Efficient and flexible representation of higher-dimensional cognitive variables with grid cells

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Entorhinal grid cells are crucial for spatial navigation and memory.
• Existing models struggle to explain how grid cells encode high-dimensional information efficiently.
• Empirical data suggest a low-dimensional structure for grid cell firing fields.

Purpose of the Study:

• To propose a novel computational model for entorhinal grid cells.
• To demonstrate how grid cells can encode high-dimensional variables efficiently.
• To reconcile the low-dimensional structure of grid cells with their potential for high-dimensional coding.

Main Methods:

• Developed a computational model combining low-dimensional random projections and hexagonal grid cell responses.
• Introduced the concept of "mixed modular coding" integrating modularity and mixed selectivity.
• Simulated the model's ability to encode multiple variables of varying dimensions without circuit reconfiguration.

Main Results:

• The model successfully encodes high-dimensional inputs using low-dimensional grid cell representations.
• Achieved flexible encoding of multiple variables with maximized coding range per dimension.
• Demonstrated that neuronal firing fields can arise from combinations of grid cell modules representing higher-dimensional spaces.

Conclusions:

• Entorhinal grid cells, via "mixed modular coding," can efficiently represent high-dimensional variables.
• The model offers a biologically plausible and flexible mechanism for neural coding.
• This framework expands the understanding of grid cells' role in memory and cognitive functions beyond spatial navigation.