In this session we will discuss the necessary ingredients and best tools available to identify and classify young transients using the Rubin alert stream. We will focus on the identification with very few datapoints, but including all the data available: image stamps, sparse light curves, host galaxy association and other crossmatches.

In particular, in this session we will imagine the required steps going from explosion to identification and classification. To do this, we will divide the session into data inputs and methods.

For the data inputs, we will discuss what would be the best format for the image stamps, the photometry, the host galaxy association and any other information in the alert stream, and how to include or access third party data for this purpose in real-time. What are the best available sources of simulated or observed data to train automated models during the first years of LSST?

For the methods section, we will discuss the best machine learning methods to process the previous data inputs. Can we do multi modal classification? How can we best include temporal and spatial information? How can we deal with missing data? How can we train models during the early years of LSST when not enough data is available?

"The main challenges ahead of massive time-domain surveys are timely recognition of interesting transients in the torrent of imaging data, and maximizing the utility of the follow-up observations."  (Tyson 2006)
 

Preliminary Session Agenda

5 min Introduction

10 min (7+3), Enhancing Rubin Science with Robust Cross-Matches in the Crowded LSST Sky, Tom Wilson

10 min (7+3), DELIGHT: Deep Learning Identification of Galaxy Hosts of Transients using Multiresolution Images, Francisco Förster

10 min (7+3), Multiscale Stamps for Real-time Classification of Alert Streams, Ignacio Reyes-Jainaga

10 min (7+3), Fast transient identification in optical survey data, Igor Andreoni

10 min (7+3), Lessons learned managing alerts in the ELasTiCC challenge, Rob Knop

10 min (7+3), Lessons learned classifying objects in the ELasTiCC challenge, Alex Malz

10 min (7+3), Early Supernova Classification using Host Galaxy Information and Shallow Learning, Alex Gagliano

15 min Discussion