Opening the Black Box of Artificial Neural Networks: An Interpretation of Multilayer Perceptron as an Instance-based Learner

This dissertation provides a transparent explanation of a foundational artificial neural network. Artificial neural networks are at the heart of state-of-the-art artificial intelligence but are generally opaque to interpretation. This dissertation benefits society by providing broadly accessible insight into what drives this technology. The dissertation interprets the mathematics of the multilayer perceptron artificial neural networks to explain its internal logic and behaviour in a non-mathematical frame of reference. The network is framed as a memory store for training patterns. A new input retrieves samples of similar patterns that are combined into idealisations from which output decisions are made.