Software tools to power machine learning development for gravitational wave physics

Low-latency gravitational wave astronomy has begun in recent years to look towards machine learning to address the problem of rapid detection and characterization of gravitational waves. Machine learning algorithms have demonstrated the ability to reduce background noise, detect a variety of signal morphologies, and provide posteriors on signal parameters with minimal latency and computing requirements for real-time deployment. As these algorithms become more prevalent, and as the benefit for these algorithms becomes more pronounced with the ever-increasing sensitivity of gravitational wave detectors, there is a critical need for standardized infrastructure capable of at-scale testing and real-time deployment. We present ML4GW, a platform designed to accelerate training and inference in the context of gravitational wave physics. We discuss how the tools we have created streamline the process of going from development to deployment by better leveraging a heterogeneous computing environment . Finally, we demonstrate how this platform has been used to create an end-to-end gravitational wave search pipeline.