domain multi-view videos. A two-stage training framework is proposed: the first stage pre-trains the backbone network through multi-view association and self-supervised learning; the second stage employs tracking-based view augmentation and weak-hard consistency learning for robust self-training.
Multi-View MOOC Quality Evaluation via Information-Aware Graph Representation Learning
Lu Jiang (Northeast Normal University), Minghao Yin (Northeast Normal University)
Recommendation SystemRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A multi-view graph representation learning framework based on information perception (IaGRL) is proposed for MOOC course quality assessment.
MultiAct: Long-Term 3D Human Motion Generation from Multiple Action Labels
Taeryung Lee (Seoul National University), Kyoung Mu Lee (Seoul National University)
GenerationPose EstimationTransformerVideo
🎯 What it does: Proposes the MultiAct framework, which achieves long-term 3D human motion generation through recursive generation of multi-action labels;
Multiagent MST Cover: Pleasing All Optimally via a Simple Voting Rule
Bo Li (Hong Kong Polytechnic University), Ruilong Zhang (City University of Hong Kong)
OptimizationGraph
🎯 What it does: This paper proposes and studies the Multi-Agent Minimum Spanning Tree Covering (MMCP) problem, which involves constructing the smallest subgraph in a graph where different agents evaluate edge weights, such that each agent can find a minimum spanning tree within that subgraph.
Multiple Robust Learning for Recommendation
Haoxuan Li (Peking University), Peng Wu (Beijing Technology and Business University)
Recommendation System
🎯 What it does: A multiple robust learning method (MR) is proposed for recommendation systems to address bias issues in the data, thereby improving the generalization ability of the recommendation model.
Multiplex Graph Representation Learning via Common and Private Information Mining
Yujie Mo (University of Electronic Science and Technology of China), Xiaofeng Zhu (University of Electronic Science and Technology of China)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: A self-supervised multi-layer graph representation learning framework called CPIM is proposed, which jointly mines the common and private information of multi-layer graphs while minimizing redundancy in node representations.
Multispectral Invisible Coating: Laminated Visible-Thermal Physical Attack against Multispectral Object Detectors Using Transparent Low-E Films
Taeheon Kim (Korea Advanced Institute of Science and Technology), Yong Man Ro (Korea Advanced Institute of Science and Technology)
Object DetectionAdversarial AttackImageBenchmark
🎯 What it does: This paper proposes an attachable, lightweight visible and thermal imaging hybrid countermeasure patch utilizing a transparent Low-e low emissivity film, creating a multispectral cloaking garment capable of rendering humans 'invisible' from multiple angles and distances, and releases a corresponding visible-thermal dual-mode pedestrian dataset.
MultiSpider: Towards Benchmarking Multilingual Text-to-SQL Semantic Parsing
Longxu Dou (Harbin Institute of Technology), Jian-Guang Lou (Microsoft Research Asia)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: A high-quality text-to-SQL semantic parsing dataset covering seven languages, MULTISPIDER, has been constructed, and a schema-based augmentation framework, SAVE, has been proposed to enhance the performance of cross-language text-to-SQL.
Multiwinner Voting with Possibly Unavailable Candidates
Markus Brill (University of Warwick), Ulrike Schmidt-Kraepelin (Technische Universitat Berlin)
Optimization
🎯 What it does: In multi-winner elections, the introduction of scenarios where candidates may be unavailable leads to the proposal of a safe query strategy to ensure the optimality and proportional representation of the final committee.
Mutual-Enhanced Incongruity Learning Network for Multi-Modal Sarcasm Detection
Yang Qiao (Shandong University), Liqiang Nie (Shandong University)
ClassificationRecognitionObject DetectionGraph Neural NetworkTransformerLarge Language ModelImageTextMultimodality
🎯 What it does: This paper proposes MILNet, which utilizes a complementary local semantic guidance and global inconsistency learning mechanism, combining graph convolution and attention models to achieve multimodal sarcasm detection.
MVCINN: Multi-View Diabetic Retinopathy Detection Using a Deep Cross-Interaction Neural Network
Xiaoling Luo (Harbin Institute of Technology), Yong Xu (Harbin Institute of Technology)
ClassificationRecognitionConvolutional Neural NetworkTransformerImage
🎯 What it does: A framework for detecting diabetic retinopathy based on multi-view retinal images is proposed, which utilizes a cross-interaction self-attention module to fuse convolutional and Transformer features, and achieves inter-view information interaction through multi-view stitching.
Mx2M: Masked Cross-Modality Modeling in Domain Adaptation for 3D Semantic Segmentation
Boxiang Zhang (Jilin University), Wenhui Li (Jilin University)
SegmentationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkAuto EncoderMultimodalityPoint Cloud
🎯 What it does: The Mx2M method is proposed to enhance the cross-domain adaptation performance of 3D semantic segmentation through masked cross-modal modeling.
NAS-LID: Efficient Neural Architecture Search with Local Intrinsic Dimension
Xin He (Hong Kong Baptist University), Xiaowen Chu (Hong Kong Baptist University)
Neural Architecture SearchImage
🎯 What it does: A lightweight single network architecture search method called NAS-LID is proposed, which utilizes the local intrinsic dimension (LID) for subnet segmentation, thereby reducing interference between different sub-networks and enhancing performance ranking correlation.
NeAF: Learning Neural Angle Fields for Point Normal Estimation
Shujuan Li (Tsinghua University), Zhizhong Han (Wayne State University)
OptimizationPoint Cloud
🎯 What it does: A point cloud normal estimation method based on Neural Implicit Angle Field (NeAF) is proposed, which learns angular deviations using random query vectors and refines the predicted normal vectors through gradient optimization during inference.
Nearest-Neighbor Sampling Based Conditional Independence Testing
Shuai Li (East China Normal University), Wang Wen (New York University Shanghai)
Tabular
🎯 What it does: A conditional independence test method based on 1-nearest neighbor sampling (NNSCIT) is proposed, which can be implemented without knowing the conditional distribution.
Neighbor Contrastive Learning on Learnable Graph Augmentation
Xiao Shen (Hainan University), Laurence T. Yang (Hainan University)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes an end-to-end unsupervised graph contrastive learning framework named NCLA, which can automatically learn graph augmentations and node representations.
Neighborhood-Regularized Self-Training for Learning with Few Labels
Ran Xu (Emory University), Carl Yang (Emory University)
ClassificationOptimizationGraph Neural NetworkTransformerLarge Language ModelTextGraph
🎯 What it does: The NeST method is proposed, which reduces pseudo-label noise and enhances training stability in semi-supervised self-training through neighborhood regularization for sample selection and multi-round prediction aggregation.
Nested Named Entity Recognition as Building Local Hypergraphs
Yukun Yan (Tsinghua University), Sen Song (Tsinghua University)
RecognitionRecurrent Neural NetworkGraph Neural NetworkTextBiomedical Data
🎯 What it does: Proposes the Local Hypergraph Builder Network (LHBN), which achieves nested named entity recognition by first predicting entity boundaries and then constructing local hypergraphs for each boundary.
Networked Anti-coordination Games Meet Graphical Dynamical Systems: Equilibria and Convergence
Zirou Qiu (University of Virginia), Anil Vullikanti (University of Virginia)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper studies the existence and search of pure Nash equilibria in networked evolutionary anti-cooperative games, as well as the convergence time of synchronous and sequential updating schemes under self-correlated and self-independent decision-making modes.
Networked Restless Bandits with Positive Externalities
Christine Herlihy (University of Maryland), John P. Dickerson (University of Maryland)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Proposes the Networked Restless Bandits model to study the positive externality problem of resource allocation in community structures;
Neural Architecture Search for Wide Spectrum Adversarial Robustness
Zhi Cheng (University of Sydney), Chang Xu (University of Sydney)
Adversarial AttackNeural Architecture SearchConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: Finding network models (WsrNets) that exhibit good robustness under various attack intensities through integrated neural architecture search.
Neural Diffeomorphic Non-uniform B-spline Flows
Seongmin Hong (Seoul National University), Se Young Chun (Seoul National University)
Flow-based ModelBiomedical DataPhysics Related
🎯 What it does: A reversible and at least second-order continuously differentiable non-uniform B-spline flow is proposed for distribution modeling and efficient sampling of physical systems.
Neural Dynamic Focused Topic Model
Kostadin Cvejoski (Fraunhofer Institute for Intelligent Analysis and Information Systems), César Ojeda (University of Potsdam)
Recurrent Neural NetworkTextTime Series
🎯 What it does: A neural dynamic focusing topic model (NDF-TM) is proposed, which can explicitly decouple topic proportions and topic activity in document sequences, thereby better capturing the time-varying topic structure.
Neural Integro-Differential Equations
Emanuele Zappala (Yale University), David van Dijk (Baylor College of Medicine)
Time SeriesBiomedical DataMagnetic Resonance ImagingStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A framework for neural network-based integral differential equations (NIDE) is proposed to learn non-local continuous dynamics and to fit, extrapolate, and decompose data.
Neural Representations Reveal Distinct Modes of Class Fitting in Residual Convolutional Networks
Michał Jamroż (AGH University of Science and Technology), Marcin Kurdziel (AGH University of Science and Technology)
ClassificationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: By constructing a class-conditional probability distribution model, this study investigates how residual convolutional networks fit feature representations of different categories;
Neural Spline Search for Quantile Probabilistic Modeling
Ruoxi Sun (Google Cloud AI), Tomas Pfister (Google Cloud AI)
Neural Architecture SearchTabularTime Series
🎯 What it does: This paper proposes Neural Spline Search (NSS), a method that automatically searches for and learns expressible quantile functions through symbolic operators and various spline transformations, thereby accurately estimating target distributions in regression and time series forecasting.
Neural TSP Solver with Progressive Distillation
Dongxiang Zhang (Zhejiang University), Gang Chen (Zhejiang University)
OptimizationKnowledge DistillationTransformerReinforcement LearningGraph
🎯 What it does: A progressive distillation framework based on curriculum learning and knowledge distillation is proposed, combining Delaunay graph action masks and a new attention decoder to address the challenges of large-scale TSP training.
Neurosymbolic Reasoning and Learning with Restricted Boltzmann Machines
Son N. Tran (University of Tasmania), Artur d'Avila Garcez (City University of London)
TabularSequentialAlzheimer's DiseaseBenchmark
🎯 What it does: This paper proposes a neural symbolic system called the Logic Boltzmann Machine (LBM), which can transform any propositional logic formula into a Restricted Boltzmann Machine (RBM) and achieve logical reasoning and learning through energy minimization.
Next POI Recommendation with Dynamic Graph and Explicit Dependency
Feiyu Yin (University of Electronic Science and Technology of China), Peng Han (University of Electronic Science and Technology of China)
Recommendation SystemRecurrent Neural NetworkGraph Neural NetworkGraphSequential
🎯 What it does: A sequence-based dynamic neighbor graph and multi-step dependency prediction model (SNPM) is proposed for next location (POI) recommendation.
NHITS: Neural Hierarchical Interpolation for Time Series Forecasting
Cristian Challu (Carnegie Mellon University), Artur Dubrawski (Carnegie Mellon University)
Time Series
🎯 What it does: A new long-period time series forecasting model, NHITS, is proposed, which achieves specialized predictions for different frequency components through multi-rate input sampling and hierarchical interpolation, thereby maintaining low variance and reducing prediction fluctuations over long periods.
NLIP: Noise-Robust Language-Image Pre-training
Runhui Huang (Shenzhen campus of Sun Yat-sen University), Xiaodan Liang (Huawei Noah's Ark Lab)
GenerationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a cross-modal pre-training framework named NLIP, specifically designed to learn from image-text pairs scraped from the web that contain errors and missing information;
Non-IID Transfer Learning on Graphs
Jun Wu (University of Illinois), Elizabeth Ainsworth (University of Illinois)
Domain AdaptationRecommendation SystemGraph Neural NetworkGraphAgriculture Related
🎯 What it does: The paper proposes a cross-network transfer learning framework that utilizes the Graph Subtree Discrepancy metric based on Weisfeiler-Leman subtrees to achieve knowledge transfer from the source graph to the target graph.
Non-reversible Parallel Tempering for Deep Posterior Approximation
Wei Deng (Morgan Stanley), Guang Lin (Purdue University)
ClassificationOptimizationComputational EfficiencyConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: This paper proposes a posterior approximation method that combines windowed irreversible parallel temperature exchange (DEO*) with fixed learning rate SGD, which can maintain low communication costs with a limited number of chains.
Non-stationary Risk-Sensitive Reinforcement Learning: Near-Optimal Dynamic Regret, Adaptive Detection, and Separation Design
Yuhao Ding (University of California Berkeley), Javad Lavaei (University of California Berkeley)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes and analyzes a risk-sensitive reinforcement learning algorithm based on exponential risk measures in non-stationary Markov Decision Processes (MDP). It presents two types of restart-based model-oriented and model-free algorithms, as well as an adaptive algorithm without prior change budget, and provides corresponding upper and lower bounds for dynamic regret.
Normalizing Flow Ensembles for Rich Aleatoric and Epistemic Uncertainty Modeling
Lucas Berry (McGill University), David Meger (McGill University)
OptimizationComputational EfficiencyFlow-based ModelTabularTime Series
🎯 What it does: This paper proposes a Normalizing Flow (NF) ensemble model constructed through a fixed dropout mask, balancing rich aleatoric uncertainty and model epistemic uncertainty, and evaluates its performance within an active learning framework.
Not All Neighbors Matter: Point Distribution-Aware Pruning for 3D Point Cloud
Yejin Lee (Seoul National University), Hongil Yoon (Google)
Autonomous DrivingOptimizationComputational EfficiencyPoint Cloud
🎯 What it does: A neighborhood voxel pruning method based on spatial point distribution is proposed for accelerating and optimizing energy consumption in sparse 3D convolutional networks.
Novel Motion Patterns Matter for Practical Skeleton-Based Action Recognition
Mengyuan Liu (Peking University), Songtao Wu (Sony)
RecognitionPose EstimationGraph Neural NetworkGraphTime Series
🎯 What it does: Proposes the Mask Graph Convolutional Network (Mask-GCN), specifically designed to handle novel motion patterns that appear during the testing phase;
Novel Ordering-Based Approaches for Causal Structure Learning in the Presence of Unobserved Variables
Ehsan Mokhtarian (École Polytechnique Fédérale de Lausanne), Negar Kiyavash (École Polytechnique Fédérale de Lausanne)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper proposes a new removable order (r-order) to learn the structure of the Maximum Ancestral Graph (MAG) containing unobserved variables, and designs three learning algorithms based on this order;
Now We’re Talking: Better Deliberation Groups through Submodular Optimization
Jake Barrett (University of Edinburgh), Ariel D. Procaccia (Harvard University)
OptimizationTabular
🎯 What it does: This paper proposes a framework for topic discussion group allocation based on submodular optimization, aimed at enhancing interaction and diversity in citizen assemblies.
NQE: N-ary Query Embedding for Complex Query Answering over Hyper-Relational Knowledge Graphs
Haoran Luo (Beijing University of Posts and Telecommunications), Kaiyang Wan (Beijing Institute of Computer Technology and Application)
TransformerGraph
🎯 What it does: Proposes the NQE model, which supports complex query answering for arbitrary n-ary first-order logic queries on hyper-relational knowledge graphs;
NuWLS: Improving Local Search for (Weighted) Partial MaxSAT by New Weighting Techniques
Yi Chu (Institute of Software, Chinese Academy of Sciences), Chuan Luo (School of Software, Beihang University)
OptimizationBenchmark
🎯 What it does: In this paper, the authors propose two new weighting techniques (soft clause weight initialization and Dist-Weighting) and based on this, design a new Stochastic Local Search (SLS) solver NuWLS to solve the (Weighted) Partial MaxSAT (W)PMS problem.
Occupancy Planes for Single-View RGB-D Human Reconstruction
Xiaoming Zhao (University of Illinois Urbana-Champaign), Alexander G. Schwing (University of Illinois Urbana-Champaign)
SegmentationPose EstimationDepth EstimationConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: This paper proposes an occupancy plane (OPlanes) representation method based on perspective frustum forward plane slicing, used to reconstruct 3D human bodies from a single RGB-D image.