Model-free and model-based reward prediction errors in EEG

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Reinforcement learning involves model-free and model-based systems with distinct mechanisms.
• Previously, these systems were assumed to be computationally dissociated in the brain.
• Recent fMRI studies suggest potential interaction, computing reward prediction errors for both value types.

Purpose of the Study:

• To investigate the neural basis of reinforcement learning systems using EEG.
• To determine if model-based reward prediction errors exist and interact with model-free signals.
• To challenge the assumption of computational dissociation between model-based and model-free learning.

Main Methods:

• Electroencephalography (EEG) was employed to capture high temporal resolution neural activity.
• The study analyzed neural signals related to reward prediction errors in a reinforcement learning task.
• EEG data was examined for temporal sequencing of state prediction errors and action value updates.

Main Results:

• EEG data provided evidence for both model-free and model-based reward prediction errors.
• The findings indicate a temporal sequence involving state prediction errors and action value updates.
• Model-based reward prediction errors were demonstrated, challenging prior assumptions.

Conclusions:

• The brain computes both model-free and model-based reward prediction errors.
• These findings suggest that model-free and model-based learning systems are not computationally dissociated.
• The study highlights the utility of EEG for dissecting temporally distinct neural signals in learning.