Oral Abstract

Oral Contribution (O4.5) Antonia Rowlinson (ASTRON & University of Amsterdam)

Theme: Data science challenges: tools from statistics to machine learning

Identifying transient and variable sources in radio images

The huge influx of data coming from the latest and next generation radio telescopes are likely to hold many surprises. Many of these facilities have wide fields of view and are capable of producing very good quality snapshot images of durations >1 second. Therefore we can do large, deep searches for the rarest transient and variable sources on many different timescales. However, the volume of data has become far too large for manual processing.
We developed the LOFAR Transients Pipeline (Swinbank et al. 2015) to automatically these datasets and output a large database containing a wealth of information on each unique source detected in the images. Mining this large database is the next challenge, as the number of candidates requiring manual confirmation are rapidly increasing beyond what is feasible In Rowlinson et al. (2019), we investigate the use of machine learning strategies to optimise the detection of rare transients hidden within thousands of other sources in the database.
This talk will outline the current state-of-the-art strategies used to conduct image plane transient surveys on big datasets. I will explain the LOFAR Transients Pipeline and machine learning strategies used to automate processing steps. Then I will demonstrate their use with the latest transient survey datasets from LOFAR, MWA and MeerKAT. Finally, I will describe some of the machine learning strategies that we intend to investigate in the future.