Year: 2019
Type: dissertation
Abstract: Modelling of dynamical systems is an important problem in many fields of science.\nIn this thesis we explore a data-driven approach to learn dynamical systems from\ndata governed by ordinary differential equations using Neural Ordinary Differential\nEquations (ODENet). ODENet is a recently introduced family of artificial neural\nnetwork architectures that parameterize the derivative of the input data with\na neural network block. The output of the full architecture is computed using any\nnumerical differential equation solver. We evaluate the modelling capabilities of\nODENet on four datasets synthesized from dynamical systems governed by ordinary\ndifferential equations. We extract a closed-form expression for the derivative\nparameterized by ODENet with two different methods: a least squares regression\napproach and linear genetic programming. To evaluate ODENet the derivatives\nlearned by the network were compared to the true ordinary differential equations\nused to synthesize the data. We found that ODENet learns a parameterization of\nthe underlying ordinary differential equation governing the data that is valid in a\nregion surrounding the training data. From this region a closed-form expression\nthat was close to the true system could be extracted for both linear and non-linear\nODEs.
Authors: Daniel Karlsson, Olle Svanström
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2952170742
