The Universal Functions Originator and Its Extensions: Can They Solve the Explanation Issue in Modern Machine Learning Applications?

Abstract

Modern machine learning (ML) tools, such as artificial neural networks (ANNs) and support vector machines (SVMs), can provide highly accurate/precise predictions and estimations. However, in terms of explainability and interpretability, they are poor. To compromise between these key performance criteria, symbolic regression (SR) techniques could be used. However, they are hard to program because they have complicated mechanisms and need special optimization algorithms. The universal functions originator (UFO) is a new ML computing system that can be used in many computationbased applications. In addition to describing the variability of data sets as pure mathematical equations, this unique computing system has a very simple structure, and it can be initiated by any known optimization algorithm, including the most primitive ones, such as random search algorithms. This article introduces the UFO and shows why the system is so important to cybernetic applications.