Global optimization of recurrent neural networks: a comparison of the genetic algorithm and Tabu search

Abstract

To present the methodology of applying Genetic Algorithm and Tabu Search in finding the global optima in Recurrent Neural Network. Then the result is compared with Backpropagation, the legacy method. The result depicts that Genetic Algorithm and Tabu Search can help Recurrent Neural Network performs better than Backpropagation. This is because the Genetic Algorithm has a cross-over operator to jump off of local optima whilst Tabu Search employs Tabu list to prevent re-cycling search as well as using long term memory to make the searching broader. However, Genetic Algorithm and Tabu Search take more time to find out the solution. In a short time running, Backpropagation can find a solution in some dataset better than others.