Advances in Intelligent Data Analysis XX: 20th International Symposium on Intelligent Data Analysis, IDA 2022, Rennes, France, April 20–22, 2022, Proceedings

Description

This book constitutes the proceedings of the 20th International Symposium on Intelligent Data Analysis, IDA 2022, which was held in Rennes, France, during April 20-22, 2022.  The 31 papers included in this book were carefully reviewed and selected from 73 submissions. They deal with high quality, novel research in intelligent data analysis.  Preface Organization Contents Multi-modal Ensembles of Regressor Chains for Multi-output Prediction 1 Introduction 2 Background and Related Work 3 Multi-modal Ensemble of Regressor Chains 3.1 Mechanism 1: Base Estimator Training 3.2 Mechanism 2: Ensemble Mode Prediction 4 Experiments 4.1 Methods 4.2 Evaluation 4.3 Datasets 5 Results and Discussion 6 Conclusions and Future Work References A Two-Step Approach for Explainable Relation Extraction 1 Introduction 2 Related Works 3 Relation Classification with Concepts of Neighbours 3.1 Concepts of Neighbours 3.2 Application to Text 3.3 Application to Relation Classification 4 A Two-Step Approach for Relation Extraction 4.1 Relation Detection with Deep Learning 4.2 Explainability 5 Experiments and Results 5.1 Relation Detection 5.2 Relation Classification 5.3 Relation Extraction 6 Conclusion References Towards Automation of Topic Taxonomy Construction 1 Introduction 2 Method 2.1 Topic Generation 2.2 Taxonomy Construction 3 Visualization 4 Evaluation and Experiments 4.1 Datasets 4.2 Evaluation Measures 4.3 Results 5 Conclusion and Perspectives References A Fault Detection Framework Based on LSTM Autoencoder: A Case Study for Volvo Bus Data Set 1 Introduction 2 Related Work 3 Problem Description 4 Fault Detection Methodology 5 Case Study 6 Experimental Results 7 Conclusions References Detection and Multi-label Classification of Bats 1 Introduction 2 Related Work 3 Approach 3.1 Input Data 3.2 Data Augmentation 3.3 Proposed Architecture 4 Results 4.1 Datasets 4.2 Evaluation 4.3 Architectures for Comparison 4.4 Performance on the Different Challenges 5 Conclusion References End-to-End Mobile System for Diabetic Retinopathy Screening Based on Lightweight Deep Neural Network 1 Introduction 2 Novel Method for DR Screening 2.1 Pre-processing and Data Processing 2.2 Nasnet-Mobile Architecture 2.3 Transfer Learning of Nasnet-Mobile Architecture 2.4 MLP Configuration 3 Experimental Results 3.1 Dataset 3.2 ML Implementation 3.3 Evaluation Metrics 3.4 Performance Evaluation of Proposed Method 3.5 Execution Time Evaluation of the Mobile-Aided Screening System 4 Conclusion References Efficient Bayesian Learning of Sparse Deep Artificial Neural Networks 1 Introduction 2 Problem Formulation 3 Bayesian Optimization 3.1 Hierarchical Bayesian Model 3.2 Hamiltonian Sampling 4 Experimental Validation 4.1 ConvNet Models 4.2 Experiment 1: Challenging Case 4.3 Experiment 2: CIFAR-10 Image Classification 5 Conclusion References Tensor Completion Post-Correction 1 Introduction 2 Method 2.1 Proposed Algorithm: TCPC 2.2 Illustrative Example 2.3 Time Complexity 3 Experimental Evaluation 3.1 Tensor Completion Algorithms 3.2 Datasets 3.3 Evaluation Metric 3.4 Experimental Configuration 3.5 Evaluation of TCPC on Improvement of TC Estimation 3.6 Sensitivity of TCPC's Unique Parameter: 4 Conclusion and Future work References s-LIME: Reconciling Locality and Fidelity in Linear Explanations 1 Introduction 2 Preliminaries 3 Locality vs. Fidelity 3.1 The Paradox of Small Bandwidth 3.2 Why do Seem Locality and Fidelity in Opposition? 3.3 What Makes a Good LIME Explanation? 4 s-LIME 4.1 Generic Algorithm 4.2 s-LIME Subsumes LIME 4.3 s-LIME and the Gradient of the Black-Box Function 4.4 s-LIME Implementations 5 Experiments 5.1 Experimental Settings 5.2 Impact of 5.3 Fidelity Analysis 6 Related Work 7 Conclusion References Quantifying Changes in Predictions of Classification Models for Data Streams 1 Introduction 2 Related Works 3 Prediction Change Measures 3.1 Monitoring Prediction Changes 3.2 Quantifying Prediction Changes 4 Results 4.1 Reference Data Streams and Streaming Classifiers 4.2 Investigating Changes in Predictions for Hyperplane Data 4.3 Quantifying Prediction Changes for Airlines Data 4.4 Summary of Results for Remaining Data Streams 5 Summary References Impact of Dimensionality on Nowcasting Seasonal Influenza with Environmental Factors 1 Introduction 2 Framework 2.1 Problem formulation 2.2 Problem Instantiations and Solutions 3 Empirical Evaluation 3.1 Data Description 3.2 Dynamic Training and Hyperparameters 3.3 Perfomance Indicators 4 Results 5 Conclusions References On Usefulness of Outlier Elimination in Classification Tasks 1 Introduction 2 Related Work 3 Research Questions and Methodology 3.1 Basic Concepts 3.2 Determine Whether Some OEMs are Potentially Useful 3.3 Identify the Most Useful Workflows with OEMs 4 Experimental Setup 5 Results 5.1 Can We Use OEMs Without Restrictions (RQ1)? 5.2 Determining Whether Some OEMs are Potentially Useful (RQ2) 5.3 Constructing a Portfolio with the Most Useful Workflows with OEMs (RQ3) 6 Future Work and Conclusions References Suitability of Different Metric Choices for Concept Drift Detection 1 Introduction 2 Problem Setup 2.1 A General Scheme for Drift Detection 2.2 Formal Setup and Research Question 3 Dissimilarity Estimators 4 Theoretical Analysis 5 Empirical Evaluation 6 Conclusion References Exploring the Geometry and Topology of Neural Network Loss Landscapes 1 Introduction 2 Preliminaries 2.1 PHATE Dimensionality Reduction and Visualization 2.2 Topological Data Analysis 3 What Is the ``Shape'' of the Loss Landscape? 3.1 Jump and Retrain Sampling 3.2 PHATE Dimensionality Reduction and Visualization 3.3 Topological Feature Extraction 4 Geometric and Topological Reflection on ANN Training and Generalization 4.1 Experimental Setup 4.2 Jump and Retrain Sampling Captures Generalization and Training Characteristics 4.3 Generalization Indicated by Visual Patterns from Loss Landscape Regions Around Optima 4.4 Generalization May Be Related to Low Topological Activity in Near-Optimum Regions 5 Discussion and Conclusion References Selecting Outstanding Patterns Based on Their Neighbourhood 1 Introduction 2 Related Work 3 Background 4 Outstanding Pattern Selector: How to Exploit Siblings 5 Experiments 5.1 Itemset Data 5.2 Comparison to Self-sufficient Itemsets 5.3 Structured Pattern Selection 5.4 Expert Analysis Upon an Outstanding Pattern and Its Family 6 Conclusion References Using Explainable Boosting Machine to Compare Idiographic and Nomothetic Approaches for Ecological Momentary Assessment Data 1 Introduction 2 Methodology 2.1 Idiographic (Person-Specific) Approach 2.2 Nomothetic (Group-Level) Approaches 3 Experimental Setup 3.1 EMA Datasets 3.2 Data Preparation 3.3 Data Analysis 4 Experimental Results 4.1 Synthetic Dataset 4.2 Dataset: Drink 4.3 Dataset: ThinkSlim2 5 Challenges of Modelling EMA Data 6 Conclusion References dunXai: DO-U-Net for Explainable (Multi-label) Image Classification 1 Introduction 1.1 Image Context Classification 2 The Classification Tasks 2.1 Data 2.2 Existing Classification Methods 2.3 Our Classification Models 3 dunXai 3.1 DO-U-Net 3.2 Explainable AI: Class Activation Maps 3.3 dunXai Metrics 4 Conclusion References AGS: Attribution Guided Sharpening as a Defense Against Adversarial Attacks 1 Introduction 2 Background 2.1 Adversarial Attack Generation 2.2 Attribution Techniques 3 Methodology 3.1 Choi and Hall Sharpening (CHSharp) 3.2 Attribution Guided Sharpening (AGS) 4 Experiments 4.1 MNIST 4.2 CIFAR-10 4.3 CIFAR-100 5 Conclusions 5.1 Looking Ahead References VAE-CE: Visual Contrastive Explanation Using Disentangled VAEs 1 Introduction 2 Related Work 3 Method: VAE-CE 3.1 Learning a Data Representation for Class Explanation 3.2 Pair-Based Dimension Conditioning 3.3 Explanation Generation 4 Experiments 4.1 Datasets 4.2 Considered Evaluations 4.3 Comparison Overview 4.4 Results 5 Conclusions References Evaluation of Uplift Models with Non-Random Assignment Bias 1 Introduction 2 Uplift Modeling and Evaluation 2.1 Definition 2.2 Uplift Modeling 2.3 Uplift Evaluation 3 Evaluation of Uplift with Biased Data 3.1 Problem Setting 3.2 Designing of the Experimental Protocol 3.3 Experiments 3.4 Results 4 Method to Reduce the NRA Bias Impact 5 Conclusion References A Generic Trace Ordering Framework for Incremental Process Discovery 1 Introduction 2 Related Work 3 Preliminaries 3.1 Event Data 3.2 Process Models 3.3 Incremental Process Discovery 4 Dynamic Trace-Ordering Strategies 4.1 General Framework 4.2 Instantiations 5 Evaluation 5.1 Experimental Setup 5.2 Results & Discussion 6 Conclusion References Bank Statements to Network Features: Extracting Features Out of Time Series Using Visibility Graph 1 Introduction 2 Related Work 3 From Bank Statements to Network Features 3.1 Visibility Graphs 3.2 Daily Time Series to Visibility Graphs 3.3 Properties of the Generated Graphs 4 Case Study 4.1 Data Set 4.2 Feature Extraction 4.3 Feature Selection 4.4 Evaluation 5 Conclusion and Future Work References Modular-Relatedness for Continual Learning 1 Introduction 2 Background Knowledge 2.1 Continual Learning Problem Definition 2.2 Modular Networks 2.3 Gradient Episodic Memory 3 General Approach 3.1 Relatedness Estimation 3.2 Modularization 4 Modular-Relatedness for Rehearsal-Based Continual Learning 4.1 Modular GEM 4.2 Network Modularization 4.3 Complexity Analysis 5 Related Work 5.1 Continual Learning 5.2 Modular Neural Networks 6 Experiments 6.1 Setting: Online Continual Learning 7 Conclusion A Hyperparameter Search References Combining Multiple Data Sources to Predict IUCN Conservation Status of Reptiles 1 Introduction 2 Related Work 3 Methodology 3.1 Preliminaries 3.2 Proposed Pipeline 4 Experimental Results 4.1 Datasets 4.2 Experimental Setup 4.3 Results and Discussion 5 Conclusions References LG4AV: Combining Language Models and Graph Neural Networks for Author Verification 1 Introduction 2 Related Work 3 Combining Language Models and Graph Neural Networks for Author Verification 3.1 Problem 3.2 Combining Language Models and Graph Neural Networks for Author Verification 4 Experiments 4.1 Baselines and Configurations of LG4AV 4.2 Implementation Details 4.3 Results and Discussion 5 Conclusion and Outlook References Efficient Subgroup Discovery Through Auto-Encoding 1 Introduction 1.1 Main Contribution 2 Related Work 2.1 Subgroup Discovery 2.2 Dimensionality Reduction 3 Preliminaries 3.1 Auto-Encoding 4 Methodology 4.1 Experimental Setup 4.2 Algorithms and Settings 4.3 Data 5 Results 6 Discussion 7 Conclusions References Simulation of Scientific Experiments with Generative Models 1 Introduction 2 Related work 3 Proposed Approach 3.1 Use Case: Biomaterials Research 3.2 Model Concept 3.3 Implementation 4 Evaluation 4.1 Dataset 4.2 Results 5 Conclusion References A Learning Vector Quantization Architecture for Transfer Learning Based Classification in Case of Multiple Sources by Means of Null-Space Evaluation 1 Introduction 2 Learning Vector Quantization 3 Null-Space Transfer Classification Learning for GMLVQ Using a Siamese-Like Architecture 4 Mathematical Justification of the T-GMLVQ 5 Exemplary Application – Analysis of Polluted Breathing Air Spectra 6 Conclusion and Future Work References MuseBar: Alleviating Posterior Collapse in Recurrent VAEs Toward Music Generation 1 Introduction 2 Background 2.1 MIDI Representation 2.2 Generative Models 3 MuseBar 4 Empirical Study 4.1 Evaluation Metrics 4.2 Performance Results 5 Conclusions References Parameter Learning in ProbLog with Annotated Disjunctions 1 Introduction 2 Preliminaries 3 Learning from Interpretations in ProbLog 4 Learning with Annotated Disjunctions 4.1 Relevant Interpretations 4.2 Directly Learning Multi-head ADs 5 Proofs 5.1 Correctness 5.2 Convergence Rate 6 Experiments 7 Related Work 8 Conclusion References Semantic-Based Few-Shot Classification by Psychometric Learning 1 Introduction 2 Related Work 3 Semantic-Based Few-Shot Learning 3.1 Problem Formulation 3.2 Self-supervised Feature Learning 3.3 Psychometric Testing 3.4 Semantic-Based Few-Shot Prediction 4 Experiments and Discussion 5 Conclusion References Author Index