Glossary

Deep learning

Machine learning method using neural networks with several layers of hidden neurons. As these algorithms have a large number of parameters, they require a very large amount of data to be trained.

Inductive bias

A priori assumption or information that a learning algorithm uses to predict the output of an input it has never encountered.

Contrainte douce/dure; soft/hard constraint

When incorporating physical knowledge into a learning model, a soft constraint encourages the model to respect the constraint, but it can still be violated. In contrast, a hard constraint is automatically verified by the model.

Différentiation automatique; automatic differentiation

Automatic differentiation is a technique that...