rmojgani Goto Github PK

followers: 50.0 following: 101.0 repos: 30.0 gists: 0.0

Name: Rambod Mojgani

Type: User

Bio: ■ scientific machine learning (SciML) ■ machine learning in dynamical systems ■ computational sciences ■ data-driven modeling ■ model reduction

Twitter: rmojgani1

Blog: rmojgani.com

Rambod Mojgani

■ computational sciences ■ data-driven modeling ■ machine learning in physical systems ■ model reduction

Select Projects:

	Project	Description	Repository	Language
🖥️	RL SGS on Korali	Theory-driven reinforcement learning for sub-grid scale LES models of 2D turbulent	GitHub	Python (JAX)
🌊	Py2D	Python-Jax solver for 2D turbulence	GitHub	Python (JAX)
🖥️	LPINNs	Lagrangian physics-informed neural network	GitHub	Python (Pytorch)
🖥️	MEDIDA	Model Error Discovery with Interpretability and Data Assimilation	GitHub	Python
🖥️	MEDIDA_QG	Model Error Discovery with Interpretability and Data Assimilation for quasi-geostrophic turbulence	GitHub	Python (PyTorch)
🖥️	Physics Aware Auto-encoder	Manifold learning for large Kolmogorov n-width PDEs	GitHub	Matlab & Python (Keras)
🦾	LTV ROM Stabilization	An optimal feedback controller for linear time-varying reduced order models	Contact me	Matlab
🌊	Incompressible flow in a Lid-driven cavity	A control-volume-based finite element method	GitHub	Fortran 90
🌊	Compressible flow on an airfoil	Roe's Riemann solver for Euler equations on unstructured grids	GitHub	Fortran 90

Summary of my resume

Highlights:

My PhD Thesis on ``Reduced order modeling of convection-dominated flows, dimensionality reduction and stabilization''

A presentation at Rice University - MECH Seminar Series, Nov 2020

Fully data-driven dimensionality reduction (Autoencoder approach):

Print: AAAI-2021
Repository: PhysicsAwareAE
Poster: MLTP-2020
Presentation: My YouTube Channel, presented at AAAI 2021, AAAI-2021

Stabilization of time-varying reduced order models:

Print: IJNME
Repository: Contact me
Presentation: My YouTube Channel, presented at MMLDT 2021

Lagrangian dimensionality reduction:

Preprint: arXiv
Repository: Contact me

Rambod Mojgani's Projects

Project I : A control-volume-based finite element method is used to solve the in compressible flow, in a lid-driven cavity. Project II : Roe's Riemann solver is used for compressible Euler equations on unstructured grids, flow on an airfoil.

cross-eof-eddy-feedback-model

This repository contains scripts/codes to calculate cross-EOF eddy-zonal flow feedbacks of the annular modes based on NCL

deeplearningswe

deepxde

Deep learning library for solving differential equations and more

delete

dominant-balance

Methods and code for J. L. Callaham, J. N. Kutz, B. W. Brunton, and S. L. Brunton (2020)

exgan

Adversarial Generation of Extreme Samples

korali

High-performance framework for uncertainty quantification, optimization and reinforcement learning.

lpinns

To address some of the failure modes in training of physics informed neural networks, a Lagrangian architecture is designed to conform to the direction of travel of information in convection-diffusion equations, i.e., method of characteristic; The repository includes a pytorch implementation of PINN and proposed LPINN with periodic boundary conditions

medida_qg

meta-pde

nmor

Deep learning framework for model reduction of dynamical systems

npe-and-ntk

physicsawareae

The unsupervised learning problem trains a diffeomorphic spatio-temporal grid, that registers the output sequence of the PDEs onto a non-uniform parameter/time-varying grid, such that the Kolmogorov n-width of the mapped data on the learned grid is minimized.