A coded block neural network system suitable for VLSI implementation using an adaptive learning-rate epoch-based back propagation technique

A coded block adaptive neural network system is presented. It is suitable for VLSI implementation using an adaptive learning-rate epoch-based backpropagation technique to train large-volume input patterns. Using this coded block neural network system, 500 frequently-used Chinese characters were successfully trained in 47.2 hours using a 28 MIPs computer. Training of the epoch-based system is much less sensitive to initial weights and irrelevant to the order of the input patterns as compared to the system using the conventional backpropagation algorithm