This repository consists of the work done as a student developer, during the Google Summer of Code (GSoC) 2021 program. The project, Machine Learning for Turbulent Fluid Dynamics is a part of the Turbulent suborganization under the Machine Learning for Science (ML4SCI) umbrella organization.
Machine Learning for Science (ML4SCI)
Generating the data directly from the quasilinear and fully nonlinear simulations is extremely computationally expensive and infeasible. We work on reducing the dimensionality of the data through various Machine Learning techniques. The current work done during GSoC mainly involved Proper Orthogonal Decomposition (POD) in the spectral space.
The covariance matrix generated has been transformed by multidimensional fourier transforms. The eigenvalues have been computed and plotted in different ways.
We would be extending the work to include other computational and visualization techniques to better understand the fluid motion. Considering the important modes and reconstructing the data we aim to develop a statistical theory describing the motion. We would be exploring various models including Autoencoders and compare the various results. Once we have good enough reduction which is measurable through some standard metric.
I have been working on documenting the idea, approach, code and experience of GSoC. You can follow the work at ameygsocblog
For any contributions, issues, suggestions, help reach out to me through [email protected]
You can also reach out to me on the below handles
You can also open issues, create pull requests on this repository. I would be actively maintaining the project and also working on further updates to this. Keep checking this space frequently.