Diagonal Linear Networks and the Lasso Regularization Path

Raphaël Berthier

Diagonal linear networks are neural networks with linear activation and diagonal weight matrices. Their theoretical interest is that their implicit regularization can be rigorously…