An Explainable Self-Labeling Grey-Box Model

Abstract

The massive success in machine learning in recent years has led to a wide spread of Artificial Intelligence (AI) models. Due to their enormous complexity, most of these AI models, notably the most effective type, Deep Learning Models, are classified as Black-box models, making them difficult to comprehend. Therefore, the goal of deployable, transparent AI models is the focus of the current research field known as Explainable Artificial Intelligence (XAI).Humans can learn how machine learning algorithms generate decisions through explanation, which leads to novel data-driven insights. In this work, we study an explanation approach so-called the Grey-Box model. The developped Grey-Box model uses a self-labeling framework based on a semi-supervised methodology. The key idea of the Grey-Box model is to exploit the benifits of black-box and white-box models. For this purpose, we implement a mechanism to increase a small initial labeled dataset. It allows to incorporate the model's most reliable predictions from a large unlabeled dataset. The proposed approach results in an efficient black box model that is accurate and interpretable.