Sparse dictionaries for the explanation of classification systems

Providing algorithmic explanations for the decisions of machine learning systems to end users, data protection officers, and other stakeholders in the design, production, commercialization and use of machine learning systems pipeline is an important and challenging research problem. Crucial motivations to address this research problem can be advanced on both ethical and legal grounds. Notably, explanations of the decisions of machine learning systems appear to be needed to protect the dignity, autonomy and legitimate interests of people who are subject to automatic decision-making. Much work in this area focuses on image classification, where the required explanations can be given in terms of images, therefore making explanations relatively easy to communicate to end users. In this paper we discuss how the representational power of sparse dictionaries can be used to identify local image properties as main ingredients for producing humanly understandable explanations for the decisions of a classifier developed on the basis of machine learning methods.