Synthesis of three-layer perceptrons for parity functions

Summary form only given, as follows. The synthesis of a three-layer perceptron has been proposed for a parity function. Since a parity function is not a linearly separable function, more than one processing element (PE) is required to realize a parity function. It turns out that the required number of PEs in the hidden layer of a three-layer perceptron is equal to n for an n-bit parity function. Since the weights of a PE in the proposed perceptron are set to either 1 or -1, the PE does not require multipliers. This will greatly facilitate actual hardware implementation of the proposed perceptrons using digital VLSI technology