On Hidden Nodes for Neural Nets
G. Mirchandani and W. Cao
IEEE Transactions on Circuits and Systems - Special Issue on
Neural Networks, Vol. 36, No.5, pp.661-664, May, 1989
(Paper)
Recent results indicate that the number of hidden nodes (H)
in a feedforward neural net depend only on the number of
input training patterns (T). There appear to be conjectures
that H is of the order of (T-1) and log_2 T. The main
contribution of this work is a proof that maximum number
of separable regions (M) in the input space is a function
of both the H and the input space dimension (d). We also
show that H = M - 1 and H = log_2 M are special cases of
that formulation. M defines a lower bound on T, the number of
input patterns that may be used for training. Application
to some experiments are investigated.