Experiments using minimal-length encoding to solve machine learning problems

Abstract

Describes a system called Emily which was designed to implement the minimal-length encoding principle for induction, and a series of experiments that was carried out with some success by that system. Emily is based on the principle that the formulation of concepts (i.e., theories or explanations) over a set of data can be achieved by the process of minimally encoding that data. Thus, a learning problem can be solved by minimising its descriptions.<>