You must login before you can run this tool.

## [Illinois] KohonenSOM

This tool implements the Kohonen self-organizing map (SOM) algorithm

#### Category

#### Published on

#### Abstract

In this tool, KohonenSOM is used to train neural networks using simple (competitive) unsupervised learning, in which only the winner is trained, and progress to full SOM(self-organizing maps) strategies in which feature maps of various types are formed. Unsupervised learning causes output units to specialize for specific input patterns. The tool determines the input patterns and the number of input units required to represent those patterns, by finding the numbers of rows and columns of the input array. The number of output units must be specified in the tool under nOut. The neighborhood is the area that stores the topological properties of the input space. It must be set to 0 when we wish to have the output unit with the largest response to a given pattern trained by itself on each iteration during competitive learning. The KohonenSOM algorithm is plausible neurobiologically and approximates the synaptic integration as the computation ny a neural unit of its weighted sum. The normalization equation in the tool keeps the total amount of synaptic input to a neural unit constant. This appears to the same role in real neurons.

#### Sponsored by

NanoBio Node, University of Illinois Champaign-Urbana

#### References

Anastasio, Thomas J. Tutorial on Neural Systems Modeling. Sunderland: Sinauer Associates, 2010. Print.

#### Cite this work

Researchers should cite this work as follows: