Machine learning in GW astrophysics

Machine learning (ML) is a booming topic in recent years. I am excited about exploring the application of ML in gravitational-wave astronomy.

First, ML is a promising way to tackle the excess low-frequency contaminations to the LIGO sensitivity as it can recognize nonlinear and nonstationary noise couplings that classical signal processing techniques fail.

With my knowledge about the interferometer, I am designing ML-based nonlinear noise regression algorithms utilizing data stored in hundreds of LIGO auxiliary channels.

Second, ML algorithms can be highly efficient in terms of computation, making them especially suitable for tasks that require low latency. This includes the early warning of a binary neutron star merger. If we integrate the noise regression and signal detection together, we may detect binary neutron star signals minutes prior to the final merger. We can thus catch not only the post-merger kilonova, but also precursor electromagnetic signals, leading to new exciting discoveries in multi-messenger astronomy.