Explainable AI Within the Digital Transformation and Cyber Physical Systems: XAI Methods and Applications

Description

This book presents Explainable Artificial Intelligence (XAI), which aims at producing explainable models that enable human users to understand and appropriately trust the obtained results. The authors discuss the challenges involved in making machine learning-based AI explainable. Firstly, that the explanations must be adapted to different stakeholders (end-users, policy makers, industries, utilities etc.) with different levels of technical knowledge (managers, engineers, technicians, etc.) in different application domains. Secondly, that it is important to develop an evaluation framework and standards in order to measure the effectiveness of the provided explanations at the human and the technical levels. This book gathers research contributions aiming at the development and/or the use of XAI techniques in order to address the aforementioned challenges in different applications such as healthcare, finance, cybersecurity, and document summarization. It allows highlighting the benefits and requirements of using explainable models in different application domains in order to provide guidance to readers to select the most adapted models to their specified problem and conditions. Includes recent developments of the use of Explainable Artificial Intelligence (XAI) in order to address the challenges of digital transition and cyber-physical systems;Provides a textual scientific description of the use of XAI in order to address the challenges of digital transition and cyber-physical systems;Presents examples and case studies in order to increase transparency and understanding of the methodological concepts. Preface Contents About the Editor 1 Prologue: Introduction to Explainable Artificial Intelligence 1.1 Explainable Machine Learning 1.2 Beyond State-of-the-Art: Contents of the Book 1.2.1 Chapter 2: Principles of Explainable Artificial Intelligence 1.2.2 Chapter 3: Science of Data: A New Ladder for Causation 1.2.3 Chapter 4: Explainable Artificial Intelligence for Predictive Analytics on Customer Turnover 1.2.4 Chapter 5: An Efficient Explainable Artificial Intelligence Model of Automatically Generated Summaries Evaluation 1.2.5 Chapter 6: On the Transparent Predictive Models for Ecological Momentary Assessment Data 1.2.6 Chapter 7: Mitigating the Class Overlap Problem in Discriminative Localization: COVID-19 and Pneumonia Case Study 1.2.7 Chapter 8: A Critical Study on the Importance of Feature Selection for Diagnosing Cyber-Attacks in Water Critical Infrastructures 1.2.8 Chapter 9: A Study on the Effect of Dimensionality Reduction on Cyber-Attack Identification in Water Storage Tank SCADA Systems References 2 Principles of Explainable Artificial Intelligence 2.1 Introduction 2.2 Motivations for XAI 2.3 Dimensions of XAI 2.4 Explanations and Explanators 2.4.1 Single Tree Approximation 2.4.2 Rules List and Rules Set 2.4.3 Partial Dependency 2.4.4 Local Rule-Based Explanation 2.4.5 Feature Importance 2.4.6 Saliency Maps 2.4.7 Prototype-Based Explanations 2.4.8 Counterfactual Explanations 2.5 Conclusions References 3 Science of Data: A New Ladder for Causation 3.1 Introduction 3.2 Related Works 3.2.1 Cognitive Architectures 3.2.2 Inferential Logic 3.2.3 Probabilistic Fuzzy Logic (PFL) 3.2.4 Neural Networks (NN) 3.2.4.1 Microscopic Neural Network (NN) 3.2.4.2 Macroscopic Neural Network (NN) 3.2.4.3 Graph Neural Networks 3.2.4.4 Hybrid Neural Networks for Reasoning 3.3 Cognitive Architecture Equipped with PFL and GNNs 3.4 Conclusion References 4 Explainable Artificial Intelligence for Predictive Analytics on Customer Turnover: A User-Friendly Interface for Non-expert Users 4.1 Introduction 4.2 Background 4.2.1 Shapley Values 4.2.2 Types of Explanation Techniques 4.3 Related Works 4.3.1 Shapley Additive Explanations 4.3.2 Contrastive Explanations 4.3.3 XAI User Interfaces 4.4 Our Explainable AI Web Interface 4.4.1 Back-End Component 4.4.2 Front-End Component 4.4.2.1 Home and Expected Loss 4.4.2.2 Local Feature Importance 4.4.2.3 Global Feature Importance 4.4.2.4 Model Recommendation 4.5 Evaluation 4.6 Conclusions References 5 An Efficient Explainable Artificial Intelligence Model of Automatically Generated Summaries Evaluation: A Use Case of Bridging Cognitive Psychology and Computational Linguistics 5.1 Introduction 5.1.1 Automatic Text Summarization 5.1.2 Evaluation Protocols of Automatically Generated Text Summaries 5.1.3 Cognitive Psychology Models for Text Comprehension 5.1.3.1 The Resonance Model 5.1.3.2 The Landscape Model 5.1.3.3 The Langston and Trabasso Model 5.1.3.4 The Construction–Integration Model 5.1.3.5 The Predication Model 5.1.3.6 The Gestalt Models 5.1.3.7 The Golden and Rumelhart Model 5.1.3.8 The Distributed Situation Space Model 5.1.3.9 The Structure Building Model 5.1.4 Originality of Our Work 5.2 CATSE: A Cognitive Automatic Text Summarization Evaluation Protocol 5.2.1 The Main Idea 5.2.2 Levels of Representation 5.2.2.1 The Surface Level 5.2.2.2 The Intermediate Level: The Textbase 5.2.2.3 The Cognitive Level: The Situation Model 5.2.3 The CATSE Protocol 5.2.3.1 The Construction Phase 5.2.3.2 The Integration Phase 5.3 Experiments and Results 5.3.1 Datasets 5.3.2 Experimental Results 5.4 Conclusion References 6 On the Transparent Predictive Models for Ecological Momentary Assessment Data 6.1 Introduction 6.1.1 Ecological Momentary Assessment (EMA) 6.1.2 Classification of EMA Data 6.1.3 Model Transparency 6.2 Dataset 6.3 Analysis Methods 6.3.1 Classification Settings 6.3.2 Tools 6.3.2.1 Own Pipeline: Model Training and Testing 6.3.3 Experiments 6.3.4 Model Interpretation and Analysis 6.3.4.1 Analysis of Categorical Features 6.3.4.2 Analysis of Continuous Features 6.3.4.3 Interpretation of the Resulting Values 6.3.4.4 Model Comparison 6.4 Results 6.4.1 Experiments with KNIME Analytics Platform 6.4.2 Experiments with Own Pipeline 6.4.3 Model Interpretation and Analysis 6.5 Discussion 6.5.1 Is Personalization Always Necessary? 6.5.2 Model Similarity 6.5.3 Model Validation 6.5.4 EMA with Other Features 6.5.5 Personalization of Class Labels 6.5.6 Method Limitations 6.6 Conclusion References 7 Mitigating the Class Overlap Problem in Discriminative Localization: COVID-19 and Pneumonia Case Study 7.1 Introduction 7.2 Related Work 7.3 Discriminative Localization 7.3.1 Class Activation Maps 7.3.2 Saliency Maps with Backpropagation 7.3.3 Amplified Directed Divergence with Ensembles 7.3.4 Scaled Directed Divergence (SDD) 7.4 Experiments 7.4.1 Method 7.4.2 COVID-19 and Pneumonia Data 7.4.3 COVID-19 AND Pneumonia Classifier 7.4.4 Scaled Directed Divergence with Natural Imagery 7.4.5 Scaled Directed Divergence Applied to Chest X-rays 7.5 Discussion 7.6 Conclusion References 8 A Critical Study on the Importance of Feature Selection for Diagnosing Cyber-Attacks in Water Critical Infrastructures 8.1 Introduction 8.2 Background 8.2.1 Infinite Feature Selection 8.2.2 Infinite Latent Feature Selection 8.2.3 Evolutionary Computation Feature Selection 8.2.4 Relief Feature Selection 8.2.5 Mutual Information 8.2.6 Maximum Relevance and Minimum Redundancy 8.2.7 Feature Selection via Concave Minimization 8.2.8 Laplacian Score 8.2.9 Multi-Cluster Feature Selection 8.2.10 Recursive Feature Elimination 8.2.11 L0-Norm 8.2.12 Fisher Score 8.3 Design of Intrusion Detection System 8.3.1 Data Collection 8.3.2 Decision-Making 8.4 Experimental Results 8.4.1 Experimental Setting 8.4.2 Results Analysis 8.4.3 Feature Analysis 8.5 Conclusion References 9 A Study on the Effect of Dimensionality Reduction on Cyber-Attack Identification in Water Storage Tank SCADASystems 9.1 Introduction 9.2 Background 9.2.1 Principal Component Analysis 9.2.2 Factor Analysis 9.2.3 Confirmatory Factor Analysis 9.2.4 Multidimensional Scaling 9.2.5 Linear Discriminant Analysis 9.2.6 Isomap 9.2.7 Semantic Mapping 9.2.8 Probabilistic Principal Component Analysis 9.2.9 Locally Linear Embedding 9.2.10 Laplacian Eigenmaps 9.2.11 Landmark Isomap 9.2.12 Hessian-based Locally Linear Embedding 9.2.13 Local Tangent Space Alignment 9.2.14 Kernel Principal Component Analysis 9.2.15 Generalized Discriminant Analysis 9.2.16 Neighborhood Preserving Embedding 9.2.17 Locality Preserving Projections 9.2.18 Diffusion Maps 9.2.19 Locally Linear Coordination 9.2.20 Manifold Charting 9.2.21 Large Margin Nearest Neighbor 9.2.22 Independent Component Analysis 9.3 Design of Intrusion Detection System 9.3.1 Data Collection 9.3.2 Decision Making 9.4 Experimental Results 9.4.1 Experiment Setting 9.4.2 Results Analysis 9.5 Conclusion References Index