This course is open to Post-doc researchers, PhD and MSc students. AIDA Members, namely university partners of ICT48 Projects: AI4Media, HumanE AI Network, VISION. PhD students, can request for ETC credits upon completion of this course.
Course Objectives: Learn to recognize when and how applying interpretability. Understand the dimensions and scopes of interpretability. Analyze and criticize the literature, definitions and evaluation methods. Implement state of the art techniques and evaluate their outcomes.
Applications: Healthcare, Multi-media indexing
Prerequisites: Confidence in linear algebra, probability, machine learning. Experience with python, numpy, tensorflow, keras.
Extra Material and Readings
https://www.barnesandnoble.com/w/making-ai-intelligible-herman-cappelen/1138471724
Transparency: Motivation and challenges
Towards Better Understanding of Gradient-based Attribution Methods for Deep Neural Networks
An Evaluation of Human Interpretability of ExplanationsUnderstanding Neural Networks via Feature Visualization: A Survey
The lecturers
Henning Müller, Professor at Hes-so Valais and University of Geneva
Mara Graziani, PhD student at Hes-so Valais and University of Geneva and member of the AIDA curriculum program
Jenny Benois-Pineau, Prof. at the University of Bordeaux
Georges Quénot, Prof. at IMAG, Grenoble, France
Introduction to AI Interpretability 