The astronomical community has entered the age of large-scale time-series datasets, with billions of observed objects and hundreds or more data points per time-series. For example, the Zwicky Transient Facility (ZTF) Public Data Release 5 already includes over 350 billion source detections, of 3.6 billion astrophysical objects. The Rubin Observatory’s Legacy Survey of Space and Time — expected to begin operating in October 2023 — is expected to deliver time-series for ~37 billion objects with a total of 30 trillion observations. Datasets at this scale are non-trivial to manage and analyze, including applying ML to recognize or identify classes of time-variable objects.
In this talk, I will present the frameworks constructed at UW’s DiRAC Institute to make the data analysis problem and application of ML techniques to Bn+ scale time-series datasets tractable to astronomy audience. I will discuss the AXS analysis framework (Zecevic et al. 2019), which allows one to perform scalable full-dataset analyses on LSST-sized datasets. I will review its applications to work with ZTF, including a project to search for Boyajian star analogues and visualize the survey along various axes. Finally, I will present some ideas for time-series classification and inference for the LSST era, from transients to supernovae.