Search⌘ K

Ground-Truth Faithfulness—Rank Correlation

Explore the concept of rank correlation as a key metric for ground-truth faithfulness in interpretability. Learn how to compare feature importance rankings between post-hoc explanations and ground-truth explanations using Spearman's rank correlation coefficient. Understand its practical application through code implementation involving vanilla, smooth, and integrated gradients on image classification models.

We'll cover the following...

The rank correlation is a metric that computes Spearman’s rank correlation coefficient to measure the agreement between feature rankings provided by a given post-hoc explanation and the corresponding ground-truth explanation. In other words, given a post-hoc explanation SS and the ground-truth explanation SgS_g, the rank correlation RA(S,Sg)RA(S, S_g) ...