5. Glossary

Machine learning algorithm; Algorithme d'apprentissage automatique

Algorithm for optimizing the parameters (weights and biases) of a neural network that calculates, from input data, output data that must best satisfy a loss function. Gradient backpropagation is an example of a learning algorithm based on gradient descent.

Gradient descent; Gradient descent

Method for finding a minimum of a differentiable function of several variables, exploiting information from the direction opposite the gradient of the function to be minimized, which gives the direction towards which the function will locally decrease. This method is applied to the determination of a minimum of an upstream chosen loss function, linking the neural network with data and/or a physical quantity of interest. The variables of the loss function are the...