Self-organizing maps (SOM) were invented by Teuvo Kohonen in the 1980s. Sometimes, they are known as Kohonen maps. So, why do they exist? The prime motive for these kind of maps is to reduce dimensionality through a neural network. The following diagram shows the different 2D patterns from the input layers:

They take the number of columns as input. As we can see from the 2D output, it transforms and reduces the amount of columns in the dataset into 2D.

The following link leads to the the 2D output: https://www.cs.hmc.edu/~kpang/nn/som.html

The depiction of the preceding diagram in 2D talks about a health of the country based on various factors. That is, it shows whether they are rich or poor. Some other factors that are taken into account are education, quality of life, sanitation, inflation, and health. Therefore, it forms a huge set of columns or dimensions. Countries such as Belgium and Sweden seem to show similar traits, depicting that they have got a good score on the health indicator.

Since this is an unsupervised learning technique, the data wasn't labeled. Based on patterns alone, the neural network is able to understand which country should be placed where.

Similar to the situation we just covered, opportunities are aplenty where self-organizing maps can be utilized. It can be thought as being similar in nature to K-means clustering.