Session moderated by Stanislas Dehaene.
Each 30' paper will be followed by a 10' discussion.
Abstract
Machines learn by optimizing prediction models. The surprise comes from the extraordinary ability of neural networks to solve problems as diverse as image recognition, language generation or the prediction of physical measurements. The mathematical principles behind this learning process remain poorly understood, which is a major challenge if we are to guarantee its reliability. Despite significant differences, there are convergences with human learning, and we must now also consider the future impact of artificial intelligence on education.