Neural Information Processing: 30th International Conference, ICONIP 2023, Changsha, China, November 20–23, 2023, Proceedings, Part XII (Communications in Computer and Information Science, 1966)

The nine-volume set constitutes the refereed proceedings of the 30th International Conference on Neural Information Processing, ICONIP 2023, held in Changsha, China, in November 2023. The 1274 papers presented in the proceedings set were carefully reviewed and selected from 652 submissions. The ICONIP conference aims to provide a leading international forum for researchers, scientists, and industry professionals who are working in neuroscience, neural networks, deep learning, and related fields to share their new ideas, progress, and achievements. Preface Organization Contents – Part XII Applications PBTR: Pre-training and Bidirectional Semantic Enhanced Trajectory Recovery 1 Introduction 2 Related Works 2.1 Trajectory Recovery 2.2 Sequence-to-Sequence Models 2.3 Pre-training Embedding 3 Preliminaries 4 Methodology 4.1 Pre-training and Bidirectional Road Representation Encoder 4.2 Attentive Recurrent Decoder 4.3 Model Training 5 Experiments 5.1 Experimental Settings 5.2 Result and Analysis 5.3 Ablation Study 6 Conclusion References Event-Aware Document-Level Event Extraction via Multi-granularity Event Encoder 1 Introduction 2 Related Work 3 Methodology 3.1 Task Definition 3.2 Candidate Argument Extractor 3.3 Multi-granularity Event Encoder 3.4 Event Record Decoder 3.5 Model Training 4 Experiments 4.1 Experimental Setup 4.2 Baselines 4.3 Experiment Results 4.4 Ablation Studies 5 Conclusion References Curve Enhancement: A No-Reference Method for Low-Light Image Enhancement 1 Introduction 2 Methodology 2.1 Brightness Boost Curve 2.2 RV-Net 2.3 No-Reference Loss Functions 3 Experiment 3.1 Implementation Details 3.2 Datasets and Comparison Methods 3.3 Experiment Results 3.4 Performance Comparison 4 Conclusion References A Deep Joint Model of Multi-scale Intent-Slots Interaction with Second-Order Gate for SLU 1 Introduction 2 Related Work 3 Approach 3.1 Char-Word Channel Layer 3.2 Fusion Layer 3.3 SF-ID Interaction Layer 3.4 Joint Loss Function 4 Experiments 4.1 Datasets and Evaluation Metrics 4.2 Experimental Setting 4.3 Baseline Models 4.4 Main Results 4.5 Ablation Study 4.6 Convergence Analysis 4.7 Effect of Interation 5 Conclusion and Future Work References Instance-Aware and Semantic-Guided Prompt for Few-Shot Learning in Large Language Models 1 Introduction 2 Problem Definition 3 ISPrompt: Instance-Aware and Semantic-Based Prompt 3.1 Select Semantic Prompt 3.2 Model Formulation 4 Experiments 4.1 Datasets 4.2 Experiment Setting 4.3 Baselines 4.4 Select the Appropriate Dep-Filter for ISPrompt 4.5 Tasks Transferability 4.6 Different K-Shot Setting 5 Conclusion References Graph Attention Network Knowledge Graph Completion Model Based on Relational Aggregation 1 Introduction 2 Related Work 2.1 Translation Model 2.2 Neural Network Model 3 GANERA: Model Description 3.1 Entity Attention Fusion Layer 3.2 Relationship-Aware Function Layer 3.3 Decoder 4 Experiments 4.1 Benchmark Datasets 4.2 Training Protocol 4.3 Baselines 4.4 Results and Analysis 4.5 Ablation Study 5 Conclusion and Future Work References SODet: A LiDAR-Based Object Detector in Bird's-Eye View 1 Introduction 2 Related Work 3 Methods 3.1 Point Cloud Encoding in Bird's-Eye View 3.2 Network Architecture 3.3 Training Strategy 4 Experiments 4.1 Implementation Details 4.2 Results 5 Conclusion References Landmark-Assisted Facial Action Unit Detection with Optimal Attention and Contrastive Learning 1 Introduction 2 Related Work 3 Proposed Framework 4 Experiments 4.1 Datasets and Evaluation Metrics 4.2 Implementation Details 5 Results Analysis 6 Conclusion and Future Work References Multi-scale Local Region-Based Facial Action Unit Detection with Graph Convolutional Network 1 Introduction 2 Related Work 3 Proposed Method 3.1 Overview of MSLR-GCN 3.2 MSLR Feature Extraction 3.3 AU Relationship Modeling with GCN 3.4 Loss Function 4 Experiments 4.1 Datasets and Implementation Details 4.2 Comparisons with the State-of-the-Art 4.3 Ablation Study 5 Conclusion References CRE: An Efficient Ciphertext Retrieval Scheme Based on Encoder 1 Introduction 2 Related Work 2.1 Semantic Search over Encrypted Data 2.2 BART (Bidirectional and Auto-Regressive Transformer) 2.3 Prompt-Based Bidirectional Encoder Representation from Transformer 2.4 Hierarchical Navigable Small World Graphs 3 Problem Statement 3.1 Notation 3.2 System Model 3.3 Threat Model 3.4 Design Goal 4 Ciphertext Retrieval Based on Encoder 4.1 Generate Embedding Set 4.2 Retrieval for Documents 5 Experiment and Performance Analysis 6 Conclusion References Sentiment Analysis Based on Pretrained Language Models: Recent Progress 1 Introduction 2 Models Based on BERT 2.1 Document-Level SA 2.2 Aspect-Level SA 2.3 General SA 3 Models Based on Variants of BERT 3.1 Aspect-Level SA Based on RoBERTa 3.2 SA Based on mBERT 3.3 Document-Level SA Based on HAdaBERT 3.4 CLSA Based on IndicBERT 4 Models Based on Other PLMs 4.1 Sentence-Level SA Based on XLNet 4.2 CLSA Based on XLM-R 4.3 Aspect-Level SA Based on XLM-Align 4.4 Cross-Domain SA Based on SentiX 4.5 Sentence-Level SA Based on GPT-3 4.6 CLSA Model Based on Hybrid PLMs 5 Challenges 6 Conclusion References Improving Out-of-Distribution Detection with Margin-Based Prototype Learning 1 Introduction 2 Related Works 3 Preliminaries 4 Method 4.1 Intra-class Compactness in Hypersphere 4.2 Inter-class Dispersion 5 Experiments 5.1 Implementation Details 5.2 Comparison with State-of-the-art Methods 5.3 Further Analysis 6 Conclusion References Text-to-Image Synthesis with Threshold-Equipped Matching-Aware GAN 1 Introduction 2 ETMA-GAN for Text-to-Image Generation 2.1 Network Architecture 2.2 Inaccurate Negative Sample Filter (INSF) 2.3 Word Fine-Grained Supervisor (WFGS) 2.4 Loss Function 3 Experiment 3.1 Quantitative Evaluation 3.2 Qualitative Evaluation 3.3 Ablation Studies 4 Conclusion and Future Work References Joint Regularization Knowledge Distillation 1 Introduction 2 Related Literature 2.1 Online Knowledge Distillation 2.2 Disagreement 3 Approach 3.1 Confidence-Based Continuous Scheduler 3.2 Joint Regularization Knowledge Distillation 4 Experiments 4.1 Experiments on Benchmarks 4.2 Ablation Experiments 4.3 Visual Analytics 5 Conclusion References Dual-Branch Contrastive Learning for Network Representation Learning 1 Introduction 2 Related Work 2.1 Network Representation Learning 2.2 Graph Contrastive Learning 3 Method Overview 3.1 Problem Definition 3.2 View Generation 3.3 Dual-Branch Graph Contrastive Learning Module 4 Experiments 4.1 Experimental Results and Analysis 4.2 Ablation Experiments 5 Conclusion References Multi-granularity Contrastive Siamese Networks for Abstractive Text Summarization 1 Introduction 2 Related Work 3 Method 3.1 Multi-granularity Data Augmentation 3.2 Self-supervised Contrastive Learning 3.3 Summary Generation 3.4 Model Training 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Baselines 4.4 Experimental Results 4.5 Ablation Study 5 Conclusion References Joint Entity and Relation Extraction for Legal Documents Based on Table Filling 1 Introduction 2 Related Work 2.1 Information Extraction in Judicial Field 2.2 Joint Entity and Relation Extraction 3 Model 3.1 Encoding Layer 3.2 Legal Feature Enhancement Module 3.3 Table Filling Module 3.4 Loss Function 4 Experiment 4.1 Datasets 4.2 Experimental Setup 4.3 Experimental Results and Analysis 4.4 Experiments Based on Different Sentence Types 5 Conclusion References Dy-KD: Dynamic Knowledge Distillation for Reduced Easy Examples 1 Introduction 2 Related Work 2.1 Knowledge Distillation 2.2 Curriculum Learning 3 Method 3.1 Preliminaries 3.2 Dynamic Knowledge Distillation 3.3 Dynamic Weighted Cross Entropy 3.4 Pacing Function 4 Experiments 4.1 Datasets and Experiments Configuration 4.2 Comparison of Results 4.3 Ablation 5 Conclusion References Fooling Downstream Classifiers via Attacking Contrastive Learning Pre-trained Models 1 Introduction 2 Related Works 3 Methodology 3.1 Notations 3.2 Our Method 4 Experiments 4.1 Experimental Setting 4.2 Comparative Results 4.3 Discussion the Influence of Attacking Different Layers 5 Conclusion References Feature Reconstruction Distillation with Self-attention 1 Introduction 2 Related Work 2.1 Knowledge Distillation 2.2 Feature-Map Distillation 3 Method 3.1 Notation 3.2 Logit-Based Knowledge Distillation 3.3 Feature-Map Distillation 3.4 Feature Rectification Distillation 3.5 Feature Fusion with Self-attention 4 Experiments 4.1 Experimental Setup 4.2 Experiments of STL-10 Dataset 4.3 Experiments of CIFAR-100 Dataset 4.4 Ablation Study 5 Conclusion References DAGAN:Generative Adversarial Network with Dual Attention-Enhanced GRU for Multivariate Time Series Imputation 1 Introduction 2 Related Work 3 Preliminary 4 Model 4.1 The Overall Structure 4.2 Generator Network 4.3 Discriminator Network 5 Experiment 5.1 Datasets 5.2 Baseline 5.3 Experimental Setup 5.4 Experimental Results and Analysis 5.5 Ablation Experiment 6 Conclusion References Knowledge-Distillation-Warm-Start Training Strategy for Lightweight Super-Resolution Networks 1 Introduction 2 Related Work 2.1 Knowledge Distillation 2.2 Training Strategy 3 Method 3.1 KDWS Training Strategy 3.2 Loss Function 4 Experiments 4.1 Setup 4.2 Quantitative Results 4.3 Ablation Study 5 Conclusion References SDBC: A Novel and Effective Self-Distillation Backdoor Cleansing Approach 1 Introduction 2 Related Work 2.1 Backdoor Attack 2.2 Backdoor Defense 3 Our SDBC Approach 3.1 Defense Setup 3.2 Overview 3.3 Self-Distillation Backdoor Cleansing 4 Experiments 4.1 Experimental Setting 4.2 Effectiveness of SDBC 4.3 A Comprehensive Understanding of SDBC 5 Conclusion References An Alignment and Matching Network with Hierarchical Visual Features for Multimodal Named Entity and Relation Extraction 1 Introduction 2 Task Formulation 3 Method 3.1 Encoding Module 3.2 Image-Text Alignment 3.3 Cross-Modal Fusion 3.4 Image-Text Matching 3.5 Classifier 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Baselines 4.4 Main Results 4.5 Analysis and Discussion 4.6 Case Study 5 Related Work 6 Conclusion References Multi-view Consistency View Synthesis 1 Introduction 2 Related Work 2.1 Novel View Synthesis 2.2 Long Short-Term Memory Mechanism 3 Method 3.1 Multi-view Consistency 3.2 Overall Architecture 4 Experiments 5 Conclusion References A Reinforcement Learning-Based Controller Designed for Intersection Signal Suffering from Information Attack 1 Introduction 2 Problem Description 2.1 Application Scenarios 3 RL Signal Controller for Intersection Suffering from Information Attack 3.1 Reinforcement Learning Model Controller 3.2 Signaling Attack Model Based on State Table 4 Simulation Result 4.1 Simulation Settings 4.2 Reinforcement Learning Q-Network Training 4.3 Reinforcement Learning Control Model Results Analysis 4.4 Analysis of Signal Attack Results Based on State Table 5 Conclusion References Dual-Enhancement Model of Entity Pronouns and Evidence Sentence for Document-Level Relation Extraction 1 Introduction 2 Related Work 3 Methodology 3.1 Problem Formulation 3.2 Entity Pronouns Enhancement Module 3.3 Text Encoding Module 3.4 Document-Aware Context Embedding 3.5 Sentence-Aware Context Embedding 3.6 Fusion and Classification Module 3.7 Loss Function 4 Experiments 4.1 Setting 4.2 Main Results 4.3 Ablation Studies 5 Conclusion References Nearest Memory Augmented Feature Reconstruction for Unified Anomaly Detection 1 Introduction 2 Related Work 2.1 Reconstruction-Based Methods 2.2 Representation-Based Methods 3 Method 3.1 Problem Definition and Basic Methodology 3.2 Feature Extraction 3.3 Nearest Memory Addressing 3.4 Loss Function 3.5 Anomaly Score 4 Experiments 4.1 Experimental Setup 4.2 Results 4.3 Ablation Study 5 Conclusion References Deep Learning Based Personalized Stock Recommender System 1 Introduction 2 Related Work 3 Methodology 3.1 Preprocessing 3.2 Collaborative Filtering Approach 3.3 Content Based Approach 3.4 Hybrid Approach 3.5 Datasets and Evaluation Metrics 4 Experiments 5 Conclusion and Future Work References Feature-Fusion-Based Haze Recognition in Endoscopic Images 1 Introduction 2 Method 2.1 Features Extraction 2.2 Classification Using SVM 3 Experiments 3.1 Dataset and Experimental Settings 3.2 Classification Results with Different Kernel and Dimension 3.3 Classification Results with Different Weights of Features 4 Discussion 5 Conclusion References Retinex Meets Transformer: Bridging Illumination and Reflectance Maps for Low-Light Image Enhancement 1 Introduction 2 Related Works 2.1 Conventional Method 2.2 Learning-Based Method 3 The Proposed Method 3.1 The Model Framework 3.2 Retinex-Based Network 3.3 Transformer-Based Network 3.4 Loss Functions 4 Experiments 4.1 Datasets 4.2 Experimental Settings 4.3 Quantitative Results 4.4 Qualitative Results 4.5 Ablation Study 5 Conclusion References Make Spoken Document Readable: Leveraging Graph Attention Networks for Chinese Document-Level Spoken-to-Written Simplification 1 Introduction 2 Related Work 3 Preliminary 3.1 Problem Formulation 3.2 Graph Attention Network 4 The Proposed Model 4.1 The Encoder-Decoder Architecture 4.2 Graph Encoder 4.3 Integration of Graph Representation 5 Experiments 5.1 Datasets 5.2 Experimental Settings 5.3 Main Results and Analysis 5.4 Results on D-Wikipedia 6 Conclusion and Future Work References MemFlowNet: A Network for Detecting Subtle Surface Anomalies with Memory Bank and Normalizing Flow 1 Introduction 2 Related Work 3 Proposed Methods 3.1 Overview 3.2 Memory Bank and Feature Domain Adaptation 3.3 Flow-Based Detection 4 INSCup Dataset and Augmentation 5 Experiments 5.1 Setup 5.2 Results 5.3 Ablation Studies 6 Conclusion References LUT-LIC: Look-Up Table-Assisted Learned Image Compression 1 Introduction 2 Related Works 2.1 Learned Image Compression 2.2 Model Compression Techniques 3 Proposed Method 3.1 RF-Constrained LIC Model 3.2 Transferring Hyper Decoder to LUT 3.3 Inference Using LUT-LIC 4 Experiments 4.1 Settings 4.2 Comparison 4.3 Ablation Studies 5 Conclusion References Oil and Gas Automatic Infrastructure Mapping: Leveraging High-Resolution Satellite Imagery Through Fine-Tuning of Object Detection Models 1 Introduction 2 State of the Art 3 An Oil and Gas Infrastructure Database 4 Object Detection Algorithms Presentation 4.1 One Stage Object Detector: YOLO 4.2 Two Stage Object Detector: FASTER-RCNN 4.3 Encoder-Decoder Object Detector: DETR 4.4 Model Evaluation 5 Results 5.1 Algorithms and Models Performance Comparisons 5.2 Algorithms Pre-training Effect 5.3 Applications 6 Conclusion References AttnOD: An Attention-Based OD Prediction Model with Adaptive Graph Convolution 1 Introduction 2 Related Works 2.1 Temporal Dependencies 2.2 Spatial Dependencies 3 Methodology 3.1 Problem Definition 3.2 The Attention-Based OD Prediction Model (AttnOD) 3.3 Adaptive Graph Convolution 3.4 Temporal Self-attention 3.5 Cross Attention 3.6 Encoder-Decoder Structure 4 Experiments 4.1 Datasets 4.2 Baselines 4.3 Experimental Setups and Computation Cost 4.4 Results 4.5 Ablation Experiments 5 Conclusion References CMMix: Cross-Modal Mix Augmentation Between Images and Texts for Visual Grounding 1 Introduction 2 Proposed Approach 2.1 Mix Image 2.2 Mix Text 2.3 Filter Samples by Loss 3 Experimental Results 3.1 Datasets and Implementation Details 3.2 Visual Grounding Results 3.3 Exploration Study 3.4 Visual Analysis 4 Conclusion References A Relation-Oriented Approach for Complex Entity Relation Extraction 1 Introduction 2 Related Work 2.1 Entity Relation Extraction 2.2 Nested Entity Relation Extraction 2.3 MRC-Based Entity Relation Extraction 3 Method 3.1 Problem Definition 3.2 Query Generation 3.3 Relation Recognition 3.4 Entity Recognition 3.5 Triples Determination 4 Experiment 4.1 Dataset 4.2 Baseline Model 4.3 Implementation Details 4.4 Result and Analysis 5 Conclusion References A Revamped Sparse Index Tracker Leveraging K–Sparsity and Reduced Portfolio Reshuffling 1 Introduction 2 Background 3 SIT Based on IT4-PGD Algorithm 3.1 Logarithmic Return 3.2 The Proposed Portfolio Design 3.3 Algorithm for IT4-PGD 3.4 Algorithm for Nonconvex Projection PC(z) 3.5 Convergence Analysis of IT4-PGD 4 Numerical Experiments 4.1 Datasets and Settings 4.2 Quantitative Measurements 4.3 Performance Comparison 5 Conclusion References Anomaly Detection of Fixed-Wing Unmanned Aerial Vehicle (UAV) Based on Cross-Feature-Attention LSTM Network 1 Introduction 2 Materials and Methods 2.1 Dataset 2.2 Data Processing 2.3 Overview 2.4 Spatial-Temporal Features Processing Module 2.5 Detection Module 3 Experiment 4 Result 5 Conclusions References Spatial and Frequency Domains Inconsistency Learning for Face Forgery Detection 1 Introduction 2 Related Work 2.1 Face Forgery Detection 2.2 Vision Transformer 3 Method 3.1 Overview 3.2 Frequency Clue 3.3 Intra-frame Inconsistency Feature Extraction 3.4 Cross-Attention Feature Fusion Module 4 Experiments 4.1 Dataset 4.2 Experimental Setting 4.3 Comparison with Other Methods 4.4 Cross-Dataset Evaluation 4.5 Ablation Study 5 Conclusion References Enhancing Camera Position Estimation by Multi-view Pure Rotation Recognition and Automated Annotation Learning 1 Introduction 2 Related Work 3 Methodology 3.1 Pure Rotational Indicators in Multiple Views 3.2 Rotation-Only Anomaly Recognition Model 4 Experiments 4.1 Experimental Comparison of Four Simulated Motion Scenarios 4.2 Experimental Results in Virtual KITTI Dataset 5 Conclusion References Detecting Adversarial Examples via Classification Difference of a Robust Surrogate Model 1 Introduction 2 Related Detection Methods 3 Methodology 3.1 Analyzing Adversarial Robustness of the Initial Surrogate Model 3.2 Enhancing Adversarial Robustness of the Surrogate Model 3.3 The Proposed Detector 4 Experimental Results 4.1 Setup 4.2 Evaluation Under the Case of Un-aware Detections 4.3 Discussion of the Security 5 Conclusion References Minimizing Distortion in Steganography via Adaptive Language Model Tuning 1 Introduction 2 Proposed Framework 2.1 Problem Definition and Optimization Goal 2.2 Adaptive Language Model Tuning 2.3 Embeding and Extraction 3 Experiment and Analysis 3.1 Experiment Settings 3.2 Metrics 3.3 Evaluation Results and Analysis 4 Conclusion References Efficient Chinese Relation Extraction with Multi-entity Dependency Tree Pruning and Path-Fusion 1 Introduction 2 Related Work 2.1 Pruning Strategy 2.2 RE Dependency-Based Models 3 Methods 3.1 Pruning Strategy 3.2 The Whole Architecture of Our Model 4 Experiments 4.1 Datasets 4.2 Dependency Graph Construction 4.3 Experimental Setup 4.4 Implementation Details 5 Results and Analysis 5.1 Overall Results 5.2 The Effect of Pruned Strategy 5.3 Case Study 6 Conclusions References A Lightweight Text Classification Model Based on Label Embedding Attentive Mechanism 1 Introduction 2 Related Work 2.1 Text Classification 2.2 Label Embedding 2.3 Metric Learning 3 Model 3.1 Model Structure 3.2 Metric Learning 4 Experiments 4.1 Experimental Setup 4.2 Experimental Result 5 Conclusion References Author Index