|
|
|
|
|
|
|
|
Hybrid Genetic Algorithm for Heterogeneous Recurrent Neural Networks Construction
|
|
|
|
|
|
Organization: | Czech Technical University in Prague |
Department: | Computational Intelligence Group Department of Computer Science and Engineering Faculty of Electrical Engineering |
Organization: | Czech Technical University in Prague |
Department: | Department of Computer Science and Engineering Faculty of Electrical Engineering |
|
|
|
|
|
|
International Mathematica User Conference 2008
|
|
|
|
|
|
Champaign, IL
|
|
|
|
|
|
Our research is focused on optimization of recurrent neural networks using evolutionary algorithms. We are working with a hybrid evolutionary algorithm called the Continual Evolution Algorithm (CEA). Application of this algorithm to the problem of recurrent neural networks construction is the main part of our work. The CEA algorithm is used in our experiments to construct the main topology of the networks and for setting up the activation functions in particular neurons to provide complex models. The behavior of the population in the CEA and the evolution control mechanism are studied. We are presenting the methods and their implementations in our Mathematica package.
|
|
|
|
|
|
|
|
|
|
|
|
http://www.wolfram.com/news/events/userconf2008
|
|
|
|
|
|
| HybridGeneticAlgorithm_Abstract.nb (251.3 KB) - Mathematica Notebook | | HybridGeneticAlgorithm_Presentation.nb (5.4 MB) - Mathematica Notebook |
|
|