Consider the problem of neural network inversion, wherein, given a fixed network and target output values, the objective is to learn the associated input patterns which, when forward- propagated through the network, produce outputs that well-approximate the targets. Sketch a method (an optimization technique ?) that approximately achieves NN inversion. Also suggest some possible applications for NN inversion.