Title: Punzi-loss: a non-differentiable metric approximation for sensitivity optimisation in the search for new particles

Journal Article · · European Physical Journal. C, Particles and Fields
 [1];  [2];  [3];  [4];  [5];  [6];  [5];  [6];  [7];  [8];  [9]; ORCiD logo [2];  [10];  [10];  [10];  [10];  [7];  [2];  [11];  [12] more »;  [2];  [10];  [10];  [6];  [11];  [3];  [13];  [3];  [2];  [6];  [14];  [10];  [11];  [15];  [9];  [10];  [3];  [15];  [2];  [9];  [16];  [17];  [11];  [18];  [19];  [20] « less
  1. Istituto Nazionale di Fisica Nucleare (INFN), Trieste (Italy); OSTI
  2. Institute of High Energy Physics, Vienna (Austria)
  3. Ludwig Maximilian University of Munich, Munich (Germany)
  4. University degli Studi di Napoli Federico II (Italy); Istituto Nazionale di Fisica Nucleare (INFN), Naples (Italy)
  5. University of Pisa (Italy); Istituto Nazionale di Fisica Nucleare (INFN), Pisa (Italy)
  6. Deutsches Elektronen-Synchrotron, Hamburg (Germany)
  7. Istituto Nazionale di Fisica Nucleare (INFN), Rome (Italy)
  8. Tel Aviv University, Tel Aviv (Israel)
  9. University of Bonn (Germany)
  10. Karlsruhe Institute of Technology (KIT) (Germany)
  11. University of Melbourne, VIC (Australia)
  12. Max-Planck-Institut für Physik, Munich (Germany)
  13. Istituto Nazionale di Fisica Nucleare (INFN), Padova (Italy)
  14. Duke University, Durham, NC (United States)
  15. High Energy Accelerator Research Organization (KEK), Tsukuba (Japan)
  16. Istituto Nazionale di Fisica Nucleare (INFN), Turin (Italy)
  17. Istituto Nazionale di Fisica Nucleare (INFN), Pisa (Italy)
  18. Aix-Marseille University, Marseille (France)
  19. Charles University, Prague (Czech Republic)
  20. Jozef Stefan Institute (IJS), Ljubljana (Slovenia)
+ Show Author Affiliations

We present the novel implementation of a non-differentiable metric approximation and a corresponding loss-scheduling aimed at the search for new particles of unknown mass in high energy physics experiments. We call the loss-scheduling, based on the minimisation of a figure-of-merit related function typical of particle physics, a Punzi-loss function, and the neural network that utilises this loss function a Punzi-net. We show that the Punzi-net outperforms standard multivariate analysis techniques and generalises well to mass hypotheses for which it was not trained. This is achieved by training a single classifier that provides a coherent and optimal classification of all signal hypotheses over the whole search space. Our result constitutes a complementary approach to fully differentiable analyses in particle physics. We implemented this work using PyTorch and provide users full access to a public repository containing all the codes and a training example.

Research Organization:
Duke Univeristy, Durham, NC (United States)
Sponsoring Organization:
European Research Council; Helmholtz Association Initiative; USDOE
OSTI ID:
1903813
Journal Information:
European Physical Journal. C, Particles and Fields, Journal Name: European Physical Journal. C, Particles and Fields Journal Issue: 2 Vol. 82; ISSN 1434-6044
Publisher:
SpringerCopyright Statement
Country of Publication:
United States
Language:
English

References (12)

The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations journal July 2014
FastBDT: A Speed-Optimized Multivariate Classification Algorithm for the Belle II Experiment journal September 2017
The Belle II Core Software: Belle II Framework Software Group journal November 2018
Optimal Statistical Inference in the Presence of Systematic Uncertainties Using Neural Network Optimization Based on Binned Poisson Likelihoods with Nuisance Parameters journal January 2021
Geant4—a simulation toolkit
  • Agostinelli, S.; Allison, J.; Amako, K.
  • Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, Vol. 506, Issue 3 https://doi.org/10.1016/S0168-9002(03)01368-8
journal July 2003
INFERNO: Inference-Aware Neural Optimisation journal November 2019
Toward Machine Learning Optimization of Experimental Design journal January 2021
Accelerator design at SuperKEKB journal January 2013
The Belle II Physics Book journal December 2019
Search for an Invisibly Decaying Z ′ Boson at Belle II in e + e − → μ + μ − ( e ± μ ∓ ) Plus Missing Energy Final States journal April 2020
Loss Scheduling for Class-Imbalanced Image Segmentation Problems conference December 2020
Search for an invisibly decaying Z′ boson and study of particle identification at the Belle II experiment text January 2021

Similar Records

Related Subjects

71 CLASSICAL AND QUANTUM MECHANICS
GENERAL PHYSICS