Neural Information Processing: 30th International Conference, ICONIP 2023, Changsha, China, November 20–23, 2023, Proceedings, Part XIII (Communications in Computer and Information Science, 1967)

Preface
Organization
Contents – Part XIII
Applications
Improve Conversational Search with Multi-document Information
	1 Introduction
	2 Related Works
		2.1 Conversational Search
		2.2 LLM for Information Retrieval
	3 Method
		3.1 Preliminaries
		3.2 Multi-document Conversation Segmentation
		3.3 LLM-Based Document Summarization
		3.4 Reranker Based on Passage-Segment-Document Post-training
	4 Experiments
		4.1 Experimental Settings
		4.2 Performance Comparison
		4.3 Experimental Results
		4.4 Ablation Experiment
	5 Conclusion
	References
Recurrent Update Representation Based on Multi-head Attention Mechanism for Joint Entity and Relation Extraction
	1 Introduction
	2 Related Work
	3 Task Formulation
	4 Methodology
		4.1 Representation Construction
		4.2 Multi-head Attention Fusion Layer
		4.3 Relation Extraction
	5 Experiments
		5.1 Experiment Settings
		5.2 Experimental Results
		5.3 Number of Heads in Multi-head Attention
		5.4 Layers of Multi-head Attention Fusion Layers
	6 Conclusion
	References
Deep Hashing for Multi-label Image Retrieval with Similarity Matrix Optimization of Hash Centers and Anchor Constraint of Center Pairs
	1 Introduction
	2 Related Work
	3 The Proposed Deep Hashing Framework
		3.1 Similarity Matrix Optimization of Hash Centers
		3.2 Loss Function of Multi-label Deep Hashing Model
	4 Experiments
		4.1 Baselines and Datasets
		4.2 Experimental Details and Hyperparameter Settings
		4.3 Evaluation Criteria
		4.4 Experimental Comparison Results
	5 Conclusion
	References
MDAM: Multi-Dimensional Attention Module for Anomalous Sound Detection
	1 Introduction
	2 Related Work
		2.1 Attention Mechanism
		2.2 Analysis of Temporal and Frequency Dimension Characteristics
	3 Method
		3.1 Multi-Dimensional Attention Module
		3.2 Domain Mixture
	4 Experiments
		4.1 Experimental Setup
		4.2 Comparison of Different Outlier Detection Algorithms
		4.3 Pooling Selection
		4.4 Comparison with Other Attention Modules
		4.5 Effectiveness of Domain Mixture
		4.6 Comparison with Other Anomaly Detection Models
	5 Conclusion
	References
A Corpus of Quotation Element Annotation for Chinese Novels: Construction, Extraction and Application
	1 Introduction
	2 Related Work
	3 Corpus Construction
		3.1 Annotation Schema
		3.2 Corpus Analysis
	4 Quotation Element Extraction
	5 Applications of Quotation Elements
		5.1 Character Recognition Based on Quotation Elements
		5.2 Gender Classification Based on Quotation Elements
	6 Conclusion
	References
Decoupling Style from Contents for Positive Text Reframing
	1 Introduction
	2 Related Works
		2.1 Text Style Transfer
		2.2 Pre-trained Language Models
		2.3 Contrastive Learning
	3 Method
		3.1 Seq2Seq Text Generation
		3.2 Decoupling Style from Contents
		3.3 Preserving Primary Content
	4 Experiment
		4.1 Experimental Settings
		4.2 Results
		4.3 Ablation Study
	5 Conclusion
	References
Multi-level Feature Enhancement Method for Medical Text Detection
	1 Introduction
	2 Related Works
	3 Methodology
		3.1 Overall Network Architecture
		3.2 Efficient Feature Enhancement Module
		3.3 Multi-scale Feature Fusion Module
		3.4 Post-processing
		3.5 Deformable Convolution and Label Generation
	4 Experimental Results
		4.1 Datasets
		4.2 Implementation Details
		4.3 Ablation Study
		4.4 Comparisons with Previous Methods
	5 Conclusion
	References
Neuron Attribution-Based Attacks Fooling Object Detectors
	1 Introduction
	2 Background
	3 Methodology
		3.1 Problem Setting
		3.2 Our Approach
	4 Experiments
		4.1 Experimental Setting
		4.2 Comparison with Other Methods
		4.3 Ablation Study
	5 Conclusion
	References
DKCS: A Dual Knowledge-Enhanced Abstractive Cross-Lingual Summarization Method Based on Graph Attention Networks
	1 Introduction
	2 Related Work
	3 The Framework
		3.1 Key Clues Extraction
		3.2 Clue-Focused Graph Encoder
		3.3 Clue Encoder
		3.4 Decoder
	4 Experimental Setup
		4.1 Datasets
		4.2 Baselines
		4.3 Experimental Settings
	5 Results and Analysis
		5.1 Automatic Evaluation
		5.2 Ablation Study
		5.3 Human Judgement
		5.4 The Impact of Article Length
	6 Discussion and Conclusion
	References
A Joint Identification Network for Legal Event Detection
	1 Introduction
	2 Model Structure
		2.1 Encoding Module
		2.2 Initial Scoring Module
		2.3 Feature Extraction Module
		2.4 Aggregation Module
	3 Model Training
		3.1 Dataset
		3.2 Revising the Loss Function by Mask
		3.3 Adversarial Training
	4 Experiment
		4.1 Dataset
		4.2 Baseline
		4.3 Evaluation Criteria
		4.4 Setting
		4.5 Benchmark Experiment
		4.6 Ablation Experiment
		4.7 Benchmark with ChatGPT
		4.8 Competition Results
	5 Related Works
		5.1 Generative-Based Approach
		5.2 Classification-Based Approach
		5.3 Graph Neural Network-Based Approach
		5.4 Span-Based Approach
	6 Conclusion
	References
YOLO-D: Dual-Branch Infrared Distant Target Detection Based on Multi-level Weighted Feature Fusion
	1 Introduction
	2 Method
		2.1 Model Architecture
		2.2 Contour Feature Extraction
		2.3 Multilevel Feature Fusion
	3 Experiments
		3.1 Datasets and Evaluation Metrics
		3.2 Experiment Setup
		3.3 Experimental Data Results
		3.4 Experimental Visual Results
		3.5 Ablation and Validation Experiments
	4 Conclusion
	References
Graph Convolutional Network Based Feature Constraints Learning for Cross-Domain Adaptive Recommendation
	1 Introduction
	2 Related Work
		2.1 Cross-Domain Recommendation
		2.2 Domain Adaptation
	3 Methodology
		3.1 Problem Formulation
		3.2 CDAR-FCL Overview
		3.3 Graph Embedding Layer
		3.4 Feature Constraint
		3.5 Feature Combination
	4 Experiments
		4.1 Experimental Settings
		4.2 Effect of Hyper-parameter
		4.3 Comparison
		4.4 Visualization
		4.5 Ablation Study
	5 Conclusion
	References
A Hybrid Approach Using Convolution and Transformer for Mongolian Ancient Documents Recognition
	1 Introduction
	2 Related Work
	3 The Proposed Approach
	4 Experimental Results
		4.1 Dataset and Baseline
		4.2 Parameter Settings
		4.3 Results and Analysis
	5 Conclusion
	References
Incomplete Multi-view Subspace Clustering Using Non-uniform Hyper-graph for High-Order Information
	1 Introduction
	2 Background
		2.1 Multi-View Subspace Clustering Preliminaries
		2.2 Tensor Preliminaries
		2.3 Hypergraph Preliminaries
	3 The Proposed NUHG-IMSC Method
	4 Experiment and Results
		4.1 Data Sets
		4.2 Methods
		4.3 Quantified Experimental Results and Variable Visualizations
	5 Conclusions
	References
Deep Learning-Empowered Unsupervised Maritime Anomaly Detection
	1 Introduction
	2 Related Work
		2.1 Data-Driven Approaches for Maritime Anomaly Detection
		2.2 Rule-Based Approaches for Maritime Anomaly Detection
		2.3 Hybrid Approaches for Maritime Anomaly Detection
	3 Fast GAN-Based Anomalous Vessel Trajectory Detection
		3.1 Problem Definition
		3.2 Generation of Vessel Trajectory Images
		3.3 Unsupervised Learning of Normal Anatomical Variability
		3.4 Detection of Anomalous Vessel Trajectory
	4 Experiments
		4.1 Data Set and Baselines
		4.2 Experimental Evaluation
	5 Conclusions
	References
Hazardous Driving Scenario Identification with Limited Training Samples
	1 Introduction
	2 Related Work
		2.1 Hazardous Driving Scenario Identification
		2.2 Motion Profile Based Models
	3 Data Operation
		3.1 Motion Profile Generation
	4 Methodology
		4.1 Data Augmentation
		4.2 Model Architecture
	5 Experiments and Results
		5.1 Experiment Settings
		5.2 Effect of Data Augmentation
		5.3 Comparison with Existing Methods
		5.4 Impact of Mixed Sample Ratio on Model Performance
	6 Conclusion and Future Work
	References
Machine Unlearning with Affine Hyperplane Shifting and Maintaining for Image Classification
	1 Introduction
	2 Method
		2.1 Preliminaries
		2.2 Affine Hyperplane Unlearning
		2.3 Affine Hyperplane Maintaining
	3 Experiments
		3.1 Experimental Settings
		3.2 Experimental Results
		3.3 Ablation Study
	4 Conclusions
	References
An Interpretable Vulnerability Detection Framework Based on Multi-task Learning
	1 Introduction
	2 Related Work
	3 Methodology
		3.1 Overview of Our Detection Model
		3.2 Encoder: Features Extraction
		3.3 Objective Task
		3.4 Training
	4 Evaluation and Analysis
		4.1 Dataset
		4.2 Experimental Results
	5 Conclusion
	References
Co-GAN: A Text-to-Image Synthesis Model with Local and Integral Features
	1 Introduction
	2 Related Work
	3 Method
		3.1 Overview
		3.2 Local Feature Enhancement Module
		3.3 Integral Structural Maintenance Module
		3.4 Loss Function
	4 Experiments and Discussion
		4.1 Set up
		4.2 Experiments and Discussion
	5 Conclusion
	References
Graph Contrastive ATtention Network for Rumor Detection
	1 Introduction
	2 Preliminary Knowledge
	3 Graph Attentive Self-supervised Learning Model
		3.1 Rumor Identification Network
		3.2 Graph Contrastive Learning Network
		3.3 Model Fusion
	4 Experiments
		4.1 Datasets
		4.2 Baselines
		4.3 Experimental Results and Analysis
		4.4 Early Detection
		4.5 Ablation Study
	5 Conclusion and Future Work
	References
E3-MG: End-to-End Expert Linking via Multi-Granularity Representation Learning
	1 Introduction
	2 Related Work
		2.1 Name Entity Linking
		2.2 Representation Learning
	3 Methodology
		3.1 Problem Definitions
		3.2 Fine-Grained Linkage via Cross Attention
		3.3 Expert Coherence via Knowledge Distillation
		3.4 Multi-granularity Representation Learning Framework
	4 Experiment
		4.1 Datasets
		4.2 Experiment and Analysis
	5 Conclusion
	References
TransCenter: Transformer in Heatmap and a New Form of Bounding Box
	1 Introduction
	2 Related Work
		2.1 Heatmap in Human Pose Estimation
		2.2 Heatmap in Object Detection
		2.3 Transformer with Heatmap
	3 Method
		3.1 Model Structrue
		3.2 Bounding Box in Heatmap Form
		3.3 Offset
		3.4 Label Assignment
		3.5 Loss Function
	4 Experiment
		4.1 Lower Limit of Model
		4.2 Upper Limit of Model
	5 Conclusion
	References
Causal-Inspired Influence Maximization in Hypergraphs Under Temporal Constraints
	1 Introduction
	2 Related Work
		2.1 Influence Maximization
		2.2 Causal Representation Learning
	3 Problem Formulation
	4 Proposed Framework
		4.1 Latency Aware Contact Process on Causal Independent Cascading Model
		4.2 Causal-Inspired Cost-Effective Balanced Selection Algorithm
	5 Experiments
		5.1 Experimental Settings
		5.2 Results and Analysis
	6 Conclusions and Future Work
	References
Enhanced Generation of Human Mobility Trajectory with Multiscale Model
	1 Introduction
	2 Related Work
	3 Methods
		3.1 Problem Definition and Basic Setting
		3.2 Generator
		3.3 Discriminator
		3.4 Model Training
	4 Experiments
		4.1 Datasets and Evaluation Metrics
		4.2 Analysis of Pre-training
		4.3 Performance Comparison
		4.4 Ablation Study
		4.5 Application: Epidemic Spreading Simulation
	5 Conclusion
	References
SRLI: Handling Irregular Time Series with a Novel Self-supervised Model Based on Contrastive Learning
	1 Introduction
	2 Related Work
		2.1 Self-supervised Learning for Time-Series
		2.2 Representation Learning on Irregular Sampled Time-Series
	3 Method
		3.1 Problem Definition
		3.2 Model Architecture
		3.3 Irregular Augmentation
		3.4 IrregTrans Contrasting
		3.5 GenDiscrim
	4 Experiment
		4.1 Time Series Classification
		4.2 ISMTS Forecasting
		4.3 Forecasting on Industrial Dataset
	5 Conclusion
	References
Multimodal Event Classification in Social Media
	1 Introduction
	2 Related Work
	3 Methodology
		3.1 Task Definition
		3.2 Overview
		3.3 CLIP-Based Text Feature Extraction
		3.4 CLIP-Based Image Feature Extraction
		3.5 Transformer-Based Multimodal Fusion
		3.6 Late Fusion Strategy
	4 Experimentation
		4.1 Experimental Settings
		4.2 Baselines and Overall Performance
		4.3 Effect of Different Layers of Transformer
		4.4 Error Analysis
	5 Conclusion
	References
ADV-POST: Physically Realistic Adversarial Poster for Attacking Semantic Segmentation Models in Autonomous Driving
	1 Introduction
	2 Related Work
		2.1 Real-Time Semantic Segmentation Networks
		2.2 Adversarial Attacks
	3 Methodology
		3.1 Preliminaries
		3.2 Adversarial Poster Attacks Overview
		3.3 Adversarial Poster Generation
		3.4 Loss Function
	4 Experiment
		4.1 Experimental Setup
		4.2 Digital Domain Experiment
		4.3 Real World Experiment
	5 Conclusion
	References
Uformer++: Light Uformer for Image Restoration
	1 Introduction
	2 Materials and Methods
		2.1 Review Uformer
		2.2 Uformer++
		2.3 NAFFA
		2.4 Loss Function
	3 Results
		3.1 Experimental Setup
		3.2 Image Denoising
		3.3 Image Deblurring
	4 Ablation Study
	5 Conclusion
	References
Can Language Really Understand Depth?
	1 Introduction
	2 Related Work
		2.1 Vision-Language Modeling
		2.2 Monocular Depth Estimation
	3 Method
		3.1 Basic Depth Estimators
		3.2 Global Context-Guided Depth Fusion
		3.3 Depth Distillation
	4 Experiment
		4.1 Datasets and Metrics
		4.2 Implementation Detail
		4.3 Main Results
		4.4 Ablation on Basic Depth Estimators
		4.5 Ablation on GCGF and Depth Distillation
		4.6 Ablation on Text Prompting
		4.7 Analysis by Scene Classification
		4.8 Visualization
	5 Conclusion
	References
Remaining Useful Life Prediction of Control Moment Gyro in Orbiting Spacecraft Based on Variational Autoencoder
	1 Introduction
	2 Related Work
	3 CMG-VAE Model
		3.1 Problem Formulation
		3.2 Overview
		3.3 Variational Autoencoder
		3.4 Time Encoding Module
		3.5 Structure Encoding Module
		3.6 Feature Fusion Module
		3.7 Decoder and RUL Prediction Module
	4 Experiments
		4.1 Dataset and Preprocessing
		4.2 Comparison Metrics
		4.3 Experimental Settings
		4.4 Evaluation Results
		4.5 Ablation Experiments
	5 Conclusion
	References
Dynamic Feature Distillation
	1 Introduction
	2 Related Work
	3 Method
		3.1 Preliminary
		3.2 Online Feature Estimation
		3.3 Adaptive Position Selection
		3.4 Implementation Procedure
	4 Experiment
		4.1 Datasets and Experimental Setting
		4.2 Main Results
		4.3 The Numbers of Projectors
		4.4 The Length of a Training Phase
		4.5 Parameter Sensitivity Test
	5 Conclusion
	References
Detection of Anomalies and Explanation in Cybersecurity
	1 Introduction
	2 Preliminaries and Related Work
		2.1 Preliminaries
		2.2 Anomaly Detection
		2.3 Outlying Aspect Mining Algorithms
	3 The Proposed Method: HMass
		3.1 Relation to HBOS and SPAD
	4 Empirical Evaluation
		4.1 Contenders and Their Parameter Settings
		4.2 Datasets and Performance Measure
		4.3 Outlier Detector Performance
		4.4 Outlying Aspect Mining Performance
	5 Conclusion
	References
Document-Level Relation Extraction with Relation Correlation Enhancement
	1 Introduction
	2 Related Work
	3 Methodology
		3.1 Task Formulation
		3.2 Overall Architecture
		3.3 Encoder Module
		3.4 Relation Correlation Module
		3.5 Classification Module
	4 Experiments
		4.1 Dataset
		4.2 Implementation Details
		4.3 Baseline Systems
		4.4 Quantitative Results
		4.5 Analysis of Relation Correlation Module
		4.6 Performance on Multi-Label Extraction
		4.7 Ablation Study
		4.8 Case Study
	5 Conclusion
	References
Multi-scale Directed Graph Convolution Neural Network for Node Classification Task
	1 Introduction
	2 Related Work
		2.1 Spectral Graph Convolution Theory
		2.2 Chebyshev Polynomials Approximate
	3 Method
		3.1 Problem Formulation
		3.2 Constructing the Magnetic Laplacian Matrix of G
		3.3 Imporved Lanczos Algorithm
		3.4 MSDGCNN Architecture
	4 Experiment
		4.1 Datasets
		4.2 Implementation Details
		4.3 Experimental Results and Analysis
	5 Conclusion
	References
Probabilistic AutoRegressive Neural Networks for Accurate Long-Range Forecasting
	1 Introduction
	2 Background
		2.1 ARIMA Model
		2.2 ARNN Model
		2.3 Ensemble and Hybrid Models
	3 Proposed Model
		3.1 PARNN Model
		3.2 Prediction Interval of the PARNN Model
	4 Experimental Analysis
		4.1 Data
		4.2 Analysing Global Characteristics in the Datasets
		4.3 Performance Measures
		4.4 Baselines
		4.5 Model Implementations
		4.6 Benchmark Comparsions
		4.7 Significance Test
	5 Conclusion and Discussion
	References
Stereoential Net: Deep Network for Learning Building Height Using Stereo Imagery
	1 Introduction
	2 Methodology
		2.1 Baseline Architecture
		2.2 Differential Shortcut Connection Module (DSCM)
	3 Results
		3.1 Dataset
		3.2 Experimental Detail
		3.3 Quantitative and Qualitative Comparison
		3.4 Error Analysis
	4 Conclusions
	References
FEGI: A Fusion Extractive-Generative Model for Dialogue Ellipsis and Coreference Integrated Resolution
	1 Introduction
	2 Data Reconstruction
	3 Methodology
		3.1 Extractive Module
		3.2 Generative Module
	4 Experimentation
		4.1 Experimental Settings
		4.2 Compared Methods
		4.3 Main Results
	5 Ablation Study
		5.1 Effect of the Auxiliary Task OMIT
		5.2 The Utility of Ellipsis Recovery and Coreference Resolution Respectively
	6 Conclusion
	References
Assessing and Enhancing LLMs: A Physics and History Dataset and One-More-Check Pipeline Method
	1 Introduction
	2 Related Works
		2.1 LLMs Evaluations
		2.2 Chain-of-Thought Improvements
	3 Dataset
		3.1 Overview
		3.2 Data Collection
		3.3 Data Verification
	4 Method
		4.1 One-More-Check(OMC)
		4.2 Merge Multiple Reasoning Paths
	5 Experiment
		5.1 Setup
		5.2 Main Results
		5.3 Answer-Checking Ability Study
		5.4 Attending Actual History Examination
	6 Conclusion
	References
Sub-Instruction and Local Map Relationship Enhanced Model for Vision and Language Navigation
	1 Introduction
	2 Related Work
		2.1 Sub-instruction
		2.2 Map in Navigation
	3 Method
		3.1 Map Module
		3.2 Language Module
		3.3 Cross-Modal Attention Module
		3.4 Navigation Module
	4 Experiments
		4.1 Experiment Setup
		4.2 Experiment Results
		4.3 Ablation Study
	5 Conclusion
	References
STFormer: Cross-Level Feature Fusion in Object Detection
	1 Introduction
	2 Related Work
	3 Method
		3.1 Overall Architecture
		3.2 Transformer Encoder-Decoder Blocks
		3.3 Patch Merge and Patch Demerge
		3.4 Hybrid Self-attention Module
		3.5 Hybrid Cross-Attention Module
	4 Experiments
		4.1 Object Detection on MS COCO Dataset
	5 Conclusion
	References
Improving Handwritten Mathematical Expression Recognition via an Attention Refinement Network
	1 Introduction
	2 Related Works
	3 Methodology
		3.1 Pyramid Data Augmentation
		3.2 Encoder and Positional Encoding
		3.3 Shift Window Attention
		3.4 Refined Coverage Attention
		3.5 Loss Function
	4 Experiments
		4.1 Dataset and Evaluation Metrics
		4.2 Implementation Details
		4.3 Experimental Results
		4.4 Ablation Study
	5 Conclusion
	References
Dual-Domain Learning for JPEG Artifacts Removal
	1 Introduction
	2 Related Work
		2.1 JPEG Artifacts Removal
		2.2 Spatial-Frequency Interaction
	3 Method
		3.1 Fourier Transform
		3.2 Network Framework
		3.3 Loss Function
	4 Experiments
		4.1 Experimental Datasets and Implementation Details
		4.2 Results
		4.3 Ablation Studies
	5 Conclusions
	References
Graph-Based Vehicle Keypoint Attention Model for Vehicle Re-identification
	1 Introduction
	2 Related Works
	3 Proposed Method
		3.1 Overall Framework
		3.2 Relation Matrix
		3.3 Keypoint Feature Extractor
		3.4 Vehicle Keypoint Graph Model
		3.5 Cross-Attention
	4 Experiment
		4.1 Datasets and Settings
		4.2 Implement Details and Evaluation Protocols
		4.3 Ablation Study
		4.4 Parameter Analysis
		4.5 Comparison Results
		4.6 Visualization of the Results
	5 Conclusion
	References
POI Recommendation Based on Double-Level Spatio-Temporal Relationship in Locations and Categories
	1 Introduction
	2 Related Work
		2.1 POI Recommendation with Spatio-Temporal Contexts
		2.2 POI Recommendation with Category Contexts
	3 Problem Formulation and LSTPM
		3.1 Problem Formulation
		3.2 LSTPM
	4 The Proposed Model
		4.1 Data Embedding
		4.2 Short-Term User Preference Learning
		4.3 Long-Term User Preference Learning Based on Double-Layer Temporal Relationship in Locations and Categories
		4.4 Long-Term User Preference Learning Based on Double-Layer Spatial Relationship in Locations and Categories
	5 Experiments
		5.1 Experiment Settings
		5.2 Performance Comparison with Baseline Models
		5.3 Ablation Study
	6 Conclusion
	References
Multi-Feature Integration Neural Network with Two-Stage Training for Short-Term Load Forecasting
	1 Introduction
	2 Framework of TCN-GRU-TEmb
		2.1 Load Feature Extraction Module
		2.2 Environment Variables Feature Extraction Module
		2.3 Temporal Embedding Self-Learning Module
		2.4 Output Module and Training Algorithm
	3 Experimental Setup
		3.1 Data Preparation
		3.2 Baselines and Evaluation Metrics
	4 Experimental Results
		4.1 Comparison Experiment
		4.2 Comparison of Different Temporal Information Coding Strategies
	5 Conclusion
	References
Author Index