Evolutionary Machine Learning: An Essay on Experimental Design

Authors

Abstract

Evolutionary Machine Learning (EML) combines Evolutionary Computation (EC) with Machine Learning (ML) to automatically search for the best structure and/or parameterisation of ML models for solving specific tasks. However the results reported by the authors in their articles detail their work, replicating the results and comparing them to other approaches are tasks that tend to be difficult. This happens mainly because of the high number of numeric parameters, and specific technical details. Another issue that prevents the approaches from being replicated is the fact that the code developed is rarely made publicly available. In this essay we discuss and provide some guidelines to address these problems. Our goal is not to provide a unique, right answer, for these issues. Rather, we aim to promote a healthy discussion that can lead to new and innovative ideas and practices.

Conference

Portuguese Conference on Pattern Recognition (RecPad) 2017