Search⌘ K
AI Features
Home
Courses
Master Explainable AI: Interpreting Image Classifier Decisions

Stability and Fairness

Explore the concepts of stability and fairness in explainable AI. Understand how the Relative Input Stability metric measures the robustness of explanations against small input changes. Learn methods to evaluate the fairness of explanations across different subgroups. This lesson equips you to implement and assess explanation algorithms for consistent and equitable AI interpretations.

We'll cover the following...

Relative input stability (RIS)

Stability refers to the robustness of an explanation with respect to small input perturbations. Relative input stability (RIS) is a popular stability metric that measures the maximum change in explanation relative to input changes.

In other words, given an image XX and its explanation SS, we generate perturbed versions {X1,X2,..,XN}\{ X_1, X_2, .., X_N\} of the original image by adding small random noise to Top-k influential features.

Here, XnX_n is the nthn^{th} perturbed image, ...