AAAI 2023 Papers — Page 5
AAAI Conference on Artificial Intelligence · 1578 papers
DropMessage: Unifying Random Dropping for Graph Neural Networks
Taoran Fang (Zhejiang University), Yang Yang (Zhejiang University)
Graph Neural NetworkGraph
🎯 What it does: A DropMessage random dropout method is proposed, which masks the message matrix element-wise during the GNN message passing process, achieving finer-grained regularization and robustness enhancement.
Dropout Is NOT All You Need to Prevent Gradient Leakage
Daniel Scheliga (Technische Universitaet Ilmenau), Marco Seeland (Technische Universitaet Ilmenau)
Federated LearningSafty and PrivacyAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: This study investigates the impact of using dropout on gradient inversion attacks in federated learning and proposes a new Dropout Inversion Attack (DIA) that can approximate the client model without the attacker knowing the specific dropout mask, thereby recovering training data.
DrugOOD: Out-of-Distribution Dataset Curator and Benchmark for AI-Aided Drug Discovery – a Focus on Affinity Prediction Problems with Noise Annotations
Yuanfeng Ji (Tencent AI Lab), Yatao Bian (Tencent AI Lab)
Drug DiscoveryGraph Neural NetworkTransformerBiomedical DataBenchmark
🎯 What it does: We propose DrugOOD - an automated OOD dataset construction and benchmarking framework, focusing on the noisy drug-target affinity prediction task;
Dual Label-Guided Graph Refinement for Multi-View Graph Clustering
Yawen Ling (University of Electronic Science and Technology of China), Lifang He (Lehigh University)
OptimizationGraph Neural NetworkAuto EncoderGraph
🎯 What it does: A dual-label guided graph improvement framework, DuoLGR, is proposed for multi-view graph clustering, significantly enhancing clustering performance on low-homogeneity graphs.
Dual Low-Rank Graph Autoencoder for Semantic and Topological Networks
Zhaoliang Chen (Fuzhou University), Wenzhong Guo (Fuzhou University)
ClassificationRepresentation LearningGraph Neural NetworkAuto EncoderGraph
🎯 What it does: This paper proposes the Dual Low-Rank Graph AutoEncoder (DLR-GAE), which simultaneously utilizes node embeddings from both semantic graphs and topological graphs, and reconstructs the adjacency matrix using a shared low-rank factor matrix.
Dual Memory Aggregation Network for Event-Based Object Detection with Learnable Representation
Dongsheng Wang (Dalian University of Technology), Huchuan Lu (Huawei Technologies Co. Ltd)
Object DetectionConvolutional Neural NetworkTime Series
🎯 What it does: This paper proposes a learnable event representation method called EventPillars and a Dual Memory Aggregation Network (DMANet) to achieve high-precision object detection on event cameras.
Dual Memory Units with Uncertainty Regulation for Weakly Supervised Video Anomaly Detection
Hang Zhou (Huazhong University of Science and Technology), Wei Yang (Huazhong University of Science and Technology)
Anomaly DetectionTransformerVideoMultimodality
🎯 What it does: A Transformer model with dual memory units and uncertainty modulation is designed for weakly supervised video anomaly detection.
Dual Mutual Information Constraints for Discriminative Clustering
Hongyu Li (Wuhan University), Kehua Su (Wuhan University)
Representation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A deep clustering framework DMICC based on dual mutual information constraints is proposed, which minimizes mutual information for redundancy at the feature level and maximizes mutual information for robust representation at the instance level, using K-means to output clustering results.
Dual-Domain Attention for Image Deblurring
Yuning Cui (Technical University of Munich), Alois Knoll (Technical University of Munich)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A dual-domain attention mechanism is proposed, embedding spatial attention modules and frequency attention modules into a U-shaped network to enhance image deblurring effects.
DUET: Cross-Modal Semantic Grounding for Contrastive Zero-Shot Learning
Zhuo Chen (Zhejiang University), Huajun Chen (Zhejiang University)
ClassificationRecognitionTransformerContrastive LearningImageMultimodality
🎯 What it does: An end-to-end Transformer-based zero-shot learning framework called DUET is proposed, which achieves fine-grained induction and classification of image attributes through cross-modal semantic alignment and attribute-level contrastive learning.
Dynamic Ensemble of Low-Fidelity Experts: Mitigating NAS “Cold-Start”
Junbo Zhao (Tsinghua University), Yu Wang (Tsinghua University)
OptimizationNeural Architecture SearchMixture of ExpertsImage
🎯 What it does: This paper proposes a Dynamic Ensemble Low-Fidelity Experts framework to enhance the ranking quality of predictors and accelerate the search during the cold start phase of predictor-based NAS by utilizing various low-cost performance estimations.
Dynamic Heterogeneous Graph Attention Neural Architecture Search
Zeyang Zhang (Tsinghua University), Wenwu Zhu (Tsinghua University)
Recommendation SystemNeural Architecture SearchGraph Neural NetworkGraphTime Series
🎯 What it does: A method for automated design of dynamic heterogeneous graph neural network architectures is proposed, aiming to automatically search for the optimal structure for different dynamic heterogeneous graph tasks without the need for human intervention.
Dynamic Multi-Behavior Sequence Modeling for Next Item Recommendation
Junsu Cho (POSTECH), Hwanjo Yu (POSTECH)
Recommendation SystemRecurrent Neural NetworkSequential
🎯 What it does: This study investigates the multi-behavior sequence recommendation problem and proposes two models, DyMuS and DyMuS+, which use dynamic routing to fuse behavior sequence encoding to capture heterogeneity and personalized information.
Dynamic Representation Learning with Temporal Point Processes for Higher-Order Interaction Forecasting
Tony Gracious (Indian Institute of Science), Ambedkar Dukkipati (Indian Institute of Science)
Representation LearningGraph Neural NetworkGraphTime Series
🎯 What it does: This paper proposes a dynamic hypergraph hyperedge prediction model based on point processes (HGDHE/HGBDHE), which can predict the types and occurrence times of multi-node interaction events in time-evolving hypergraphs.
Dynamic Structure Pruning for Compressing CNNs
Jun-Hyung Park (Korea University), SangKeun Lee (Korea University)
CompressionNeural Architecture SearchConvolutional Neural NetworkImage
🎯 What it does: A dynamic structure pruning method is proposed, which automatically learns the kernel grouping of different convolutional layers, significantly compressing the network while maintaining accuracy.
DyRRen: A Dynamic Retriever-Reranker-Generator Model for Numerical Reasoning over Tabular and Textual Data
Xiao Li (Nanjing University), Gong Cheng (Nanjing University)
GenerationRetrievalOptimizationRecurrent Neural NetworkTransformerTextTabularFinance RelatedRetrieval-Augmented Generation
🎯 What it does: A dynamic retrieval-re-ranking-generation (DyRRen) framework is proposed to address numerical reasoning problems involving tables and long texts.
EASAL: Entity-Aware Subsequence-Based Active Learning for Named Entity Recognition
Yang Liu (Shenzhen Research Institute of Big Data), Tsung-Hui Chang (Chinese University of Hong Kong)
RecognitionTransformerLarge Language ModelTextBiomedical Data
🎯 What it does: This paper proposes an active learning method based on entity-aware subsequences, EASAL, for named entity recognition tasks, significantly reducing annotation costs.
Easy Begun Is Half Done: Spatial-Temporal Graph Modeling with ST-Curriculum Dropout
Hongjun Wang (Southern University of Science and Technology), Boyuan Zhang (Huawei Technologies)
Graph Neural NetworkGraphTime Series
🎯 What it does: This paper proposes a node difficulty-based ST-Curriculum Dropout strategy to improve the training of spatial-temporal graph models.
Echo of Neighbors: Privacy Amplification for Personalized Private Federated Learning with Shuffle Model
Yixuan Liu (Renmin University of China), Hong Chen (Renmin University of China)
Federated LearningSafty and PrivacyImage
🎯 What it does: This paper proposes a personalized private federated learning framework based on a shuffling model, APES and S-APES, which can achieve strong central privacy protection while satisfying different local privacy budgets of users.
ECO-3D: Equivariant Contrastive Learning for Pre-training on Perturbed 3D Point Cloud
Ruibin Wang (Peking University), Jinfa Yang (Peking University)
ClassificationSegmentationRepresentation LearningGraph Neural NetworkAuto EncoderContrastive LearningPoint Cloud
🎯 What it does: The ECO-3D framework is proposed, which uses VAE to map randomly perturbed point clouds to approximate undisturbed point embeddings and separated disturbed embeddings, and then conducts contrastive learning based on mixed embeddings, incorporating equivariant learning.
Edge Structure Learning via Low Rank Residuals for Robust Image Classification
Xiang-Jun Shen (JiangSu University), Sirui Tian (Nanjing University of Science and Technology)
ClassificationSupervised Fine-TuningImage
🎯 What it does: This paper proposes a method called ESL-LRR, which utilizes graph regularization information in low-rank residuals to adaptively learn projections, preserving low-rank representations while maintaining image edges, achieving robust image classification.
Editing Boolean Classifiers: A Belief Change Perspective
Nicolas Schwind (National Institute of Advanced Industrial Science and Technology), Pierre Marquis (University of Artois)
Classification
🎯 What it does: This paper proposes an editing framework for Boolean Classifiers, defining forward editing, backward editing, and complete editing, and provides corresponding axiomatic constraints and representation theorems.
EffConv: Efficient Learning of Kernel Sizes for Convolution Layers of CNNs
Alireza Ganjdanesh (University of Pittsburgh), Heng Huang (University of Pittsburgh)
ClassificationOptimizationComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkSupervised Fine-TuningImage
🎯 What it does: An efficient convolution kernel size learning framework, EffConv, has been designed to predict the optimal kernel size and generate corresponding weights within a small number of training epochs, achieving end-to-end differentiable optimization of CNN kernel sizes.
Effective and Stable Role-Based Multi-Agent Collaboration by Structural Information Principles
Xianghua Zeng (Beihang University), Angsheng Li (Zhongguancun Laboratory)
Reinforcement LearningTabularBenchmark
🎯 What it does: A role discovery method based on the principle of structural information, SIRD, is proposed and embedded in the MARL framework SR-MARL to achieve automatic role decomposition and learning for multi-agent collaboration.
Effective Continual Learning for Text Classification with Lightweight Snapshots
Jue Wang (Zhejiang University), Gang Chen (Zhejiang University)
ClassificationKnowledge DistillationTransformerText
🎯 What it does: A continuous learning framework based on lightweight adapters is proposed, which constructs compressed snapshots by freezing and saving adapters after each task training, utilizing knowledge distillation to allow the global model to continuously review old task knowledge while learning new tasks, thereby alleviating catastrophic forgetting.
Effective Integration of Weighted Cost-to-Go and Conflict Heuristic within Suboptimal CBS
Rishi Veerapaneni (Carnegie Mellon University), Maxim Likhachev (Carnegie Mellon University)
OptimizationComputational EfficiencyBenchmark
🎯 What it does: This paper introduces two variants of weighted cost-to-go heuristics and conflict heuristics in CBS low-level planning, significantly improving the solving speed of suboptimal CBS.
Effective Open Intent Classification with K-center Contrastive Learning and Adjustable Decision Boundary
Xiaokang Liu (China Automotive Technology and Research Center), Benyou Wang (Chinese University of Hong Kong)
ClassificationRepresentation LearningTransformerContrastive LearningText
🎯 What it does: A two-stage open-source intent classification method called CLAB is designed, which first obtains robust and balanced semantic representations through K-center contrastive learning, and then utilizes a tunable spherical decision boundary to achieve precise classification of known intents and effective recognition of unknown intents.
Efficient and Accurate Learning of Mixtures of Plackett-Luce Models
Duc Nguyen (University of Pennsylvania), Anderson Y. Zhang (University of Pennsylvania)
Recommendation SystemComputational EfficiencyTabular
🎯 What it does: This paper proposes a new Spectral EM algorithm for efficiently and accurately learning mixtures of the Plackett-Luce model.
Efficient Answer Enumeration in Description Logics with Functional Roles
Carsten Lutz (Leipzig University), Marcin Przybyłko (Leipzig University)
🎯 What it does: This paper studies the efficient enumeration of answers to ontology-mediated queries (OMQ) in description logic with functional roles (ELIHF). It presents a linear preprocessing and constant delay enumeration method using general models and Horn formulas under the conditions of connectivity and acyclicity, and proves that this method is optimal when it is acyclic and satisfies the connectivity condition.
Efficient Distributed Inference of Deep Neural Networks via Restructuring and Pruning
Afshin Abdi (Georgia Institute of Technology), Tushar Krishna (Georgia Institute of Technology)
OptimizationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A hierarchical reconstruction and pruning method called RePurpose is proposed for efficiently executing trained deep neural networks in a multi-node distributed environment.
Efficient Distribution Similarity Identification in Clustered Federated Learning via Principal Angles between Client Data Subspaces
Saeed Vahidian (University of California San Diego), Bill Lin (University of Central Florida)
Federated LearningComputational EfficiencyImage
🎯 What it does: This paper proposes a one-shot clustering federated learning framework called PACFL, which utilizes each client to perform truncated SVD on local data to extract a set of principal vectors, and measures the similarity of data distribution through the principal angles between the principal subspaces, thereby performing clustering on the server side.
Efficient Edge-Preserving Multi-View Stereo Network for Depth Estimation
Wanjuan Su (Huazhong University of Science and Technology), Wenbing Tao (Huazhong University of Science and Technology)
Depth EstimationPoint Cloud
🎯 What it does: An efficient multi-view stereo network EPNet that preserves edge details is proposed.
Efficient Embeddings of Logical Variables for Query Answering over Incomplete Knowledge Graphs
Dingmin Wang (University of Oxford), Bernardo Cuenca Grau (University of Oxford)
Computational EfficiencyKnowledge DistillationRepresentation LearningGraph Neural NetworkNeural Radiance FieldGraph
🎯 What it does: A new method is proposed to address the problem of answering complex first-order logic queries on incomplete knowledge graphs.
Efficient End-to-End Video Question Answering with Pyramidal Multimodal Transformer
Min Peng (Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences), Xiang-Dong Zhou (Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences)
RetrievalComputational EfficiencyTransformerVideoTextMultimodalityBenchmark
🎯 What it does: An end-to-end video question answering model PMT is proposed, which interacts with text through a pyramid multimodal Transformer on multi-scale spatio-temporal features to achieve semantic reasoning of video content.
Efficient Enumeration of Markov Equivalent DAGs
Marcel Wienöbst (University of Lübeck), Maciej Liskiewicz (University of Lübeck)
Graph Neural NetworkGraph
🎯 What it does: This paper designs and implements a method that can enumerate all DAGs in all Markov equivalence classes with linear time delay, and provides an enumeration sequence that achieves a structural Hamming distance of no more than three.
Efficient Exploration in Resource-Restricted Reinforcement Learning
Zhihai Wang (University of Science and Technology of China), Jie Wang (Institute of Artificial Intelligence)
Reinforcement Learning
🎯 What it does: A resource-constrained exploration framework is proposed in reinforcement learning with limited resources, and a Resource-Aware Exploration Bonus (RAEB) is designed to allow agents to seek novel states while conserving resources.
Efficient Explorative Key-Term Selection Strategies for Conversational Contextual Bandits
Zhiyong Wang (Chinese University of Hong Kong), John C. S. Lui
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes a new dialog-based linear UCB framework called ConLinUCB, and based on it, designs two exploratory keyword selection strategies, ConLinUCB-BS and ConLinUCB-MCR, for more efficient extraction of user preferences.
Efficient Extraction of EL-Ontology Deductive Modules
Hui Yang (CNRS Universite Paris-Saclay), Nicole Bidoit (CNRS Universite Paris-Saclay)
OptimizationComputational EfficiencyGraph
🎯 What it does: This paper proposes a SAT algorithm based on forest structure, ForMod, for efficiently extracting pseudo-minimal modules and complete modules of EL ontologies.
Efficient Gradient Approximation Method for Constrained Bilevel Optimization
Siyuan Xu (Pennsylvania State University), Minghui Zhu (Pennsylvania State University)
OptimizationTabularBiomedical Data
🎯 What it does: A gradient approximation method for constrained bilevel optimization is proposed, which determines the descent direction by approximating the Clarke generalized derivative;
Efficient Image Captioning for Edge Devices
Ning Wang (Huawei Inc), Linlin Li (Huawei Inc)
Object DetectionGenerationComputational EfficiencyKnowledge DistillationTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Implement lightweight image captioning on mobile edge devices, proposing the LightCap model;
Efficient Mirror Detection via Multi-Level Heterogeneous Learning
Ruozhen He (City University of Hong Kong), Rynson W.H. Lau (City University of Hong Kong)
Object DetectionSegmentationComputational EfficiencyConvolutional Neural NetworkImageBenchmark
🎯 What it does: An efficient mirror detection network HetNet is proposed, utilizing multi-layer heterogeneous learning.
Efficient Top-K Feature Selection Using Coordinate Descent Method
Lei Xu (Northwestern Polytechnical University), Xuelong Li (Northwestern Polytechnical University)
OptimizationTabular
🎯 What it does: This paper proposes a non-parametric coordinate descent framework CD-LSR for efficiently solving the feature selection problem with ℓ₂⁰-norm constraints.
Electrophysiological Brain Source Imaging via Combinatorial Search with Provable Optimality
Guihong Wan (Massachusetts General Hospital), Feng Liu (University of Texas at Dallas)
OptimizationTime SeriesBiomedical DataElectrocardiogram
🎯 What it does: A combinatorial search framework based on graph search (ESIA*) is proposed for non-invasive EEG/MEG source imaging, and provable optimality is provided within this framework.
Eliciting Structural and Semantic Global Knowledge in Unsupervised Graph Contrastive Learning
Kaize Ding (Arizona State University), Huan Liu (Arizona State University)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A S3-CL framework is proposed for self-supervised learning of graph node representations through structural and semantic comparisons.
Eliminating the Impossible, Whatever Remains Must Be True: On Extracting and Applying Background Knowledge in the Context of Formal Explanations
Jinqiang Yu (Monash University), Joao Marques-Silva (VMWare Research)
Explainability and InterpretabilityTabularBenchmark
🎯 What it does: By automatically extracting high-confidence rules from data and using them as background knowledge, combined with formal reasoning techniques, more concise and credible 'why' (AXp) and 'why not' (CXp) explanations are generated;
EMEF: Ensemble Multi-Exposure Image Fusion
Renshuai Liu (Xiamen University), Xuan Cheng (Xiamen University)
Image TranslationOptimizationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: An integrated multi-exposure image fusion method EMEF is proposed, which mimics and optimizes the fusion results of various existing MEF methods to obtain a better fused image than a single method.
Emergence of Punishment in Social Dilemma with Environmental Feedback
Zhen Wang (Northwestern Polytechnical University), Shuyue Hu (Kyushu University)
🎯 What it does: A third-party punishment public goods game evolutionary model incorporating environmental feedback is proposed and analyzed, studying the conditions and dynamics of the co-evolution of punishment and cooperation.
Emergent Quantized Communication
Boaz Carmeli (Technion Israel Institute of Technology), Yonatan Belinkov (Technion Israel Institute of Technology)
Convolutional Neural NetworkReinforcement LearningImageText
🎯 What it does: This paper proposes a quantization-based discrete communication framework for emergent communication tasks in multi-agent systems.
Enabling Knowledge Refinement upon New Concepts in Abductive Learning
Yu-Xuan Huang (Nanjing University), Zhi-Hua Zhou (Nanjing University)
Tabular
🎯 What it does: The ABL nc framework is proposed, which combines new concept detection, rule learning, and conflict resolution to achieve adaptive updates of the knowledge base, and enhances the performance of the perception model through inductive logic programming and reasoning.
End-to-End Deep Reinforcement Learning for Conversation Disentanglement
Karan Bhukar (IBM Research), Ajay Gupta (IBM Research)
Recurrent Neural NetworkGraph Neural NetworkTransformerReinforcement LearningText
🎯 What it does: An end-to-end reinforcement learning framework is proposed to directly optimize the multi-party dialogue segmentation task using global metrics;
End-to-End Entity Linking with Hierarchical Reinforcement Learning
Lihan Chen (Fudan University), Yanghua Xiao (Fudan University)
Recurrent Neural NetworkTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Modeling the entity linking task as a hierarchical reinforcement learning framework, first detecting mentions through high-level decisions, and then completing entity disambiguation with a low-level generative strategy;
End-to-End Learning for Optimization via Constraint-Enforcing Approximators
Rares Cristian (Massachusetts Institute of Technology), Ioannis Spantidakis (Massachusetts Institute of Technology)
Optimization
🎯 What it does: A ProjectNet neural network architecture is designed to achieve end-to-end learning by approximating projections to solve linear optimization problems, embedding it into a prediction-optimization pipeline.
End-to-End Zero-Shot HOI Detection via Vision and Language Knowledge Distillation
Mingrui Wu (Xiamen University), Xiaoshuai Sun (Xiamen University)
Object DetectionKnowledge DistillationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes an end-to-end zero-shot human-object interaction detection framework EoID, which utilizes CLIP's visual-language knowledge for distillation.
Energy-Motivated Equivariant Pretraining for 3D Molecular Graphs
Rui Jiao (Tsinghua University), Yang Liu (Tencent)
Representation LearningDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: A novel energy-driven equivariant pre-training framework (3D-EMGP) is proposed, achieving unsupervised pre-training of 3D molecular structures through the construction of equivariant force prediction and layer noise scale classification tasks for 3D molecular graphs.
Enhanced Multi-Relationships Integration Graph Convolutional Network for Inferring Substitutable and Complementary Items
Huajie Chen (Meituan Group), Yuanyuan Qiao (Beijing University of Posts and Telecommunications)
Recommendation SystemGraph Neural NetworkMixture of ExpertsGraph
🎯 What it does: This study investigates how to simultaneously infer the substitutable and complementary relationships of products through Graph Neural Networks (GNNs), and further identify the strength of these relationships to enhance the accuracy and interpretability of recommendation systems.
Enhanced Tensor Low-Rank and Sparse Representation Recovery for Incomplete Multi-View Clustering
Chao Zhang (Nanjing University), Chunlin Chen (Nanjing University)
Multimodality
🎯 What it does: This paper proposes an IMVC method called ETLSRR based on tensor low-rank and sparse representation recovery, which completes multi-view clustering by jointly learning incomplete similarity graphs and complete tensors.
Enhancing the Antidote: Improved Pointwise Certifications against Poisoning Attacks
Shijie Liu (University of Melbourne), Benjamin I. P. Rubinstein (University of Melbourne)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Proposes point-level provable robustness defense in the context of data poisoning attacks, ensuring projection invariance for individual test samples using differential privacy and sampling Gaussian mechanisms.
Entity-Agnostic Representation Learning for Parameter-Efficient Knowledge Graph Embedding
Mingyang Chen (Zhejiang University), Huajun Chen (University of Edinburgh)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes an entity-agnostic representation learning framework (EARL) that addresses the problem of linear growth in parameter count with the number of entities in knowledge graph embedding by encoding distinguishable information of entities (edge relationships, k-nearest reserved entities, multi-hop neighbors) instead of storing vectors for each entity individually.
Entropy Regularization for Population Estimation
Ben Chugg (Carnegie Mellon University), Daniel E. Ho (Stanford University)
OptimizationReinforcement LearningTabular
🎯 What it does: A sampling strategy based on entropy regularization and KL divergence is proposed to simultaneously maximize rewards and accurately estimate the overall mean in the optimization-estimation structured multi-armed bandit problem.
Epistemic Disjunctive Datalog for Querying Knowledge Bases
Gianluca Cima (Sapienza University of Rome), Antonella Poggi (Sapienza University of Rome)
Knowledge Distillation
🎯 What it does: The Data K log query language is proposed, allowing the use of branching Datalog rules with cognitive operators on knowledge bases to achieve recursive and non-monotonic queries.
Equi-Tuning: Group Equivariant Fine-Tuning of Pretrained Models
Sourya Basu (IBM Research), Payel Das (IBM Research)
ClassificationGenerationConvolutional Neural NetworkRecurrent Neural NetworkTransformerSupervised Fine-TuningImageText
🎯 What it does: A method called Equi-Tuning is proposed, which transforms pre-trained models into group equivariant models and validates its effectiveness in image classification, speech synthesis, and fairness tasks in natural language generation.
Equity Promotion in Public Transportation
Anik Pramanik (New Jersey Institute of Technology), Yifan Xu (Southeast University)
OptimizationTabular
🎯 What it does: An optimization model is proposed to study how to integrate traditional public transportation infrastructure with ride-hailing services to promote social equity, particularly to provide better public transport access for low-income families.
ESL-SNNs: An Evolutionary Structure Learning Strategy for Spiking Neural Networks
Jiangrong Shen (Zhejiang University), Huajin Tang (Zhejiang University)
Spiking Neural NetworkImageTime Series
🎯 What it does: A sparse training framework for SNNs called ESL-SNNs is proposed, which utilizes dynamic pruning and regeneration mechanisms to achieve adaptive evolution of sparse network structures from scratch.
ESPT: A Self-Supervised Episodic Spatial Pretext Task for Improving Few-Shot Learning
Yi Rong (Wuhan University of Technology), Shengwu Xiong (Wuhan University of Technology)
ClassificationMeta LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a self-supervised episodic spatial pretext task (ESPT) that enhances few-shot learning performance in image classification by utilizing the consistency of local spatial relationships between original images and images transformed by random geometric transformations in few-shot episodes.
Estimating Average Causal Effects from Patient Trajectories
Dennis Frauen (LMU Munich), Stefan Feuerriegel (LMU Munich)
Recurrent Neural NetworkTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: An end-to-end deep learning model, DeepACE, was designed and implemented to estimate the average causal effect (ACE) over time from follow-up electronic medical records (temporal patient trajectories). This was achieved through joint learning of iterative G-computation and sequential targeting to obtain doubly robust and asymptotically efficient estimates.
Estimating Reflectance Layer from a Single Image: Integrating Reflectance Guidance and Shadow/Specular Aware Learning
Yeying Jin (National University of Singapore), Robby T. Tan (National University of Singapore)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a two-stage network for estimating the diffuse layer that removes shadows and highlights from a single image. Initially, an initial reflection layer is obtained through constraints based on shadow/highlight-free images, and then a Shadow/Specular-Aware network with a classifier is used to adaptively focus attention on and refine the shadow/highlight regions.
Estimating Regression Predictive Distributions with Sample Networks
Ali Harakeh (University of Toronto), Liam Paull (University of Toronto)
OptimizationMultimodalityTabular
🎯 What it does: A neural network architecture named SampleNet is proposed, which represents the probability distribution of regression predictions using samples instead of parameterized distributions, and is trained using Energy Score.
Estimating Treatment Effects from Irregular Time Series Observations with Hidden Confounders
Defu Cao (University of Southern California), Yan Liu (University of Southern California)
Recurrent Neural NetworkTime SeriesBiomedical DataElectronic Health RecordsStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes a Lipschitz regularized neural control differential equation (LipCDE) model for estimating individualized treatment effects (ITE) under conditions of hidden confounding and irregularly observed time series;
Evaluating and Improving Interactions with Hazy Oracles
Stephan J. Lemmer (University of Michigan), Jason J. Corso (University of Michigan)
Object TrackingVideoText
🎯 What it does: This paper proposes a reasoning delay framework for AI systems with noisy human inputs and introduces the Deferred Error Volume (DEV) metric to measure the trade-off between error and human cost.
Evaluating Epistemic Logic Programs via Answer Set Programming with Quantifiers
Wolfgang Faber (University of Klagenfurt), Michael Morak (University of Klagenfurt)
OptimizationTabularBenchmark
🎯 What it does: Compiles epistemic logic programs (ELP) into quantified answer set programs ASP(Q) to use ASP(Q) solvers for determining world views.
Event Process Typing via Hierarchical Optimal Transport
Bo Zhou (University of Chinese Academy of Sciences), Jun Zhao (University of Chinese Academy of Sciences)
TransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes an event process annotation model based on hierarchical optimal transport, which combines event hierarchy and label hierarchy to infer action and object labels of the process.
Evidential Conditional Neural Processes
Deep Shankar Pandey (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
Anomaly DetectionOptimizationImageTime Series
🎯 What it does: This paper proposes a condition neural process model based on evidential learning (ECNP) for fine-grained uncertainty decomposition in few-shot learning.
Experimental Observations of the Topology of Convolutional Neural Network Activations
Emilie Purvine (Pacific Northwest National Laboratory), Youjia Zhou (Scientific Computing and Imaging Institute and School of Computing, University of Utah)
ClassificationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper utilizes topological data analysis methods (persistent homology and Mapper) to model and visualize the activation vectors of hidden layers in convolutional neural networks, thereby revealing the relationship between the internal representations of the network and the classification results.
Explaining (Sarcastic) Utterances to Enhance Affect Understanding in Multimodal Dialogues
Shivani Kumar (Indraprastha Institute of Information Technology Delhi), Tanmoy Chakraborty (Indian Institute of Technology Delhi)
RecognitionGenerationTransformerTextMultimodality
🎯 What it does: Proposes the MOSES model, which utilizes multimodal code-mixed dialogue generation for sarcastic explanations and is used for sentiment, humor, and sarcasm detection;
Explaining Model Confidence Using Counterfactuals
Thao Le (University of Melbourne), Liz Sonenberg (University of Melbourne)
Explainability and InterpretabilityTabular
🎯 What it does: This study investigates the use of counterfactual explanations to elucidate the confidence of machine learning models and evaluates their impact on user understanding and trust through two rounds of user research.
Explaining Random Forests Using Bipolar Argumentation and Markov Networks
Nico Potyka (Imperial College London), Francesca Toni (Imperial College London)
Explainability and InterpretabilityTabular
🎯 What it does: This paper proposes the use of a bipolar argumentation framework and Markov networks to explain random forests, constructing interpretable models.
Explicit Invariant Feature Induced Cross-Domain Crowd Counting
Yiqing Cai (East China Normal University), Gaoqi He (East China Normal University)
Domain AdaptationGraph Neural NetworkImage
🎯 What it does: The IF-CKT framework is proposed for cross-domain crowd counting, explicitly separating and aligning domain-invariant and domain-specific features, and enhancing target domain adaptability through graph neural networks and pseudo-labeling.
Exploit Domain-Robust Optical Flow in Domain Adaptive Video Semantic Segmentation
Yuan Gao (University of Science and Technology of China), Junjie Li (University of Science and Technology of China)
SegmentationDomain AdaptationOptical FlowVideo
🎯 What it does: This paper studies domain adaptive video semantic segmentation and proposes to improve the model using domain-robust optical flow through segmentation-flow consistency supervision.
Exploiting Multiple Abstractions in Episodic RL via Reward Shaping
Roberto Cipollone (Universita degli Studi di Roma La Sapienza), Fabio Patrizi (Banca d'Italia)
Reinforcement Learning
🎯 What it does: This paper proposes a reward shaping method using multi-layer abstract MDPs, which significantly improves sample efficiency by guiding finer-grained RL learning through the optimal value function learned at the abstract level.
Exploration via Epistemic Value Estimation
Simon Schmitt (DeepMind), Hado van Hasselt (University College London)
Reinforcement LearningTabular
🎯 What it does: A scalable empirical value uncertainty estimation (EVE) method is proposed and applied to Q-Learning for efficient exploration.
Exploratory Inference Learning for Scribble Supervised Semantic Segmentation
Chuanwei Zhou, Jian Yang (Nanjing University of Science and Technology)
SegmentationConvolutional Neural NetworkReinforcement LearningContrastive LearningImage
🎯 What it does: This paper proposes an Exploratory Inference Learning (EIL) framework that utilizes a continuous policy searcher and evaluator for adaptive label exploration in unlabeled regions, significantly improving semantic segmentation performance based on scribble annotations through contrastive reward-guided learning for the segmenter.
Exploring CLIP for Assessing the Look and Feel of Images
Jianyi Wang (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
ClassificationRecognitionTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: Using the CLIP model, the quality and abstract perception of images are directly evaluated through adversarial prompt comparison and the removal of positional embeddings.
Exploring Faithful Rationale for Multi-Hop Fact Verification via Salience-Aware Graph Learning
Jiasheng Si (Southeast University), Deyu Zhou (Southeast University)
Explainability and InterpretabilityGraph Neural NetworkTextTabular
🎯 What it does: A multi-hop fact verification explanation method based on graph convolutional networks is proposed, utilizing learnable sparse graph perturbations to achieve subgraph extraction for obtaining trustworthy rationales.
Exploring Non-target Knowledge for Improving Ensemble Universal Adversarial Attacks
Juanjuan Weng (Xiamen University), Shaozi Li (Xiamen University)
Knowledge DistillationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A min-max training framework is proposed that only considers KL loss of non-target classes and adaptive weights, enhancing the transferability of Universal Adversarial Perturbations (UAP).
Exploring Self-Distillation Based Relational Reasoning Training for Document-Level Relation Extraction
Liang Zhang (Xiamen University), Yidong Chen (Xiamen University)
Knowledge DistillationRepresentation LearningTransformerSupervised Fine-TuningTextBiomedical Data
🎯 What it does: In document-level relation extraction, a reasoning module that combines four reasoning patterns and a self-distillation training framework is proposed to explicitly model the relationship reasoning process.
Exploring Stochastic Autoregressive Image Modeling for Visual Representation
Yu Qi (Tsinghua University), Wei Li (SenseTime Research)
Object DetectionSegmentationRepresentation LearningTransformerImage
🎯 What it does: This paper proposes a self-regressive image modeling method based on random sequences, called SAIM, which achieves visual self-supervised pre-training by randomly permuting patch sequences and using a parallel encoder-decoder architecture.
Exploring Stroke-Level Modifications for Scene Text Editing
Yadong Qu (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
Image TranslationDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the MOSTEL framework, which can replace and edit scene text while keeping the background texture unchanged, and trains using both synthetic pairs and unlabelled real images through semi-supervised mixed learning.
Exploring Temporal Information Dynamics in Spiking Neural Networks
Youngeun Kim (Yale University), Priyadarshini Panda (Technology Innovation Institute)
ClassificationOptimizationComputational EfficiencySpiking Neural NetworkImage
🎯 What it does: This paper analyzes the phenomenon of Temporal Information Concentration in SNN training by measuring the dynamic Fisher information of temporal information.
Exploring the Interaction between Local and Global Latent Configurations for Clustering Single-Cell RNA-Seq: A Unified Perspective
Nairouz Mrabah (University of Quebec at Montreal), Abdoulaye Banire Diallo (University of Quebec at Montreal)
ClassificationRepresentation LearningGraph Neural NetworkAuto EncoderBiomedical Data
🎯 What it does: A single-cell RNA-seq clustering method called scTPF based on graph autoencoders is proposed, which gradually adjusts pseudo-supervised and self-supervised tasks by utilizing the interaction between local and global latent spaces to achieve high-quality cell clustering.
Exploring Tuning Characteristics of Ventral Stream’s Neurons for Few-Shot Image Classification
Lintao Dong (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
ClassificationSegmentationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper simulates the shape and color neurons in the visual pathway from lower to upper layers (V1→V2→V4) using a computer model, and incorporates them as hierarchical feature regularization into a few-shot classification model. Additionally, an unsupervised foreground segmentation algorithm is designed to simulate belongingness features to further enhance the model's ability to identify foreground objects.
Exposing the Self-Supervised Space-Time Correspondence Learning via Graph Kernels
Zheyun Qin (Shandong University), Jianbing Shen (University of Macau)
RecognitionSegmentationGraph Neural NetworkContrastive LearningVideo
🎯 What it does: Proposes the VideoHiGraph framework, which utilizes self-supervised graph kernel learning to generate hidden graphs and perform subgraph matching and node-level temporal consistency by capturing the spatial-temporal correspondence in videos.
Expressive Optimal Temporal Planning via Optimization Modulo Theory
Stefan Panjkovic (Fondazione Bruno Kessler), Andrea Micheli (Fondazione Bruno Kessler)
OptimizationSequentialBenchmark
🎯 What it does: An optimization model theory (OMT)-based temporal planning encoding is proposed to solve optimal temporal plans (minimizing completion time or action cost).
Extracting Low-/High- Frequency Knowledge from Graph Neural Networks and Injecting It into MLPs: An Effective GNN-to-MLP Distillation Framework
Lirong Wu (Westlake University), Stan Z. Li (Zhejiang University)
Knowledge DistillationGraph Neural NetworkGraph
🎯 What it does: This paper studies GNN-to-MLP knowledge distillation, proposing to separate the low-frequency and high-frequency knowledge learned by the pre-trained GNN and inject them into the MLP respectively to enhance model performance.
Extracting Semantic-Dynamic Features for Long-Term Stable Brain Computer Interface
Tao Fang (Zhejiang University), Gang Pan (Zhejiang University)
Domain AdaptationRecurrent Neural NetworkGenerative Adversarial NetworkBiomedical Data
🎯 What it does: This paper proposes an unsupervised domain adaptation method for the recalibration of long-term stable brain-machine interfaces.
Facility Location Games with Entrance Fees
Mengfan Ma (University of Electronic Science and Technology of China), Bakh Khoussainov (University of Electronic Science and Technology of China)
Optimization
🎯 What it does: This paper introduces location-dependent entry fees in facility location games, studying the strategy-proof mechanism under this new model and providing deterministic and randomized mechanisms for both single and double facility scenarios.
FacT: Factor-Tuning for Lightweight Adaptation on Vision Transformer
Shibo Jie (Peking University), Zhi-Hong Deng (Peking University)
CompressionOptimizationTransformerSupervised Fine-TuningImage
🎯 What it does: A parameter-efficient fine-tuning method called Factor-Tuning (FacT) is proposed for pre-trained Vision Transformers, which updates only a small set of factors instead of the complete weights.
Factual and Informative Review Generation for Explainable Recommendation
Zhouhang Xie (University of California), Bodhisattwa Prasad Majumder (University of California)
Recommendation SystemExplainability and InterpretabilityTransformerTextRetrieval-Augmented Generation
🎯 What it does: The PRAG model is proposed, which combines a personalized retriever with a question-answering reader, utilizing historical reviews to generate factual and diverse recommendation explanations.
Fair Division with Prioritized Agents
Xiaolin Bu (Shanghai Jiao Tong University), Biaoshuai Tao (Shanghai Jiao Tong University)
🎯 What it does: A new concept of fair allocation problem is proposed - the fairness of priority agents (EFPRIOR), and the properties of this concept in terms of existence and algorithmic implementation are studied, providing polynomial-time algorithms (partial allocation) under additive value functions, specific homogenous values, and general value functions.
Fair Generative Models via Transfer Learning
Christopher T.H. Teo (Singapore University of Technology and Design), Ngai-Man Cheung (Singapore University of Technology and Design)
GenerationGenerative Adversarial NetworkImage
🎯 What it does: Using a transfer learning framework to train GANs, pre-trained on a large-scale biased dataset, and then fine-tuned with a small-scale fair reference dataset to achieve fairness in generated samples.
Fair Representation Learning for Recommendation: A Mutual Information Perspective
Chen Zhao (Hefei University of Technology), Meng Wang (Hefei University of Technology)
Recommendation SystemRepresentation LearningGraph Neural NetworkContrastive LearningTabular
🎯 What it does: Proposes the FairMI framework, which utilizes a second-order mutual information objective (minimizing sensitive information MI and maximizing non-sensitive information MI) to learn user/item embeddings that are both fair and highly predictive;