The hippocampus as a predictive map.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The hippocampus is traditionally viewed as a cognitive map, with place cells encoding spatial locations.
• Emerging evidence suggests place cells also exhibit predictive coding, reward sensitivity, and policy dependence, challenging the purely spatial view.

Purpose of the Study:

• To investigate the functional role of hippocampal representations from a reinforcement learning perspective.
• To determine the optimal spatial representation for maximizing future rewards.

Main Methods:

• Utilized a reinforcement learning framework to analyze hippocampal function.
• Modeled the relationship between spatial representations and reward maximization.

Main Results:

• Demonstrated that a predictive representation, rather than a purely spatial one, is most effective for maximizing future rewards.
• Showed that this predictive representation accounts for various place cell response characteristics beyond spatial mapping.
• Proposed that entorhinal grid cells provide a basis set for this predictive representation, aiding in noise reduction and hierarchical planning.

Conclusions:

• The hippocampal representation is likely predictive, integrating spatial information with reward and policy.
• Entorhinal grid cells support this predictive coding by providing a structured basis for complex navigation and planning.
• This perspective reconciles existing data and offers a new framework for understanding hippocampal function in decision-making.