In a recent study of evolutionary artificial neural networks (EANNs), it has been argued that a partial training process after an architectural mutation plays an important role in maintaining the behavioural link between parents and their offspring and thus is beneficial to the simulated evolution. This paper investigates the issue further through a number of experiments. The experimental results show that a closer behavioural link between parents and their offspring due to the partial training process does lead to better performance, i.e., evolved ANNs generalize better. The results also illustrate that given a fixed amount of time there is an optimal balance of time between evolution and training (learning).
|Title of host publication
|Proceedings of the IEEE Conference on Evolutionary Computation, ICEC
|Number of pages
|Published - 1997