skip to main content
DOE PAGES title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: (Machine) learning to do more with less

Abstract

Determining the best method for training a machine learning algorithm is critical to maximizing its ability to classify data. In this paper, we compare the standard “fully supervised” approach (which relies on knowledge of event-by-event truth-level labels) with a recent proposal that instead utilizes class ratios as the only discriminating information provided during training. This so-called “weakly supervised” technique has access to less information than the fully supervised method and yet is still able to yield impressive discriminating power. In addition, weak supervision seems particularly well suited to particle physics since quantum mechanics is incompatible with the notion of mapping an individual event onto any single Feynman diagram. We examine the technique in detail — both analytically and numerically — with a focus on the robustness to issues of mischaracterizing the training samples. Weakly supervised networks turn out to be remarkably insensitive to a class of systematic mismodeling. Furthermore, we demonstrate that the event level outputs for weakly versus fully supervised networks are probing different kinematics, even though the numerical quality metrics are essentially identical. This implies that it should be possible to improve the overall classification ability by combining the output from the two types of networks. For concreteness,more » we apply this technology to a signature of beyond the Standard Model physics to demonstrate that all these impressive features continue to hold in a scenario of relevance to the LHC. Example code is provided on GitHub.« less

Authors:
 [1];  [1]; ORCiD logo [1]
  1. Univ. of Oregon, Eugene, OR (United States)
Publication Date:
Research Org.:
Univ. of Oregon, Eugene, OR (United States)
Sponsoring Org.:
USDOE
OSTI Identifier:
1512295
Grant/Contract Number:  
SC0018191; SC0011640
Resource Type:
Accepted Manuscript
Journal Name:
Journal of High Energy Physics (Online)
Additional Journal Information:
Journal Name: Journal of High Energy Physics (Online); Journal Volume: 2018; Journal Issue: 2; Journal ID: ISSN 1029-8479
Publisher:
Springer Berlin
Country of Publication:
United States
Language:
English
Subject:
72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS; Beyond Standard Model; Hadron-Hadron scattering (experiments); Particle correlations and fluctuations; Supersymmetry

Citation Formats

Cohen, Timothy, Freytsis, Marat, and Ostdiek, Bryan. (Machine) learning to do more with less. United States: N. p., 2018. Web. doi:10.1007/jhep02(2018)034.
Cohen, Timothy, Freytsis, Marat, & Ostdiek, Bryan. (Machine) learning to do more with less. United States. doi:10.1007/jhep02(2018)034.
Cohen, Timothy, Freytsis, Marat, and Ostdiek, Bryan. Tue . "(Machine) learning to do more with less". United States. doi:10.1007/jhep02(2018)034. https://www.osti.gov/servlets/purl/1512295.
@article{osti_1512295,
title = {(Machine) learning to do more with less},
author = {Cohen, Timothy and Freytsis, Marat and Ostdiek, Bryan},
abstractNote = {Determining the best method for training a machine learning algorithm is critical to maximizing its ability to classify data. In this paper, we compare the standard “fully supervised” approach (which relies on knowledge of event-by-event truth-level labels) with a recent proposal that instead utilizes class ratios as the only discriminating information provided during training. This so-called “weakly supervised” technique has access to less information than the fully supervised method and yet is still able to yield impressive discriminating power. In addition, weak supervision seems particularly well suited to particle physics since quantum mechanics is incompatible with the notion of mapping an individual event onto any single Feynman diagram. We examine the technique in detail — both analytically and numerically — with a focus on the robustness to issues of mischaracterizing the training samples. Weakly supervised networks turn out to be remarkably insensitive to a class of systematic mismodeling. Furthermore, we demonstrate that the event level outputs for weakly versus fully supervised networks are probing different kinematics, even though the numerical quality metrics are essentially identical. This implies that it should be possible to improve the overall classification ability by combining the output from the two types of networks. For concreteness, we apply this technology to a signature of beyond the Standard Model physics to demonstrate that all these impressive features continue to hold in a scenario of relevance to the LHC. Example code is provided on GitHub.},
doi = {10.1007/jhep02(2018)034},
journal = {Journal of High Energy Physics (Online)},
number = 2,
volume = 2018,
place = {United States},
year = {2018},
month = {2}
}

Journal Article:
Free Publicly Available Full Text
Publisher's Version of Record

Save / Share:

Works referenced in this record:

Ensemble-based classifiers
journal, November 2009