The calculation of the output also occurs in two stages. First the input sub-vectors for each module are selected from the applied input vector according to the permutation function and the intermediate output is calculated by all modules. In a second stage all the outputs from the input layer are used as input to the decision network; then the final result is computed.

The mapping of the whole network is denoted by:

$\Phi o\Psi :\; RlRk$

The response $r$ for a given test input
$(a$_{1},a_{2},&ldots;,a_{l}) is determined by the following function:

$r\; =\; \Psi (\Phi (a$_{1}, a_{2}, &ldots;, a_{l}))

The k-dimensional output of the decision module is used to determine the class number for the given input. In the experiments the output neuron with the highest response was chosen as the calculated class. The differences between this neuron and the runner-up may be taken as a measure of accuracy.

Mit Okt 4 16:45:34 CEST 2000