SELF-ORGANIZED FORMATION OF A SET OF SCALING FILTERS AND THEIR NEIGHBORING CONNECTIONS

A set of scaling feedforward filters is developed in an unsupervised w ay via inputting pixel-discretized extended objects into a winner-take -all artificial neural network. The system discretizes the input space by both position and size. Depending on the distribution of input sam ples and below a certain number of neurons the spatial filters may for m groups of similar filter sizes with each group covering the whole in put space in a quasiuniform fashion. Thus a multi-discretizing system may be formed. Interneural connections of scaling filters are also dev eloped with the help of extended objects. It is shown both theoretical ly and with the help of numerical simulation that competitive Hebbian learning is suitable for defining neighbours for the multi-discretizin g system. Taking into account the neighbouring connections between fil ters of similar sizes only, i.e. within the groups of filters, the sys tem may be considered as a self-organizing multi-grid system.