1. Reinforcement learning

The design of an artificial agent with an advanced degree of autonomy in carrying out its mission relies on the implementation of reasoning and decision-making capabilities, which have been at the heart of artificial intelligence work since its inception. However, as it is impossible for the designer to foresee every possible situation that could arise in even the most complex use cases, this autonomy must also be based on learning capabilities that enable it to integrate new information and acquire new action capabilities.

Machine learning distinguishes three main classes of learning algorithms.

1. The aim of unsupervised learning is to learn how to identify statistical regularities in an input data stream, in order to reveal any hidden structures that may have generated this stream. It can be used to categorize these inputs,...