XAI - Explaining Multi-Layer Networks : The effect of complexity on the interpretability of multi-layer perceptron networks

Abstract

Machine learning systems and neural networks are often described as black boxes, or closed systems that are given an input and spit out an answer after performing several tasks to the input values. XAI, or explainable AI aims to reveal what is happening inside the so-called black box and explain the process in an intelligible way using measurements. Explainable AI is an important step in making machine learning more widely adaptable to several integral systems, such as health care. This thesis focuses on multi-layer perceptron networks and how their explainability changes as their complexity increases.