Reinforcement-based decision making in corticostriatal circuits: mutual constraints by neurocomputational and...

Area of Science:

• Computational neuroscience
• Cognitive psychology
• Reinforcement learning

Background:

• Reinforcement-based decision-making involves complex neural computations.
• Existing models, like the basal ganglia (BG) model and diffusion models, offer different levels of analysis.
• Bridging these models can enhance understanding of decision-making mechanisms.

Purpose of the Study:

• To integrate neural models of corticostriatal circuits (BG model) with diffusion models for decision-making.
• To improve the fit of diffusion models to data from reinforcement learning tasks.
• To capture the dynamics of the subthalamic nucleus in decision processes.

Main Methods:

• Simulated data from the BG model were used to fit diffusion models.
• Diffusion model parameters (nondecision time, decision thresholds) were adjusted to account for conflict.
• Human reinforcement learning experimental data were analyzed using modified diffusion models.

Main Results:

• Standard diffusion models underestimated response times under high conflict.
• Modified diffusion models, incorporating increased nondecision time and time-varying decision thresholds, fit simulated and human data well.
• Modulations in BG circuitry, specifically the subthalamic nucleus, correlated with decision boundary changes.

Conclusions:

• Modified diffusion models provide a better account of reinforcement-based choice data than individual models.
• This work demonstrates a method for summarizing BG model computations within diffusion models.
• The findings offer a convergent theoretical account of decision-making consistent with empirical data.