Genetic Algorithms as Optimisers for Feedforward Neural Networks

Currently, multilayer feedforward neural networks with backpropagation learning rule are the most popular artificial neural networks for handling optimisation and learning problems. This success is largely based on the simplicity to implement these networks. However, there are some major drawbacks: there is no guarantee for training in finite time and there is no guarantee that the optimal solution will be reached. In the sequel, the application of genetic algorithms as learning rule (thus, as optimiser) is compared to the error backpropagation learning rule (i.e., gradient descent optimiser).