Stability analysis of a neural field self-organizing map

Area of Science:

• Computational neuroscience
• Neural network modeling

Background:

• Self-organizing maps (SOMs) are unsupervised learning algorithms inspired by biological neural networks.
• Area 3b of the somatosensory cortex exhibits complex spatiotemporal activity relevant to sensory processing.

Purpose of the Study:

• To derive theoretical conditions for the efficient development of input space representations by SOMs.
• To investigate the stability and formation mechanisms of SOMs using a neural field model.

Main Methods:

• Utilized a neural field model simulating spatiotemporal activity in the primary somatosensory cortex (area 3b).
• Applied Lyapunov's theory for neural fields to establish theoretical conditions for stability.
• Verified theoretical findings through numerical experiments.

Main Results:

• Identified theoretical conditions that guarantee efficient input space representation by SOMs.
• Demonstrated the critical role of the excitation-inhibition balance in lateral synaptic coupling.
• Highlighted the importance of synaptic gain strength in SOM formation and maintenance.

Conclusions:

• The study provides a theoretical framework for understanding SOM development.
• Neural field dynamics, particularly synaptic coupling and gain, are key to stable and effective SOMs.
• Findings have implications for both artificial neural networks and understanding biological sensory processing.