Modeling sequential cognitive states via population level cortical dynamics
Abstract
In this work, we propose a mathematical framework for cyclic and sequential patterns of brain activity, combining heteroclinic dynamics with discrete neural-field models.
Under explicit structural and positivity assumptions on the prescribed activity patterns, we show that the resulting spatially discrete neural-field system supports neither heteroclinic cycles nor nonconstant periodic orbits.
Heteroclinic dynamics, on the other hand, arise in Lotka--Volterra-type systems, but these equations do not directly correspond to neuronal processes.
To bridge these descriptions, we use a $C^1$ version of the Universal Approximation Theorem to approximate a target vector field by a neural network interpretable as a high-dimensional Amari-type neural-field system.
Each periodic solution of the reduced dynamics admits a unique periodic realization at the population level with the same minimal period.
For three-dimensional heteroclinic targets, we establish sufficient conditions ensuring that the approximating vector field admits a unique hyperbolic attracting periodic orbit in a trapping channel surrounding the original heteroclinic cycle.
As a proof of concept, we apply this construction to a Lotka--Volterra-type target phenomenologically representing sequential network activations associated with focused-attention meditation.