Machine classification of spatiotemporal patterns: automated parameter search in a rebounding spiking network

Area of Science:

• Computational Neuroscience
• Machine Learning Applications
• Neural Network Dynamics

Background:

• Cortical electrical activity exhibits diverse patterns, including traveling waves, observed in both animal and human studies.
• Spiking neuronal networks are crucial for modeling complex brain dynamics and emergent network behaviors.

Purpose of the Study:

• To investigate spatiotemporal patterns in a spiking neuronal network with inhibition-induced firing using machine learning.
• To identify the parameter regimes that govern pattern formation, such as synchrony and traveling waves.
• To develop an automated method for characterizing network activities in a novel cortical model.

Main Methods:

• Utilized a spiking neuronal network model with inhibition-induced firing (rebounding).
• Employed machine learning techniques, specifically a computationally efficient classifier based on statistical, textural, and temporal features.
• Varied parameters like inhibition level, coupling strength, and kernel geometry of a Gaussian derivative coupling kernel to control pattern formation.

Main Results:

• The model generated various network activities, including synchrony, target waves, and traveling wavelets.
• Switching between synchrony and traveling waves was observed to be transient, spontaneous, noise-dependent, or stimulus-induced under moderate parameter values.
• Target wave speed was found to be most sensitive to coupling strength perturbations.

Conclusions:

• This study presents an automated method to characterize complex activities in a novel spiking network model of cortical dynamics.
• The findings offer insights into the mechanisms underlying the emergence and transitions of different electrical activity patterns in the cortex.
• The developed machine classifier effectively links specific parameter regimes to distinct spatiotemporal patterns.