Parameter genetic learning of perceptron networks
Roman Neruda, Stanislav Slušný · International Conference on Systems · 2006
This paper reviews different combinations between the most widely used type of neural networks -- a multi-layer perceptron -- and evolutionary algorithms. Several methods to train the weights of the network are tested using a real-world classification problems from Proben1 benchmark suite. It is shown, that combining evolutionary algorithms with neural networks can lead to better results than relying on neural networks alone. Comparison to gradient algorithms is discussed.