About SOM

This is an article about SOM.

Hello!

Today, we’re going to talk about SOM.

Self-Organizing Map (SOM) is an unsupervised learning algorithm used to embedding high-dimensional data into low-dimensional lattice structures.

I will explain the SOM below.

principle of operation

competitive learning

The nearest neuron (unit) to the input data is selected, and the selected neuron and its surrounding neurons compete and learn from the input data.

Adjusting Neighbourhood Relationships

In the early stages of learning, the connection with the surrounding neurons is strong, and the connection with the surrounding neurons is gradually adjusted to weaken over time.

formation of a map

Through this process, we envision the characteristics of the input data in a low-dimensional lattice structure while preserving it, and form a map that allows us to visually grasp the similarities between the input data.

Key Features

a non-linear idea

SOM performs nonlinear ideas, allowing it to project in low dimensions while preserving the structure of complex data.

Visualization

As we picture the data in a low-dimensional lattice structure, we form a map that allows us to visually grasp the similarities between the data.

Utilization

pattern recognition

It is used to recognize patterns in data, such as images and speech.

clustering

It is used for visual representation by clustering based on the similarity of the data.

Precautions

hyperparameter settings

Care must be taken as hyperparameter settings, such as SOM’s learning rate, neighbor relationship adjustment, etc., can affect the results.

Dimension down

The loss of information can occur in the process of mapping high-dimensional data to low-dimensionality, which requires consideration.

at the end of the day

Self-organizing maps are utilized to identify and visually represent the structure of data in various fields, and are powerful unsupervised learning algorithms that project data into low dimensions while preserving its characteristics.

Thank you!