[Invited] Quality Assurance of Machine Learning Software
Shin Nakajima · 2018
Functionalities of machine learning software are dependent on a set of data input to them; a slight change in a training dataset has much impact on learning parameter values and thus on inference results. ML-based systems bring about a new challenge to quality assurance methods. This paper reviews two traditional views of service and product qualities. Furthermore, it introduces a platform view, in which co-creation of value is a major concern.