Rapid processing and unsupervised learning in a model of the cortical macrocolumn.

Area of Science:

• Computational neuroscience
• Neural modeling

Background:

• Single-neuron models exhibit limitations in robustness and computational efficiency.
• Cortical organization involves macrocolumns composed of interconnected minicolumns.

Purpose of the Study:

• To develop and analyze a computational model of cortical macrocolumns using inhibitorily coupled minicolumns.
• To demonstrate the functional capabilities of minicolumns as monolithic units for decision-making and learning.

Main Methods:

• Utilized a spiking neuron model with refractory periods and fixed random excitatory within-minicolumn connections.
• Implemented instantaneous inhibition within macrocolumns and analyzed system dynamics.
• Employed Hebbian plasticity for unsupervised organization of inputs and receptive fields.

Main Results:

• Minicolumns demonstrated stability as functional units for fast decisions and learning.
• Oscillating inhibition (gamma frequency) induced phase-coupled population rate coding and high sensitivity.
• The model successfully organized inputs into selective receptive fields, acting as pattern classifiers.

Conclusions:

• The proposed minicolumn model overcomes limitations of single-neuron systems, offering improved robustness and computational speed.
• Minicolumns can autonomously organize inputs via Hebbian plasticity, functioning as classifiers.
• The system exhibits distributed neural coding capabilities, validated by the bars test.