Charles Explorer logo
🇬🇧

Multi-objective evolution of machine learning workflows

Publication at Faculty of Mathematics and Physics |
2017

Abstract

In this paper we describe a multi-objective genetic programming algorithm which can be used to create complete machine learning workflows. The algorithm is an extension of a single-objective one.

In a series of test on four datasets, we show that the additional objectives can be used to search for smaller or faster models. The algorithm is also in some cases much faster than the single-objective one while obtaining results of similar quality.