From MSE to Correntropy, a friendly survey

Correntropy is a metric that has been widely used in place of the root mean square error in problems where it is intended to minimize the divergence between data and models. In particular, machine learning is currently in focus, where increasingly complex models require increasingly statistically heterogeneous data. In this article we will give an introduction to correntropy in a friendly and intuitive way. Contrary to purely technical summaries, we will try to balance a precisely technical language with a freer and more informal text. We will present the history of how correntropy came to be developed, in order to lead the reader to a coherent temporal sequence that will facilitate the precise understanding of this new metric.