Bos, Hidde van den (2022) Using physics-informed neural networks to reduce data dependence. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI_2022_HiddevdBos.pdf Download (9MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (123kB) |
Abstract
Nowadays, with computational power no longer being the bottleneck in many deep learning problems, one of the most prohibitive factors in machine learning has become data acquisition. Physics-Informed Neural Networks (PINNs) incorporate physical information, in the form of a differential equation, into neural networks to reduce the amount of data required to achieve accurate results. In this thesis the performances of an Extreme Learning Machine (ELM), a Linear Physics-Informed ELM (L-PIELM), a Non-Linear Physics-Informed ELM (NL-PIELM), a Multilayer Perceptron (MLP), and a Physics-Informed MLP (PIMLP) are compared on a linear ordinary differential equation and a non-linear ordinary differential equation given varying amounts of data. Their performance is also compared to three algorithms for numerical integration. Furthermore, the performance of the NL-PIELM and PIMLP is analyzed on a non-linear partial differential equation. The PINNs are demonstrated to outperform the ELM and MLP when data is limited, with the L-PIELM and NL-PIELM achieving higher accuracies than the PIMLP. Compared to the algorithms for numerical integration, PINNs are a small improvement accuracy wise, and they also provide a more flexible alternative to the strict data requirements of the algorithms for numerical integration. PINNs, with the exception of the NL-PIELM, are also shown to have the additional capability of reducing the effects of noise.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Netten, S.M. van |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 11 Oct 2022 14:13 |
| Last Modified: | 11 Oct 2022 14:13 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/28831 |
Actions (login required)
 |
View Item |