THE NEURAL NETWORK

The neural network that constitutes the basis of the inference engine is based on the architecture of the A.R.T and Fuzzy A.R.T [17], [18], [19], [20], [26], [37], [47], and is therefore a competitive (unsupervised or self-organised) net. It is shaped initially with three nodes that are expected to increase in real-time if needed. The input, as well as the output, is an n x m vector or a grid of n x m pixels. (click to see picture)

Each node learns to detect a specific form within the grid. The net is basically a "rule detector", trying to discover patterns within the data that constitute rules for the configuration of space. A rule is a fuzzy group of transformations: if the input is contained to this group then the group is enhanced, otherwise it remains unchangeable. So the function and purpose of this net is two-fold: to detect rules and to learn rules. To this end we may think of the weight vectors as archetypal patterns that coordinate the formation of data clusters.