Deep Learning Projects
We use transformer based models to combine different modalities (time-domain, spectral).
In this work we translated MRI images to CT images using Pix2Pix (cGAN) as well as a diffusion based model with Controlnet.
Upcoming work applying ML methods to improve PDE methods for general relativity.
Work in numerical relativity
Published by PRL
In this work, we used simulations of highly boosted black holes to study the aftermath and if non-linear effects are needed to describe the process.
Published in CQG
with generated data available here
Using new ideas and developments, we study and perform some of the longes simulations of a boson-star Ringdown ever performed. We find that the resulting gravitational waves are potential smoking gun signals for the existence of boson stars, an exotic cousin of neutron stars.
Published in PRD
Using smart assumptions on the physics, we reduce the cost of simulations drastically, which allows us to simulate with high fidelity the motion of a black hole through a medium, which is a scalar field in this case.
Accepted by PRL
Accepted by PRD
Two studies present various aspects of superradiance, which is a process that extracts energy from the spin of a black hole and causes the exponential growth of a scalar field cloud.