Chris Ick, Vincent Lostanlen
Proceedings of the 1st DeepMath conference, New York City, NY, USA, November 2020
Publication year: 2020

This article proposes a machine learning method to discover a nonlinear transformation which maps a collection of source vectors onto a collection of target vectors. The key idea is to learn the Lie algebra associated to the underlying one-parameter subgroup of the general linear group. This method has the advantage of not requiring any human intervention other than collecting data samples by pairs, i.e., before and after the action of the group.