Parity Madeline: a neural net with complete Boolean repertoire capable of one-pass learning

The authors explore n-input Madelines in which the fixed logic function in the output stage is chosen as the parity function. If the of neurons in the front slab of these parity Madelines is chosen to be the number of disjoint sets in a certain decomposition of the input hypercube, the resulting net has a complete Boolean repertoire. A parameter setting and a learning rule are found which enable the net to learn any Boolean function of n variables in a single training pass over the input hypercube. The net can be provided with an automatically varying generalizing ability, while retaining the full Boolean repertoire, by using a somewhat broader learning rule.<>