2021-11-11

Present:

  • Ramesh Bhat
  • Sammy McSweeney
  • Nick Swainston
  • Susmita Sett
  • Isaac Colleran
  • Willem van Straten
  • Shi Dai
  • Ryan Shannon
  • Ian Morrison
  • Mengyao Xue
  • Marcin Sokolowski

Agenda:

  1. Overview and general updates (Ramesh)
  2. Processing updates + candidate scrutiny (Nick)
  3. MWAX and commissioning plans (Sammy)
  4. ML testing and integration plans (Isaac)
  5. Update on papers in prep. (Ramesh/Sammy) 
  6. Publication policy discussion (all) 
  7. Discussion time (all) 

Minutes:

  • Ramesh: Overview
    • Nothing has changed in terms of observings
      • ...because MWA is still undergoing upgrade
      • Science commissioning tasks are imminent
      • Then SMART survey will continue, while still in compact config
    • PSR #3: J1002-2036
      • Found using “SMART”er (har har) candidate sifting process, including clustering algorithms that take advantage of complex tied-array beam (for compact config)
      • First estimate: ~10 mJy, possibly steep spectrum
      • 30-sigma detection over 80 mins
    • PSR #2: J0026-1955
      • Active proposals at Parkes and GMRT for follow-up
      • Various issues with localisation
      • Parkes (P1134) followup: Shi managed to pull a detection out of a magic hat, where Ramesh and Nick both had trouble. Found in 2nd harmonic.
      • GMRT Cycle 41 followup: TAC suggested trying full polarimetry. Requires a bit more overhead for calibration, but is an acceptable risk, so we will go for it.
        • Nulling makes confirmation of source seen in GMRT image a little less straightforward, so we are taking conservative approach.
        • Observation campaign starting soon!
        • Flux of candidate GMRT source suggests possibly steep spectrum source!
      • Priorities
        • Papers (Nick’s, Sam’s, Ramesh’s) all underway
        • Commissioning the new system
        • Continue SMART observing as soon as system allows!
        • Continue processing and candidate searching (= “pulsar hunting”)
    • Discussion:
      • WvS: offers generic help, e.g. for subpulse drifting analysis, polarimetric calibration
        • SM: I might reach out about MWA polarimetry
        • Ramesh: GMRT polarimetric is also new, so might need extra help/advice with that.
      • RS: phase-connected timing solution for PSR #2?
        • Ramesh: Not yet. This is the main objective for Parkes followup data.
        • RS: I can help with that.
        • SD: RFI was really bad, so careful cleaning might help with higher frequencies. Optimistic about timing.
  • Nick’s update:
    • Full 80-minute detection shown for PSR #3
    • Not yet found in other obs, but still have other compact config observations to process
    • It was found by exploiting compact config tied-array beam pattern, with its grating lobes
    • Introduced some smarts to eliminate large number of known pulsar detections by flagging candidates coincident in DM and in nearby tied-array beams
    • A few more obs searched, processing is chugging along
    • Ramesh: Will be re-applying for supercompute (round is open now, closes in a couple of weeks). Should still be well justified for the next couple of quarters.
      • Nick concurs
    • Ramesh: We will ask for a similar request as last time.
  • Sam’s update:
    • (Will fill in later)
  • Isaac’s update:
    • Showing results for trained networks. ~85% accuracy (on average, across networks).
    • SGAN best model even better
    • Possibly even better than that because some candidates in validation set were incorrectly labelled.
    • Will be implementing trained network into SMART pipeline.
    • SD: What is F1-score?
      • IC: Just one of many metrics.
  • Paper updates:
  • Discussion:
    • WvS: Recall might be most important metric for ML
    • WvS: van Leeuwin 0809 (reference to PSR #2)
    • WvS: Looking forward to polarimetry on PSR #2
    • SD: Can all SMART detections be detected in continuum imaging? Would be interesting to get spectrum.
      • RB: Should be possible for subset of pulsars. Susmita has a “golden” image used for testing this kind of thing. Short answer, ~0.5 of pulsars might be able to see in imaging.
      • MS: compact vs extended configuration will make a difference!
      • RB: GMRT probably still best way to get positions