A NOVEL APPROACH FOR LEARNING RATE IN SELF ORGINIZING MAP (SOM)
The performance of resultant topological structure of Kohonen Self Organizing Map SOM is highly dependent of the learning rate and neighborhood parameters. In literature there are plenty many different types of approaches to and proposals for those parameters. It has been investigated that in genera...
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Anadolu University
2018-03-01
|
| Series: | Anadolu University Journal of Science and Technology. A : Applied Sciences and Engineering |
| Subjects: | |
| Online Access: | http://dergipark.gov.tr/aubtda/issue/36292/363606?publisher=anadolu |
| Summary: | The performance of resultant topological structure of
Kohonen Self Organizing Map SOM is highly dependent of the learning rate and
neighborhood parameters. In literature there are plenty many different types of
approaches to and proposals for those parameters. It has been investigated that
in general the learning rate and neighborhood parameters are data independent
and predefined before the training period. Here in this paper a novel approach
has been proposed to change the learning rate parameter according to the
interaction of neurons with data. During training, the worst matching neuron
also tracked and used to trace the formation of topological structure of SOM. A
slight modification on conventional learning rate with proposed method has a
considerable influence on resultant topologies in a positive way. The effects
of this approach has been tested with the real world problem and different
synthetic data. |
|---|---|
| ISSN: | 1302-3160 2146-0205 |