Related Experiment Videos
A predictive reinforcement model of dopamine neurons for learning approach behavior
J L Contreras-Vidal1, W Schultz
1Motor Control Laboratory, Arizona State University, Tempe 85287-0404, USA. pepe@asu.edu
Journal of Computational Neuroscience
|July 16, 1999
Summary
This study proposes a neural network model for how dopamine and prefrontal cortex guide reward learning. The model predicts distinct neuronal signals for timing and amount/type prediction errors in cortico-striatal circuits.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Cognitive Science
Background:
- Dopamine and prefrontal cortex (PFC) play crucial roles in reward-related learning and decision-making.
- Cortico-striatal circuits are central to processing information guiding approach behavior.
- Understanding the neural mechanisms underlying reward prediction errors is key to explaining learning.
Purpose of the Study:
- To propose a neural network model explaining how dopamine and PFC activity guide information processing in cortico-striatal circuits during reward learning.
- To predict distinct types of neuronal responses related to reward prediction errors.
- To investigate the model's ability to account for dopamine and PFC responses in various learning scenarios.
Main Methods:
- Development of a neural network model based on adaptive resonance theory principles.
- Simulations of the model under different reward delivery conditions: free food, timed contingent stimuli, and a delay response task.
- Analysis of model predictions regarding neuronal responses to reward timing and expectancy.
Main Results:
- The model predicts two distinct reward-related neuronal signals: one for reward timing prediction errors (akin to dopamine responses) and another for reward amount/type prediction errors (akin to PFC responses).
- The proposed neural network architecture successfully accounts for dopamine responses related to novelty, generalization, and discrimination of stimuli.
- Simulations validate the model's ability to explain observed neural activities in reward-based learning tasks.
Conclusions:
- A unified neural network model can explain the differential roles of dopamine and PFC in reward-based learning.
- The model provides a computational framework for understanding how prediction errors in timing and expectancy are processed.
- The findings support the adaptive resonance theory's principles in modeling complex cognitive functions like reward learning.