Deep learning has seen great success in astronomy, yet few models are set for online deployment during the upcoming O4 observing run of the LIGO-Virgo-KAGRA collaboration. A key barrier has been the lack of standardized tools for quickly developing and deploying reliable machine learning solutions. We address this gap with the ml4gw and hermes libraries, designed to improve efficiency, production readiness, and robustness in gravitational-wave detection. We'll showcase their use in Aframe, a low-latency pipeline for detecting compact binaries, and DeepClean, a deep-learning-based gravitational-wave denoising method for sources like binary neutron stars (BNS), neutron star-black hole (NSBH), and binary black hole (BBH) mergers.
We also introduce "Gravitational Wave Anomalous Knowledge" (GWAK), an anomaly detection technique using deep recurrent autoencoders with semi-supervised/self-supervised learning. GWAK targets gravitational-wave signals from unmodeled astrophysical sources such as core-collapse supernovae and cosmic strings, which matched-filter methods cannot easily detect. We'll present GWAK's latest results from LIGO's third observing run (O3) and initial results from the ongoing fourth run (O4), highlighting new anomalous waveforms identified. These advances pave the way for deployment of deep learning technologies in gravitational-wave astronomy.