Lessons learned from the AutoML challenge

Abstract

We give a brief account of the main findings of our post-hoc analysis of the first AutoML challenge (2015-2016). This competition, which took place in 2015-2016 challenged the participants to submit code that solve classification and regression problems from fixed length feature representations, without any human intervention. This paper is a digest of a book chapter to be published in the Springer Series on Challenges in Machine Learning. All datasets, code of the winners, and challenge results are found at: http://automl.chalearn.org.

Tasks

Reproductions