Supervised Learning Using The Vector Memory Array Method

Abstract

The Vector Memory Array (VMA) is a novel neural network architecture. The principles of VMA are presented here and it is applied to data gathered by an Electronic nose in response to five simple odours (alcohols) and three complex odours (coffees). VMA achieved 100% accuracy on the alcohol data-set (40 samples) and 92% accuracy on the coffee data-set (90 samples) in just a few seconds. These results suggest a superior generalisation capability and learning speed compared to other neural paradigms, such as backpropagation, Alpaydin ’s constructive learning and logical neurons. Although VMA requires the assignment of the input vectors to an input hidden array layer, the associated memory cost may be offset in applications where fast processing and easy changes in training set are the principal requirements.