Irrelevance by inhibition: Learning, computation, and implications for schizophrenia

Area of Science:

• Neuroscience
• Computational Neuroscience

Background:

• Schizophrenia symptoms may stem from impaired cortical filtering of irrelevant stimuli.
• Disruptions in cortical inhibitory interneurons are linked to schizophrenia, suggesting their role in relevance determination.

Purpose of the Study:

• To develop a neural network model demonstrating how the cortex learns to ignore irrelevant inputs via plasticity in inhibitory circuits.
• To explore the role of inhibitory interneurons in assigning relevance to sensory information.

Main Methods:

• A neural network model was created based on the hypothesis that excitatory output encodes "relevance" (expected reward/punishment).
• Temporal difference learning was used to train feedforward inputs to inhibitory interneurons.
• The model simulated the effects of disruptions to inhibitory units on activity levels and relevance coding.

Main Results:

• Irrelevant stimuli produced lower excitatory activity than relevant stimuli in the model.
• Disruptions to inhibitory units abolished the difference in activity levels between relevant and irrelevant stimuli.
• The model successfully recapitulated experimental data linking frontal cortex inhibition to fear learning and expression.

Conclusions:

• The developed model provides a theory for how the cortex learns to selectively inhibit inputs.
• This mechanism offers insights into how relevance assignment problems may emerge in schizophrenia.
• Relevance learning can occur in parallel with other learning processes through plasticity in inhibitory and excitatory components.