Machine Learning for Cosmic Explosions: Algorithms to Applications - Edo Berger - 05/25/2022
This is a high-level research talk designed for professional astronomers. It is part of the Caltech Astronomy Colloquium Series, hour-long presentations given weekly by top scientists from around the world to the members of our department. All are welcome to watch these colloquium presentations; however, for talks specifically designed for the general public, please see our Caltech Astro Outreach channel: / caltechastro . Timestamps and abstract are included below.
Date: 05/25/22
Speaker: Edo Berger, Professor of Astronomy, Harvard University
Title: Machine Learning for Cosmic Explosions: Algorithms to Applications
Abstract : Supernova explosions play a central role in many areas of astrophysics: shedding light on the late stages of stellar evolution, giving birth to compact objects, synthesizing chemical elements, and tracking cosmic expansion. Such studies are enabled by the classification of supernovae into different classes that track distinct progenitor systems, evolutionary paths, and explosion physics. Already at the present, only 10% of supernovae are classified spectroscopically, and starting with the Vera C. Rubin Observatory LSST survey in late 2023, only 0.1% will be classified. This simple fact necessitates a new approach of photometric classification. In this talk I will present our recent work on developing and using various machine learning approaches, trained on real rather than simulated data, to classify supernovae. I will describe both broad classification approaches (using multiple supernova classes) and targeted approaches tailored to specific rare types of supernovae. In each case I will also give specific examples of scientific results enabled by these approaches.