Three types of remapping with linear decoders: A population-geometric perspective

Area of Science:

• Neuroscience
• Computational Neuroscience

Background:

• Hippocampal remapping is crucial for spatial memory, with place cells forming distinct activity maps in different environments.
• Existing theories offer competing explanations, such as minimizing memory interference or reflecting latent state shifts.
• The relationship between these interpretations and compatible neural activity changes remains unclear.

Purpose of the Study:

• To unify and elucidate the mechanisms behind hippocampal remapping.
• To adopt a neural coding and population geometry perspective to understand remapping.
• To provide a framework for comparing remapping theories and experimental findings.

Main Methods:

• Assumed hippocampal population activity is decodable through a latent space.
• Identified three potential mechanisms for inducing remapping based on neural coding principles.
• Simulated and visualized remapping types in a network model.

Main Results:

• Demonstrated three distinct mechanisms for hippocampal remapping: changes in neural-to-latent space mapping, non-spatial mixed selectivity modulation, and neural variability in the latent space null space.
• Related simulated remapping behaviors to existing models and experimental data.
• Provided a unifying framework for understanding remapping.

Conclusions:

• The study offers a unifying framework for visualizing, understanding, and comparing diverse remapping theories and experimental observations.
• The proposed framework can serve as a testbed for investigating neural response variability.
• This work clarifies the relationship between different remapping interpretations and their underlying neural mechanisms.