HEP.TrackX Weekly

Monday 21 Nov 2016, 13:30 → 14:30 US/Pacific

vidyo

Paolo Calafiura, Steven Farrell

Please join the meeting by clicking this link:
https://vidyoportal.cern.ch/flex.html?roomdirect.html&key=UoHjbOgfcsrBsDfBwOlOsCObR3E 

If you want to join by phone, please use one of the phone numbers listed in the link below: http://information-technology.web.cern.ch/services/fe/howto/users-join-vidyo-meeting-phone and enter the meeting extension 103958871 in order to join.

Present:  Josh, Jim, Giuseppe, Paolo, Steve, Mayur, Jean-Roch

Apologies :

Hot items:

• ML on HPC workshop: limits on multi-node training. Connectivity limitation (sync after SGD for each batch), and overtraining with increasing batch size.

• Conference contributions (poster, potentially paper):

  • Connecting the Dots/Intelligent Detector - LAL Mar 2017 (all plenary workshop)

    • Deadline Dec 9

  • International conference of learning representations - ICLR, 2017

    • Deadline for submissions to workshops Dec ?

  • International conference of machine learning - icml 2017

    • Deadline for submissions Feb 5

• Discussed collaboration with Parallel CKF group:

  • Will work together with them when benchmarking algorithms

Outstanding Action Items:

• Pietro → introduce group activities

• Maria: I want to try to put all the agendas and the links and the minutes in a Basecamp : do you people use Basecamp? Pietro Shall we try? Others?

  • PC: Indico+google seems to work but I am open to evaluate basecamp

• Data Storage: JBK

• Paolo: add acknowledgment of PCKF on HEP.TrkX web page https://heptrkx.github.io

• Computing platform:

  • NERSC ERCAP allocation:  Lali said we can start when we want

    • NERSC account request form

  • PC submitted NERSC ERCAP request for 1.2M hours 5+5 TB storage

  • CalTech GPU servers. Jean-Roch contact person if you need access.

New Action Items:

 

Round the Table for Status Reports:

LBL

Paolo: fixed bug in ACTS track generation example (was generating neutral “muons”). Moved to cylindrical coordinates. Throwing away second hit on double-sided detector layer.

Test files, and a conversion script, uploaded to our google drive.

 

Mayur: single track “Kalman Filter” working reasonably well. Will try on 1M track training sample.

 

Steve: Multi-track background handled OK in hit probability LSTM network. Trying again with trapezoid detector, using remapping layer. Seem to work single track, more checking to be done, working on generating multi-track samples.

Next Meeting Monday Dec 5