Binding in hippocampal-entorhinal circuits enables compositionality in cognitive maps.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The hippocampal formation is crucial for spatial representation and memory.
• Existing models often struggle to explain the computational efficiency and representational capacity observed in the hippocampus.
• Understanding the neural mechanisms of spatial coding is a key challenge in neuroscience.

Purpose of the Study:

• To propose a normative model for spatial representation in the hippocampal formation.
• To integrate optimality principles with an algebraic framework for distributed computation.
• To explain how the hippocampus encodes spatial information and associates it with context.

Main Methods:

• Developed a normative model employing a residue number system for spatial encoding.
• Utilized high-dimensional, complex-valued vectors to represent individual residues.
• Implemented a conjunctive vector-binding operation for composing positional representations.
• Used a modular attractor network to enforce self-consistency and model entorhinal cortex grid cell modules.

Main Results:

• The model achieves normative goals, including maximizing coding range and spatial information per neuron.
• Demonstrated superlinear scaling of patterns with dimension and robust error correction.
• Showcased hexagonal, carry-free encoding of spatial position, enabling efficient path integration.
• The vector binding operation allows for the association of different contexts with spatial representations.

Conclusions:

• The proposed model provides a formal framework for compositional computations in the hippocampal formation.
• It explains efficient spatial encoding, path integration, and contextual association.
• The model generates testable experimental predictions for future research in hippocampal and entorhinal cortex function.