[Theory-seminars] Theory seminar on Monday Jan. 11th at 1pm

[Christos Kallidonis]

Dear all,

Wishing everyone a Happy and Prosperous New Year!

Tomorrow we will have this year’s first theory seminar, please see below for details:

_____________________________________________________________
Bluejeans connection: https://bluejeans.com/801786278
Date and time: Monday January 11th, 13:00
Speaker: Patrick Komiske (MIT)
Title: Simultaneously Unfolding All Observables with Deep Learning
Abstract: Unfolding is the procedure by which the "recorded" detector-level distribution of an observable is corrected for detector effects and other sources of noise to obtain the "true" particle-level distribution. It is a ubiquitous step in the measurement of particle physics quantities, but many procedures in common use suffer acutely from the curse of dimensionality when trying to unfold multi-differential distributions due to their explicit reliance on histograms. In this talk, I will discuss a method, OmniFold, that uses deep learning to take advantage of the full phase-space information at both detector and truth levels to find the maximum likelihood solution to the unfolding problem. OmniFold learns a universal weighting of truth events such that the distribution of any observable can be calculated. The method is demonstrated on simulated Z+jet events at the LHC using the Pythia and Herwig event generators and the Delphes detector package. It is compared to Iterated Bayesian Unfolding, the maximum likelihood solution restricted to histograms, on a variety of jet substructure observables, showing equal or improved robustness in all cases. Prospects and challenges for improving future measurements at colliders like the LHC and EIC will be discussed.
_____________________________________________________________

See you all there!

Best regards,
Astrid, Christos, Filippo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/theory-seminars/attachments/20210110/8bb524d0/attachment.html>