Learning the Activation Function for the Neurons in Neural Networks
Ever since the sigmoid replaced the threshold as the main activation function used in artificial neural networks, the properties of the activation function have been largely ignored. Most research aimed at improving the quality of the hypothesis or the speed at which it is obtained has concentrated on the topology or enhancing the learning algorithm (normally back propagation).
Unable to display preview. Download preview PDF.
- Fletcher, G.P., Hinde, C.J., West, A.A. & Williams, D.J., (1994) Neural networks as a paradigm for knowledge elicitation, ibid.Google Scholar