What is Self-Organizing Maps (SOM)

SOMs or Kohonen networks have a grid topology, with unequal grid weights. The topology of the grid provides a low dimensional visualization of the data distribution.

Also known as Kohonen maps, these are a grid of prototype units. Values in these units are adjusted using the SOM learning algorithm over a dataset.

A self-organizing map (SOM) is a low-dimensional (typically two-dimensional) type of neural network that has been trained using unsupervised learning to represent training instances as a discrete representation of the input field. Self-organizing maps differ from other neural networks in that they implement competitive learning versus error-correction learning (such as backpropagation with reverse descent) and use a neighborhood function to preserve the topological characteristics of the entrance area. SOM can also be used to detect security threats in computer networks. Each network package that creates regular network traffic is analyzed by software and a self-organizing map is created with certain features on the package. This neural network creates a certain pattern and the learning process begins. If a packet examined in network traffic does not match the general pattern, it is detected as a threat and included in the learning process. As long as the situation that was initially defined as a threat is repeated, it may come out of being a threat and be compatible with the pattern.