Optimization of Experimental Stimuli for Data-Driven Discovery of Neural Dynamics

Sammanfattning

Mathematical and computational modeling of biological systems has been a focus in the field of computational neuroscience for a long time. In the current work, the main focus is on a mathematical model of a neuron, the FitzHugh-Nagumo model, which is a system of two first-order ordinary differential equations (ODEs), and the goal is to find the optimal stimuli of the neuron in order to correctly learn the governing equations of the neuronal dynamics via the sparse identification of nonlinear dynamics (SINDy) algorithm. This is done via anactive learning approach, using Particle Swarm Optimization (PSO) as a global optimization algorithm throughout the learning process. Three types of stimuli are optimized, triangular, rectangular, and multi-sine pulses, for two different dynamical regimes, one excitable and one oscillatory regime. The optimization algorithm is used to minimize the condition number of the SINDy algorithm’s linear system for the stable identification of neuronal dynamics. The results indicate that optimized stimuli outperform the random noise stimulus in recovering the governing equations of the FitzHugh-Nagumo system and, thereby, the identification of dynamics.