Comparing different neural network architectures for classifying handwritten digits.

I. Guyon, I. Poujaud, L. Personnaz, G. Dreyfus, J. Denker, and Y. Le Cun .
In Proceedings of the International Joint Conference on Neural Networks , volume II, pages 127--132, Washington DC, IEEE.

We have evaluated several neural network classifiers, comparing their performance on a typical problem, namely handwritten digit recognition. For this purpose, we use a database of handwritten digits, with relatively uniform handwriting styles. We propose a novel way of organizing the network architecture by training several small networks so as to deal separately with subsets of the problem, and then combine the results. This approach work in conjunction with various techniques including: layered networks, with one or several layers of adaptive connections, fully connected recursive networks, ad hoc networks with no adaptive connections, and architectures with second degree polynomial decision surfaces.

[ next paper ]