Explainable Artificial Intelligence

2.

Taxonomy and Classification of XAI Methods

2.1.

Scope of Explanation

2.1.1.

Global Explanations

2.1.1.1.

Model-Level Understanding

2.1.1.2.

Feature Importance at Model Level

2.1.1.3.

Model Behavior Summarization

2.1.1.4.

Decision Boundary Analysis

2.1.2.

Local Explanations

2.1.2.1.

Instance-Level Predictions

2.1.2.2.

Feature Attribution for Single Instances

2.1.2.3.

Local Decision Boundaries

2.1.3.

Cohort Explanations

2.1.3.1.

Group-Level Analysis

2.1.3.2.

Subpopulation Behavior

2.2.

Timing of Explanation Generation

2.2.1.

Ante-hoc Methods

2.2.1.1.

Intrinsically Interpretable Models

2.2.1.2.

Design-Time Interpretability

2.2.1.3.

Trade-offs with Model Performance

2.2.2.

Post-hoc Methods

2.2.2.1.

Explanation of Pre-trained Models

2.2.2.2.

Retrofitting Interpretability

2.2.2.3.

Explanation Generation Algorithms

2.3.

Model Dependency

2.3.1.

Model-Agnostic Methods

2.3.1.1.

Universal Applicability

2.3.1.2.

Black-Box Treatment of Models

2.3.1.3.

Flexibility and Limitations

2.3.2.

Model-Specific Methods

2.3.2.1.

Architecture-Tailored Approaches

2.3.2.2.

Leveraging Internal Model Structures

2.3.2.3.

Deep Learning Specific Methods

2.4.

Type of Explanation Output

2.4.1.

Feature Attribution Methods

2.4.2.

Example-Based Explanations

2.4.3.

Rule-Based Explanations

2.4.4.

Concept-Based Explanations

2.4.5.

Counterfactual Explanations

2.4.6.

Natural Language Explanations

Previous

1. Foundations of Explainable AI

Go to top

Next

3. Intrinsically Interpretable Models