Description
In this course, you will learn:
- A thorough knowledge of the necessity for and advantages of Explainable AI.
- The capacity to create and implement well-known explanation algorithms.
- Experiment with mixing current explanation approaches to create more solid explanations.
- An understanding of explainers used to interpret a neural network's decision.
- The capacity to assess and measure the accuracy of neural network explanations.
Syllabus :
1. Introduction to Explainable AI
- Explainable AIPrimer on Gradients
- Coding Exercise on Derivatives
- Introductory Quiz on Explainable AI
2. Saliency Maps
- Introduction to Saliency Maps—The Vanilla Gradient
- Guided Backpropagation
- Smooth Gradient Saliency
- Integrated Gradients
- LIME
- Layer-wise Relevance Propagation
- LRP-γ and LRP-Ɛ Rules
- SHAP
3. Class Activation Maps
- Introduction to Class Activation Maps
- Grad
- CAMX-Grad
- CAM
- Eigen-CAM
4. Miscellaneous Methods
- Counterfactual Explanations
- Perturbation-Based Explanations
- Concept Explanations
- Prototypical Explanations
- Coding Exercise on Miscellaneous Methods
5. Metrics of Interpretability
- Ground-Truth Faithfulness—Feature Agreement
- Ground-Truth Faithfulness—Rank Correlation
- Predictive Faithfulness
- Stability and Fairness
- Coding Exercise on Metrics of Interpretability