Accepted:
12 Jul 2001
| |
/vol09/vasech01/ | © Copyright 2001 | |||
| Volume 09 | Received: Accepted: |
06 Apr 2000 12 Jul 2001 |
|||
|
Evolving polynomial neural network by means of genetic algorithm: some application examples
Vasechkina, E. F. & Yarin, V. D. |
|
| Abstract | |
| The new technique for evolving feed-forward NN architecture by self-organising process is proposed. The approach doesn't require learning as separate process since evaluation of weights is carried out simultaneously with the architecture construction. The NN architecture is built by adding hidden layers to the network, while configuration of connections between neurons is defined by means of GA. GA runs for each neuron searching its optimal set of connections with the preceding layer. The output of the neuron is represented as polynomial function of the inputs with coefficients evaluated using the least-squares method. The proposed method was applied in a suite of tasks. It allows evolving compact feed-forward NN structures giving good decision to different real world problems and does not require large computational costs. It is able to generate an equation directly expressing the algebraic dependence of the output on the inputs. So it can be successfully applied to different simulation problems arising in biology, ecology and other natural sciences. | |
| Full Text |
|
|
Multimedia Links (none) Reference Links Citation Reference Vasechkina, E. F. & Yarin, V. D. (2001), Evolving polynomial neural network by means of genetic algorithm: some application examples, Complexity International, Volume 09, Paper ID: vasech01, URL: http://www.complexity.org.au/vol09/vasech01/ |
Get viewers for PS & PDF   |
