Deep reinforcement learning to study spatial navigation, learning and memory in artificial and biological agents.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Cognitive Science

Background:

• Artificial neural networks (ANNs) and AI have historical ties to neuroscience and psychology.
• Deep learning and reinforcement learning (RL) are key AI approaches linked to understanding brain mechanisms.
• Current ANNs lack biological features that could improve AI and neuroscience insights.

Purpose of the Study:

• Summarize Botvinick et al.'s arguments on deep reinforcement learning (DRL) for neuroscience.
• Frame DRL as a framework for studying learning, representations, and decision-making in the brain.
• Apply DRL to spatial navigation to understand brain-based world representations.

Main Methods:

• Review and synthesis of arguments from Botvinick et al. (Neuron, 2020).
• Conceptual framing of DRL within neuroscience research.
• Application of DRL principles to the domain of spatial navigation.

Main Results:

• DRL is crucial for advancing neuroscience research on learning and decision-making.
• DRL provides a framework to study how the brain forms and uses representations.
• Studying spatial navigation with DRL can illuminate how the brain processes external information.

Conclusions:

• Integrating AI, particularly DRL, with neuroscience is vital for progress.
• DRL offers a promising avenue for understanding complex cognitive functions like spatial navigation.
• This interdisciplinary approach can deepen our knowledge of brain function and representation.