A computational architecture incorporating shallow brain networks: integrating parallel cortical and subcortical...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Artificial neural networks (ANNs) often mimic hierarchical brain structures.
• Non-hierarchical brain aspects, like subcortical pathways, are under-represented in ANNs.
• Understanding cortico-subcortical interactions is crucial for brain computation.

Purpose of the Study:

• To develop a computational model integrating cortical hierarchy and subcortical pathways.
• To investigate the roles of these structures in decision-making.
• To explore neuroanatomically informed computational principles.

Main Methods:

• Developed a computational model combining hierarchical cortical processing with subcortical pathways.
• Implemented cortical hierarchy using a convolutional feedforward network and a predictive coding network.
• Tested the model on a perceptual context-dependent decision-making task.

Main Results:

• Both model variants replicated human and monkey behavioral data.
• Subcortical structures were found to drive decisions in easier trials.
• Complex hierarchical networks were necessary for harder trials.

Conclusions:

• Parallel cortico-subcortical processing is a fundamental computational property of the brain.
• Integrating subcortical pathways enhances the biological plausibility of ANNs.
• This model advances our understanding of brain computation and decision-making.