Visualization for Artificial Intelligence

Acknowledgements Contents 1 Introduction 1.1 Generalization and Interpretability of AI 1.2 Visualization for AI 1.3 The Development of VIS4AI 1.4 Conceptual Framework and Method Overview 1.5 Book Motivation and Structure 1.5.1 Book Motivation 1.5.2 Book Structure 2 Fundamentals 2.1 Data 2.1.1 Tabular Data 2.1.2 Sequential Data 2.1.3 Multi-dimensional Array Data 2.1.4 Graph Data 2.1.5 Multi-modal Data 2.2 Machine Learning Models 2.2.1 Classical Models 2.2.2 Deep Models 2.2.3 Foundation Models 2.3 Relationships Between Data and Models 3 Techniques for Data Preparation 3.1 Instance Diagnosis 3.1.1 Inaccurate Instances 3.1.2 Insufficient Instances 3.1.3 Inexact Instances 3.2 Annotation Diagnosis 3.2.1 Inaccurate Annotations 3.2.2 Insufficient Annotations 3.2.3 Inexact Annotations 3.3 Feature Engineering 3.3.1 Insufficient Features 3.3.2 Inexact Features 3.4 Summary 4 Techniques for Model Development 4.1 Model Understanding 4.1.1 Node-Link Diagrams 4.1.2 Scatterplots 4.1.3 Parallel Coordinate Plots (PCPs) 4.1.4 Heat Maps 4.1.5 Glyphs 4.2 Model Diagnosis 4.2.1 Chart Visualizations 4.2.2 Matrix Visualizations 4.2.3 Tree Visualizations 4.2.4 Sankey Diagrams and Parallel Sets 4.2.5 Customized Visualizations 4.3 Model Steering 4.3.1 Model Refinement 4.3.2 Model Selection and Ensembling 4.4 Summary 5 Techniques for Model Deployment 5.1 Decision Explanation 5.1.1 Local Explanations 5.1.2 Global Explanations 5.2 Model Monitoring and Maintenance 5.2.1 Robustness 5.2.2 Fairness 5.3 Summary 6 Research Challenges and Opportunities 6.1 Data Preparation 6.1.1 Data Quality Research in Weakly Supervised Learning 6.1.2 Data Quality Research in Multi-Modal Learning 6.1.3 Active Selection of Training/Fine-Tuning/Test Data 6.1.4 Explainable Feature Engineering 6.2 Model Development 6.2.1 Understanding Multi-Modal Learning Models 6.2.2 Model-Agnostic Explanations 6.2.3 Online Training Diagnosis 6.2.4 Interactive Model Refinement 6.2.5 Interactive Performance Evaluation 6.3 Model Deployment 6.3.1 Evaluation of XAI Explanations 6.3.2 Fitting the Dynamic Nature of AI Systems 6.4 Generic Challenges and Opportunities 6.4.1 Building Trust from XAI Explanations 6.4.2 Cross-Cultural and Ethical Considerations 6.4.3 Dynamic Explanations 6.5 Foundation Models 6.5.1 Pre-Training Diagnosis 6.5.2 Adaptation Steering Conclusion Appendix References