Speaker
Description
The CMS Phase-2 upgrade integrates machine learning (ML) throughout the Level-1 Trigger, marking a transition toward differentiable detector systems in which multiple ML models collaboratively reconstruct physics objects in real time. As detector conditions and physics goals evolve, these models require continuous retraining, validation, and deployment, transforming trigger algorithms into dynamic, data-driven systems.
This shift introduces major challenges in reproducibility, versioning, and deployment. Trigger algorithms must remain consistent across software models, intermediate representations, firmware implementations, and detector emulators, while preserving datasets, preprocessing steps, training environments, compiler settings, and firmware-generation workflows. The integration of trained ML models into deployable firmware further increases system complexity.
TriggerFlow addresses these challenges with a standardized framework for end-to-end ML development and deployment within the CMS Level-1 Trigger. Its core TriggerModel abstraction unifies software, firmware, and emulation artifacts within a single versioned object. TriggerFlow also provides reproducible pipelines for data preparation, training, validation, compilation, and deployment across local and cloud environments. By treating a trigger algorithm as a complete, versioned workflow rather than an isolated model or firmware block, TriggerFlow provides a scalable foundation for the next generation of ML-driven trigger systems in CMS.