Deep Learning for Corrupted Gravitational-Wave Time Series in LISA
The Laser Interferometer Space Antenna (LISA) will open a new observational window on the Universe through low-frequency gravitational waves, enabling precision studies of supermassive black hole binaries, extreme mass-ratio inspirals, and the Galactic foreground. However, LISA data will present unique challenges, including non-stationary noise, transient glitches (localised bursts), and irregular data gaps arising from instrumental and operational effects.
This PhD project will develop deep learning methods for modelling, mitigating, and reconstructing corrupted LISA time series, with the goal of improving Bayesian gravitational-wave inference, including parameter estimation and uncertainty quantification, under realistic data corruption. The student will explore modern machine learning approaches such as variational methods, flow matching posterior estimation, transformers, and diffusion-based models to realistic simulated LISA data, making extensive use of GPU-accelerated computing.
We are looking for a highly motivated student with a background in computer science, statistics, applied mathematics or astrophysics, an interest in developing deep learning methods for challenging scientific data, and a strong programming background. Prior background in astrophysics is not required, but an interest in scientific applications is desirable.
Skillset: solid programming background, solid understanding of machine/deep learning and Bayesian methods, experience with PyTorch, JAX, and CUDA is desirable.
Contact: Matt Edwards and Kate Lee, University of Auckland, New Zealand