CaltechAUTHORS
Published September 2025 | Version Published
Journal Article Open

Fast jet tagging with MLP-Mixers on FPGAs

Creators

  • 1. ROR icon California Institute of Technology
  • 2. ROR icon Fermilab
  • 3. ROR icon European Organization for Nuclear Research

Abstract

We explore the innovative use of MLP-Mixer models for real-time jet tagging and establish their feasibility on resource-constrained hardware like FPGAs. MLP-Mixers excel in processing sequences of jet constituents, achieving state-of-the-art performance on datasets mimicking Large Hadron Collider conditions. By using advanced optimization techniques such as High-Granularity Quantization and Distributed Arithmetic, we achieve unprecedented efficiency. These models match or surpass the accuracy of previous architectures, reduce hardware resource usage by up to 97%, double the throughput, and half the latency. Additionally, non-permutation-invariant architectures enable smart feature prioritization and efficient FPGA deployment, setting a new benchmark for machine learning in real-time data processing at particle colliders.

Copyright and License

© 2025 The Author(s). Published by IOP Publishing Ltd. Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Acknowledgement

C S acknowledges partially supported by the NSF ACCESS Grant Number PHY240298. C S and M S acknowledge partial support from the U.S. Department of Energy (DOE), Office of Science, Office of High Energy Physics Grant DE-SC0011925. J N, M S, and C S are partially supported by the U S. Department of Energy (DOE), Office of Science, Office of High Energy Physics ‘Designing efficient edge AI with physics phenomena’ Project (DE-FOA-0002705). J N is partially supported by the AI2050 program at Schmidt Futures (Grant G-23-64934).

Data Availability

The data used in this study are openly available at Zenodo at reference [27] under DOI: https://doi.org/10.5281/zenodo.3602260. The software used in this study is also available at Zenodo at reference [39] under DOI: https://doi.org/10.5281/zenodo.16687452.

Files

Sun_2025_Mach._Learn.__Sci._Technol._6_035025.pdf

Files (6.5 MB)

Additional details

Related works

Featured in
Journal Issue: https://iopscience.iop.org/collections/mlst-240321-512 (URL)
Is new version of
Discussion Paper: arXiv:2503.03103 (arXiv)
Is supplemented by
Dataset: 10.5281/zenodo.3602260 (DOI)
Software: 10.5281/zenodo.16687452 (DOI)

Funding

National Science Foundation
PHY240298
United States Department of Energy
DE-SC0011925
United States Department of Energy
DE-FOA-0002705
Schmidt Futures
G-23-64934

Dates

Accepted
2025-07-29
Available
2025-08-07
Published online

Caltech Custom Metadata

Caltech groups
CMS@Caltech, Division of Physics, Mathematics and Astronomy (PMA)
Publication Status
Published