Approximation capabilities of multilayer feedforward networks
Neural Networks
Genetic Programming III: Darwinian Invention & Problem Solving
Genetic Programming III: Darwinian Invention & Problem Solving
Evolutionary Computation: Towards a New Philosophy of Machine Intelligence
Evolutionary Computation: Towards a New Philosophy of Machine Intelligence
Data Mining and Knowledge Discovery with Evolutionary Algorithms
Data Mining and Knowledge Discovery with Evolutionary Algorithms
| Hi-index | 0.00 |
Evolutionary optimization algorithms contain, due to their heuristic inspiration, many heuristic parameters, which need to be empirically tuned for the algorithm to work most properly. This paper deals with tuning those parameters in situations when the values of the objective function have to be obtained in a costly experimental way. It suggests to use a neural-network based approximation of the objective function for parameter tuning in such situations. In this way, the convergence speed of the algorithm and the density of the population of points can be investigated for many various combinations of heuristic parameters. To construct the approximating neural network, some initial amount of data is needed, usually obtained from running the algorithm for several generations with default values. The feasibility of the approach is demonstrated on a case study in materials science.