Robust Associative Learning Is Sufficient to Explain the Structural and Dynamical Properties of Local Cortical...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Neural network activity association is key to cognitive functions.
• Ubiquitous structural and dynamical properties of cortical networks are observed across species.
• Synaptic connectivity is known to be shaped by experience and learning.

Purpose of the Study:

• To test the hypothesis that associative learning shapes cortical network properties.
• To investigate if recurrent neural networks trained on memories exhibit experimentally observed features.
• To determine if learning is a primary driver of local cortical network structure and dynamics.

Main Methods:

• Trained recurrent neural networks of excitatory and inhibitory neurons on associative memories.
• Varied the number of associations loaded into the network.
• Compared emergent network properties (connectivity, motifs, firing activity, excitation-inhibition balance) with experimental data.

Main Results:

• Networks loaded with near-maximum associations developed properties matching experimental observations.
• These properties include connection probabilities, weight distributions, motif over/underexpression, and firing patterns.
• Learned memories were retrievable even with significant noise.

Conclusions:

• Many structural and dynamical properties of local cortical networks are byproducts of associative learning.
• Continual learning, not just genetics, may produce these ubiquitous network features.
• Predicts a relationship between excitatory-excitatory and inhibitory-excitatory connection patterns.