On 2 June 2025, the NGT ATLAS & CMS MLOps Workshop took place at CERN’s IdeaSquare, marking the first installment in a new series of workshops & events looking to bring together intra and inter-experiment groups to start discussion on the effective deployment of Machine Learning algorithms into the hardware trigger systems and beyond in anticipation of the HL-LHC.
With 57 people attending, both in person and online, the workshop brought NGT ATLAS and CMS colleagues to address common infrastructure challenges — across software, firmware, and hardware — and to identify solutions that support the effective deployment of ML in trigger systems.
The morning session focused on experiment-specific discussions, with parallel sessions exploring CMS and ATLAS developments in firmware integration, data generation and monitoring, infrastructure for ML deployment, and more. From software metadata management to pythonisation of training data and real-time inference at scale, speakers shed light on the complexity and momentum building behind ML in hardware triggers.
In the afternoon, participants joined a joint open-floor session structured around three core MLOps themes:
- Training: Strategies for robust ML under evolving detector conditions
- Model Logistics: Managing models through software and firmware transitions
- Data Flow: Handling conversions in distributed, heterogeneous computing environments
This workshop underscores NGT’s commitment to fostering cross-experiment collaboration and advancing real-world ML integration into trigger systems. With widespread ML usage expected at the HL-LHC, today’s conversations will directly shape the road ahead.
You can check all the presentations by clicking here.
And here is the reflection of the organizers Christopher Edward Brown, Ioannis Xiotidis, Maciej Mikolaj Glowacki, Rimsky Alejandro Rojas Caballero after the event.