Compositional Sequence Generation in the Entorhinal-Hippocampal System

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Neurons in the medial entorhinal cortex possess periodically organized firing fields, forming an internal representation of space.
• Grid coding is observed in other cortical areas, suggesting it's a general neural principle.
• Previous work linked grid coding to sequential reactivations of hippocampal place cells via dynamical systems.

Purpose of the Study:

• Extend the sequence generation model to support compositional cognitive computations using multiple dynamical systems.
• Empirically validate the model through simulations of spatial and temporal compositionality.
• Describe neural network architectures for grid-coding-based compositionality and link to machine learning.

Main Methods:

• Analysis through the lens of dynamical systems theory.
• Simulations of two experiments demonstrating compositionality in space and time.
• Description of neural network architectures based on grid coding.

Main Results:

• The synthesis of multiple dynamical systems supports compositional cognitive computations.
• Simulations successfully demonstrated compositionality in space and time during sequence generation.
• Identified neural network architectures and connections to machine learning techniques.

Conclusions:

• Grid coding, through synthesized dynamical systems, can support complex compositional computations.
• The model provides a framework for understanding sequential reactivations and cognitive map traversals.
• Highlights potential for analogous techniques in artificial intelligence and machine learning.