Powered by Indico
v3.1.1
Heavy ion collisions produce a medium of strongly interacting quarks and gluons called Quark Gluon Plasma (QGP). Studying the properties of the QGP medium with external probes is difficult due to its short lifetime of approximately 10^-23 seconds. Internally generated probes, such as jets, provide a built-in way to study the properties of the QGP. Jets are collections of collimated particles produced when quarks or gluons scatter off each other with a large momentum transfer. The scattered partons traverse the medium, losing energy in a manner dependent on the transport properties of the QGP. However, access to this energy loss depends on the ability of experimenters to accurately reconstruct the jets in the particle dense environment produced in heavy ion collisions. Reconstruction results in particles related to the hard scattering being clustered with particles from soft processes (combinatorial background) and clusters entirely composed of soft particles (combinatorial jets). This talk explores different methods for dealing with this background by studying the problem with several different Monte Carlo models including Pythia, Angantyr, and a custom-made heavy ion background generator, TennGen. Machine learning techniques including deep neural nets, symbolic regression, and random forests are explored in these models with promising results for jet background subtraction and mitigation in data. Finally, the implications of these techniques are briefly discussed with emphasis on finding a balance between the predictive power of black-box machine learning methods with the clarity of more physics inspired and human-understandable methods.
Recording:
https://bnl.zoomgov.com/rec/share/zxzzzDccjFGHfRAx2S92d6LtZT6yAcVz4VwUYA6ivAJ0Ddawt-Sup57SGkvc7Ntn.uCHoI-3un5K56rCk
Passcode: Lfc^.^1&