Arnab Lahiry

I am a third-year PhD student working at the intersection of extragalactic astrophysics, and machine learning, co-supervised by Dr. Jean-Luc Starck and Dr. Tanio Díaz-Santos (FORTH-IA).

I primarily work with spectral data cubes of high-redshift galaxies and use advanced signal-processing techniques and deep learning to extract astrophysical information. I work with observational data from the Atacama Large Millimeter/submillimeter Array as well as state-of-the-art zoom-in cosmological simulations - Feedback in Realistic Environments (FIRE) in close collaboration with Dr. Daniel Anglez-Alcazar and Niranjan Roy of the University of Connecticut, USA.

My PhD work includes:

• Constructing a large, variable, physically-motivated toy data set of mock spectral data cubes of rotating disk galaxies
• Developing an iterative re-weighted 2D-1D wavelet-based de-noising framework for high-z data cubes dominated by spatially correlated interferometric noise.
• Building a three-dimensional U-Net architecture trained on the toy dataset to recognise patterns in spatially correlated noise and learn to separate the astrophysical signal - observed to generalise well to advanced simulations and observations from the ALMA-CRISTAL survey, which have a similar broad physical basis as the toy data.
• Working with the spectral axis of simulated and observed spectral cubes to go beyond the Moment-0 map-based scaling relations to compute galaxy properties with greater precision.

I have also worked in collaboration with Dr. Francisco Villaescusa-Navarro and Dr. Adrian Bayer of the Center for Computational Astrophysics, Flatiron Institute, USA, on a project to dissect and understand the origins of cosmological information from two-dimensional simulated maps of the total matter density cosmic web from the CAMELS project, by investigating the various scales and morphological structures as case studies.

Education :