A reinforcement learning approach to model interactions between landmarks and geometric cues during spatial learning.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Behavioral studies challenge associative learning theory regarding geometric cues and landmarks.
• Neurobiological findings suggest independent memory systems with limited interaction.
• Existing models do not fully explain the interplay between different spatial cues.

Purpose of the Study:

• To propose a computational model of spatial learning based on the multiple memory systems theory.
• To integrate neurobiological and behavioral findings on spatial navigation.
• To explain phenomena like lack of competition, potentiation, and blocking between geometric cues and landmarks.

Main Methods:

• Developed a computational model incorporating distinct spatial strategies (locale and taxon).
• Modeled hippocampal place cells for locale strategy and dorsolateral striatum for taxon strategy.
• Implemented a strategy selection network in the prefrontal cortex using reinforcement learning.
• Simulated a behavioral experiment investigating landmark and geometric cue interactions.

Main Results:

• The model successfully reproduced behavioral results, explaining the absence of competition between landmark and geometric cues.
• Demonstrated how cooperation between memory systems during learning potentiates geometry learning by landmarks.
• Explained blocking as a result of competition between memory systems during action selection.

Conclusions:

• The multiple memory systems model provides a neurobiologically grounded explanation for spatial learning dynamics.
• Cooperation and competition between independent memory systems are key to understanding spatial navigation and cue interactions.
• The model offers a framework for future research on spatial memory and decision-making.