Research Experience

1) Evolution of Large-Scale Structures Using the gevolution Code 

a) Read the papers about N-body codes and gevolution, run gevolution simulation and try to modify the code.

b) Learning the advanced C++ language to go through the code and be able to get a big picture of the relationships between
    the functions and various files(headers and main files).

c) Programming in Python and analyze gevolution data; focusing on matter power spectrum and density snapshots.

d) Compare the data with theoretical predictions and explicitly analyze discrepancies between Newtonian and Relativistic
    Cosmologies.

Figure1: density snapshots for different boxsizes in Lambda-CDM model; using gevolution simulation which is compeletly relativistic.
Figure2: Mean relative error between Newtonian and GR cosmology density fields in different redshifts.
2) Generating Relativistic Cosmology N-Body Simulations with CNNs

a) Get to know Machine Learning(ML) techniques such as Linear Regression and Random Forests to predict the position
and velocity of gevolution particles.

b) Work on Deep Neural Networks particularly, Convolutional Neural Network (CNNs); design a CNN model to convert the
Newtonian density simulation boxes to Relativistic one.

c) Examining and changing the performance of each of the ML parameters such as Loss function and Learning rate to
increase the accuracy of the ML model.

Figure3: Machine Learning Training Algorithm


 

For more information see the article and github that is related to this project

Article
github
3) Distinguishing Standard and Modified Gravity Cosmologies with ML

a) Run gevolution and k-evolution simulations at the university of Oslo cluster computing and analyze the data of both
Cosmologies.

b) Design A Generative Adversarial Neural Networks(GANs) for distinguishing both cosmologies.

c) Compare both Cosmologies using statistical cosmology tools such as Bi-spectrum and halo mass function.

Figure4: The deviation of matter power spectrum from Lambda-CDM model for different w in modified gravity (k-essence) Cosmology
4) Generating Modified Gravity simulations with CycleGANs
a) Design A Cycle-Consistent Generative Adversarial Neural Networks(CycleGANs) to convert the ΛCDM density
simulation boxes to Modified Gravity one.
Figure5: Generative Adversarial Networks
Figure6: Cycle GAN
Github