A taxonomy of neural network optimality

The authors discuss some of the criteria which can be used to decide when one network topology is better than another. The focus is on neural networks which generate their topology during training. The taxonomy presented can be used to determine methods for comparison of different neural network paradigms. The criteria for determining what is an optimum network is highly application specific. The criteria were developed with the idea of applying them to the field of ontogenic neural networks. This means, for example, that the criteria will be used to determine whether it is better to add new layers or new nodes to a layer. The criteria may also help to determine when the optimum topology has been achieved. It could further be used to help evaluate different ontogenic paradigms. This taxonomy has been discussed on the Internet. The comments of the responding researchers are incorporated.<
>