Robustness of learning that is based on covariance-driven synaptic plasticity

Area of Science:

• Neuroscience
• Computational Neuroscience
• Learning Theory

Background:

• Learning is linked to synaptic plasticity, with covariance-based rules proposed for various learning forms.
• Covariance-based plasticity is sensitive to parameter mistuning, questioning its biological relevance.

Purpose of the Study:

• Investigate the impact of parameter mistuning in covariance-based synaptic plasticity on decision-making behavior.
• Determine the biological plausibility of approximate covariance-based plasticity in operant conditioning.

Main Methods:

• Developed a decision-making model incorporating synaptic plasticity driven by reward and neural activity covariance.
• Analyzed the effects of parameter mistuning on synaptic efficacies and behavioral outcomes (matching law).

Main Results:

• Slight parameter mistuning caused large changes in synaptic efficacies but small behavioral effects, demonstrating robustness of matching behavior.
• Mistuned covariance rules led to undermatching, consistent with experimental observations.
• Mistuning mean subtraction in the plasticity rule increased behavioral sensitivity to network property changes.

Conclusions:

• Approximate covariance-based synaptic plasticity is a plausible mechanism for operant conditioning.
• Matching behavior exhibits robustness to plasticity rule variations but sensitivity to network property changes.
• A trade-off exists between robustness to plasticity rule changes and robustness to network property changes.