Separable vs. End-to-End Learning: A Critical Examination of Learning Paradigms
Résumé
Machine learning is undoubtedly becoming more and more integrated into medical image computing research and practice. As with any conceptual or technological tool, this exposure requires the community to better understand the logical and philosophical foundations underpinning its use. One of the these (largely unexamined) areas is the learning paradigm that motivates everything from problem decomposition to architecture selection to clinical regulation. This article examines one of the spectra that defines the learning paradigm, specifically the spectrum between end-to-end and separable learning, in order to analyse what values drive solutions towards or away from either extreme. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.