gw-mist
Machine Learning Tools to Detect Model Mispsecification in Gravitational Astronomy.
thumbnail image credit: Sora “Generate me an image of a binary black hole merger”
Lay Summary
This project formed the summer of my 3rd undergraduate year, and was carried out as part of an IoA summer studentship at Cambridge. I developed neural networks to process complex-valued frequency-domain gravitational wave data and detect both localised and model-wide distortions within a frequentist statistical framework by using simulation based inference (inference from simulation-trained neural density estimators to evaluate the posterior predictive distribution). Use of torch, jax, ripple, jimgw and multiprocessing to carry out high-performance GPU computing. Following my end-of-project presentation I was invited back to speak at the IoA’s 30-year anniversary event, alongside Prof Nikku Madhusudhan and Dr Will Handley. The project is an extension of the Anau-Montel, Alvey & Weniger library mist
Project Details
Supervisor: Dr James Alvey
Key Skills: Machine Learning - Neural Architecture - jax & torch - Gravitational Wave Astronomy
This project is still in progress! Stay tuned for future results.