A neural network model with dopamine-like reinforcement signal that learns a spatial delayed response task

Area of Science:

• Computational Neuroscience
• Reinforcement Learning
• Cognitive Psychology

Background:

• Dopamine neurons signal reward prediction errors, a key component in learning theories.
• Spatial delayed response tasks are crucial for assessing frontal cortex and basal ganglia functions.
• Understanding dopamine's role in learning can inform artificial intelligence and neurological disorder research.

Purpose of the Study:

• To investigate the use of simulated dopamine neuron responses as a reinforcement signal for learning.
• To model a spatial delayed response task using a neural network and dopamine-like reinforcement.
• To compare computational learning outcomes with biological dopamine neuron activity and animal learning.

Main Methods:

• A neural network model implementing the temporal difference algorithm was developed.
• The reinforcement signal was modeled on dopamine neuron responses to rewards and prediction errors.
• The model learned a simulated spatial delayed response task, including subtasks for spatial choice and temporal delay.

Main Results:

• The model's reinforcement signal mimicked natural dopamine neuron responses during learning.
• The neural network successfully learned the spatial delayed response task, replicating animal learning curves.
• Simulations showed that dopamine-like prediction errors are crucial for effective learning and behavior.

Conclusions:

• Dopamine-like reward prediction error signals are effective teaching signals for learning complex behaviors.
• Computational models using dopamine-like signals can replicate primate cognitive learning and behavior.
• This approach offers insights into dopamine's role in learning and potential applications for neurological conditions.