- Fast Transients and Pulsars/ Data Intensive Astronomy
The Southern-Sky MWA Rapid Two-Metre (SMART) pulsar survey is an ambitious program currently in its early stages, slated to discover hundreds of new pulsars — with three discoveries having been made so far (at the time of this writing), and many more around the corner. A major challenge is the rapid identification of true pulsar signals amidst millions of spurious candidates. To this end, two machine learning solutions have been explored. The first was developed for a different pulsar survey (LOTAAS), and is being actively used in the SMART pipeline. The second is a bespoke Semi-supervised Generative Adversarial Network (SGAN) developed by a former Honours student.
In this project, the student will (1) integrate the SGAN into the SMART pipeline, (2) benchmark its performance on the Garrawarla supercomputing cluster (at Pawsey), and (3) investigate the effect of retraining the network on a much larger sample set. An important test will be to determine the SGAN’s efficacy, by determining whether it would have successfully detected the SMART survey’s first few confirmed pulsar discoveries.
Enrollment in any Physics or Computing course is appropriate. The applicant should have an interest in astrophysics, but it is not required to be enrolled in the astronomy stream.
Unix/Linux operating systems, Python
Tensorflow, Nextflow, using supercomputing systems
- Week 1 Inductions and project introduction
- Week 2 Initial Presentation
- Week 3 Training & running the SGAN on local computer
- Week 4 Training & running the SGAN on Garrawarla
- Week 5 Train on large numbers of labelled inputs
- Week 6 Benchmark & improve SGAN; compare with other ML
- Week 7 Benchmark & improve SGAN; compare with other ML
- Week 8 Incorporate the SGAN into the SMART Nextflow pipeline
- Week 9 Final Presentation
- Week 10 Final Report
Dr Ramesh Bhat
Senior Research FellowRead More