CDS&E: Collaborative Research: Development and Application of Machine Learning Classification of Optical Transients

This project will develop, test, and use, a range of machine learning (ML) algorithms and pipelines for the photometric classification of optical transients from current and future surveys. The discovery rate of optical transients already outpaces traditional spectroscopic classification methods, and with future surveys only a tiny fraction of their discoveries can be observed spectroscopically.

Photometric classification is therefore essential, for identifying rare transients in real time, for classifying transients to allow population studies, and for discovering new classes of transient. Each of these goals requires different ML and algorithmic approaches. Now that appropriate data are in hand and usable for initial classification tests, this is the right time to test the pipelines to be used for the real-time discovery of rare transients, in preparation for future much larger data volumes.

Students and postdocs will gain experience developing and implementing ML algorithms, carrying out spectroscopic and multi-wavelength studies of astronomical transients. This experience will feed into undergraduate education, connecting classroom learning and hands-on research and involving non-computer science majors, including student observing with large-aperture telescopes, and science fair experiences for K-12 students.

This project builds on recent successes by this team in creating initial classification pipelines using a range of ML algorithms, which were trained on, and then applied to, real data. It draws on a combination of large survey data, ML techniques, and active multi-wavelength follow-up, to prepare students and postdocs for Big Data scientific techniques in astronomy.

This project will develop pipelines to: (i) combine time-series based light curve classification with image-based host galaxy classification; (ii) develop, test, and implement ML pipelines targeted at specific classes of known rare transients; and (iii) design algorithms for anomaly detection to discover new types of rare transients. The classification tools produced by this work will be used for a wide range of time-domain astrophysics applications.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.