Self-Organizing Maps: Unsupervised Learning in Action |…

What Self-Organizing Maps Are

Self-Organizing Maps (SOMs), also known as Kohonen maps, are a type of artificial neural network used for unsupervised learning. They map multidimensional data into a lower-dimensional space while preserving the topological relationships between the data points.

Developed by Teuvo Kohonen in 1982, SOMs are particularly useful for clustering and visualizing high-dimensional datasets, making complex patterns more accessible.

How Self-Organizing Maps Work

The process of training a SOM involves competitive learning. Each neuron in the map has weights that correspond to the dimensions of the input data. During each iteration, a random data point is selected and compared against all neurons in the map. The neuron with the closest weight vector (the winning neuron) adjusts its weights based on the difference between its current values and those of the input data.

Additionally, neighboring neurons also adjust their weights to reflect the similarity among data points. This process continues over many iterations until the map stabilizes, resulting in a 2D representation where similar data points are close to each other.

live demo · related simulation● LIVE

Why It Matters

SOMs are invaluable for exploratory data analysis and feature extraction. They can help identify patterns, clusters, and outliers in large datasets that might be difficult to discern otherwise.

In practical applications, SOMs have been used in fields such as bioinformatics, image processing, and financial forecasting, providing insights into complex systems.

Real-World Examples

SOMs are widely applied in recommendation systems. For instance, Netflix uses similar techniques to recommend movies based on user preferences.

In medical research, SOMs can be used to analyze gene expression data and identify groups of genes with similar functions.

Frequently asked questions

How does the neighborhood decay affect the training process?

Neighborhood decay ensures that only nearby neurons are updated during each iteration, gradually reducing the influence as training progresses. This prevents overfitting and helps the map converge to a stable configuration.

Can SOMs be used for supervised learning tasks?

While SOMs are primarily designed for unsupervised learning, they can sometimes be adapted for supervised tasks by using the trained map as an input layer in a supervised neural network.

What is the U-matrix and how does it help interpret the map?

The U-matrix (Unified Distance Matrix) represents the distances between neighboring neurons. It helps visualize the topological structure of the SOM, making it easier to identify clusters and patterns in the data.

How do learning rate and neighborhood radius affect training speed and accuracy?

A higher learning rate speeds up initial training but can lead to less accurate maps. A smaller neighborhood radius focuses updates on nearby neurons, promoting local convergence, while a larger radius encourages global adjustments.

Try it live

Everything above runs in your browser — open Self-Organizing Map: 2D Kohonen Training and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.

▶ Open Self-Organizing Map: 2D Kohonen Training simulation