On Membership of Black-box or White-box of Artificial Neural Network Models

Artificial Neural network (ANN) has been widely used in solving complex real world problems. It is commonly known as a black-box approach in the sense that examples (input and output pairs) are learned in the process of establishing an ANN model. However, there is a long debate in literature on whether ANN is a white-box or a black-box approach. In this article, we take a more conceptual view towards the nature of modelling with ANN, leading to a conclusion that ANN can be a black-box or grey-box approach, depending on the way to create an ANN model, in particular whether the knowledge of the problem and/or domain of the problem to be modeled by ANN is used in the creation of the ANN model. Another contribution of this paper is the outline of a methodology for designing ANN models, which is based on our proposed view in this paper that designing an ANN model is like designing a product and then the design theory and methodology for general products is learned to designing an ANN model.