I’m using techniques at the forefront of AI research to classify real-time data from wide-field surveys. My goal is to identify rare and unusual sources within these multi-messenger alert streams, as well as create pure samples of type Ia supernovae for cosmology with LSST. I use a combination of ML techniques to accomplish this including random forests, gradient boosting and much else. These days though, I’m using neural networks, particularly recurrent neural networks (RNNs) with Gated Recurrent Units (GRUs) or Long-Short Term Memory (LSTM) layers. I am working on using convolutional neural nets (CNNs) and generative adverserial networks (GANs) to move us from classifying time-series to classifying image-series.<\/p>\n\n
I’m lead algorithms developer on the Arizona-NOAO Transient Alert and Response to Events System (ANTARES<\/a>) broker, and my work is part of a special issue<\/a> of the ApJ Supplement. I also lead the validation team for the Photometric LSST Astronomical Time-series Classification Challenge (PLAsTiCC<\/a>) that is live now! I’m also involved in using deep learning to find supernovae within the TESS satellite’s footprint, using lessons from my time as part of the Kepler Extragalactic Survey (KEGS). My work has involved three students, and I’m guiding them towards their theses and publications.<\/p>\n"},{"title":"Forensics on Stellar Corpses","link":"https:\/\/gnarayan.github.io\/project\/02snia\/","pubDate":"Thu, 01 Nov 2018 00:00:00 -0500","guid":"https:\/\/gnarayan.github.io\/project\/02snia\/","description":" I’m part of the KEGS team investigating the progenitor systems of supernovae with high-cadence light curves from Kepler. We’ve found some of the first supernovae that have early-time excess flux, indicating a possible main sequence companion, or surface nickel mixing. I’m leading the analysis of one of the KEGS SNIa, and I am advising a student on their analysis of an unusual rapid transient. I’ll be applying the lessons learned from Kepler to TESS, NASA’s next flagship exoplanet mission. I am particularly interested in combining my work on deep learning to classify transients in alert streams to ground-based searches of the TESS field. I am working on combining machine learning and citizen science and building hybrid active learning systems for supernova studies. We will discover some of the earliest supernovae ever, gleaning insights into their explosions. With JWST, and Roman Space Telescope after that, we will have new windows with which to peer into the infrared and understand supernova host-environments. These advances will in turn help refine models of SNIa and improve their precision as cosmological probes with LSST.<\/p>\n"},{"title":"Standards for Future Surveys","link":"https:\/\/gnarayan.github.io\/project\/03calib\/","pubDate":"Thu, 01 Nov 2018 00:00:00 -0500","guid":"https:\/\/gnarayan.github.io\/project\/03calib\/","description":"