Self-Organizing Maps

What are Self-Organizing Maps?

Self-Organizing Maps (SOMs), also known as Kohonen maps, are a type of artificial neural network that is trained using unsupervised learning to produce a low-dimensional (typically two-dimensional), discretized representation of the input space of the training samples. This makes SOMs useful for visualizing high-dimensional data in a way that highlights the relationships and structures within the data.

How Self-Organizing Maps Work

SOMs consist of nodes, or neurons, arranged in a grid. Each node is associated with a weight vector of the same dimensionality as the input data. During training, the SOM undergoes a process that ensures these nodes self-organize to resemble the input data space.

The training process involves presenting each input vector to the SOM and finding the node with the weight vector most similar to the input vector (usually using the Euclidean distance measure). This node is known as the Best Matching Unit (BMU). The weights of the BMU and its neighboring nodes are then adjusted to be closer to the input vector, with the degree of adjustment decreasing with distance from the BMU. Over time, the map learns to fit the data in such a way that spatially close nodes in the map represent input vectors that are similar to each other.

Training Algorithm

The training of a Self-Organizing Map involves the following steps:

  1. Initialization: The weight vectors of the SOM nodes are initialized, often randomly.
  2. Competition: For each input vector, the BMU is found by comparing the input vector to all weight vectors.
  3. Cooperation: The neighborhood radius around the BMU is determined, which defines which nodes will be updated.
  4. Adaptation: The weight vectors of the BMU and its neighbors are adjusted to be more like the input vector.
  5. Continuation: Steps 2-4 are repeated for a large number of iterations and for all input vectors, with the neighborhood radius and learning rate decreasing over time.

Applications of Self-Organizing Maps

SOMs have been applied in various domains for tasks such as:

  • Data visualization: Reducing the dimensionality of data to visualize complex data structures.
  • Clustering: Grouping data based on similarity, where each node in the SOM can represent a cluster.
  • Feature extraction: Identifying patterns and relationships in the data that can be useful for other machine learning tasks.
  • Anomaly detection: Identifying data points that do not conform to the general distribution of the data.

Advantages and Disadvantages

Advantages of SOMs include:

  • They provide a way to visualize high-dimensional data in two dimensions.
  • They can help reveal the intrinsic clustering and structure of the data.
  • They are topology preserving, meaning similar data points are mapped close to each other in the grid.

However, SOMs also have some disadvantages:

  • The quality of the map can be sensitive to the choice of learning rate and neighborhood function.
  • The grid structure may impose a limit on the representation of the data topology.
  • They can be computationally intensive, especially with very large datasets.


Self-Organizing Maps offer a unique approach to the problem of high-dimensional data visualization and clustering. By transforming data into a two-dimensional grid of nodes, SOMs provide an intuitive way to understand and explore complex data structures. While they are not without their challenges, SOMs remain a valuable tool in the machine learning practitioner's toolkit, especially for exploratory data analysis.


Teuvo Kohonen, the Finnish professor who invented Self-Organizing Maps, has written extensively on the subject. His book "Self-Organizing Maps" is considered the seminal text on SOMs and provides a comprehensive overview of the theory and application of these neural networks.

Please sign up or login with your details

Forgot password? Click here to reset