Abstract: Based-on a single-layer pattern recognition system (WISARD) and referred to the concept of a sparse distributed memory (SDM) by P.Kanerva, this paper presents a novel multi-layer adaptive pattern recognition system model and describes in detail its processing procedures and main features. The results of primitive multi-font printed Chinese character recognition experiment are given.Unlike WISARD, MAPR replaces the normal RAM array with sparse RAM array, and the directly setting bit strategy with counting frequency training for n-tuple pattern, so that not only MAPR maintains the main advantages of WISARD, but its performance is obviously improved as well in the aspect of recognizing large-dimensional or non-deterministic pattern data.

Key words: pattern recognition system, sparse distrobuted memory