Hebbian self-organizing integrate-and-fire networks for data clustering.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Computer Vision

Background:

• Data clustering is crucial for pattern recognition and data analysis.
• Existing methods like k-means and Ward's linkage have limitations in handling arbitrary cluster shapes and noisy data.

Purpose of the Study:

• To develop a Hebbian learning-based data clustering algorithm utilizing spiking neurons.
• To enable the segmentation of visual scenes into arbitrarily shaped homogeneous regions.
• To compare the proposed algorithm's performance against standard clustering techniques.

Main Methods:

• Implementation of a novel data clustering algorithm based on Hebbian learning principles.
• Utilizing spiking neurons for data representation and processing.
• Systematic comparison with k-means and Ward's linkage clustering algorithms.

Main Results:

• The proposed algorithm successfully distinguishes between clusters and noisy background data.
• It can identify an arbitrary number of clusters with arbitrary shapes.
• Demonstrated superior clustering ability and more modest time complexity compared to k-means and Ward's linkage.

Conclusions:

• The Hebbian learning-based spiking neuron algorithm offers a powerful and efficient approach for data clustering.
• Its capability to handle arbitrary cluster shapes makes it highly suitable for visual scene segmentation.
• The method presents a promising alternative to conventional clustering techniques.