CaltechAUTHORS
  A Caltech Library Service

Autoencoders on field-programmable gate arrays for real-time, unsupervised new physics detection at 40 MHz at the Large Hadron Collider

Govorkova, Ekaterina and Puljak, Ema and Aarrestad, Thea and James, Thomas and Loncar, Vladimir and Pierini, Maurizio and Pol, Adrian Alan and Ghielmetti, Nicolò and Graczyk, Maksymilian and Summers, Sioni and Ngadiuba, Jennifer and Nguyen, Thong Q. and Duarte, Javier and Wu, Zhenbin (2022) Autoencoders on field-programmable gate arrays for real-time, unsupervised new physics detection at 40 MHz at the Large Hadron Collider. Nature Machine Intelligence, 4 (2). pp. 154-161. ISSN 2522-5839. doi:10.1038/s42256-022-00441-3. https://resolver.caltech.edu/CaltechAUTHORS:20220324-730041000

[img] PDF - Submitted Version
See Usage Policy.

4MB
[img] Image (JPEG) (Extended Data Fig. 1: Network architectures) - Supplemental Material
See Usage Policy.

410kB
[img] Image (JPEG) (Extended Data Fig. 2: TPR ratios for different bit width) - Supplemental Material
See Usage Policy.

131kB

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20220324-730041000

Abstract

To study the physics of fundamental particles and their interactions, the Large Hadron Collider was constructed at CERN, where protons collide to create new particles measured by detectors. Collisions occur at a frequency of 40 MHz, and with an event size of roughly 1 MB it is impossible to read out and store the generated amount of data from the detector and therefore a multi-tiered, real-time filtering system is required. In this paper, we show how to adapt and deploy deep-learning-based autoencoders for the unsupervised detection of new physics signatures in the challenging environment of a real-time event selection system at the Large Hadron Collider. The first-stage filter, implemented on custom electronics, decides within a few microseconds whether an event should be kept or discarded. At this stage, the rate is reduced from 40 MHz to about 100 kHz. We demonstrate the deployment of an unsupervised selection algorithm on this custom electronics, running in as little as 80 ns and enhancing the signal-over-background ratio by three orders of magnitude. This work enables the practical deployment of these networks during the next data-taking campaign of the Large Hadron Collider.


Item Type:Article
Related URLs:
URLURL TypeDescription
https://doi.org/10.1038/s42256-022-00441-3DOIArticle
https://rdcu.be/cJKZQPublisherFree ReadCube access
https://doi.org/10.5281/zenodo.5055454DOIData
https://doi.org/10.5281/zenodo.5046446DOIData
https://doi.org/10.5281/zenodo.5061633DOIData
https://doi.org/10.5281/zenodo.5061688DOIData
https://doi.org/10.5281/zenodo.5046389DOIData
http://github.com/google/qkerasRelated ItemQKeras library
https://github.com/fastmachinelearning/hls4ml/tree/AE_L1_paperRelated Itemhls4ml library
https://arxiv.org/abs/2108.03986arXivDiscussion Paper
ORCID:
AuthorORCID
Govorkova, Ekaterina0000-0003-1920-6618
Puljak, Ema0000-0002-6011-9965
Aarrestad, Thea0000-0002-7671-243X
Loncar, Vladimir0000-0003-3651-0232
Pierini, Maurizio0000-0003-1939-4268
Pol, Adrian Alan0000-0002-9034-0230
Summers, Sioni0000-0003-4244-2061
Ngadiuba, Jennifer0000-0002-0055-2935
Nguyen, Thong Q.0000-0003-3954-5131
Duarte, Javier0000-0002-5076-7096
Alternate Title:Autoencoders on FPGAs for real-time, unsupervised new physics detection at 40 MHz at the Large Hadron Collider
Additional Information:© 2022 Nature Publishing Group. Received 12 August 2021; Accepted 06 January 2022; Published 23 February 2022. This work is supported by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 772369) and the ERC-POC programme (grant no. 996696). Data availability: The data used in this study are openly available at Zenodo57,58,59,60,62. Code availability: The QKeras library is available at github.com/google/qkeras, where the work presented here is using QKeras version 0.9.0. The hls4ml library with custom layers used in the paper are under AE_L1_paper branch and available at https://github.com/fastmachinelearning/hls4ml/tree/AE_L1_paper. Contributions: V.L., M.P., A.A.P., N.G., M.G., S.S., J.D. and Z.W. conceived and designed the hls4ml software library. M.P., T.Q.N. and Z.W. designed and prepared the dataset format. E.G., E.P., T.A., T.J., V.L., M.P., J.N., T.Q.N. and Z.W. designed and implemented autoencoders in hls4ml. E.G., E.P., T.A., T.J., M.P. and J.D. wrote the paper. The authors declare no competing interests. Peer review information: Nature Machine Intelligence thanks the anonymous reviewers for their contribution to the peer review of this work.
Funders:
Funding AgencyGrant Number
European Research Council (ERC)772369
European Research Council (ERC)996696
Subject Keywords:Computer science; Experimental particle physics; Scientific data
Issue or Number:2
DOI:10.1038/s42256-022-00441-3
Record Number:CaltechAUTHORS:20220324-730041000
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20220324-730041000
Official Citation:Govorkova, E., Puljak, E., Aarrestad, T. et al. Autoencoders on field-programmable gate arrays for real-time, unsupervised new physics detection at 40 MHz at the Large Hadron Collider. Nat Mach Intell 4, 154–161 (2022). https://doi.org/10.1038/s42256-022-00441-3
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:114066
Collection:CaltechAUTHORS
Deposited By: Tony Diaz
Deposited On:24 Mar 2022 21:57
Last Modified:24 Mar 2022 21:57

Repository Staff Only: item control page