Theoretical note: the C/T ratio in artificial neural networks

Area of Science:

• Computational Neuroscience
• Artificial Intelligence

Background:

• Reinforcement learning models are crucial for understanding learning mechanisms.
• Dopaminergic mechanisms play a key role in reinforcement learning.
• The Gibbon-Balsam model provides a theoretical framework for associative learning.

Purpose of the Study:

• To investigate the effect of the C/T ratio on acquisition rate using artificial neural networks.
• To explore how variations in reinforcement probability and intertrial interval impact learning.
• To compare simulation results with existing theoretical models and experimental findings.

Main Methods:

• Computer simulations of artificial neural networks with a neurocomputational learning rule.
• Manipulation of reinforcement probability (C) and intertrial interval (T) under various conditions.
• Analysis of acquisition rates in response to changes in C/T ratios and absolute values of C and T.

Main Results:

• Acquisition rate generally increased with the C/T ratio.
• Invariant C/T ratios showed no consistent effect on acquisition in initial simulations.
• Larger absolute values of C and T significantly slowed acquisition, suggesting a bounded effect of invariant ratios.

Conclusions:

• The dynamics of neural networks can replicate findings from established learning models like the Gibbon-Balsam model.
• The C/T ratio is a significant factor influencing acquisition rate in artificial neural networks.
• The effect of invariant C/T ratios appears to be bounded, particularly with larger parameter values.