Abstract Details
| Name: Ashish Mahabal Affiliation: Caltech Conference ID: ASI2018_888 Title : Deep Learning the Time Domain Authors and Co-Authors : CRTS and ZTF collaboration Abstract Type : Poster Abstract Category : Instrumentation and Techniques Abstract : Deep Learning has fast become a buzzword in the Data Science world. Astronomy too has seen a few applications especially where some structure in the inputs is discernible e.g images of galaxies and streaking asteroids. Much of time domain astronomy relies on light-curves where identifying structure is the basis of classification and yet non-trivial at best. As a result classification with sparse light-curves continues to be a hard problem. We show how the one-dimensional light curves can be converted to a two-dimensional image representation and directly used with convolutional networks for effective classification without going through the steps of feature extraction and dimensionality reduction. We use examples from CRTS and PTF, but the technique can be applied to any set of light curves. We also caution about the pitfalls at overusing canned architectures. The deep-learning based platform being set-up now will be indispensable once transient rates zoom beyond hundreds of thousands per night with surveys like ZTF and LSST, both of which will have publicly announced transients. |