Recent developments in supervised learning
Nicolaos B. Karayiannis · 2003
The author presents a generalized criterion proposed recently for the fast training of feedforward neural networks. A strategy is also described which allows the development of learning algorithms capable of determining the architecture of feedforward neural networks, i.e., the number of hidden layers and the number of hidden units per layer, while performing their training. It was experimentally verified that the proposed algorithms provide the simplest possible network for a given training task.>