Back propagation and forward propagation

The authors analyze the validity of the pessimistic views of M.L. Minsky and S.A. Papert (Perceptrons: An Intro. to Comput. Geom., MIT Press, Cambridge, MA, 1969) on the possibility of training multi-layer perceptrons. In particular, they show that it was possible, using the simplex algorithm of J.A. Nelder and R. Mead (Comput. Journ., vol.8, pp.308-13, 1965), to train these networks at the time of writing of Minsky's and Papert's book, and demonstrate this experimentally on three real-life classification problems, each involving a specific data set.<
>