Classification without labels: learning from mixed samples in high energy physics

Modern machine learning techniques can be used to construct powerful models for difficult collider physics problems. In many applications, however, these models are trained on imperfect simulations due to a lack of truth-level information in the data, which risks the model learning artifacts of the...

Full description

Bibliographic Details
Main Authors: Nachman, Benjamin, Metodiev, Eric Mario, Thaler, Jesse
Other Authors: Massachusetts Institute of Technology. Center for Theoretical Physics
Format: Article
Language:English
Published: Springer International Publishing AG 2018
Online Access:http://hdl.handle.net/1721.1/114621
https://orcid.org/0000-0002-2406-8160
_version_ 1826217954241413120
author Nachman, Benjamin
Metodiev, Eric Mario
Thaler, Jesse
author2 Massachusetts Institute of Technology. Center for Theoretical Physics
author_facet Massachusetts Institute of Technology. Center for Theoretical Physics
Nachman, Benjamin
Metodiev, Eric Mario
Thaler, Jesse
author_sort Nachman, Benjamin
collection MIT
description Modern machine learning techniques can be used to construct powerful models for difficult collider physics problems. In many applications, however, these models are trained on imperfect simulations due to a lack of truth-level information in the data, which risks the model learning artifacts of the simulation. In this paper, we introduce the paradigm of classification without labels (CWoLa) in which a classifier is trained to distinguish statistical mixtures of classes, which are common in collider physics. Crucially, neither individual labels nor class proportions are required, yet we prove that the optimal classifier in the CWoLa paradigm is also the optimal classifier in the traditional fully-supervised case where all label information is available. After demonstrating the power of this method in an analytical toy example, we consider a realistic benchmark for collider physics: distinguishing quark- versus gluon-initiated jets using mixed quark/gluon training samples. More generally, CWoLa can be applied to any classification problem where labels or class proportions are unknown or simulations are unreliable, but statistical mixtures of the classes are available. Keyword: jets
first_indexed 2024-09-23T17:11:41Z
format Article
id mit-1721.1/114621
institution Massachusetts Institute of Technology
language English
last_indexed 2024-09-23T17:11:41Z
publishDate 2018
publisher Springer International Publishing AG
record_format dspace
spelling mit-1721.1/1146212022-10-03T11:05:04Z Classification without labels: learning from mixed samples in high energy physics Nachman, Benjamin Metodiev, Eric Mario Thaler, Jesse Massachusetts Institute of Technology. Center for Theoretical Physics Massachusetts Institute of Technology. Department of Physics Metodiev, Eric Mario Thaler, Jesse Modern machine learning techniques can be used to construct powerful models for difficult collider physics problems. In many applications, however, these models are trained on imperfect simulations due to a lack of truth-level information in the data, which risks the model learning artifacts of the simulation. In this paper, we introduce the paradigm of classification without labels (CWoLa) in which a classifier is trained to distinguish statistical mixtures of classes, which are common in collider physics. Crucially, neither individual labels nor class proportions are required, yet we prove that the optimal classifier in the CWoLa paradigm is also the optimal classifier in the traditional fully-supervised case where all label information is available. After demonstrating the power of this method in an analytical toy example, we consider a realistic benchmark for collider physics: distinguishing quark- versus gluon-initiated jets using mixed quark/gluon training samples. More generally, CWoLa can be applied to any classification problem where labels or class proportions are unknown or simulations are unreliable, but statistical mixtures of the classes are available. Keyword: jets United States. Department of Energy (Contract DE-SC-00012567) United States. Department of Energy (Contract DE-SC-0001547) 2018-04-09T15:29:08Z 2018-04-09T15:29:08Z 2017-10 2017-09 2017-11-10T05:57:26Z Article http://purl.org/eprint/type/JournalArticle 1126-6708 1029-8479 http://hdl.handle.net/1721.1/114621 Metodiev, Eric M. et al. "Classification without labels: learning from mixed samples in high energy physics." Journal of High Energy Physics 2017 (October 2017): 174 © 2017 The Author(s) https://orcid.org/0000-0002-2406-8160 en http://dx.doi.org/10.1007/JHEP10(2017)174 Journal of High Energy Physics Creative Commons Attribution http://creativecommons.org/licenses/by/4.0/ The Author(s) application/pdf Springer International Publishing AG Springer Berlin Heidelberg
spellingShingle Nachman, Benjamin
Metodiev, Eric Mario
Thaler, Jesse
Classification without labels: learning from mixed samples in high energy physics
title Classification without labels: learning from mixed samples in high energy physics
title_full Classification without labels: learning from mixed samples in high energy physics
title_fullStr Classification without labels: learning from mixed samples in high energy physics
title_full_unstemmed Classification without labels: learning from mixed samples in high energy physics
title_short Classification without labels: learning from mixed samples in high energy physics
title_sort classification without labels learning from mixed samples in high energy physics
url http://hdl.handle.net/1721.1/114621
https://orcid.org/0000-0002-2406-8160
work_keys_str_mv AT nachmanbenjamin classificationwithoutlabelslearningfrommixedsamplesinhighenergyphysics
AT metodievericmario classificationwithoutlabelslearningfrommixedsamplesinhighenergyphysics
AT thalerjesse classificationwithoutlabelslearningfrommixedsamplesinhighenergyphysics