Neural network enhanced rapid interpolation of gravitational-waveforms

Binary neutron stars can be detected via early warning pipelines, which can distribute General Coordinate Notices to the broader astrophysical community by the time of merger. To maximize the likelihood of additional multi-messenger discoveries with gravitational waves, it is crucial to rapidly localize and classify the properties of the progenitor. This is typically done via Bayesian inference-based pipelines that rely on evaluating the likelihood function, which depends on the noise-weighted inner product between the data and the gravitational wave. In this work, we describe a machine-learning based interpolation scheme that can accelerate waveform generation. The accuracy, speed-up, and simplicity of the model suggest it is a promising way to accelerate the overlap calculations needed to evaluate the likelihood. The model we present directly utilizes detection pipeline output, and is a step towards ultra-low-latency parameter estimation.