Evidential Deep Learning for uncertainty quantification in jet tagging deep neural network model
Wang, Xiwei
This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.
Permalink
https://hdl.handle.net/2142/124591
Description
Title
Evidential Deep Learning for uncertainty quantification in jet tagging deep neural network model
Author(s)
Wang, Xiwei
Issue Date
2024-04-30
Director of Research (if dissertation) or Advisor (if thesis)
Kindratenko, Volodymyr
Neubauer, Mark
Department of Study
Electrical & Computer Eng
Discipline
Electrical & Computer Engr
Degree Granting Institution
University of Illinois at Urbana-Champaign
Degree Name
M.S.
Degree Level
Thesis
Keyword(s)
Evidential Deep Learning
Jet Tagging
High Energy Physics
Abstract
Evidential Deep Learning (EDL) is an uncertainty-aware deep learning method used in order to provide confidence about the input data. The learning based on the input data is treated as an evidence acquisition process and more evidence is interpreted as the increased predictive confidence. In this way, the model with EDL is able to quantify the epistemic uncertainty (uncertainty) of the input data and detect the anomaly in the data. In this study, we explore the integration of Evidential Deep Learning with the Particle Flow Identification Network (PFIN), a deep neural network model tailored for jet tagging in high-energy physics. We adapted EDL principles to enhance the PFIN model, enabling it not only to make predictions but also to estimate the confidence level of those predictions and detect possible anomaly. This adaptation involved developing an evidence layer within the DNN architecture, allowing the model to dynamically assess and quantify uncertainty by interpreting the input data's reliability and relevance. Our results show that the EDL-enhanced PFIN model significantly outperforms conventional models in uncertainty quantification without sacrificing predictive accuracy. This improvement is particularly notable in the detection of anomalous data, where the model's ability to quantify uncertainty provides a robust mechanism for identifying out-of-distribution data. Such capabilities are critical in high-energy physics experiments, where precise and reliable data interpretation can lead to groundbreaking discoveries.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.
