SCSC2003 Abstract S91675

Beyond limits in Kohonen's self-organized clustering

Submitting Author: Prof. Miroslav Snorek
Abstract:
A lot of applications (e.g. inductive, or data driven modeling) require the knowledge how are the input data spread in the (usually multidimensional) input data space. Without knowing it any inductive model wil produce corrupted results. A lot of statistical methods commonly referred as methods for Exploratory Data Analysis exist. The objective of it is to produce simplified description and summaries of large data sets. One standard method used in EDA is clustering, another alternative is projection of high dimensional data sets as points on a low-dimensional, usually 2D display. The aim of this article is the publication of an extended method for both clustering and projection based on Kohonen’s Self-Organizing Maps. We use learned SOM as a model (tool) for interactive simulation.


Back to SCSC2003 Abstracts