Abstract
A systolic array implementation for a Reconfigurable Learning Machine (RLM) in the Occam 2 language (executed on transputers) is proposed. The theoretical model, Evolving Transformation Systems, which forms the basis for RLM, is outlined. This is a general adaptive model unifying existing pattern learning models. Central to our implementation of RLM is both the computation of a general Weighted Levenshtein Distance (WLD) using a known systolic array algorithm and substring matching. We propose a new algorithm for substring matching based on this systolic array as well as a general method to represent the properties of systolic arrays, in particular the two dimensional, hexagonal systolic array in the Occam 2 language. This method is defined as Systolic Array Mapping. Some initial timings are also included.