Introduction
This multidisciplinary research project aims to take advantage of modern Machine Learning (ML) architectures for data processing tasks relevant to High-Energy Physics (HEP) experiments. One example is the task of subatomic particle trajectory reconstruction, a.k.a., tracking, for accelerator collision events. Tracking is crucial to understanding subatomic particle behaviour.
The approach is to perform exploration of ML-assisted data processing solution designs. These solutions are expected to deliver higher capability, ultimately replacing currently fielded classical algorithms. The ambition is to deliver solutions that are sufficiently flexible to be detector agnostic.
One of the adopted strategies is the systematic simplification of the problem at hand through Reduced-Order Models (ROMs) and complexity-reduced simulations. In doing so, we have developed the REDuced VIrtual Detector (REDVID), a particle propagation simulator and synthetic data generator for HEP research.
TrackCore-F addresses FPGA deployment for ML models used in ML-assisted particle track reconstruction. The focus is on converting and accelerating Transformer-based model components to reduce inference latency, supporting future low-power and potentially on-site tracking applications.
Dedicated pages for our developed tools:
- REDVID
- REDVID-Gen (future work)
- TrueTrack (under development)
- TrackCore-F
- TrackCore-A (future work)
Resources
For background information, foundational knowledge, and detailed context related to this research, the following resources may be helpful:
- Wikipedia: Particle Physics – General introduction to particle physics and its core concepts.
- ATLAS Inner Detector – A brief overview of the inner detector installed within the ATLAS detector.
- Phase 2 Upgrade of the ATLAS Inner Tracker – A detailed overview of the ATLAS experiment Inner Tracking upgrades intended for High-Luminosity LHC operation.
- A Living Review of Machine Learning for Particle Physics - A categorised collection of references, listing modern machine learning applications, intended for particle physics.
Contact
Initially (2023-2025), this research was being carried out as a multidisciplinary collaboration between Radboud University (RU), the Dutch National Institute for Subatomic Physics (Nikhef) and the University of Twente (UTwente). From 2025 onward, the research is being carried out at UTwente.
We usually offer Master-level Student Projects in relevant topics. In case of enquiries, you can reach us at: aW5mb0B2aXJ0dWFsZGV0ZWN0b3IuY29t