Probabilistic Learning by Rodent Grid Cells.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Mammalian brains are increasingly understood as probabilistic computers.
• Grid cells in the hippocampal formation are crucial for spatial cognition, but their varied responses lack explanation.
• Existing models fail to account for stable grid fields in darkness or phenomena like grid fragmentation and rescaling.

Purpose of the Study:

• To develop a unified model explaining diverse grid cell response properties.
• To investigate the role of probabilistic learning in grid cell function.
• To reconcile seemingly contradictory grid cell dynamics and their response to environmental changes.

Main Methods:

• Reinterpreting published electrophysiological data using a novel probabilistic learning model.
• Analyzing grid cell responses across various experimental manipulations, including changes in environment and darkness.
• Statistically comparing model predictions with experimental observations from rat grid cells.

Main Results:

• The probabilistic learning model accurately predicts grid cell responses, including stable fields in darkness.
• The model explains grid fragmentation, partial rescaling in resized arenas, and low-dimensional attractor dynamics.
• It reconciles single-cell oscillatory dynamics with cell ensemble attractor dynamics and proposes a role for boundary cells in spatial learning.

Conclusions:

• Grid cell function is explained by a coherent set of probabilistic computations.
• These findings offer a parsimonious and unified explanation for grid cell behavior.
• Grid cells serve as a neuronal readout for probabilistic spatial computations.