Preface
Organization
Contents
XAI andย Machine Learning
To Pay or Not to Pay Attention: Classifying and Interpreting Visual Selective Attention Frequency Features
1 Introduction
2 State of the Art
3 Approach or Method
4 Results and Discussions
5 Conclusions
References
GridEx: An Algorithm for Knowledge Extraction from Black-Box Regressors
1 Introduction
2 State of the Art
2.1 Knowledge Extraction
2.2 The Iter Algorithm
3 GridEx
3.1 The Algorithm
3.2 An Example
3.3 GridEx Adaptive Splitting
3.4 Parameter Tuning
4 Assessment of GridEx
4.1 Iter Experimental Analysis
4.2 GridEx Experimental Analysis
4.3 Comparison of Iter and GridEx
5 Conclusions
References
Comparison of Contextual Importance and Utility with LIME and Shapley Values
1 Introduction
2 Background and Definitions
2.1 Core Definitions
2.2 LIME
2.3 Shapley Values
3 Contextual Importance and Utility (CIU)
4 Experiments
4.1 Classification with Continuous Inputs
4.2 Regression with Continuous Inputs
4.3 Classification with Mixed Discrete and Continuous Inputs
5 Conclusion
References
ciu.image: An R Package for Explaining Image Classification with Contextual Importance and Utility
1 Introduction
2 Contextual Importance and Utility for Images
3 Installation and Use
4 Results
4.1 ImageNet Classification
4.2 Gastro-Enterological Image Explanation
5 Conclusions
References
Shallow2Deep: Restraining Neural Networks Opacity Through Neural Architecture Search
1 Introduction
2 Background
2.1 Neural Architecture Search
2.2 eXplainable AI vs. Neural Networks
3 Shallow2Deep
3.1 Architecture Design
3.2 Search Algorithm
4 Discussion
5 Experiments
5.1 Shallow2Deep Architecture
5.2 Shallow2Deep vs. State-of-the-Art
6 Conclusion
References
Visual Explanations for DNNs with Contextual Importance
1 Introduction
2 Related Work
3 Method
3.1 Contextual Importance
3.2 Contextual Importance Explanations for DNNs
4 Experimental Results
4.1 Visual Comparisons
4.2 Explanations on Contrastive Cases
4.3 Explanations on Distorted Images
4.4 Explanations on Misclassification
5 Conclusion
References
Towards Explainable Recommendations of Resource Allocation Mechanisms in On-Demand Transport Fleets
1 Introduction
2 About the Need for Explainability
3 AV-OLRA Metamodel
4 Explainable MAS for AV-OLRA Recommendation
4.1 AV Agents' Behavior
4.2 Monitor Agent's Behavior
4.3 Computing the Recommendations
4.4 Creating and Communicating the Explanations
5 ExDARP: A Use Case of Explainable Decisions in Decentralized DARP
6 Related Work
7 Conclusion
References
XAI Vision, Understanding, Deployment andย Evaluation
A Two-Dimensional Explanation Framework to Classify AI as Incomprehensible, Interpretable, or Understandable
1 Introduction
2 Background
2.1 Problem
2.2 Transparency
2.3 Related Work
3 A Two-Dimensional Framework to Classify AI
3.1 Transparency vs. Explainability
3.2 Interpretability vs. Understandability
3.3 Two-Dimensional Framework to Classify AI
3.4 Extended Framework
4 Discussion
4.1 Evaluation of Main Framework
4.2 Evaluation of Extended Framework
4.3 Application of Framework
4.4 Future Work
5 Conclusion
References
Towards Explainable Visionary Agents: License to Dare and Imagine
1 Introduction
2 State of the Art
2.1 Imagination in Humans
2.2 Cognitive Agents
2.3 Mechanisms for Imagination
2.4 Computational Creativity
3 Imagination in Cognitive Agents
3.1 Acquiring Knowledge via Learning and Induction
3.2 Synthesizing Knowledge via Deduction, Abduction, and Generative Methods
3.3 Speculating via Simulation and Planning
3.4 Adaptively Revising Knowledge via Biochemical Coordination
4 Open Challenges
5 Conclusions
References
Towards an XAI-Assisted Third-Party Evaluation of AI Systems: Illustration on Decision Trees
1 Introduction
2 State of the Art
2.1 Issues in the Evaluation of Artificial Intelligence
2.2 XAI for Third-Party Evaluation
3 Methodology
3.1 Context
3.2 Experimental Data
3.3 Experiments
4 Results
4.1 Health Data
4.2 Bank Data
5 Conclusions and Outlook
References
What Does It Cost to Deploy an XAI System: A Case Study in Legacy Systems
1 Introduction
2 Dimensions of Explainability
3 Steps in the Innovation Process
3.1 Custom Code Analysis
3.2 Multi-modal Conversational Interface
3.3 Source-Code Transformation
4 Explainability Requirements and Costs
4.1 Explainability Requirements
4.2 Explainability Costs
5 Conclusions
References
XAI Applications
Explainable AI (XAI) Models Applied to the Multi-agent Environment of Financial Markets
1 Introduction
1.1 Related Works
1.2 Contribution
1.3 Why GBDT?
2 Methodology
2.1 GBDT Hyperparameters
2.2 Features Used
2.3 Process of Features Selection
3 Results
3.1 Model Presentation
3.2 AUC Performance
4 Understanding the Model
4.1 Shapley Values
4.2 Shapley Interpretation
4.3 Joint Features and Shapley Values Distribution
4.4 Local Explanation of the Covid March 2020 Meltdown
5 Conclusion
References
Toward XAI & Human Synergies to Explain the History of Art: The Smart Photobooth Project
1 Introduction
2 Background
2.1 AI & Art
2.2 Styles in Modern and Contemporary Art Painting
2.3 Neural Style Transfer
2.4 XAI
3 The Smart Photobooth Project
4 Architecture
5 Open Challenges and Research Directions
6 Conclusions
References
Assessing Explainability in Reinforcement Learning
1 Introduction
2 Background
2.1 RL Main Application Domains
2.2 XAI Types
2.3 Explainable Reinforcement Learning
3 Methodology
3.1 Analysis of Application Domains
4 Key Criteria
4.1 The Intended User
4.2 Means of Interaction
4.3 Industry Sector
4.4 Urgency/Time-Restraint
4.5 Legal
4.6 Responsibility
5 Assessment
5.1 A Box-Moving Robot
5.2 Cloud Computing Resource Allocation
5.3 Frogger Videogame
5.4 Surgical Robot
6 Discussion
7 Conclusion
References
XAI Logic andย Argumentation
Schedule Explainer: An Argumentation-Supported Tool for Interactive Explanations in Makespan Scheduling
1 Introduction
2 Preliminaries
2.1 Makespan Scheduling
2.2 Abstract Argumentation (AA)
2.3 ArgOpt
3 Algorithms
3.1 Notation
3.2 Constructing AFs
3.3 Verifying Stability
3.4 Generating Explanations
4 Tool
5 Related Work
6 Conclusions and Future Work
References
Towards Explainable Practical Agency
1 Introduction
2 Algebraic Foundations
3 LogA PR Languages
4 Monotonic Logical Consequence
5 Graded Consequence
6 Explanations
7 Related Work
8 Concluding Remarks
References
Explainable Reasoning in Face of Contradictions: From Humans to Machines
1 Introduction
2 Human Intelligence: Bounded Rationality and Reasoning Backwards
3 Levels of Intelligent Reasoning in Face of Contradictions
3.1 Clear Preferences
3.2 Consistent Preferences
3.3 Explainable Backwards Reasoning''
3.4 Evidence-Based Principle Revision
4 Examples: Abstract Argumentation
4.1 Clear Preferences
4.2 Consistent Preferences
4.3 ExplainableBackwards Reasoning''
4.4 Evidence-Based Principle Revision
5 Research Directions
5.1 Consistent Preferences and Undecided Beliefs
5.2 Burdens of Persuasion
5.3 Intuitive Rationality
5.4 Neuro-Symbolic Artificial Intelligence
6 Conclusion
References
Towards Transparent Legal Formalization
1 Introduction
2 The Legislation Editor
2.1 The Annotation Editor
3 Legal Formalization
3.1 Article 7 of GDPR
3.2 Formalization of Article 7 Using the Legislation Editor
3.3 The Comprehensibility of Formalization - Future Work
4 Conclusion
References
Applying Abstract Argumentation to Normal-Form Games
1 Introduction
2 Backgrounds
2.1 Normal-Form Games
2.2 Abstract Argumentation Framework
3 Game-Based Argumentation Framework
3.1 Applying Argumentation to Games
3.2 Game-Based Argumentation Framework
4 Properties of a Game-Based Argumentation Framework
4.1 Correspondences Between Game-Theoretical Solution Concepts and Argument Extensions
4.2 Towards Explanation for Normal-Form Games
5 Related Works
6 Results and Future Work
References
Decentralized andย Heterogeneous XAI
Expectation: Personalized Explainable Artificial Intelligence for Decentralized Agents with Heterogeneous Knowledge
1 Background and Motivations
2 State of the Art
2.1 Explainable Agency
2.2 Agreement Technologies
2.3 AI Ethics
3 The Expectation Approach
3.1 Research Method
4 Discussion
References
Author Index