Self-Organizing Maps
Self organizing maps use processing units (neurons) to place centroids on an adjustable map. The model self organizes based on learning rules and interactions. Processing units maintain proximity relationships as they grow.
Introduction
SOM project the data while keeping the topology N dimensional space are placed next to each other in the 2D map
SOM performs two functions:
(i) the modeling of the source data with a smaller number of vectors that attempt to represent the source data as closely as possible (e.g. minimizing some measure of deviation from the observations)
(ii) the production of a topology in the SOM grid whereby similar models are close together and dissimilar models are far apart.
Explanation
A Self-Organizing Map (SOM) is a neural network that creates a low-dimensional (usually 2D grid) representation of high-dimensional data. Following is the work process:
- Define the grid size and neuron type (hexagonal or rectangular) for your SOM.
- Initialize each neuron with random weights or a subset of the data.
- Train the network in an unsupervised way:
- Present a data sample.
- Find the neuron (BMU) whose weights are closest to the data sample.
- Update the weights of the BMU and its neighbors to become more similar to the data sample.
- Repeat step 3 for all data samples, progressively adjusting weights.
Through this process, the SOM learns to map similar data points to nearby locations on the grid, creating a visual representation of the data's relationships.
Examples
How to
Outgoing relations
- Self-Organizing Maps is a kind of Artificial neural networks
Incoming relations
- Sammon's projection is a kind of Self-Organizing Maps
- U-matrix is a kind of Self-Organizing Maps
Contributors
- Sandra