02.03.2013 Views

TOR VERGATA UNIVERSITY UNSUPERVISED CLASSIFICATION ...

TOR VERGATA UNIVERSITY UNSUPERVISED CLASSIFICATION ...

TOR VERGATA UNIVERSITY UNSUPERVISED CLASSIFICATION ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Chapter 1 19<br />

This interaction determines a modification of the weights dependent on<br />

neighborhood function and the response of the neuron. Gauss‘s function can be<br />

used for this purpose:<br />

where rc is the position vector of the winning neuron, rj is the position vector of the<br />

jth neuron of the map (Fig. 1.2) and σ is named proximity parameter.<br />

Fig. 1.2: radius of interaction<br />

When the process starts, σ parameter has a high value, thus the area of bubble is<br />

wide. During the learning phase, the bubble decreases its dimension, until a certain<br />

user-defined value.<br />

The weights of the neurons in the bubble are, therefore, updated according to the<br />

following formula:<br />

where is the learning rate and decreases gradually during the learning phase.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!