| 103D1 |
|---|
| Acquisition of 2-layer structure in a growing neural network |
| AUTHORS |
| Ryusuke Kurino (oita university) |
| ABSTRACT |
| Neural networks are broadly used to approximate non-linear functions. Conventionally some methods to decide the structure have been proposed. However, in these methods, some structure, such as 3-layer, is assumed and some parameters are acquired. Accordingly, the degree of freedom to decide the structure is not large enough. In this paper, ¡Ègrowing neural network¡Éis proposed as an extension of Back Propagation (BP) learning for freer acquisition of the structure. This algorithm gets a hint from NGF (nerve growth factor) that is thought to contribute to promote the growth of axons and maintain the connection between neurons in the natural nerve system. It was examined through a simulation whether the algorithm worked appropriately in a simple problem. There were 4 input neurons and 2 output neurons. One output was required to output "AND" of two inputs. The other output was required to output "OR" of the other two inputs. Even though two input neurons were located farther from the corresponding output neuron than from the other output, their axons grew to the corresponding output neuron, and the appropriate structure could be obtained. |
Back to Program
Time Table
Top of This Site
Annual Conference 2002