Latent representations in hippocampal network model co-evolve with behavioral exploration of task structure

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Animals learn complex behaviors by integrating sensory information and internal logic.
• The hippocampus forms
• cognitive maps
• representing task structures.
• The biophysical mechanisms for creating these task-relevant maps are not fully understood.

Purpose of the Study:

• To model the emergence of task-relevant neural representations in the hippocampus.
• To investigate the interplay between single-cell learning mechanisms and reinforcement learning in shaping cognitive maps.
• To explain how latent representational structures evolve with behavior.

Main Methods:

• A computational model combining plateau-based learning at the single-cell level with reinforcement learning in an agent.
• Simulating the co-dependent evolution of neural representations and behavioral policy.
• Analyzing the emergence of cue-dependent
• splitters
• in the model.

Main Results:

• The model successfully develops latent representational structures crucial for task-solving.
• Cue-dependent
• splitters
• , essential for distinguishing task-relevant information, emerged in the model.
• Irrelevant representational structures were excluded, aligning with experimental observations.

Conclusions:

• Plateau-based learning and reinforcement learning can drive the formation of task-specific cognitive maps.
• The model provides a framework for understanding how neural representations and behavior co-evolve.
• The study offers testable predictions on the interaction between neural representations and behavioral policies.