Matrix realisations of multilayer perceptron ANN
R.F. Chidzonga · 2002
The backpropagation neural network training algorithm is formulated via matrix transformations as opposed to the usual indexed algebraic scalar approach. This formulation allows for easy visualisation and understanding of error propagation and the consequent weight adjustment. The so configured network can be readily unravelled for further study if required. Illustrative results on simple and concise Matlab simulation are presented.