Reliability modeling and design criteria for the backpropagation artificial neural network

The author introduces a methodology which takes into account parametric behavior that contributes to reliable outputs. An approach to modeling the reliability of 'reliable' networks is developed. A major contributing point to the theory behind the model is that the weights connecting the hidden and output layers of the network must follow a normal distribution. This model is valid when and only when the normal distribution criterion is met. The model is applicable when a single connecting weight is set to 0 by component failure.<
>