Improving the learning speed of 2-layer neural networks by choosing initial values of the adaptive weights
The authors describe how a two-layer neural network can approximate any nonlinear function by forming a union of piecewise linear segments. A method is given for picking initial...