Neural Information Processing: 28th International Conference, ICONIP 2021, Sanur, Bali, Indonesia, December 8–12, 2021, Proceedings, Part I (Theoretical Computer Science and General Issues)

The four-volume proceedings LNCS 13108, 13109, 13110, and 13111 constitutes the proceedings of the 28th International Conference on Neural Information Processing, ICONIP 2021, which was held during December 8-12, 2021. The conference was planned to take place in Bali, Indonesia but changed to an online format due to the COVID-19 pandemic. The total of 226 full papers presented in these proceedings was carefully reviewed and selected from 1093 submissions. The papers were organized in topical sections as follows: Part I: Theory and algorithms; Part II: Theory and algorithms; human centred computing; AI and cybersecurity; Part III: Cognitive neurosciences; reliable, robust, and secure machine learning algorithms; theory and applications of natural computing paradigms; advances in deep and shallow machine learning algorithms for biomedical data and imaging; applications; Part IV: Applications. Preface Organization Contents – Part I Theory and Algorithms Metric Learning Based Vision Transformer for Product Matching 1 Introduction 2 Method 2.1 ML-VIT 2.2 Inference for Matching 3 Experiment 3.1 Implementation 3.2 Comparison with State-of-Art Models 3.3 Inspection of ML-VIT 4 Conclusion References Stochastic Recurrent Neural Network for Multistep Time Series Forecasting 1 Introduction 2 Related Works 2.1 Recurrent Neural Networks 2.2 Stochastic Gradient Variational Bayes 2.3 State Space Models 3 Stochastic Recurrent Neural Network 3.1 Problem Statement 3.2 Stochastic GRU Cell 3.3 Generative Model 3.4 Inference Model 3.5 Model Training 3.6 Model Prediction 4 Experiments 5 Conclusion References Speaker Verification with Disentangled Self-attention 1 Introduction 2 Disentangled Self-attention 2.1 Disentangled Non-local Network 2.2 Disentangled Self-attention Network 2.3 Proof of Disentangled Function 3 The Proposed Model 3.1 Front-End Feature Extraction Layer 3.2 Disentangled Self-attention Layer 3.3 Utterance Embedding Layer 4 Experiments 4.1 Experimental Setting 4.2 Experimental Results 4.3 Ablation Study 4.4 Comparison with Other Methods 5 Conclusions References Multi Modal Normalization 1 Introduction 2 Related Work 3 Multi-Modal Normalization - Theory 4 Multi Modal Normalization in Audio to Video Generation 4.1 Architectural Design 5 Multi Modal Normalization in Video Emotion Detection 6 Experiments 6.1 Implementation Details 6.2 Implementation Results 6.3 Ablation Study 7 Conclusion References A Focally Discriminative Loss for Unsupervised Domain Adaptation 1 Introduction 2 Method 2.1 Motivation 2.2 Weight Class-Wise MMD 2.3 Discriminability Strategy 2.4 Total Network 3 Experimental Evaluation 3.1 Result and Discussion 3.2 Model Analysis 4 Conclusion References Automatic Drum Transcription with Label Augmentation Using Convolutional Neural Networks 1 Introduction 2 Drum Transcription Architecture 2.1 Feature Representation 2.2 Label Augmentation Training 2.3 Network Architecture 2.4 Event Segmentation 3 Experimental Setup 3.1 Tasks and Data Sets 3.2 Implementation Detail 3.3 Result and Discussion 4 Conclusion References Adaptive Curriculum Learning for Semi-supervised Segmentation of 3D CT-Scans 1 Introduction 2 Adaptive Curriculum Semi-supervised Segmentation (ACuSS) 3 Experiments and Results 3.1 Datasets 3.2 Implementation Details 3.3 Results and Discussion 4 Conclusion References Genetic Algorithm and Distinctiveness Pruning in the Shallow Networks for VehicleX 1 Introduction and Related Work 2 Methods 2.1 Input Coding Techniques 2.2 Feature Selection Based on Genetic Algorithm 2.3 Distinctiveness Pruning 2.4 Hybrid Approach 3 Results and Discussion 3.1 Feature Selection Based on Genetic Algorithm 3.2 Distinctiveness Pruning 3.3 Hybrid Approach 4 Conclusion and Future Work References Stack Multiple Shallow Autoencoders into a Strong One: A New Reconstruction-Based Method to Detect Anomaly 1 Introduction 2 Related Work 3 Our Method 3.1 Stacked Architecture 3.2 Weighted Reconstruction Error 3.3 Objective Loss 3.4 Regularity Score 4 Experiment 4.1 Experiment on MNIST Dataset 4.2 Experiment on UCSD-ped2 and CUHK Avenue Datase 4.3 Ablation Studies 5 Conclusion References Learning Discriminative Representation with Attention and Diversity for Large-Scale Face Recognition 1 Introduction 2 Related Work 3 Approach 3.1 Overall Network Architecture 3.2 Attention Module 3.3 Response Maps Diversity Loss 4 Experiments 4.1 Implementation Details and Evaluation 4.2 Ablation Study 4.3 Compared with State-of-the-Art Face Recognition Methods 4.4 Visualization 5 Conclusions References Multi-task Perceptual Occlusion Face Detection with Semantic Attention Network 1 Introduction 2 Related Work 2.1 Occlusion Face Detection 2.2 Multi-task Learning 3 Method 3.1 Overall Framework 3.2 Multi-task Prediction Head 3.3 Semantic Learning Task 3.4 SAN Semantic Attention Network 3.5 Network Loss Function 4 Experiment 4.1 Data Set 4.2 Result on MAFA 4.3 Result on WIDER FACE 4.4 Ablation Study on MAFA 5 Conclusion References RAIDU-Net: Image Inpainting via Residual Attention Fusion and Gated Information Distillation 1 Introduction 2 Related Work 3 Proposed Approach 3.1 Network Architecture 3.2 Residual Attention Fusion Block 3.3 Gated Information Distillation Block 4 Experiments 4.1 Qualitative Comparison 4.2 Quantitative Comparisons 4.3 Ablation Study 5 Conclusion References Sentence Rewriting with Few-Shot Learning for Document-Level Event Coreference Resolution 1 Introduction 2 Related Work 3 Event Coreference Resolution on Sentence Rewriting 3.1 Candidate Sentence Selection 3.2 Sentence Rewriting 3.3 Event Coreference Prediction 4 Experimentation 4.1 Experimental Settings 4.2 Experimental Results 5 Analysis 5.1 Ablation Study 5.2 Analysis on Few-Shot Learning 5.3 Case Study 6 Conclusion References A Novel Metric Learning Framework for Semi-supervised Domain Adaptation 1 Introduction 2 Related Work 3 Adaptive Metric Learning Framework for Semi-supervised Domain Adaptation 3.1 Formulation of the Model 3.2 Optimization of Adaptive Objective Function 4 Experiments 5 Results and Discussion 6 Conclusion and Future Work References Generating Adversarial Examples by Distributed Upsampling 1 Introduction 2 Related Work 2.1 Adversarial Examples 2.2 Generative Adversarial Networks (GANs) 2.3 Checkerboard Artifacts 3 Proposed Approach 3.1 Problem Definition 3.2 Proposed Framework 3.3 Reconsider Deconvolution 4 Experiments 4.1 Experimental Setup 4.2 Attack in White-Box Setting 4.3 Transferability in Black-Box Setting 4.4 Comparison of Upsampling Methods 5 Conclusion References CPSAM: Channel and Position Squeeze Attention Module 1 Introduction 2 Related Work 3 Method 3.1 Overview 3.2 Channel Squeeze Attention Module 3.3 Position Squeeze Attention Module 3.4 Combination of Two Attention Modules 4 Experiments 4.1 Image Classification 4.2 Ablation Studies on CIFAR-100 4.3 Comparing with NL-Block and State-of-the-Art on ImageNet1k 5 Conclusion References A Multi-Channel Graph Attention Network for Chinese NER 1 Introduction 2 Related Work 3 Model 3.1 Text Encoder Layer 3.2 Graph Layer 3.3 Scalar Fusion Layer 3.4 Tag Decoder Layer 4 Experiments 4.1 Experimental Settings 4.2 Results and Analysis 4.3 Combing Pre-trained Model 4.4 Ablation Study 4.5 Efficiency Study 4.6 Sentence Length Analysis 5 Conclusion References GSNESR: A Global Social Network Embedding Approach for Social Recommendation 1 Introduction 2 Preliminary for Matrix Factorization Recommendation Framework 2.1 Low Rank Matrix Factorization Model 2.2 The Social Recommendation Model 3 The Proposed Embedding Framework 3.1 Global Social Latent Factor 3.2 Unbiased Weight Vector for All Customers 3.3 Personalized Trust Preference Weight Matrix 4 Experimental Results 4.1 Datasets 4.2 Metrics 4.3 Experimental Comparisons 4.4 Result and Analysis 5 Conclusions and Future Work References Classification Models for Medical Data with Interpretative Rules 1 Introduction 2 Dataset 2.1 SARS-CoV-1 2.2 SARS-CoV-2 3 Methods 3.1 Classification Models 3.2 Rules Generation 4 Experimental Analysis 4.1 Prediction 4.2 Rules Generation Results 5 Discussion and Future Work 6 Conclusion References Contrastive Goal Grouping for Policy Generalization in Goal-Conditioned Reinforcement Learning 1 Introduction 2 Related Works 2.1 Goal-Conditioned RL 2.2 Self-supervised Learning for State Representation Learning in RL 2.3 Goal Representation Learning in Goal-Conditioned RL 3 Background 3.1 Reinforcement Learning 3.2 Goal-Conditioned Reinforcement Learning 4 Contrastive Goal Grouping 4.1 Sufficient Condition for Optimal Policy Similarity 4.2 Contrastive Goal Grouping 5 Experiments 5.1 Results and Analysis 6 Conclusion References Global Fusion Capsule Network with Pairwise-Relation Attention Graph Routing 1 Introduction 2 The Proposed GFCN with GraMPA 2.1 Multi-block Attention for Capsule 2.2 Multi-block Feature Fusion 2.3 Graph Routing Based on Multi-head Pairwise-Relation Attention 2.4 Multi-block Ensemble Classification 3 Experiments 3.1 Implementation 3.2 Classification Performance 3.3 Robustness to Adversarial Examples 3.4 Visualization 3.5 Reconstruction and disentangled Representations References MA-GAN: A Method Based on Generative Adversarial Network for Calligraphy Morphing 1 Introduction 2 Related Work 2.1 Laplacian Pyramid of Adversarial Networks 2.2 Markov Discriminator (Patch GAN) 3 Our Method: Mode Averaging Generative Adversarial Network 3.1 The Algorithm of Mode Averaging 3.2 Network Structure 3.3 Loss Function 4 Experiment 4.1 Character Synthesis 4.2 Character Synthesis with Specific Weight 4.3 The Effect of Noise Amplification Rate on Generation 4.4 Compare with Other Generative Model 5 Conclusion References One-Stage Open Set Object Detection with Prototype Learning 1 Introduction 2 Related Work 3 Method 3.1 Open Set Object Detection 3.2 Open Set Object Detection Using YOLO V3 3.3 Prototype Based Open Set Object Detection 3.4 Model Training 4 Experiments and Results 4.1 Experimental Setting 4.2 Implementation Details 4.3 Main Results 4.4 Configuration of Prototype on Different Scales 5 Conclusion References Aesthetic-Aware Recommender System for Online Fashion Products 1 Introduction 2 Related Work 3 Aesthetic-Aware Recommendation Model 3.1 Model Construction and Training 3.2 Computing Recommendations 4 Experiments and Results 5 Conclusion References DAFD: Domain Adaptation Framework for Fake News Detection 1 Introduction 2 Related Work 2.1 Fake News Detection 2.2 Transfer Learning 3 Methodology 3.1 Overview 3.2 Textual Feature Extractor 3.3 Domain Adaptation Component 3.4 Adversarial Examples Generator 4 Experiments 4.1 Experiments Setup 4.2 Baselines 4.3 Fake News Detection Performance (EQ.1) 4.4 Effects of Framework Components (EQ.2) 4.5 Analysis of Domain Adaptation Effectiveness 4.6 Impact of the Amount of Labeled Data on the DAFD (EQ.3) 5 Conclusions and Future Work References Document Image Classification Method Based on Graph Convolutional Network 1 Introduction 2 Related Work 3 Proposed Approach 3.1 Graph Node Feature Extraction 3.2 Graph Convolutional Network Module 4 Experiments 4.1 Datasets Description 4.2 Model Training and Evaluation 5 Results and Discussion 6 Conclusion References Continual Learning of 3D Point Cloud Generators 1 Introduction 2 Related Work 2.1 Point Cloud Generation 2.2 Continual Learning for Generative Models 2.3 Unsupervised Continual Learning 3 Our Method 3.1 Continual Learning Setting 3.2 Parameter Isolation Techniques 3.3 Autonomous Branch Construction 4 Evaluation Protocol 4.1 Point Cloud Generation 4.2 Model Evaluation 5 Experiments 5.1 Implementation Details 5.2 Task Setups 5.3 Results 6 Conclusion References Attention-Based 3D ResNet for Detection of Alzheimer's Disease Process 1 Introduction 2 Related Works 3 Proposed Method 3.1 Attention Mechanism 3.2 Instance-Batch Normalization 3.3 Probabilistic Fusion Method 4 Experiments and Results 4.1 Datasets 4.2 Implementation Details 4.3 Comparison to Other Methods 5 Conclusion References Generation of a Large-Scale Line Image Dataset with Ground Truth Texts from Page-Level Autograph Documents 1 Introduction 2 Related Works 3 Hardness and Significance of Autograph Documents 4 Proposed Method 4.1 Alignment and Its Formulation 4.2 Alignment Solver 5 Experiments 5.1 Datasets 5.2 Implementation Details 5.3 Metrics 5.4 Experimental Results 6 Conclusion References DAP-BERT: Differentiable Architecture Pruning of BERT 1 Introduction 2 Related Work 2.1 Network Pruning for BERT 2.2 Neural Architecture Search 3 Methodology 3.1 Definition of Search Space 3.2 Differentiable Architecture Pruning 3.3 Fine-Tuning with Two-Stage Knowledge Distillation 3.4 Summary of the Method 4 Experiments 4.1 Experiment Setup 4.2 Experiment Results 4.3 Comparison with State-of-the-arts 4.4 Discussion 5 Conclusion References Trash Detection on Water Channels 1 Introduction 2 Dataset 2.1 Collection and Annotation 2.2 Challenges in Dataset 2.3 Comparison with Existing Datasets 3 Trash Object Identification 3.1 Analysis on Object Sizes 3.2 Handling Smaller and Partially Visible Objects 4 Trash Pixel Identification 4.1 Reducing Model Size for Real Time Applications 4.2 Addressing Class Imbalance 5 Conclusions and Future Work References Tri-Transformer Hawkes Process: Three Heads are Better Than One 1 Introduction 2 Related Works 3 The Proposed Tri-THP Framework 3.1 Tri-Transformer Hawkes Process 3.2 Conditional Intensity Function 3.3 Loss Function for Forecasting Occurring Times and the Types of Events 3.4 Objective Function 4 Experimental Results 4.1 Datasets 4.2 Experimental Results and Comparison 5 Conclusions and Future Work References PhenoDeep: A Deep Learning-Based Approach for Detecting Reproductive Organs from Digitized Herbarium Specimen Images 1 Introduction 2 Related Work 3 Specimen Dataset Collection 4 Methodology 4.1 Mask Scoring RCNN 5 Experiments and Discussion 5.1 The Impact of Data Augmentation Approaches on Segmentation Results 5.2 Experimental Results and Evaluation of Multiple Segmentation Approaches 6 Discussion 7 Conclusion and Future Directions References Document-Level Event Factuality Identification Using Negation and Speculation Scope 1 Introduction 2 Related Work 2.1 Event Factuality Identification 2.2 Negation and Speculation Scope Detection 3 DEFI with Scope 3.1 Overview 3.2 Scope Detection 3.3 DEFI 3.4 Model Structure 4 Experimentation 4.1 Dataset 4.2 Implementation Details 4.3 Migration Application 4.4 Baselines 4.5 Results and Analysis 4.6 Discussion 5 Conclusion References Dynamic Network Embedding by Time-Relaxed Temporal Random Walk 1 Introduction 2 Related Work 3 Notations and Problem Definitions 4 Time-Relaxed Temporal Random Walk(TxTWalk) 5 Experiments 5.1 Experimental Setup 5.2 Experimental Results 6 Conclusions and Discussions References Dual-Band Maritime Ship Classification Based on Multi-layer Convolutional Features and Bayesian Decision 1 Introduction 2 Decision-Level Fusion Classification Model 2.1 Feature Extraction 2.2 Feature Dimensionality Reduction Using PCA 2.3 Single-Band Multi-layer Features Fusion 2.4 Dual-Band Bayesian Decision Fusion Model Construction 3 Experimental Results 3.1 Dataset 3.2 Implementation Details 3.3 Analysis of Experimental Results 4 Conclusion References Context-Based Anomaly Detection via Spatial Attributed Graphs in Human Monitoring 1 Introduction 2 Related Work 2.1 Anomaly Detection in Image Regions 2.2 Anomaly Detection in Attributed Networks 3 Target Problem 4 Context-Basded Anomaly Detection Auto-Encoder 4.1 Constructing Spatial Attributed Graph 4.2 The Proposed Model 5 Experiments 5.1 Datasets 5.2 Experimental Settings 5.3 Experimental Results and Analysis 5.4 Parameter-Sensitivity Investigation 6 Conclusion References Domain-Adaptation Person Re-Identification via Style Translation and Clustering 1 Introduction 2 Proposed Method 2.1 Cross-Domain Image Translation 2.2 Unsupervised Domain Adaptation for Person Re-ID 3 Experiments 3.1 Datasets and Evaluation Metrics 3.2 Implementation Details 3.3 Comparison with the State-of-the-art 3.4 Ablation Study 4 Conclusion References Multimodal Named Entity Recognition via Co-attention-Based Method with Dynamic Visual Concept Expansion 1 Introduction 2 Related Work 2.1 Multimodal NER 2.2 Multimodal Representation 3 Proposed Method 3.1 Input Representations 3.2 Dynamic Visual Concept Expansion 3.3 Multimodal Co-attention 4 Experiments 4.1 Datasets and Experimental Settings 4.2 Results 4.3 Ablation Studies and Case Studies 4.4 Conclusion References Ego Networks 1 Introduction 2 Related Work 3 Method 3.1 Architecture 4 Experiments 4.1 Ablation Study 4.2 Comparison with Other Methods 5 Conclusion References Cross-Modal Based Person Re-identification via Channel Exchange and Adversarial Learning 1 Introduction 2 Related Work 3 The Proposed Method 3.1 The Exchange of Channel 3.2 The Discriminator and Adversarial Loss 3.3 Optimization 4 Experiments 4.1 Experimental Setting 4.2 Comparison with State-of-the-art Methods 5 Conclusion References SPBERT: an Efficient Pre-training BERT on SPARQL Queries for Question Answering over Knowledge Graphs 1 Introduction 2 Related Work 3 SPBERT 3.1 Pre-training Data 3.2 Input Representation 3.3 Model Architecture 3.4 Pre-training Setup 3.5 Fine-Tuning SPBERT 4 Experiments 4.1 Datasets 4.2 Experimental Setup 4.3 Results 4.4 Discussion 5 Conclusion References Deep Neuroevolution: Training Neural Networks Using a Matrix-Free Evolution Strategy 1 Introduction 2 Proposed Method 2.1 Groundwork 2.2 nDES Method Formulation 3 Experiments 3.1 Training CNNs 3.2 Training RNNs 4 Closing Remarks References Weighted P-Rank: a Weighted Article Ranking Algorithm Based on a Heterogeneous Scholarly Network 1 Introduction 2 Article Ranking Model 2.1 Heterogeneous Scholarly Network 2.2 Link Weighting in Paper Citation Graph (GP) 2.3 Link Weighting in Paper-Author Graph (GPA) 2.4 Link Weighting in Paper-Journal Graph (GPJ) 2.5 Weighted P-Rank Algorithm 3 Experiments 3.1 Datasets and Experimental Settings 3.2 Evaluation Metrics 3.3 Experimental Results 4 Conclusion References Clustering Friendly Dictionary Learning 1 Introduction 2 Literature Review 3 Proposed Approach 3.1 K-means Friendly Dictionary Learning 3.2 Sparse Subspace Clustering Friendly Dictionary Learning 4 Experimental Results 5 Conclusion References Understanding Test-Time Augmentation 1 Introduction 2 Preliminaries 2.1 Problem Formulation 2.2 TTA: Test-Time Augmentation 3 Theoretical Results for the Test-Time Augmentation 3.1 Re-formalization of TTA 3.2 Upper Bounds for the TTA 3.3 Weighted Averaging for the TTA 3.4 Existence of the Unnecessary Transformation Functions 3.5 Error Decomposition for the TTA 3.6 Statistical Consistency 4 Related Works 5 Conclusion and Discussion 5.1 Future Works References SphereCF: Sphere Embedding for Collaborative Filtering 1 Introduction 2 Related Work 2.1 Metric Learning 2.2 Collaborative Filtering 2.3 Collaborative Metric Learning (CML) 3 Our Model 3.1 Cosine Metric Learning 3.2 Triplet Loss 3.3 Hypersphere Manifold 3.4 Hybrid Loss 4 Experiments 4.1 Experimental Settings 4.2 Performance Comparison (RQ1) 4.3 What Role Does the Triplet Loss in SphereCF Play in the Model for Recommendation (RQ2) 4.4 How Does the Proposed Combine Loss in SphereCF Affect Model Performance (RQ3) 5 Conclusion References Concordant Contrastive Learning for Semi-supervised Node Classification on Graph 1 Introduction 2 Related Works 2.1 Graph Neural Networks 2.2 Contrastive Learning for Graph Neural Networks 3 Method 3.1 Problem Definition 3.2 Graph Convolutional Networks with High-Order Propagation 3.3 Prototypical Prediction 3.4 Concordant Contrastive Learning 4 Experiment 4.1 Datasets 4.2 Experimental Settings 4.3 Baseline Methods 4.4 Overall Performance Comparison 4.5 Ablation Study 4.6 Generalization Analysis 4.7 Robustness Analysis 4.8 Parameter Sensitivity 5 Conclusion References Improving Shallow Neural Networks via Local and Global Normalization 1 Introduction 2 Related Works 2.1 Network Normalization 2.2 Model Compression 2.3 Convolutional Neural Networks 3 Approch 3.1 Batch Normalization 3.2 Local Feature Normalization 3.3 How to Construct LFBN 4 Construct the LFBN-Net 5 Experiments 5.1 Experiments of LFBN-Net 5.2 Applied LFBN to Various CNNs 5.3 Compared to Others Normalization Algorithms 6 Conclusion References Underwater Acoustic Target Recognition with Fusion Feature 1 Introduction 2 Method 2.1 Model Overview 2.2 Frequency-Phase Spectrum Analysis 2.3 Frequency-Phase Feature Fusion Recognition 3 Experiments 3.1 Dataset and Experiment Platform 3.2 Implementation Details 3.3 Experimental Results 4 Conclusion References Evaluating Data Characterization Measures for Clustering Problems in Meta-learning 1 Introduction 2 Meta-learning and Clustering 3 Data Characterization Measures for Clustering Problems 4 Experimental Methodology 4.1 Building the Meta-dataset 4.2 Preliminary Analysis of Meta-dataset with Beta Regression 4.3 ML Meta-learners 5 Experimental Results 5.1 Statistical Significance and Influence of Data Characterization Measures 5.2 Performance Prediction and Importance of Meta-Features 6 Conclusion References ShallowNet: An Efficient Lightweight Text Detection Network Based on Instance Count-Aware Supervision Information 1 Introduction 2 Related Work 3 Proposed Method 3.1 Network Architecture 3.2 Optimization 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Ablation Study 4.4 Comparisons with Previous Methods 5 Conclusion References Image Periodization for Convolutional Neural Networks 1 Introduction 2 Methods 2.1 Image Extension and Resampling 2.2 Circular Convolution 3 Experiments 3.1 Datasets Description 3.2 Parameters Setting 3.3 Evaluation and Comparison 3.4 Visualized Results 4 Conclusion References BCN-GCN: A Novel Brain Connectivity Network Classification Method via Graph Convolution Neural Network for Alzheimer's Disease 1 Introduction 2 Related Work 2.1 fMRI Image Based Methods 2.2 Brain Connectivity Network Based Methods 3 The Proposed Method 3.1 Subjects and Image Pre-processing 3.2 Brain Connectivity Network Construction 3.3 BCN Classification by the Proposed GCN 4 Experimental Results and Analysis 5 Conclusion References Triplet Mapping for Continuously Knowledge Distillation 1 Introduction 2 Related Work 2.1 Knowledge Distillation 2.2 Triplet Networks 3 Triplet Networks 4 Triplet Mapping Knowledge Distillation 4.1 Training the Multilayer Perceptron 4.2 Triplet Mapping Knowledge Distillation 4.3 Trained by Hard Labels 4.4 Total Objective Function 5 Experiments 5.1 Experiments on Benchmark Datasets 5.2 Compared to Other Methods 5.3 Ablation Experiments 6 Conclusion References A Prediction-Augmented AutoEncoder for Multivariate Time Series Anomaly Detection 1 Introduction 2 Method 2.1 Problem Formulation 2.2 Model Framework 3 Experimental Setup 3.1 Datasets 3.2 Hyperparameters Configurations 3.3 Evaluation Metrics 4 Experiments and Results 4.1 Overall Performance 4.2 Ablation Study 4.3 Parameters Sensitivity 5 Conclusion and Outlook References Author Index The three volume set LNCS 8226, LNCS 8227, and LNCS 8228 constitutes the proceedings of the 20th International Conference on Neural Information Processing, ICONIP 2013, held in Daegu, Korea, in November 2013. The 180 full and 75 poster papers presented together with 4 extended abstracts were carefully reviewed and selected from numerous submissions. These papers cover all major topics of theoretical research, empirical study and applications of neural information processing research. The specific topics covered are as follows: cognitive science and artificial intelligence; learning theory, algorithms, and architectures; computational neuroscience and brain imaging; vision, speech and signal processing; control, robotics and hardware technologies; and novel approaches and applications.