Discovering lensed SNe through deep learning multi-epoch LSST imaging data

Deep learning, particularly Convolutional Neural Networks
(CNNs), has demonstrated remarkable success in identifying non-variable
gravitationally lensed systems using multi-band static images. With the
advent of time-domain surveys like the Rubin Observatory’s Legacy Survey
of Space and Time (LSST), which will image the sky locations at multiple
epochs, there is a unique opportunity to exploit temporal variations
alongside spatial features in 2D images to classify lensed supernovae
(SNe) among other transient phenomena. To achieve this, we employ a
Convolutional Long Short-Term Memory (ConvLSTM) network designed to
capture spatial and temporal correlations simultaneously. Our approach
incorporates real galaxies from the Hyper Suprime-Cam (HSC) dataset as
lenses, onto which we simulate synthetic lensed SN images using a
dedicated simulation pipeline. These lensed SN images are overlaid on
multi-band HSC galaxy cutouts at different epochs, representing the
evolution of the SN over time. Negative samples include a diverse set of
transient events, such as variable stars, active galactic nuclei (AGNs),
and unlensed SNe, ensuring robust classification. As HSC data matches LSST
in depth, resolution, and band throughput, it provides an excellent
testbed for this methodology. In this talk, I will present the details of
our simulation pipeline, the design of the ConvLSTM network, and the
initial results, highlighting the potential of this technique for
discovering lensed SNe in the LSST era.