Analysis on Approaches and Structures of Automated Machine Learning Frameworks

Abstract

Due to the explosive and increasing demands of machine learning, automated machine learning is developed to handle machine learning tasks for non-experts. Lots of AutoML frameworks are introduced in past decades. Each of them has their unique contributions towards AutoML field. In this paper, four popular open source AutoML frameworks are selected and reviewed to show current development directions and common features for frameworks. This paper also analyzes innovative structures and designs of selected frameworks. The result shows that one of the newest frameworks, AutoGluon, has extraordinary performance and innovative structures when compared to others.