Towards Foundation Models on Hardware Accelerators for Particle Physics
Abstract
Bandwidth constraints require many particle physics experiments to make real-time decisions on custom hardware or firmware running simplified algorithms, unlike offline analysis, where latency is usually not a limiting factor. For example, for particle jet tagging at colliders, state-of-the-art performance is achieved by foundation models with hundreds of millions of parameters pre-trained with billions of jets. We use knowledge distillation to transfer what such models have learned into efficient networks towards deployment in hardware accelerators. The teacher is the OmniLearned foundation model fine-tuned on top quark jet tagging; the student is an attention-free Deep Sets network. We demonstrate three ways the student's performance improves: adding a message-passing layer to the Deep Sets architecture, training on the teacher's soft labels rather than on ground-truth labels alone, and distilling from a pretrained teacher rather than from the same architecture trained from scratch. In each case the gain is largest in the background rejection at low signal efficiency, the regime that is most relevant for a trigger.
Related
- Leveraging Industrial Foundation Models at the Edge of Particle Physics Detectors via Distillation Learning and Hardware Co-design
- SensOFormer: Foundation Models for Particle Accelerators
- Track finding and particle identification performance with particle physics foundation models
- Panda Diplomacy: Foundation Model Pre-training across Particle Imaging Detectors for High Energy and Nuclear Physics
- Machine-learned particle flow as a foundation model for collider physics
BibTeX
@misc{benyas2026towards,
title = {{Towards Foundation Models on Hardware Accelerators for Particle Physics}},
author = {Maya Benyas and Julia Gonski and Qibin Liu and P. Alex May and Vinicius Mikuni and Benjamin Nachman and Tanvi Wamorkar and Liangyu Wu},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.26899},
eprint = {2609.26899},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.26899},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Towards Foundation Models on Hardware Accelerators for Particle Physics
AU - Maya Benyas
AU - Julia Gonski
AU - Qibin Liu
AU - P. Alex May
AU - Vinicius Mikuni
AU - Benjamin Nachman
AU - Tanvi Wamorkar
AU - Liangyu Wu
PY - 2026
DA - 2026/09/22
PB - arXiv
DO - 10.48550/arxiv.2609.26899
UR - https://arxiv.org/abs/2609.26899
AB - Bandwidth constraints require many particle physics experiments to make real-time decisions on custom hardware or firmware running simplified algorithms, unlike offline analysis, where latency is usually not a limiting factor. For example, for particle jet tagging at colliders, state-of-the-art performance is achieved by foundation models with hundreds of millions of parameters pre-trained with billions of jets. We use knowledge distillation to transfer what such models have learned into efficient networks towards deployment in hardware accelerators. The teacher is the OmniLearned foundation model fine-tuned on top quark jet tagging; the student is an attention-free Deep Sets network. We demonstrate three ways the student's performance improves: adding a message-passing layer to the Deep Sets architecture, training on the teacher's soft labels rather than on ground-truth labels alone, and distilling from a pretrained teacher rather than from the same architecture trained from scratch. In each case the gain is largest in the background rejection at low signal efficiency, the regime that is most relevant for a trigger.
ER -