Adaptive learning circuit based on the Walsh transform

A novel adaptive function learning circuit based on appropriately configured Walsh matrices and the Walsh transform compares favourably with artificial neural networks. A Walsh-vector generator produces parallel outputs corresponding to the circuit input. A single layer of weights connected to the Walsh-vector generator outputs is summed to produce the circuit output and is updated using a simple algorithm. Fast and reliable learning is obtained with the weights themselves converging towards the coefficients of the Walsh transform of the desired output function