Choose timezone
Your profile timezone:
VBSCan - WG1 periodic meeting
→
Europe/Zurich
CERN
CERN
,
Description
- Join Zoom Meeting
- https://uni-freiburg.zoom.us/j/65801799581?pwd=SFVOdTZUMDFFcVZHZTcwSFk4WUpBUT09
- Meeting ID: 658 0179 9581
- Passcode: 4cjUTyAWC
VBSCan - WG1 periodic meeting
Friday 4 Jun 2021, 13:00 → 14:40
Up to 18 participants
1. Biswajit Das
- can you define more in detail what you mean by "prototype diagrams"?
- they are the minimum set of diagrams one has to calculate, that can be used for other diagram calculations.
for instance exchanging momenta one can go from prototype to another diag.
- how is this calculation different from qqbar?
- in that case you don't have the diagrams with the bbh yukawa interactions (even if you consider mb=0).
- did you apply any cuts/selection on final state? did you check how this changes?
- we are not decaying the W so we are not applying any cut. in principle we could but we haven't looked.
2. Andrei Popescu
- how the intereference contributions are depending on the order of the computation (LO, NLO, NNLO)?
- these are quite stable at different orders.
- how is changing the polarisation fraction depending on the order of the computation (LO, NLO, NNLO)?
- it is also rather stable.
3. Olivier Mattelaer
- did you try to redo the comparison for pp->WZjj against VBFNLO (done 10 years ago)?
- no. comparisons were done only against previous MG versions.
- did you look at the pp->4j2l or even pp->6j processes?
- no, but there are some comparison for off-shell pp->2j4l processes.
4. Taegyun Kim
- did you understand how the network works out the polarizations?
- it could be related to how the subjets are located. the theta* angle in the rest frame, after boost becomes more an opening angle. this is the same principle that is used in N-subjettiness.
- about training: you chose 2 different processes. why not stick to one only?
- this is supervised learning so you have to give it all the information, such that it learns the characteristics of all the classes. you cannot learn only one sample and then apply it to another. the alternative would be to used unsupervised learning, such as autoencoders.
- here you train on 2 specific processes and then look at another one. but say you train on W+1j and apply to W+2j. is it still stable?
- yes. even though we are studying W+1j we really study only the W. so it doesn't matter
- but is the pattern truly universal? i.e. can the image be modified by the presence of other partons?
- we haven't seen significant distortions yet in or tests. we tested on WH and it worked correctly.
- about energies: you train on a certain pT window?
- yes, in the end we will have 2 different trained network for each pt window.
There are minutes attached to this event.
Show them.