Two-stage clustering via neural networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Clustering algorithms often struggle with initialization sensitivity and the need to pre-define the number of clusters.
• Existing methods can be computationally intensive and lack flexibility in implementation.

Purpose of the Study:

• To present a novel two-stage clustering approach for fast and efficient data grouping.
• To develop a method that overcomes common limitations of traditional clustering techniques.

Main Methods:

• Utilizing a competitive neural network (CNN) to identify local density centers by harmonizing mean squared error and information entropy.
• Employing a Gravitation neural network (GNN) where identified centers are used as initial weights for unsupervised cluster formation.
• Implementing a Gravitation-like update process within the GNN, where nodes are attracted to nearby centroids within a defined radius.

Main Results:

• The proposed two-stage approach effectively performs fast clustering.
• The method is free from initialization problems and does not require pre-specification of the number of clusters.
• The approach demonstrates computational efficiency and implementation flexibility, with potential for parallel hardware implementation.

Conclusions:

• The novel two-stage clustering method offers a robust and efficient alternative to existing techniques.
• The Gravitation neural network's unique update mechanism simplifies complex computations while ensuring effective clustering.
• This approach holds significant promise for applications requiring rapid and adaptable data clustering.