Machine Learning and Science Forum
Date: Monday, November 23, 2020
Time: 11:00 AM - 12:00 PM Pacific Time
Location: Participate remotely using this Zoom link
Set and Sequence Machine Learning for Particle Identification at the Large Hadron Collider
Speaker: Nicole Hartman, SLAC
Abstract: One task that is important for the physics program at the Large Hadron Collider is identifying which particle gave rise to an energy deposition in the detector. Identifying b-quarks by leveraging their displaced decay is colloquially called “b-tagging” or more generally “flavor tagging." Extracting the relevant characteristics from the observed b-quark decay products is naturally a high dimensional task that’s been an area of active ML development in the experimental HEP community. I’ll review some of the traditional algorithms that the ATLAS experiment has used for this task, and then motivate the use of recurrent neural networks (RNNs). I’ll highlight where RNNs for b-tagging have already led to improvements in analysis sensitivity. I’ll then introduce the use of Deep Sets for this application, which uses a different set of assumptions that allow for faster training and inference. I’ll summarize some of the metrics used to understand what the network is learning and conclude with a discussion of the impact of incorporating such a new tagger into our workflow. Although I’ll cover relevant material from earlier results, the focus will be the Deep Sets work which is summarized in this note: https://cds.cern.ch/record/2718948/files/ATL-PHYS-PUB-2020-014.pdf.
The BIDS Machine Learning and Science Forum (formerly the Berkeley Statistics and Machine Learning Forum) was launched in Spring 2018 and currently meets biweekly (during the spring and fall semesters) to discuss current applications across a wide variety of research domains in the physical sciences and beyond. Hosted by BIDS Faculty Affiliate Uroš Seljak (professor of Physics at UC Berkeley) and BIDS Research Affiliate Ben Nachman (Physicist and Staff Scientist at Lawrence Berkeley National Laboratory), these active sessions bring together domain scientists, statisticians, and computer scientists who are either developing state-of-the-art methods or are interested in applying these methods in their research. All interested members of the UC Berkeley and Berkeley Lab communities are welcome and encouraged to attend. To receive email notifications about upcoming meetings, or to request more information, please contact email@example.com.