Difference: 2018-03-29 ( vs. 1)

Agenda:

• ZTF survey updates
• IPAC updates
• Updates on RB using RF (including those from students)
• Updates on SG separation
• Campaigns to run with marshal and zooniverse: plans
• papers
• Brief updates on asteroids/streaks
• Bryan’s presentation on RB using deep learning.
• Other items from the floor

Attendees: Ashish, Richard, Matthew, Vaidehi, Brian, Umaa, Adam, Charlotte, Rahul, Frank, Quan-Zhi, David Shupe

ZTF updates (Matthew): changes to alerts (SNR > 5) Alerts seen last few nights: 11K, 16K, 29K Increasing as number of ref images increase

Many alerts not passing filters counter to go on the marshal page (of detected alerts) until middle of next week with g filter.

Frank: New SG going in next week More bogus events coming through newer metadata in the alerts packets

Adam: matching looks good - so we should start next week two billion rows - 25% data have matches. Some may be spurious transient events there may be some poor source matching form Pan-STARRS end. Dima to run a ps1 X ps1 crossmatch to look for duplicates

Umaa: Single digit FP and FN bimodal distribution starting to appear [slides]

Charlotte: feature importance elongation, magnr important classstar is bad (but should improve with new SG separation) RA of nearest object being used (should not be!) Frank: zpref and zpdiff are constant for the survey (and should not appear as discriminators) look at test set with only meaningful objects

Ashish: Papers. Survey paper draft due by May 1

Brian: summarized deep learning on stamps for real-bogus. Seems to work better than RF, but takes more time training.

-- AshishMahabal - 04 Apr 2018