Introduction to AI Interpretability

Crash Lecture Series. Semester of Spring 2021

This lecture series is a preview of the full course of 16 hours starting in Autumn 2021.

Crash course Overview:

The “where and why” of interpretability. Reading of taxonomy papers and perspectives.

The “three dimensions” of interpretability. Activation maximizations, LIME surrogates, Class Activation maps and Concept Attribution methods seen in details with hands-on exercises.

Evaluation of interpretability methods. Reading of papers and hands-on exercises.


Confidence in linear algebra, probability, machine learning. Experience with Python, numpy, tensorflow.

Course registration: email mara dot graziani at hevs dot ch

Lecture 1: The Where and Why of Interpretability

Prof. Henning Müller and Mara Graziani (PhD student), Hes-so Valais and University of Geneva

May 20, Microsoft Teams

Photo by Pixabay on

Lect 2: From attention models and saliency maps to the explainability of Deep Neural Networks. 

Prof. Jenny Benois-Pineau, University of Bordeaux

May 27, Microsoft Teams

Lect. 3: The three dimensions of interpretability. Activation Maximization and Feature Attribution

May 28, Microsoft Teams

Lect. 4:  

June 4, Microsoft Teams

Lect. 5: Evaluating interpretability

%d blogger hanno fatto clic su Mi Piace per questo: