Modeling hippocampal spatial cells in rodents navigating in 3D environments.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Neural correlates of 3D navigation present challenges, with differing place cell responses observed in rodents and bats.
• Existing models often struggle to capture the complexity of spatial cell function in three-dimensional environments.

Purpose of the Study:

• To develop a computational model for spatial cells (place and grid cells) in rodents navigating 3D environments.
• To utilize deep neural networks for modeling neural representations of spatial information.

Main Methods:

• A deep autoencoder network was proposed to model spatial cells.
• The model's input layer (HD layer) encodes head direction using azimuth and pitch angles.
• Path Integration (PI) layer computes displacement, with the bottleneck layer encoding spatial cell-like responses.

Main Results:

• The deep autoencoder model successfully generated both grid cell and place cell-like responses.
• Model performance was validated using simulations in two distinct 3D environments.

Conclusions:

• The proposed deep neural network model offers a novel approach to understanding spatial cells in 3D navigation.
• This work provides a foundation for using AI to holistically model neural mechanisms of spatial cognition.