Neural ODEs for phytoplankton modeling

Description

Neural ordinary differential equations (NeuralOdes) define a dynamic system that includes neural networks. They offer a versatile way of modeling different phytoplankton based processes. In these processes, irradiance is fundamental, but there is no general consensus in the literature on how to include light in a dynamical model explaining the evolution of biomass inside a photobioreactor. We investigate the effect of including neural networks in classical model growth of microorganism. The adjoint method is used to train the neural network inside the dynamical system, together with classical techniques as mini-batch and data augmentation.

Abstract

International audience

Additional details