If we have a neural network such as the multilayer perceptron back propagation neural network that uses sigmodial logistic activation functions is it possible to feed the network outputs and have it compute back a set of inputs? Since we can reverse the activation function by using the natural logarithm and inverse operations until we have a sum value that is made up of all the weights multiplied by their inputs i would think that it would be possible to at least get sets of possible inputs that will generate the specified output value.
If we have a neural network such as the multilayer perceptron back propagation neural
Share
As there is an infinite number of possibilities to distribute the result to the feeding neurons this is only possible for each layer where the number of receiving neurons equals the number of the feeding neurons. There it is a simple linear equation which boils down to a Matrix multiplication/inversion problem, that can easily be solved.
If the number of neurons where the number of the receiving neurons (n) is smaller than the number of the feeding neurons (m), you have as solution space a manifold with m – n dimensions.