Autoencoders on field-programmable gate arrays for real-time, unsupervised new physics detection at 40 MHz at the Large Hadron Collider
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.
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.Attached Files
Submitted - 2108.03986.pdf
Supplemental Material - 42256_2022_441_Fig4_ESM.webp
Supplemental Material - 42256_2022_441_Fig5_ESM.webp
Files
Name | Size | Download all |
---|---|---|
md5:d259664bd29935c6aff31e767c383db9
|
131.5 kB | Download |
md5:38213d784933364c3395b426fc0b3aa2
|
4.3 MB | Preview Download |
md5:cf9d6b6bec3b5f5564eb9544f6550b7c
|
410.8 kB | Download |
Additional details
- Alternative title
- Autoencoders on FPGAs for real-time, unsupervised new physics detection at 40 MHz at the Large Hadron Collider
- Eprint ID
- 114066
- Resolver ID
- CaltechAUTHORS:20220324-730041000
- 772369
- European Research Council (ERC)
- 996696
- European Research Council (ERC)
- Created
-
2022-03-24Created from EPrint's datestamp field
- Updated
-
2022-03-24Created from EPrint's last_modified field