Backpropagation Algorithm for LOGic Oriented Neural Networks with Quantized Weights and Multilevel Threshold Neurons

摘要

Multilayer feedforward neural network (MFNN) trained by thebackpropagation (BP) algorithm is one of the most significant modelsin artificial neural networks. MFNNs have been used in many areas ofsignals and image processing due to high applicability. Although theyhave been implemented as ana- log, mixed analog-digital and fullydigital VLSI circuits, it is still difficult to realize theirhardware implementation with the BP learning function efficiently.