Teacher

Gloria Menegaz

Referent

Gloria Menegaz

Credits

2.5

Language

Inglese

Class attendance

Free Choice

Location

VERONA

Moodle Seminars 0

Learning objectives

Artificial Intelligence has become a fundamental instrument in fields like biomedicine and neurosciences, from the discovery of new numerical biomarkers to support to the diagnosis. In the last decades, large multidisciplinary collaborations and long-term multimodal studies as, e.g., ADNI, ENIGMA, and UK Biobank, made possible to access big repositories of different type of data like images and genetics information. Such an availability designates deep learning as particularly attractive to represent the complex relationships underlying many biological processes. However, despite the undeniable advantages of deep learning models like deep neural networks, the complexity of their architecture makes mandatory to obtain explanations favoring the interpretability, especially in medicine, healthcare and neuroscience fields. For this reason, the eXplainable Artificial Intelligence (XAI) is fundamental to explain how the model reached a specific outcome, how the features contributed, and to what extent the model is confident about the decision. This course aims at providing the students knowledge about the explainable models at the state of the art that are mostly exploited in the field, providing both theoretical bases and implementation tools. Especially, the students will learn how to deal with explainable AI when applied to interpret deep learning models assessed for extracting information from multi-dimensional heterogeneous and noisy data.

Prerequisites and basic notions

Fundamentals of signal and image processing, fundamentals of machine learning. Python.

Program

Program:
Part 1: Promises
• eXplainable Artificial Intelligence from scratch
• A hint into agnostic post-hoc explainability models (e.g. feature importance, occlusions)
• Neural networks in the light of explainability (e.g. visualization methods, attribution methods, gradient-based methods)
Part 2: Challenges
• Validation of XAI outcomes

Bibliography

Didactic methods

In person.

Learning assessment procedures

Project related to the topics covered.

Assessment

Critical analysis of the methods learned
Ability to apply the methods learned to concrete problems Quality of presentation

Criteria for the composition of the final grade

Pass/Fail