Display options
Cite
Iiyama Y, Cerminara G, Gupta A, et al. Distance-Weighted Graph Neural Networks on FPGAs for Real-Time Particle Reconstruction in High Energy Physics. Front Big Data. 2021;3:598927doi: 10.3389/fdata.2020.598927.
Iiyama, Y., Cerminara, G., Gupta, A., Kieseler, J., Loncar, V., Pierini, M., Qasim, S. R., Rieger, M., Summers, S., Van Onsem, G., Wozniak, K. A., Ngadiuba, J., Di Guglielmo, G., Duarte, J., Harris, P., Rankin, D., Jindariani, S., Liu, M., Pedro, K., Tran, N., Kreinar, E., & Wu, Z. (2020). Distance-Weighted Graph Neural Networks on FPGAs for Real-Time Particle Reconstruction in High Energy Physics. Frontiers in big data, 3598927. https://doi.org/10.3389/fdata.2020.598927
Iiyama, Yutaro, et al. "Distance-Weighted Graph Neural Networks on FPGAs for Real-Time Particle Reconstruction in High Energy Physics." Frontiers in big data vol. 3 (2020): 598927. doi: https://doi.org/10.3389/fdata.2020.598927
Iiyama Y, Cerminara G, Gupta A, Kieseler J, Loncar V, Pierini M, Qasim SR, Rieger M, Summers S, Van Onsem G, Wozniak KA, Ngadiuba J, Di Guglielmo G, Duarte J, Harris P, Rankin D, Jindariani S, Liu M, Pedro K, Tran N, Kreinar E, Wu Z. Distance-Weighted Graph Neural Networks on FPGAs for Real-Time Particle Reconstruction in High Energy Physics. Front Big Data. 2021 Jan 12;3:598927. doi: 10.3389/fdata.2020.598927. eCollection 2020. PMID: 33791596; PMCID: PMC8006281.
Copy Download .nbib Format: NLM
Share it on

Front Big Data. 2021 Jan 12;3:598927. doi: 10.3389/fdata.2020.598927. eCollection 2020.

Distance-Weighted Graph Neural Networks on FPGAs for Real-Time Particle Reconstruction in High Energy Physics.

Frontiers in big data

Yutaro Iiyama, Gianluca Cerminara, Abhijay Gupta, Jan Kieseler, Vladimir Loncar, Maurizio Pierini, Shah Rukh Qasim, Marcel Rieger, Sioni Summers, Gerrit Van Onsem, Kinga Anna Wozniak, Jennifer Ngadiuba, Giuseppe Di Guglielmo, Javier Duarte, Philip Harris, Dylan Rankin, Sergo Jindariani, Mia Liu, Kevin Pedro, Nhan Tran, Edward Kreinar, Zhenbin Wu

Affiliations

  1. International Center for Elementary Particle Physics, University of Tokyo, Tokyo, Japan.
  2. Experimental Physics Department, European Organization for Nuclear Research (CERN), Geneva, Switzerland.
  3. Institute of Physics Belgrade, Belgrade, Serbia.
  4. Manchester Metropolitan University, Manchester, United Kingdom.
  5. University of Vienna, Vienna, Austria.
  6. Department of Physics, Math and Astronomy, California Institute of Technology, Pasadena, CA, United States.
  7. Department of Computer Science, Columbia University, New York, NY, United States.
  8. Department of Physics, University of California, San Diego, San Diego, CA, United States.
  9. Laboratory for Nuclear Science, Massachusetts Institute of Technology, Cambridge, MA, United States.
  10. Department of Physics and Astronomy, Purdue university, West Lafayette, IL, United States.
  11. Department of Electrical and Computer Engineering, Northwestern University, Evanston, IL, United States.
  12. HawkEye360, Herndon, VA, United States.
  13. Department of Physics, University of Illinois at Chicago, Chicago, IL, United States.

PMID: 33791596 PMCID: PMC8006281 DOI: 10.3389/fdata.2020.598927

Abstract

Graph neural networks have been shown to achieve excellent performance for several crucial tasks in particle physics, such as charged particle tracking, jet tagging, and clustering. An important domain for the application of these networks is the FGPA-based first layer of real-time data filtering at the CERN Large Hadron Collider, which has strict latency and resource constraints. We discuss how to design distance-weighted graph networks that can be executed with a latency of less than one μs on an FPGA. To do so, we consider a representative task associated to particle reconstruction and identification in a next-generation calorimeter operating at a particle collider. We use a graph network architecture developed for such purposes, and apply additional simplifications to match the computing constraints of Level-1 trigger systems, including weight quantization. Using the hls4ml library, we convert the compressed models into firmware to be implemented on an FPGA. Performance of the synthesized models is presented both in terms of inference accuracy and resource usage.

Copyright © 2021 Iiyama, Cerminara, Gupta, Kieseler, Loncar, Pierini, Qasim, Rieger, Summers, Van Onsem, Wozniak, Ngadiuba, Di Guglielmo, Duarte, Harris, Rankin, Jindariani, Liu, Pedro, Tran, Kreinar and Wu.

Keywords: deep learning; fast inference; field-programmable gate arrays; graph network; imaging calorimeter

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

References

  1. Eur Phys J C Part Fields. 2017;77(7):466 - PubMed
  2. Nature. 2020 Sep;585(7825):357-362 - PubMed

Publication Types