TRAINING COURSE "KOHONEN MAPS in Data Mining"

Kohonen Maps enjoy a high level of popularity...in books. In the field, they are one of the least used, and yet standard techniques in Data Mining.

Whose fault is that ? Certainly not their fault, as they do very well that for which they were designed (dimensionality reduction). On top of that, they are now increasingly popular in clustering because of their unique "toplogy preservation" properties.

The reason may be found in the fact that, although they are now often found in Data Mining software, explanations given about their working principles remain somewhat obscure. It is about time that something is done about that, so that users can benefit form Kohonen Maps' quire remarkable properties.

This 1 day training course (see outline below) presents the basic working principles of Kohonen Maps, their applications, and how to use them efficiently.

Course outline

What is a Kohonen Map doing ?

Clustering

Dimensionality reduction

SOM and ACP : similarities and differences

How does a Kohonen Map work ?

Architecture of a Kohonen Map

Training algorithm of a Kohonen Map

How to measure the quality of a Kohonen Map

Applications of Kohonen Maps

Dimensionality reduction and multidimensional data visualization

Dimensionality reduction and data preprocessing. Classification as a special case

Typology

Neurons and micro-classes

From micro- to macro-classes

Typologie and map quality

Missing values estimation

Regression with a Kohonen Map

Identification of outliers with a Kohonen Map

Beyond Kohonen Maps

Curvilinear Components analysis (CCA)

Independant Components Analysis (ICA)