Balancing learning and evolution
Michael Hüsken, Christian Igel · 2002
Finding the right coupling of learning and evolution in a hybrid algorithm is an open problem. In this article, we present a strategy to adjust the time spent on learning during evolutionary optimization of neural networks. The proposed adaptation scheme leads to a significant improvement in performance. It is empirically shown that suitable learning strategies strongly depend on the problem and that it is advantageous to adapt the time spent on learning during evolution. 1