Abstract:
We use a simple network that uses negative feedback of activation and simple Hebbian learning to self organise in such a way as to produce a feature map which has the property of identifying the relative proportions of the components of the input data. Thus, it evaluates the angular properties of the input data space and ignores the magnitude of the input data. We have previously shown that we can easily find a one dimensional mapping of input data; we now present an improved algorithm which finds a three or more dimensional sphere of the input data. By creating a non- rectangular two-dimensional map, we can observe convergence to the shape of a sphere which exhibits the scale invariant classifying properties previously seen in the network.