The paper delves into the epistemic justification for a simplicity preference in inductive inference, derived from the machine learning framework of statistical learning theory. It combines elements from earlier arguments suggesting and rejecting such justification, presenting a qualified means-ends and model-relative justificatory argument. The argument is built on statistical learning theory’s central mathematical learning guarantee for the method of empirical risk minimization. The paper also addresses the ongoing debate on the validity of Occam’s razor in machine learning methods and presents a qualified epistemic justification for Occam’s razor in statistical learning theory.
Publication date: 21 Dec 2023
Project Page: https://arxiv.org/abs/2312.13842v1
Paper: https://arxiv.org/pdf/2312.13842
