LIGO Seminar

https://caltech.zoom.us/j/87546916051?pwd=aTByZU5WamhUKzlkRVY2bW05bytDdz09

Speaker: Niklas Houba

Title: "Deep source, noise, and glitch separation for the LISA global fit, with applications to LIGO"

Abstract: The Laser Interferometer Space Antenna (LISA) will detect a wide range of gravitational-wave sources, many of which overlap in time and frequency and are embedded in complex, non-stationary noise. This talk presents a deep learning-based framework for separating these signals in the time-delay interferometry (TDI) space. A shared encoder compresses noisy TDI inputs into a latent representation, which is then decoded into distinct components such as massive black hole binaries, galactic binaries, and instrumental glitches. Inspired by advances in audio source separation, the model accommodates an unknown number of overlapping signals and scales efficiently. Results on synthetic data demonstrate successful disentanglement of sources, offering a promising path for future parameter estimation pipelines  and real-data applications.