Evolutionary Machine Learning: An Essay on Bechmarking

Authors

Abstract

The search for adequate structures and parameters for Machine Learning (ML) models is problem specific and time consuming. Often, researchers follow an iterative trial-and-error process, where suitable values for multiple parameters are tested. One way to address this issue is the application of Evolutionary Computation (EC) to search, optimise and tune the ML models. Selecting appropriate benchmarks for comparing different approaches is not always trivial and is a common problem of both EC and ML. However, when combining both fields it is possible to use the evolutionary process to our advantage, speeding up the evaluation stage. In this paper we discuss what can be done to mitigate some of the issues of benchmarking in Evolutionary Machine Learning (EML). The positions herein presented denote the point of view of the authors and should not be seen as a strict methodology, but rather as a set of guidelines.

Conference

Portuguese Conference on Pattern Recognition (RecPad) 2017