Mapping machine-learned physics into a human-readable space

Mapping machine-learned physics into a human-readable space
复制标题

将机器学习的物理映射到人类可读的空间

DOI:
10.1103/physrevd.103.036020
复制
发表时间:
2021
期刊:
影响因子:
5
通讯作者:
Whiteson, Daniel
Whiteson, Daniel
中科院分区:
物理与天体物理2区
文献类型:
--
作者:
Faucett, Taylor;Thaler, Jesse;Whiteson, Daniel

文献摘要

参考文献

被引文献

相似文献

我们提出了一种技术,将运行在高维输入空间上的黑盒机器学习的分类器转换为一组人类可解释的可观测数据,这些可观测数据可以组合在一起做出相同的分类决策。我们通过找到与黑盒具有最高决策相似性的那些高级别判别式来迭代地从大量高级判别式空间中选择这些可观测参数,这些黑盒是通过我们引入的评估输入对的相对排序的度量来量化的。连续迭代只关注被当前可观测集合错乱排序的输入对的子集。这种方法可以简化机器学习策略,根据已被充分理解的物理概念解释结果,验证物理模型,并有可能对问题本身的性质有新的见解。作为演示,我们将我们的方法应用于对撞机物理中的喷注分类基准任务,在该任务中,作用于量热计喷注图像的卷积神经网络的性能优于一组六个众所周知的喷射子结构观察值。我们的方法将卷积神经网络映射成一组称为能量流多项式的可观测数据,它通过识别一类具有有趣的物理解释的可观测数据来缩小性能差距,这在以前的JET子结构文献中被忽视了。
We present a technique for translating a black-box machine-learned classifier operating on a high-dimensional input space into a small set of human-interpretable observables that can be combined to make the same classification decisions. We iteratively select these observables from a large space of high-level discriminants by finding those with the highest decision similarity relative to the black box, quantified via a metric we introduce that evaluates the relative ordering of pairs of inputs. Successive iterations focus only on the subset of input pairs that are misordered by the current set of observables. This method enables simplification of the machine-learning strategy, interpretation of the results in terms of well-understood physical concepts, validation of the physical model, and the potential for new insights into the nature of the problem itself. As a demonstration, we apply our approach to the benchmark task of jet classification in collider physics, where a convolutional neural network acting on calorimeter jet images outperforms a set of six well-known jet substructure observables. Our method maps the convolutional neural network into a set of observables called energy flow polynomials, and it closes the performance gap by identifying a class of observables with an interesting physical interpretation that has been previously overlooked in the jet substructure literature.
DOI: 10.1007/jhep10(2017)174
发表时间: 2017-08
影响因子: 5.4
作者:
E. Metodiev;B. Nachman;J. Thaler
通讯作者: E. Metodiev;B. Nachman;J. Thaler
$m{sigma (p ar p o Z) cdot}$Br$m{(Z o au au)}$ 在 $m{sqrt{s}=}$1.96 处的测量
DOI: --
发表时间: 2005
期刊:
影响因子: --
作者:
V. Abazov;B. Abbott;M. Abolins;B. Acharya;M. Adams;T. Adams;M. Agelou;J. L. Agram;S. Ahn;M. Ahsan;G. Alexeev;G. Alkhazov;A. Alton;G. Alverson;G. Alves;M. Anastasoaie;T. Andeen;S. Anderson;B. Andrieu;Y. Arnoud;A. Askew;B. Åsman;O. Atramentov;C. Autermann;C. Avila;F. Badaud;A. Baden;B. Baldin;P. Balm;S. Banerjee;E. Barberis;P. Bargassa;P. Baringer;C. Barnes;J. Barreto;J. Bartlett;U. Bassler;D. Bauer;A. Bean;S. Beauceron;M. Begel;A. Bellavance;S. Beri;G. Bernardi;I. Bertram;M. Besançon;R. Beuselinck;V. Bezzubov;P. Bhat;V. Bhatnagar;M. Binder;C. Biscarat;K. Black;I. Blackler;G. Blazey;F. Blekman;S. Blessing;D. Bloch;U. Blumenschein;A. Boehnlein;O. Boeriu;T. Bolton;F. Borcherding;G. Borissov;K. Bos;T. Bose;A. Brandt;R. Brock;G. Brooijmans;A. Bross;N. Buchanan;D. Buchholz;M. Buehler;V. Buescher;S. Burdin;T. H. Burnett;E. Busato;J. Butler;J. Bystrický;W. Carvalho;B. Casey;N. Cason;H. Castilla;S. Chakrabarti;D. Chakraborty;K. Chan;A. Chandra;D. Chapin;F. Charles;E. Cheu;L. Chevalier;D. Cho;S. Choi;B. Choudhary;T. Christiansen;L. Christofek;D. Claes;B. Clement;C. Clement;Y. Coadou;M. Cooke;W. Cooper;D. Coppage;M. Corcoran;A. Cothenet;M. Cousinou;B. Cox;S. Crépé;M. Cristetiu;D. Cutts;H. Motta;B. Davies;G. Davies;G. Davis;K. De;P. D. Jong;S. D. Jong;E. Cruz;C. D. Martins;S. Dean;F. Deliot;M. Demarteau;R. Demina;P. Demine;D. Denisov;S. Denisov;S. Desai;H. Diehl;M. Diesburg;M. Doidge;H. Dong;S. Doulas;L. Dudko;L. Duflot;S. Dugad;A. Duperrin;J. Dyer;A. Dyshkant;M. Eads;D. Edmunds;T. Edwards;J. Ellison;J. Elmsheuser;J. Eltzroth;V. Elvira;S. Eno;P. Ermolov;O. Eroshin;J. Estrada;D. Evans;H. Evans;A. Evdokimov;V. Evdokimov;J. Fast;S. Fatakia;L. Feligioni;T. Ferbel;F. Fiedler;F. Filthaut;W. Fisher;H. Fisk;M. Fortner;H. Fox;William R. Freeman;S. Fu;S. Fuess;T. Gadfort;C. Galea;E. Gallas;E. Galyaev;C. Garcia;A. Garcia;J. Gardner;V. Gavrilov;D. Gelé;R. Gelhaus;K. Genser;C. Gerber;Y. Gershtein;G. Ginther;T. Golling;B. Gomez;K. Gounder;A. Goussiou;P. Grannis;S. Greder;H. Greenlee;Z. Greenwood;E. Gregores;P. Gris;J. Grivaz;L. Groer;S. Grünendahl;M. Grünewald;S. N. Gurzhiev;G. Gutiérrez;P. Gutierrez;A. Haas;N. Hadley;S. Hagopian;I. Hall;R. E. Hall;C. Han;L. Han;K. Hanagaki;K. Harder;R. Harrington;J. Hauptman;R. Hauser;J. Hays;T. Hebbeker;D. Hedin;J. Heinmiller;A. Heinson;U. Heintz;C. Hensel;G. Hesketh;M. Hildreth;R. Hirosky;J. Hobbs;B. Hoeneisen;M. Hohlfeld;S. Hong;R. Hooper;P. Houben;Y. Hu;J. Huang;I. Iashvili;R. Illingworth;A. Ito;S. Jabeen;M. Jaffré;S. Jain;V. Jain;K. Jakobs;A. Jenkins;R. Jesik;K. Johns;Mikkel B. Johnson;A. Jonckheere;Per Jönsson;H. Jöstlein;A. Juste;D. Käfer;W. Kahl;S. Kahn;E. Kajfasz;A. Kalinin;J. Kalk;D. Karmanov;J. Kasper;D. Kau;R. Kaur;R. Kehoe;S. Kermiche;S. Kesisoglou;A. Khanov;A. Kharchilava;Y. Kharzheev;Ki;B. Klima;M. Klute;J. Kohli;M. Kopal;V. Korablev;J. Kotcher;B. Kothari;A. Koubarovsky;A. Kozelov;J. Kozminski;S. Krzywdzinski;S. Kuleshov;Y. Kulik;A. Kumar;S. Kunori;A. Kupco;T. Kurca;S. Lager;N. Lahrichi;G. Landsberg;J. Lazo;A. Bihan;P. Lebrun;S. W. Lee;W. Lee;A. Leflat;F. Lehner;C. Leonidopoulos;P. Lewis;J. Li;Q. Li;J. Lima;D. Lincoln;S. Linn;J. Linnemann;V. Lipaev;R. Lipton;L. Lobo;A. Lobodenko;M. Lokajicek;A. Lounis;H. Lubatti;L. Lueking;M. Lynker;A. Lyon;A. Maciel;R. Madaras;P. Mättig;A. Magerkurth;A. Magnan;N. Makovec;P. Mal;S. Malik;V. Malyshev;H. Mao;Y. Maravin;M. Martens;S. Mattingly;A. Mayorov;R. Mccarthy;R. McCroskey;D. Meder;H. Melanson;A. Melnitchouk;A. Mendes;M. Merkin;K. Merritt;A. Meyer;M. Michaut;H. Miettinen;J. Mitrevski;N. Mokhov;J. Molina;N. Mondal;R. Moore;G. Muanza;M. Mulders;Y. Mutaf;E. Nagy;M. Narain;N. A. Naumann;H. Neal;J. Negret;S. Nelson;P. Neustroev;C. Noeding;A. Nomerotski;S. Novaes;T. Nunnemann;E. Nurse;V. O’dell;D. O’Neil;V. Oguri;N. Oliveira;N. Oshima;G. O. Y. Garzon;P. Padley;N. Parashar;Jae;S. Park;J. Parsons;R. Partridge;N. Parua;A. Patwa;P. Perea;E. Perez;P. Pétroff;M. Petteni;L. Phaf;R. Piegaia;M. Pleier;P. Podesta;V. Podstavkov;Y. Pogorelov;B. Pope;W. D. Silva;H. Prosper;S. Protopopescu;J. Qian;A. Quadt;B. Quinn;K. Rani;K. Ranjan;P. Rapidis;P. Ratoff;N. Reay;S. Reucroft;M. Rijssenbeek;I. Ripp;F. Rizatdinova;C. Royon;P. Rubinov;R. Ruchti;V. Rud;G. Sajot;A. Sanchez;M. Sanders;A. Santoro;G. Savage;L. Sawyer;T. Scanlon;D. Schaile;R. Schamberger;H. Schellman;P. Schieferdecker;C. Schmitt;A. Schukin;A. Schwartzman;R. Schwienhorst;S. Sengupta;H. Severini;E. Shabalina;M. Shamim;V. Shary;W. Shephard;R. K. Shivpuri;D. Shpakov;R. Sidwell;V. Šimák;V. Sirotenko;P. Skubic;P. Slattery;R. P. Smith;K. Smolek;G. Snow;J. Snow;S. Snyder;S. Söldner;X. Song;Y. Song;L. Sonnenschein;A. Sopczak;M. Sosebee;K. Soustruznik;M. Souza;B. Spurlock;N. Stanton;J. Stark;J. Steele;G. Steinbrück;K. Stevenson;V. Stolin;A. Stone;D. Stoyanova;J. Strandberg;M. Strang;M. Strauss;R. Ströhmer;D. Strom;M. Strovink;L. Stutte;S. Sumowidagdo;A. Sznajder;M. Talby;P. Tamburello;W. Taylor;P. Telford;J. Temple;E. Thomas;B. Thooris;M. Tomoto;T. Toole;J. Torborg;S. Towers;T. Trefzger;S. Trincaz;B. Tuchming;C. Tully;A. Turcot;P. Tuts;L. Uvarov;S. Uvarov;S. Uzunyan;B. Vachon;R. Kooten;W. V. Leeuwen;N. Varelas;E. Varnes;I. Vasilyev;M. Vaupel;P. Verdier;L. Vertogradov;M. Verzocchi;F. Villeneuve;J. Vlimant;E. Toerne;M. Vreeswijk;T. Anh;H. Wahl;R. Walker;Long Wang;Z. M. Wang;J. Warchoł;M. Warsinsky;G. Watts;M. Wayne;M. Weber;H. Weerts;M. Wegner;N. Wermes;A. White;V. White;D. Whiteson;D. Wicke;D. A. Wijngaarden;G. Wilson;S. Wimpenny;J. Wittlin;M. Wobisch;J. Womersley;D. Wood;T. Wyatt;Q. Xu;N. Xuan;S. Yacoob;R. Yamada;M. Yan;T. Yasuda;Y. Yatsunenko;Y. Yen;K. Yip;S. Youn;J. Yu;A. Yurkewicz;A. Zabi;A. Zatserklyaniy;M. Zdrazil;C. Zeitnitz;D. Zhang;X. Zhang;T. Zhao;Z. Zhao;B. Zhou;J. Zhu;M. Zielinski;D. Zieminska;A. Zieminski;R. Zitoun;V. Zutshi;E. Zverev;A. Zylberstejn
通讯作者: A. Zylberstejn
DOI: 10.1103/physrevlett.124.182001
发表时间: 2020-05-07
影响因子: 8.6
作者:
Andreassen, Anders;Komiske, Patrick T.;Thaler, Jesse
通讯作者: Thaler, Jesse
DOI: 10.1103/physrevd.95.034001
发表时间: 2017-02-01
期刊: PHYSICAL REVIEW D
影响因子: 5
作者:
de Lima, Danilo Ferreira;Petrov, Petar;Spannowsky, Michael
通讯作者: Spannowsky, Michael
利用深度神经网络在 CMS 中识别重口味
DOI: --
发表时间: 2020
期刊:
影响因子: --
作者:
S. L. Staggs;M. L. White;Paul A. Schewe;Erica B Davis;Ebony M. Dill
通讯作者: Ebony M. Dill