|
Abstract
|
|
|---|---|
| Metaplasticity concept was defined in 1996 by W.C. Abraham and presently is a biological concept widely known in the fields of biology and medicine: neuroscience, physiology, neurology and others. Inspired in it, outstanding improvements have been achieved in artificial neural networks design applied to pattern classification. The proposed training algorithm is inspired by the biological metaplasticity property of neurons and Shannon?s information theory. The concept is applicable to Artificial Neural Networks in general, although in this presentation it is centered on Multilayer Perceptrons (MLP). During the training phase, the Artificial Metaplasticity Multilayer Perceptron (AMMLP) algorithm gives higher values for updating the weights in the less frequent activations than in the more frequent ones. AMMLP achieves a more efficient training, while improving MLP performance. Tested in standard, well known and easy available Databases, its results are superior to the rest of algorithms, no matter what multidisciplinary application used as case study | |
|
International
|
Si |
|
|
978-1-4577-1122-0 |
|
Entity
|
IEEE SMC |
|
Entity Nationality
|
E.E.U.U. DE AMERICA |
|
Place
|
Salamanca, España |