Self-Organizing Maps (SOM), or Kohonen maps as you may have heard, are one of the basic types of self-organizing neural networks. The ability to self-organize provides adaptation to formerly unseen input data. It has been theorized as one of the most natural ways of learning, like that which is used by our brains, where no predefined patterns are thought to exist. Those patterns take shape during the learning process and are incredibly gifted at representing multidimensional data at a much lower level of dimensionality, such as 2D or 1D. Additionally, this network stores information in such a way that any topological relationships within the training set remain preserved.

More formally, an SOM is a clustering technique that will help us uncover interesting data categories in large datasets. It's a type of unsupervised neural network where neurons are arranged in a single, two-dimensional grid. The grid must be rectangular, as in, a pure rectangle or a hexagon. Throughout the iterations (which we will specify), the neurons in our grid will gradually coalesce around areas with a higher density of data points (the left-hand side of our display called Points). As the neurons move, they bend and twist the grid until they move more closely to the points of interest and reflect the shape of that data.

In this chapter we will cover the following topics:

• Kohonen SOM
• Working with AForge.NET