Selecting Algorithms by Using ATD

Abstract

The Anthropological Theory of the Didactic (ATD) has proven its usefulness to describe human learning, and has improved the methods used to acquire new knowledge. However, although the same framework could be used in the context of machine learning, no attempts have appeared in the literature to apply ATD within computing environments. In this paper we argue that we can also take profit from ATD in machine learning. In particular, we will show how we can use it to describe a simple system where we need to learn to select an algorithm among a set of predefined options. The approach will be used to analyze both human learning and machine learning, studying the similarities and differences between both of them.