Improving neural network performance by adapting node nonlinearities

It is known that using an infinite number of hidden layer nodes feedforward neural networks can approximate any continuous function with compact support arbitrarily well using very simple node nonlinearities. We investigate whether network architectures can be found that use more complicated node nonlinearities to achieve better approximation using a restricted number of nodes. Two methods are proposed, one based on modifying standard backpropagation networks, and one based on Kolmogorov's theorem. The feasibility of these networks is evaluated by considering their performance when predicting chaotic time series and memorizing the XOR mapping.