Versatile Framework for Physics-Informed Machine Learning with NVIDIA Modulus

Overview

Integrating physics principles into machine learning algorithms enhances their adaptability and generalization. Physics Informed neural networks (or PINNs) pivotal for this fusion, leverage physics laws to improve model accuracy and interpretability. This project aims to harness the capabilities of the popular NVIDIA Modulus framework to develop PINN-based machine learning algorithms tailored for computer vision tasks. By leveraging Modulus' computational power and flexibility, we seek to advance the performance and robustness of these algorithms. Subsequently, new (or improved) algorithms will be proposed and evaluated through the established framework.

Research engagement

This project will develop a PINN based framework for evaluating ML algorithms. This will primarily involve creating of the framework and evaluating ML algorithms on public dataset.

Research activities include:

1) Development of physics-informed models to simulate complex phenomena (e.g., fluid dynamics, medical imaging).

2) Integration of ML algorithms (e.g., computer vision) and create framework for testing and validation.

3) To propose and evaluate improvements in existing algorithms (optional)

4) Writing up, publishing, and presenting research outcome

Outcomes

The aim of the project is to develop a versatile framework for evaluating Physics Informed Machine Learning algorithms based on NVIDIA Modulus platform.

Skills and experience

Strong math background

Programming experience in python

Some Machine Learning and/ or computer vision experience is desired.

Start date

1 November, 2024

End date

28 February, 2025

Location

QUT, Gardens point

Keywords

• Physics informed Machine Learning
• Physics Informed Neural Network (PINN)
• Computer vision

Contact

Dr. Chayan Banerjee, c.banerjee@qut.edu.au