Explainable and Reliable Artificial Intelligence

Learning Goals

The course will introduce the students to techniques in of explainable Artificial Intelligence and machine learning, ranging from interpretable-by-design models to post-hoc explanations. It will also introduce methods to measure and enforce reliability of machine learning models, including verification and methods to build robust predictors. Students will learn which are the best techniques to provide explanations and how to assess and enforce reliability depending on data and criticality of applications.

Program in pills

Introduction to Explainable AI. Transparent-by-design models: rule based and logic based models. Post-hoc explainability: feature importance, LIME, SHAP, relevance propagation and gradient based methods, global explanation approaches. Reliable AI: adversarial training, robustness of bayesian models, verification of ML models and neural networks.i

Area

Machine Learning and Artificial Intelligence

Curriculum Foundations
TAF Type

C

Curriculum Industry
TAF Type

D

Curriculum Health
TAF Type

D

Curriculum Economy
TAF Type

D

SSD

INF/01

ECTS

6

Semester

1

Lecturers

Luca Bortolussi
Tatjana Petrov