Astrophysics Colloquium: Circumventing Bottlenecks and Taking Better Data in the Time Domain

Abstract: The growing corpora of streaming, multimodal astrophysical data taxes our ability to use traditional inference frameworks for discovery and insight. Identifying computational and human bottlenecks in such frameworks becomes an invitation to employ performant AI approaches instead. Here I discuss the development of a simulation-based inference (SBI) application to Roman microlensing events, which has led to a surprising theoretical discovery. Next, I introduce multimodal SBI with an application to stellar binaries at scale. AI is also moving upstream, impacting how data itself is obtained. Here I discuss both a reinforcement-learning (RL) approach to optimizing LSST target of opportunity observations of gravitational wave events and an AI-based system for active wavefront control of Rubin Observatory itself.