Astrophysical applications of machine learning at scale and under duress

Summary form only given: The universe is teeming with change on timescales from billions of years to milliseconds. A major goal of modern synoptic imaging surveys is to categorize this change over the entire sky to infer the diverse physical origins of variability. However, event discovery is only the beginning in the quest to extract the deepest insights: expensive follow-up resources (telescopes and people) are required, often in a time constrained environment. Viewing discovery and scientific insight through a resource-maximization lens, I discuss how machine learning is being applied to some modern astrophysics challenges. Here, the surfacing of parallelized feature engineering and machine learning into production-quality (scalable and fault tolerant) frameworks is the frontier for our field.