A spiking self-organizing map combining STDP, oscillations, and continuous learning.

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Machine learning

Background:

• The self-organizing map (SOM) is an unsupervised learning algorithm for creating topographic data representations.
• Traditional SOMs lack key functional properties of biological neural networks, such as spike-based communication and localized information processing.
• Mammalian cortices exhibit feature maps, inspiring SOMs but highlighting areas for biological realism.

Purpose of the Study:

• To present a novel spiking neural network model that addresses limitations of conventional SOMs.
• To incorporate biologically plausible mechanisms like spike timing and continuous input into the SOM algorithm.
• To advance the understanding of brain self-organization and provide a new SOM implementation method.

Main Methods:

• Developed a network of integrate-and-fire neurons incorporating solutions for global information access, spike coding, and continuous input.
• Designed a network structure that modifies learning rate and lateral connectivity during training.
• Employed relative spike timing across synaptic connections for learning.
• Simulated experiments using artificial, Iris, and Wisconsin Breast Cancer datasets.

Main Results:

• The novel spiking neural network implementation successfully formed maps.
• The model maintained fundamental properties of the conventional SOM.
• Demonstrated the feasibility of the model for data representation and unsupervised learning tasks.

Conclusions:

• The presented spiking neural network SOM is a significant step toward understanding brain self-organization.
• This model offers a new method for implementing SOMs in software and specialized spiking neuron hardware.
• The biologically plausible approach bridges artificial neural networks and neuroscience.