Binary backpropagation in content addressable memory

Binary backpropagation, a variation of the standard continuous backpropagation network learning model, is introduced as an efficient associative memory for binary patterns. Binary backpropagation employs local computation for corrections to bit connection weights. Restriction to binary inputs, outputs, and weights allows several-orders-of-magnitude faster learning convergence. Binary backpropagation is based on content-addressable memory and has similar hardware requirements. A pseudoanalog extension of binary backpropagation allowing arbitrary bit-level significance is also presented