AAAI 2023 Papers — Page 8
AAAI Conference on Artificial Intelligence · 1578 papers
Improving Interpretability via Explicit Word Interaction Graph Layer
Arshdeep Sekhon (University of Virginia), Yanjun Qi (University of Virginia)
ClassificationExplainability and InterpretabilityGraph Neural NetworkTransformerText
🎯 What it does: A learnable word interaction graph layer (WIGRAPH) is designed and inserted into the text classification model to enhance the model's interpretability and predictive performance.
Improving Long-Horizon Imitation through Instruction Prediction
Joey Hejna (Stanford University), Lerrel Pinto (New York University)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningReinforcement LearningText
🎯 What it does: By having agents additionally predict task instructions during the imitation learning process, their performance in long-term planning tasks is improved.
Improving Pareto Front Learning via Multi-Sample Hypernetworks
Long P. Hoang (VinUniversity), Tran Ngoc Thang (Hanoi University of Science and Technology)
ClassificationOptimizationHyperparameter SearchImageText
🎯 What it does: Construct a multi-sample hypernetwork (PHN-HVI) that learns the complete Pareto front by generating multiple sets of solutions at once and using hypervolume maximization and cosine similarity regularization.
Improving Robotic Tactile Localization Super-resolution via Spatiotemporal Continuity Learning and Overlapping Air Chambers
Xuyang Li (Xidian University), Guangming Shi (Xidian University)
Super ResolutionRobotic IntelligenceConvolutional Neural NetworkTime Series
🎯 What it does: A flexible tactile sensor based on overlapping air cavities is proposed, and super-resolution tactile localization is achieved through spatiotemporal continuity learning.
Improving Scene Text Image Super-resolution via Dual Prior Modulation Network
Shipeng Zhu (Southeast University), Hui Xue (Southeast University)
RecognitionSuper ResolutionTransformerImageBenchmark
🎯 What it does: A pluggable Dual Prior Modulation Network (DPMN) is proposed, which utilizes two types of image-level priors—text masks and graphic recognition results—through a dual-branch design to progressively improve and integrate super-resolution results, thereby enhancing the clarity of scene text images and downstream recognition performance.
Improving Simultaneous Machine Translation with Monolingual Data
Hexuan Deng (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)
Knowledge DistillationTransformerReinforcement LearningText
🎯 What it does: This paper studies the improvement of the quality of synchronous machine translation (SiMT) models by first generating pseudo-targets from monolingual data using sequence-level knowledge distillation, and then merging them with bilingual data for training. It also proposes a monolingual sampling strategy for SiMT that considers chunk length and monotonicity to reduce hallucinations and enhance robustness.
Improving the Cross-Lingual Generalisation in Visual Question Answering
Farhad Nooralahzadeh (University of Zurich), Rico Sennrich (University of Zurich)
TransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: Improving the performance of pre-trained multilingual audiovisual models in cross-lingual visual question answering tasks, three strategies are proposed: language prior loss, sparse fine-tuning (SFT), and code mixing (CDM), and evaluated on the xGQA benchmark.
Improving Uncertainty Quantification of Deep Classifiers via Neighborhood Conformal Prediction: Novel Algorithm and Theoretical Analysis
Subhankar Ghosh (Washington State University), Janardhan Rao Doppa (Washington State University)
ClassificationComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: The paper proposes the Neighborhood Conformal Prediction (NCP) algorithm to improve the uncertainty quantification of deep classifiers.
Incentive-Boosted Federated Crowdsourcing
Xiangping Kang (Shandong University), Jinglin Zhang (Shandong University)
OptimizationFederated LearningSafty and PrivacyTabularTime Series
🎯 What it does: This paper designs an incentive mechanism for federated crowdsourcing scenarios, utilizing mobile terminals to locally collect new data and train local models, which are then encrypted and uploaded to a central server for aggregation, ensuring data privacy while improving the quality of the global model.
Incomplete Multi-View Multi-Label Learning via Label-Guided Masked View- and Category-Aware Transformers
Chengliang Liu (Harbin Institute of Technology), Yong Xu (Harbin Institute of Technology)
ClassificationRepresentation LearningTransformerImage
🎯 What it does: A Transformer-based multi-view multi-label learning framework named LMVCAT is proposed, which can simultaneously handle data with missing views and missing labels, and improves classification performance through cross-view feature aggregation, category association learning, and label manifold constraints.
Inconsistent Cores for ASP: The Perks and Perils of Non-monotonicity
Johannes K. Fichte (TU Wien), Stefan Szeider (TU Wien)
Graph
🎯 What it does: This paper studies the concept of Inconsistent Core (IC) in Answer Set Programming (ASP), systematically analyzes the complexity of finding IC, Minimal Inconsistent Core (MIC), and Minimal Strongly Minimal Inconsistent Core (SMIC), and provides an implementation based on ASP-Core-2.
Incremental Image De-raining via Associative Memory
Yi Gu (Alibaba Cloud Computing), Jie Li (Shanghai Jiao Tong University)
RestorationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an incremental raindrop removal method based on Associative Memory (AM), aimed at addressing the problem of catastrophic forgetting that occurs when continuously learning new datasets.
Incremental Reinforcement Learning with Dual-Adaptive ε-Greedy Exploration
Wei Ding (National Taiwan University), Ming-Syan Chen (National Taiwan University)
Reinforcement LearningBenchmark
🎯 What it does: This paper proposes the Incremental Reinforcement Learning (Incremental RL) problem, where the state and action space of the environment continuously expand during the training process. It designs a Dual-Adaptive ε-greedy Exploration algorithm (DAE) that dynamically adjusts the ε value through a Meta Policy and estimates the least tried actions to achieve efficient exploration. Additionally, two benchmark environments, Expanding World and Incremental Atari, are released to validate the method.
Incremental-DETR: Incremental Few-Shot Object Detection via Self-Supervised Learning
Na Dong (National University of Singapore), Gim Hee Lee (National University of Singapore)
Object DetectionKnowledge DistillationTransformerImage
🎯 What it does: This paper studies an incremental few-shot object detection method called Incremental-DETR, which achieves incremental learning under the condition of having only a few new class samples and no base class samples.
IndicSUPERB: A Speech Processing Universal Performance Benchmark for Indian Languages
Tahir Javed (Indian Institute of Technology Madras), Mitesh M. Khapra (Indian Institute of Technology Madras)
ClassificationRecognitionRetrievalTransformerSupervised Fine-TuningBenchmarkAudio
🎯 What it does: A labeled speech corpus of 1,684 hours in 12 Hindi languages (Kathbath) was constructed, and based on this, a multi-task SLU evaluation benchmark (IndicSUPERB) was released, covering ASR, speaker verification/recognition, language recognition, query retrieval, and keyword spotting.
Inferential Knowledge-Enhanced Integrated Reasoning for Video Question Answering
Jianguo Mao (Chinese Academy of Sciences), Yajuan Lyu (Baidu Inc.)
RecognitionGraph Neural NetworkTransformerVideoTextMultimodality
🎯 What it does: Proposes a video question-answering model enhanced by reasoning knowledge.
Inferring Patient Zero on Temporal Networks via Graph Neural Networks
Xiaolei Ru (Tongji University), Gang Yan (Tongji University)
Graph Neural NetworkGraphTime Series
🎯 What it does: This paper proposes a backward inference model based on graph neural networks to infer patient zero from the final snapshot of an epidemic on temporal networks.
InfoCTM: A Mutual Information Maximization Perspective of Cross-Lingual Topic Modeling
Xiaobao Wu (Nanyang Technological University), Anh Tuan Luu (Nanyang Technological University)
Auto EncoderContrastive LearningText
🎯 What it does: A cross-language topic model based on information maximization, InfoCTM, is proposed, which can automatically discover aligned cross-language topics and generate high-quality topic distributions.
Information-Theoretic Causal Discovery and Intervention Detection over Multiple Environments
Osman Mian (CISPA Helmholtz Center for Information Security), Jilles Vreeken (CISPA Helmholtz Center for Information Security)
GraphTabular
🎯 What it does: The study simultaneously learns shared causal networks and intervention targets for each environment under multi-environment data, proposing an information-theoretic MDL score and the ORION search algorithm.
Infusing Definiteness into Randomness: Rethinking Composition Styles for Deep Image Matting
Zixuan Ye (Huazhong University of Science and Technology), Hao Lu (Huazhong University of Science and Technology)
SegmentationData SynthesisImage
🎯 What it does: This study investigates the training data generation process for depth image matting and proposes a new reasonable foreground combination operator RCF and ternary/quaternary combination styles to enhance the diversity of training samples and model generalization.
InParformer: Evolutionary Decomposition Transformers with Interactive Parallel Attention for Long-Term Time Series Forecasting
Haizhou Cao (Chinese Academy of Sciences), Yangang Wang (North China Electric Power University)
TransformerTime Series
🎯 What it does: Proposes the InParformer model, which utilizes interactive parallel attention and evolutionary seasonal-trend decomposition to enhance long-term time series forecasting.
Instance Smoothed Contrastive Learning for Unsupervised Sentence Embedding
Hongliang He (Zhejiang University), Yue Zhang (Westlake University)
RetrievalRepresentation LearningTransformerContrastive LearningText
🎯 What it does: By constructing a dynamic memory buffer to retrieve instances similar in semantic meaning to sentences, and using self-attention aggregation to obtain smoothed positive samples, the generalization ability of sentence representations is enhanced within an unsupervised contrastive learning framework.
InstanceFormer: An Online Video Instance Segmentation Framework
Rajat Koner (Ludwig Maximilian University of Munich), Volker Tresp (Ludwig Maximilian University of Munich)
Object DetectionObject TrackingSegmentationTransformerContrastive LearningVideo
🎯 What it does: A single-stage online video instance segmentation framework called InstanceFormer is proposed.
Integer Subspace Differential Privacy
Prathamesh Dharangutte (Rutgers University), Fang-Yi Yu (George Mason University)
Safty and PrivacyTabular
🎯 What it does: This paper proposes the Integer Subspace Differential Privacy framework, specifically addressing how to construct differential privacy mechanisms under external invariants (such as linear constraints like total population) while ensuring the output is an integer.
Integrating Reward Maximization and Population Estimation: Sequential Decision-Making for Internal Revenue Service Audit Selection
Peter Henderson (Stanford University), Daniel E. Ho (Stanford University)
OptimizationReinforcement LearningTabularFinance Related
🎯 What it does: Proposed an optimize-and-estimate structured bandit framework that unifies the maximization of IRS audit returns with tax gap estimation in a single decision; developed the Adaptive Bin Sampling (ABS) algorithm within this framework, which ensures unbiased overall estimation while allowing for the adjustment of the reward-variance trade-off; experiments were conducted on IRS NRP random sampling audit data from 2006 to 2014 to validate its feasibility in actual tax enforcement.
Intensity-Aware Loss for Dynamic Facial Expression Recognition in the Wild
Hanting Li (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
RecognitionConvolutional Neural NetworkTransformerVideo
🎯 What it does: This paper proposes a Global Convolutional Attention block (GCA) and Intensity-Aware Loss (IAL) to enhance the performance of dynamic facial expression recognition in wild videos.
Inter-image Contrastive Consistency for Multi-Person Pose Estimation
Xixia Xu (Beijing Jiaotong University), Qi Zou (Beijing Jiaotong University)
Pose EstimationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A cross-image contrastive consistency (ICON) framework is proposed, which includes single keypoint contrastive consistency (SKCC) and pairwise relationship contrastive consistency (PRCC), enhancing keypoint localization accuracy and structural robustness in multi-person pose estimation through cross-image feature consistency.
Interactive Concept Bottleneck Models
Kushal Chauhan (Google Research), Krishnamurthy Dvijotham (Google Research)
ClassificationObject DetectionImage
🎯 What it does: Developed Interactive Concept Bottleneck Models, which actively ask humans for labels on specific concepts during inference to improve prediction accuracy.
Interpolating Graph Pair to Regularize Graph Classification
Hongyu Guo (National Research Council Canada), Yongyi Mao (University of Ottawa)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Mixup-based graph input interpolation regularization method called ifMixup for graph classification tasks;
Interpreting Unfairness in Graph Neural Networks via Training Node Attribution
Yushun Dong (University of Virginia), Jundong Li (University of Georgia)
ClassificationExplainability and InterpretabilityComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This paper studies the fairness bias that arises in node classification tasks using Graph Neural Networks (GNNs) and proposes the BIND framework, which combines the influence measurement of training nodes with Probabilistic Distribution Disparity (PDD) to explain and quantify model bias.
Intersection Coordination with Priority-Based Search for Autonomous Vehicles
Jiaoyang Li (Carnegie Mellon University), Sven Koenig (University of Southern California)
Autonomous DrivingOptimization
🎯 What it does: This paper proposes a three-layer algorithm PSL (Priority-Based Search + Safe Interval Path Planning + Linear Programming) to address the coordination problem of unmanned vehicles at signal-less intersections, aiming to minimize the total time for vehicles to leave the intersection.
Interventional SHAP Values and Interaction Values for Piecewise Linear Regression Trees
Artjom Zern (SCHUFA Holding AG), Gjergji Kasneci (University of Tuebingen)
Explainability and InterpretabilityComputational EfficiencyTabular
🎯 What it does: This paper proposes an efficient exclusive SHAP computation method, extending the TreeSHAP algorithm to piecewise linear regression trees and implementing background data aggregation.
Intriguing Findings of Frequency Selection for Image Deblurring
Xintian Mao (East China Normal University), Yan Wang (East China Normal University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A residual block for frequency-selective ReLU in the frequency domain (Res FFT-ReLU Block) is proposed, which is integrated into existing deblurring networks to fuse frequency domain and spatial domain features for restoring clear images.
Invariant Representations with Stochastically Quantized Neural Networks
Mattia Cerrato (Johannes Gutenberg University Mainz), Stefan Kramer (Johannes Gutenberg University Mainz)
Representation LearningAuto EncoderTabularStochastic Differential Equation
🎯 What it does: This paper proposes a method for directly calculating the mutual information between neurons and sensitive attributes using stochastically quantized binary neural networks to achieve fair representation learning.
Inverse-Reference Priors for Fisher Regularization of Bayesian Neural Networks
Keunseo Kim (Samsung Advanced Institute of Technology), Heeyoung Kim (KAIST)
ClassificationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Proposes the Inverse Reference Prior (IR prior) to regularize the Fisher information matrix in Bayesian neural networks, thereby enhancing the model's generalization and robustness against adversarial attacks.
Ising-Traffic: Using Ising Machine Learning to Predict Traffic Congestion under Uncertainty
Zhenyu Pan (University of Rochester), Tony Geng
OptimizationComputational EfficiencyTime Series
🎯 What it does: A dual-mode framework called Ising-Traffic based on the Ising model is proposed for real-time traffic congestion prediction under observational uncertainty.
Isolation and Impartial Aggregation: A Paradigm of Incremental Learning without Interference
Yabin Wang (Xi'an Jiaotong University), Xiaopeng Hong (Peng Cheng Laboratory)
ClassificationDomain AdaptationTransformerSupervised Fine-TuningImageBenchmark
🎯 What it does: A non-replay phase-isolated incremental learning framework ESN is designed, utilizing a pre-trained ViT backbone and independent classifiers, and achieving cross-phase knowledge aggregation through energy self-normalization and voting inference, eliminating phase interference and performance imbalance.
Isometric Manifold Learning Using Hierarchical Flow
Ziqi Pan (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)
Anomaly DetectionOptimizationFlow-based ModelImage
🎯 What it does: Design and train a Hierarchical Flow (HF) model to achieve learning, dimensionality reduction, projection, sampling, and density estimation of data manifolds.
IterDE: An Iterative Knowledge Distillation Framework for Knowledge Graph Embeddings
Jiajun Liu (Southeast University), Chenxiao Wu (Southeast University)
Knowledge DistillationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: The IterDE framework is proposed, which gradually compresses knowledge graph embedding models through iterative knowledge distillation, maintaining high performance at low dimensions.
Joint Multimodal Entity-Relation Extraction Based on Edge-Enhanced Graph Alignment Network and Word-Pair Relation Tagging
Li Yuan (South China University of Technology), Qing Li (Hong Kong Polytechnic University)
RecognitionObject DetectionGraph Neural NetworkTransformerVision Language ModelTextMultimodality
🎯 What it does: This paper proposes a joint task of multimodal named entity recognition and relation extraction, called JMERE, and designs a word pair relation labeling and edge-enhanced graph alignment network to achieve joint extraction.
Jointly Imputing Multi-View Data with Optimal Transport
Yangyang Wu (Zhejiang University), Jianwei Yin (Zhejiang University)
GenerationData SynthesisOptimizationAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A multi-view generative missing feature imputation model named Git is proposed, which utilizes a joint autoencoder for joint learning of all observed views and achieves feature-level imputation based on this.
JR2Net: Joint Monocular 3D Face Reconstruction and Reenactment
Jiaxiang Shang (Hong Kong University of Science and Technology), Hongbo Fu (City University of Hong Kong)
RecognitionGenerationPose EstimationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a joint framework for facial reconstruction and reenactment called JR2Net, which utilizes a reconstruction network to provide 3D information to enhance reenactment quality, while also using the image-expression coefficient pairs generated from reenactment to further optimize the reconstruction model.
Just Noticeable Visual Redundancy Forecasting: A Deep Multimodal-Driven Approach
Wuyuan Xie (Shenzhen University), Miaohui Wang (Shenzhen University)
Depth EstimationCompressionTransformerImageMultimodality
🎯 What it does: This paper proposes an end-to-end multimodal visibility threshold prediction network, hmJND-Net, which improves the accuracy of human visual visibility threshold (JND) prediction by integrating salient, depth, and semantic segmentation information from the same source.
Kalman Bayesian Neural Networks for Closed-Form Online Learning
Philipp Wagner (Fraunhofer Institute for Manufacturing Engineering and Automation), Marco F. Huber (University of Stuttgart)
Tabular
🎯 What it does: Proposes the Kalman Bayesian Neural Network (KBNN), which implements gradient-free, online single-sample learning of Bayesian neural networks through Bayesian filtering/smoothing.
KerPrint: Local-Global Knowledge Graph Enhanced Diagnosis Prediction for Retrospective and Prospective Interpretations
Kai Yang (Zhongguancun Laboratory), Bing Xie (Peking University)
ClassificationExplainability and InterpretabilityGraph Neural NetworkTransformerSupervised Fine-TuningTabularBiomedical DataElectronic Health Records
🎯 What it does: The KerPrint model is proposed, which combines local and global knowledge graphs, and achieves retrospective and prospective explanations for diagnostic predictions through a time-aware attention mechanism and element-level attention.
Key Feature Replacement of In-Distribution Samples for Out-of-Distribution Detection
Jaeyoung Kim (VUNO), Kyu-Hwan Jung (Samsung Advanced Institute for Health Sciences and Technology)
Anomaly DetectionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes an auxiliary OOD data generation method called KIRBY, based on key feature replacement and patching, to train a rejection network for OOD detection without modifying the original classifier.
KICE: A Knowledge Consolidation and Expansion Framework for Relation Extraction
Yilin Lu (Zhejiang University), Siliang Tang (Zhejiang University)
Large Language ModelPrompt EngineeringText
🎯 What it does: The KICE framework is proposed, which enhances the performance of relation extraction under a small amount of labeled data through knowledge integration and iterative expansion.
Knowledge Amalgamation for Multi-Label Classification via Label Dependency Transfer
Jidapa Thadajarassiri (Worcester Polytechnic Institute), Elke Rundensteiner (Worcester Polytechnic Institute)
ClassificationKnowledge DistillationRecurrent Neural NetworkTextMultimodality
🎯 What it does: Knowledge fusion for multi-label classification (MLC) problems: In the absence of labeled data, multiple pre-trained teacher models (each covering different sets of labels) are fused into a single student model, enabling the student to predict the union of all teacher label sets.
Knowledge Graph Embedding by Normalizing Flows
Changyi Xiao (University of Science and Technology of China), Yixin Cao (Singapore Management University)
Flow-based ModelGraphBenchmark
🎯 What it does: This paper proposes embedding knowledge graph entities/relations into the permutation group and introduces uncertainty through normalizing flow, forming a new knowledge graph embedding model called NFE.
Knowledge-Bridged Causal Interaction Network for Causal Emotion Entailment
Weixiang Zhao (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
TransformerText
🎯 What it does: Proposes the Knowledge-Bridged Causal Interaction Network (KBCIN) to identify the causal sentences that trigger non-neutral emotions in dialogues.
Knowledge-Constrained Answer Generation for Open-Ended Video Question Answering
Yao Jin (Hangzhou Dianzi University), Jun Yu (Hangzhou Dianzi University)
GenerationTransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: The KcGA framework is proposed, utilizing an encoder-decoder structure to achieve open VideoQA, integrating external commonsense knowledge and a multi-stream information control mechanism to generate free-form answers.
KPT: Keyword-Guided Pre-training for Grounded Dialog Generation
Qi Zhu (Tsinghua University), Minlie Huang (Tsinghua University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: An unsupervised keyword-guided pre-training (KPT) framework has been constructed to enhance the performance of dialogue generation models for different types of knowledge in low-resource scenarios.
KT-Net: Knowledge Transfer for Unpaired 3D Shape Completion
Zhen Cao (Wuhan University), Bisheng Yang (Tsinghua University)
Data SynthesisDomain AdaptationKnowledge DistillationAdversarial AttackGenerative Adversarial NetworkPoint Cloud
🎯 What it does: A KT-Net based on a teacher-assistant-student network is proposed, utilizing knowledge transfer to achieve unpaired 3D point cloud completion.
Label-Specific Feature Augmentation for Long-Tailed Multi-Label Text Classification
Pengyu Xu (Beijing Jiaotong University), Jian Yu (Beijing Jiaotong University)
ClassificationAuto EncoderContrastive LearningTextBenchmark
🎯 What it does: To address the long-tail label problem in multi-label text classification, this paper proposes a dual-layer framework based on Label-Specific Feature Augmentation (LSFA), which first learns decoupled label-specific representations and then generates positive samples for tail labels in the feature space using a prototype-driven VAE.
LaCAM: Search-Based Algorithm for Quick Multi-Agent Pathfinding
Keisuke Okumura (Tokyo Institute of Technology)
OptimizationTabularBenchmark
🎯 What it does: A new complete and fast multi-agent path planning algorithm LaCAM is proposed, which utilizes a two-layer search structure to add positive constraints at the lower level and generate configurations at the upper level, significantly reducing the branching factor;
LADA-Trans-NER: Adaptive Efficient Transformer for Chinese Named Entity Recognition Using Lexicon-Attention and Data-Augmentation
Jiguo Liu (Institute of Information Engineering Chinese Academy of Sciences), Dali Zhu (Institute of Information Engineering Chinese Academy of Sciences)
RecognitionTransformerGenerative Adversarial NetworkText
🎯 What it does: This paper proposes a Transformer model that integrates dictionary attention and data augmentation for Chinese named entity recognition.
LagNet: Deep Lagrangian Mechanics for Plug-and-Play Molecular Representation Learning
Chunyan Li (Xiamen University), Chenxi Huang (Xiamen University)
Representation LearningDrug DiscoveryRecurrent Neural NetworkGraph Neural NetworkGraphOrdinary Differential Equation
🎯 What it does: LagNet simulates molecular force fields using discrete-time Lagrangian mechanics, learning 3D conformations directly from SMILES and using them as molecular representations.
LANCER: A Lifetime-Aware News Recommender System
Hong-Kyun Bae (Hanyang University), Sang-Wook Kim (Hanyang University)
Recommendation SystemText
🎯 What it does: This study investigates the news lifecycle in news recommendation and proposes the LANCER method, which utilizes the news lifecycle to enhance recommendation effectiveness during training and recommendation.
Language Model Pre-training on True Negatives
Zhuosheng Zhang (Shanghai Jiao Tong University), Eiichiro Sumita (National Institute of Information and Communications Technology)
TransformerLarge Language ModelText
🎯 What it does: This study investigates the issue of false negative samples in MLM pre-training and proposes two methods, hard correction and soft regularization, to improve the training quality of PLM.
Language-Assisted 3D Feature Learning for Semantic Scene Understanding
Junbo Zhang (Tsinghua University), Li Yi (Shanghai Artificial Intelligence Laboratory)
Object DetectionSegmentationRepresentation LearningRecurrent Neural NetworkSupervised Fine-TuningTextPoint Cloud
🎯 What it does: Achieve semantic scene understanding through weakly supervised text description-guided 3D feature learning.
Large-State Reinforcement Learning for Hyper-Heuristics
Lucas Kletzander (Technical University of Vienna), Nysret Musliu (Technical University of Vienna)
OptimizationReinforcement LearningTabular
🎯 What it does: A general hyper-heuristic method LAST-RL based on large state reinforcement learning is proposed to dynamically select low-level heuristic chains on unknown instances.
Latent Autoregressive Source Separation
Emilian Postolache (Sapienza University of Rome), Emanuele Rodolà (Sapienza University of Rome)
GenerationTransformerAuto EncoderImageAudio
🎯 What it does: Using a pre-trained VQ-VAE and autoregressive model, source separation is achieved through Bayesian inference in the quantized latent space without gradient optimization or model modification.
Latent Constraints on Unsupervised Text-Graph Alignment with Information Asymmetry
Jidong Tian (Shanghai Jiao Tong University), Yaohui Jin (Shanghai Jiao Tong University)
Graph Neural NetworkTransformerAuto EncoderText
🎯 What it does: A model is proposed to introduce information asymmetry in unsupervised text-image alignment through three types of VAE constraints (transferable, dependent, and non-transferable latent variables).
Layer-Wise Adaptive Model Aggregation for Scalable Federated Learning
Sunwoo Lee (University of Southern California), A. Salman Avestimehr
Federated LearningImage
🎯 What it does: Proposes FedLAMA, a federated learning scheme with layer-wise adaptive aggregation intervals, which dynamically adjusts the synchronization frequency of each layer to reduce communication costs and accelerate convergence;
Layout Generation as Intermediate Action Sequence Prediction
Huiting Yang (South China University of Technology), Shengfeng He (Singapore Management University)
GenerationTransformerContrastive LearningImage
🎯 What it does: A layout generation method based on action sequences is proposed, which directly predicts actions such as copy and margin instead of border coordinates;
Layout Representation Learning with Spatial and Structural Hierarchies
Yue Bai (Northeastern University), Yun Fu (Northeastern University)
RetrievalRepresentation LearningGraph Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a self-supervised hierarchical autoencoder (SSH-AE) that treats layouts as a tree hierarchy, learning features from both spatial and structural aspects simultaneously for layout retrieval.
Layout-Aware Dreamer for Embodied Visual Referring Expression Grounding
Mingxiao Li (KU Leuven), Marie-Francine Moens (KU Leuven)
Robotic IntelligenceTransformerReinforcement LearningVision Language ModelImageText
🎯 What it does: Designed and implemented an agent named Layout-aware Dreamer (LAD) that can navigate unknown indoor environments and locate remote targets by utilizing layout learning and goal imagination modules upon receiving high-level natural language instructions.
Learn from Yesterday: A Semi-supervised Continual Learning Method for Supervision-Limited Text-to-SQL Task Streams
Yongrui Chen (Southeast University), Yang Dong (Alibaba Group)
Data-Centric LearningPrompt EngineeringTextBenchmark
🎯 What it does: In the supervised limited text-to-SQL task flow, two approaches that combine semi-supervised learning (SSL) with continual learning (CL) are proposed, namely VANILLA and the improved SFNET.
Learn More for Food Recognition via Progressive Self-Distillation
Yaohui Zhu (Beijing Normal University), Jiang Tian (Lenovo Research)
RecognitionKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: A Progressive Self-Distillation (PSD) mechanism is proposed, which conducts multiple rounds of self-distillation from a teacher network to a student network without explicitly locating multiple regions. It gradually masks the areas that have been attended to, forcing the network to extract more fine-grained information for more detailed recognition of food images.
Learnable Blur Kernel for Single-Image Defocus Deblurring in the Wild
Jucai Zhai (Peking University), Yong Zhao (Peking University)
RestorationGenerative Adversarial NetworkImage
🎯 What it does: This study addresses the depth-of-field deblurring problem for single images, proposing a learnable blur kernel estimation for defocus images and utilizing a multi-scale GAN guided by this image for deblurring. During training, an unsupervised defocus image estimation is achieved using dual-pixel views, and inference requires only a single image input.
Learnable Path in Neural Controlled Differential Equations
Sheo Yon Jhin (Yonsei University), Noseong Park (Yonsei University)
ClassificationOptimizationRecurrent Neural NetworkTime SeriesSequentialBiomedical DataOrdinary Differential Equation
🎯 What it does: The LEAP model is proposed, which learns interpolated paths through an encoder-decoder architecture, enhancing the performance of NCDE on irregular time series.
Learnable Spectral Wavelets on Dynamic Graphs to Capture Global Interactions
Anson Bastos (Indian Institute of Technology Hyderabad), Manish Singh (Indian Institute of Technology Hyderabad)
Representation LearningRecurrent Neural NetworkGraph Neural NetworkTransformerGraph
🎯 What it does: The DEFT framework is proposed, utilizing learnable spectral wavelets to capture global interactions in dynamic graphs and integrating them with spatial features to enhance representation learning effectiveness.
Learned Distributed Image Compression with Multi-Scale Patch Matching in Feature Domain
Yujun Huang (Tsinghua University), Shu-Tao Xia (Tsinghua University)
CompressionAutonomous DrivingAuto EncoderImage
🎯 What it does: A distributed image compression method based on Multi-Scale Feature Domain Block Matching (MSFDPM) is proposed, achieving alignment of side information and multi-scale feature fusion at the decoding end.
Learning a Generalized Gaze Estimator from Gaze-Consistent Feature
Mingjie Xu (Beihang University), Feng Lu (Beihang University)
Domain AdaptationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A gaze estimation method based on gaze-invariant features for domain generalization is proposed. By applying adversarial perturbations or data augmentation to gaze-invariant factors such as identity, expression, lighting, and hue, the model learns only the features related to gaze.
Learning Adversarially Robust Sparse Networks via Weight Reparameterization
Chenhao Li (University of Chinese Academy of Sciences), Xueqi Cheng (Chinese Academy of Sciences)
ClassificationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Achieving robust sparse network training and pruning through weight reparameterization.
Learning by Applying: A General Framework for Mathematical Reasoning via Enhancing Explicit Knowledge Learning
Jiayu Liu (University of Science and Technology of China), Qi Liu (University of Illinois at Urbana-Champaign)
Graph Neural NetworkAuto EncoderText
🎯 What it does: A general learning-application framework (LeAp) is proposed, which allows existing mathematical word problem (MWP) solvers to learn and utilize the relationships between words, words and operators through an explicit knowledge graph, thereby enhancing reasoning capabilities.
Learning Chemical Rules of Retrosynthesis with Pre-training
Yinjie Jiang (Zhejiang University), Zhihua Wang (Zhejiang University)
TransformerContrastive LearningGraph
🎯 What it does: This paper proposes a pre-trained single-step inverse synthesis model (PMSR) and learns chemical rules through three pre-training tasks: molecular recovery, autoregression, and contrastive classification.
Learning Compact Features via In-Training Representation Alignment
Xin Li (Wayne State University), Dongxiao Zhu (Wayne State University)
ClassificationOptimizationRepresentation LearningConvolutional Neural NetworkSupervised Fine-TuningImageText
🎯 What it does: A regularization method (ITRA) is proposed to align the features of different mini-batches through MMD during the training process, in order to reduce the overfitting of SGD to mini-batches and learn more compact feature representations.
Learning Compositional Tasks from Language Instructions
Lajanugen Logeswaran (LG AI Research), Honglak Lee (University of Michigan)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningText
🎯 What it does: This study investigates how reinforcement learning agents can achieve compositional generalization in photo-realistic kitchen environments based on natural language instructions through attention and additive action value decomposition.
Learning Conflict-Noticed Architecture for Multi-Task Learning
Zhixiong Yue (Southern University of Science and Technology), Jie Liang (University of Technology Sydney)
Neural Architecture SearchReinforcement LearningImageText
🎯 What it does: A multi-task network architecture search method based on conflict detection, CoNAL, is proposed, which automatically switches between shared modules and purely exclusive modules to alleviate gradient conflicts, and incorporates conflict detection operations during the search process.
Learning Context-Aware Classifier for Semantic Segmentation
Zhuotao Tian (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
SegmentationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This study proposes a Context-Aware Classifier (CAC) that dynamically generates classifier weights for each image to adapt to the potential distribution of different images, thereby improving semantic segmentation accuracy.
Learning Continuous Depth Representation via Geometric Spatial Aggregator
Xiaohang Wang (Shanghai Jiao Tong University), Hang Wang (Shanghai Jiao Tong University)
Depth EstimationSuper ResolutionTransformerImage
🎯 What it does: This paper proposes a continuous depth representation-based RGB-guided depth super-resolution network, GeoDSR, which can perform super-resolution on low-resolution depth maps at any scaling factor and any shape.
Learning Control Policies for Stochastic Systems with Reach-Avoid Guarantees
Đorđe Žikelić (Institute of Science and Technology Austria), Krishnendu Chatterjee (Institute of Science and Technology Austria)
OptimizationRobotic IntelligenceReinforcement Learning
🎯 What it does: This paper proposes a framework that simultaneously learns control strategies and formal reach-avoid guarantees, utilizing neural networks to learn controllers and generate verifiable Reach-Avoid Stochastic Markov (RASM) certificates, completing a closed-loop process from learning to verification.
Learning Decomposed Spatial Relations for Multi-Variate Time-Series Modeling
Yuchen Fang (Shanghai Jiao Tong University), Dongsheng Li (Central South University)
Recurrent Neural NetworkGraph Neural NetworkTime Series
🎯 What it does: This paper proposes a Spatial Relationship Decomposition (SRD) framework for multivariate time series modeling, which automatically learns a global prior graph and a dynamic graph for each sample, ensuring the differentiation between the two graphs through a minimization/maximization learning mechanism to enhance prediction performance.
Learning Deep Hierarchical Features with Spatial Regularization for One-Class Facial Expression Recognition
Bingjun Luo (Tsinghua University), Yue Gao (Tsinghua University)
RecognitionAnomaly DetectionAuto EncoderImage
🎯 What it does: This paper proposes HS-OCFER, a single-class facial expression recognition network trained using only a single normal expression category, aimed at detecting anomalous expressions that were not present during training.
Learning Dynamic Latent Spaces for Lifelong Generative Modelling
Fei Ye (University of York), Adrian G. Bors (University of York)
GenerationData SynthesisAuto EncoderImage
🎯 What it does: This paper studies a generative model suitable for lifelong learning without task labels—Online Recursive Variational Autoencoder (ORVAE), which achieves knowledge transfer and prevents forgetting through recursive expansion and attention mechanisms.
Learning Event-Relevant Factors for Video Anomaly Detection
Che Sun (Beijing Institute of Technology), Yuwei Wu (Beijing Institute of Technology)
Anomaly DetectionAuto EncoderVideo
🎯 What it does: This paper proposes an explicit learning method for event-related factors through causal generative models and counterfactual learning to eliminate event-irrelevant interference in video anomaly detection.
Learning Explicit Credit Assignment for Cooperative Multi-Agent Reinforcement Learning via Polarization Policy Gradient
Wubing Chen (Nanjing University), Yang Gao (Nanjing University)
Reinforcement Learning
🎯 What it does: A new multi-agent reinforcement learning algorithm called MAPPG is proposed to address the credit assignment problem in cooperative multi-agent systems.
Learning Fractals by Gradient Descent
Cheng-Hao Tu (Ohio State University), Wei-Lun Chao (Ohio State University)
RestorationGenerationOptimizationRecurrent Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a gradient descent-based iterative function system (IFS) parameter learning method that can generate visually similar fractal images for any target image (not necessarily fractal images).
Learning from Good Trajectories in Offline Multi-Agent Reinforcement Learning
Qi Tian (Zhejiang University), Baoxiang Wang (Chinese University of Hong Kong)
Graph Neural NetworkReinforcement LearningSequential
🎯 What it does: This paper proposes the Shared Individual Trajectories (SIT) framework for offline multi-agent reinforcement learning, which first uses an attention-based reward decomposition network to split the global reward into individual rewards, then reconstructs and shares high-quality individual trajectories through prioritized experience replay, and finally employs a graph attention network and CRR constraints for conservative policy training.
Learning from the Wisdom of Crowds: Exploiting Similar Sessions for Session Search
Yuhang Ye (Huawei), Zhao Cao (Huawei)
RetrievalGraph Neural NetworkSequential
🎯 What it does: A retrieval model based on similar session enhancement, SSR, is proposed, which utilizes queries and session information from historical sessions that are similar to the current query intent to improve session search ranking.
Learning from Training Dynamics: Identifying Mislabeled Data beyond Manually Designed Features
Qingrui Jia (Beihang University), Dejing Dou (BCG Greater China)
Anomaly DetectionData-Centric LearningRecurrent Neural NetworkSupervised Fine-TuningImageTime SeriesSequential
🎯 What it does: A noise detector based on LSTM is trained using the probability sequences during the training process to identify mislabeled samples in the training set.
Learning Instrumental Variable from Data Fusion for Treatment Effect Estimation
Anpeng Wu (Zhejiang University), Fei Wu (Shanghai AI Laboratory)
Representation LearningTabular
🎯 What it does: The Meta-EM algorithm is proposed, which automatically reconstructs source labels through representation learning and EM iteration in multi-source data fusion, using group instrumental variables (GIV) to achieve causal effect estimation without prior instrumental variables.
Learning Interpretable Temporal Properties from Positive Examples Only
Rajarshi Roy (Max Planck Institute for Software Systems), Ufuk Topcu (Arizona State University)
Explainability and InterpretabilityComputational EfficiencyTime SeriesSequential
🎯 What it does: This paper proposes an algorithm for learning interpretable temporal models (DFA and LTLf) given only positive examples;
Learning Logic Programs by Discovering Where Not to Search
Andrew Cropper (University of Oxford), Céline Hocquette (University of Oxford)
🎯 What it does: This paper proposes a method to automatically discover and utilize constraints from background knowledge (BK) to limit the hypothesis space of Inductive Logic Programming (ILP), thereby 'discovering areas that do not need to be searched' before the search, significantly accelerating the learning process.
Learning Markov Random Fields for Combinatorial Structures via Sampling through Lovász Local Lemma
Nan Jiang (Purdue University), Yexiang Xue (Purdue University)
GenerationOptimizationGraphTabular
🎯 What it does: This paper proposes a differentiable sampler NELSON based on the Lovász Local Lemma (LLL) and embeds it into a Contrastive Divergence (CD) learning framework to train constrained Markov Random Fields (MRF), thereby achieving high-quality generation of combinatorial structures that satisfy hard constraints (such as K-SAT, acyclic directed graphs, and vehicle routing paths).
Learning Motion-Robust Remote Photoplethysmography through Arbitrary Resolution Videos
Jianwei Li (Xi'an Jiaotong University), Jingang Shi (Xi'an Jiaotong University)
Convolutional Neural NetworkOptical FlowVideo
🎯 What it does: Two pluggable modules, PFE and TFA, are proposed to enhance the robustness of remote photoacoustic blood pressure measurement under arbitrary resolution video and head motion.
Learning Noise-Induced Reward Functions for Surpassing Demonstrations in Imitation Learning
Liangyu Huo (Beihang University), Mai Xu (Beihang University)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: This paper proposes a noise-reward learning-based imitation learning method called LERP, which can train a policy that surpasses the demonstrator even with only suboptimal demonstrations.
Learning Optimal Features via Partial Invariance
Moulik Choraria (University of Illinois at Urbana-Champaign), Lav R. Varshney (University of Illinois at Urbana-Champaign)
ClassificationDomain AdaptationText
🎯 What it does: This paper proposes a 'Partial Invariance' framework (P-IRM) in the context of concept drift, which relaxes the complete invariance constraint of traditional IRM by partitioning or conditioning the training environment, thereby enhancing the model's generalization ability in external environments.
Learning Pessimism for Reinforcement Learning
Edoardo Cetin (King's College London), Oya Celiktutan (King's College London)
Reinforcement LearningSequential
🎯 What it does: This paper proposes Generalized Pessimism Learning (GPL), which dynamically adjusts overestimation bias through a learnable penalty term and dual TD learning, combined with SAC and DrQ, achieving more robust and sample-efficient offline policy learning.