An integrated deep learning‐based model of spatial cells that combines self‐motion with sensory information

Abstract

A special class of neurons in the hippocampal formation broadly known as the spatial cells, whose subcategories include place cells, grid cells, and head direction cells, are considered to be the building blocks of the brain's map of the spatial world. We present a general, deep learning‐based modeling framework that describes the emergence of the spatial‐cell responses and can also explain responses that involve a combination of path integration and vision. The first layer of the model consists of head direction (HD) cells that code for the preferred direction of the agent. The second layer is the path integration (PI) layer with oscillatory neurons: displacement of the agent in a given direction modulates the frequency of these oscillators. Principal component analysis (PCA) of the PI‐cell responses showed the emergence of cells with grid‐like spatial periodicity. We show that the Bessel functions could describe the response of these cells. The output of the PI layer is used to train a stack of autoencoders. Neurons of both the layers exhibit responses resembling grid cells and place cells. The paper concludes by suggesting the wider applicability of the proposed modeling framework beyond the two simulated studies.

Modeling Hippocampal Spatial Cells in Rodents Navigating in 3D Environments

“…Hence, following the work of (Soman, Chakravarthy and Yartsev, 2018) this study proposes an autoencoder-based deep learning approach to model place and grid cells in speci c environmental paradigms. This approach has been successfully employed to model place and grid cells in a 2D environment (Aziz et al, 2022).…”

“…For the implementation, we use concatenation of sine and cosine terms as input to the autoencoder model i.e., concatenate( (Aziz et al, 2022).…”

“…The PCA model for grid cells (Soman, Muralidharan and Chakravarthy, 2018; Aziz et al, 2022), shows that the PI's covariance matrix is circulant. It is well known that the eigenvectors of the circulant matrix are sinusoids, hence, giving rise to periodic ring in PI.…”

“…However, recent studies have shown that even the activation patterns of the layers of CNNs can be gainfully compared to the activation patterns of cortical areas in primary visual and auditory systems(Kell et al, 2018). These modelling studies have inspired applying deep learning networks to model spatial cells in 2D and related navigational problems(Banino et al, 2018;Cueva and Wei, 2018;Aziz et al, 2022). However, a paucity of studies applies deep learning techniques to model hippocampal responses in 3D.…”

“…2) (Soman, Chakravarthy and Yartsev, 2018) uses Lateral Anti-Hebbian Network (LAHN) to model place cells in 3D for bat and rodent navigation. This LAHN layer is updated to an autoencoder layer by(Aziz et al, 2022) which shows emergence of place and grid cells in 2D. This advancement paves a way for multisensory integration using realistic visual cues, real-time navigation using reinforcement learning signals and comprehensive modelling of most spatial cells such as vector-based navigation and memory.…”