The current profusion of pattern recognition demonstrations using data driven methods tends to obscure the preferability in many applications of statistical methods. In place of machine learning, these approaches exploit the knowledge of a subject matter expert, who devises a reliable description of the relevant phenomenology along with a statistical model of signal variations. In the simplest cases a detection algorithm can be devised that is optimal, subject to the accuracy of the statistical model.

© 2019 The Author(s)

PDF Article