[From Grid Cells to Place Cells:A Gauss Distribution Activation Function Model].

Area of Science:

• Neuroscience
• Computational Neuroscience
• Mathematical Biology

Background:

• Simple linear models fail to capture the complex mapping from grid cell activity to place cell output in the rodent hippocampus.
• Existing models like competitive learning, ICA, and Bayesian methods have limitations in accurately representing this neural computation.

Purpose of the Study:

• To develop a novel mathematical model for feature mapping from multiple grid cells to a single place cell.
• To improve the accuracy and simplicity of models describing hippocampal spatial coding in rodents.

Main Methods:

• Introduction of a Gaussian distribution activation function to model grid cell-place cell interactions.
• Utilizing the localization properties of the Gaussian function to filter grid cell inputs and connection weights.
• Comparing the proposed model against competitive learning, independent component analysis, and Bayesian methods.

Main Results:

• The Gaussian-based model successfully replicates the feature mapping relationship observed in biological studies.
• The model demonstrates superior performance, achieving higher accuracy in place cell output.
• The proposed method requires fewer grid cell inputs and results in a simpler algorithm.

Conclusions:

• The Gaussian distribution activation function provides a neurophysiologically plausible and computationally efficient model for hippocampal spatial coding.
• This model offers a significant improvement over existing methods for understanding how rodents navigate and represent space.