PPP 06.03.2025

• Mu2e: neutrinoless mu->e conversion.

  • Aim to be background-free in ~105MeV region.
  • Relies on art framework, so bound by the version of ROOT packaged by art.
  • Interest also in using the event display.

• ANNs training for track quality determination and particle ID (differentiate electrons from muons), identify hit in tracker is likely to be from background or signal.

  • Use SOFIE for inference

Q&A

• Vincenzo: RDataFrame should work out of the box with nested vectors, but what is your EDM?

  • Andy: std::vector<std::vector\<mystruct>>

• Stephan: are they same length? For the momenta for example, momenta of segments, always 3 segments now?

  • Andy: these can now be different due to how in the reconstruction we extrapolate the stopping target and the target might get through different numbers of target disks and we store these different numbers.

• Jonas Rembser: The RooUtil looks clean and usable, do you have a similar example with PyUtils? We can also understand what kind of interfaces are expected also for RDataFrame.

  • Sophie: we use awkward arrays in PyUtils, but we try to make function names similar

• Jonas Rembser: do you know what users prefer?

  • Sophie: kind of 50/50 according to a survey run roughly one year ago.

• Vincenzo: you mention SOFIE. Is this run per event on CPU?

  • Andy: run once per track, on CPU. We have object we can write the event ntuple afterwards.
• Vincenzo: run workflow as part of writing the evnt ntuple?
  • Andy: yes, but also we have a loop. Read data from event ntuple, do the training.

• Vincenzo: if you're also interested in re-training, there is a recent development in ROOT of a tool to generate batches of events from ROOT data so they can be read directly by ML models (e.g. TensorFlow, PyTorch).

  • Andy: sounds interesting, let us know.

• Vincenzo: ROOT trainings?

  • Sophie: yes, we're interested.

• Danilo: Any feedback on REve?

  • Sophie: We've been using it for a while, now we use it as a finished product, for us it's working quite well.

• Danilo: Any suggestion regarding ROOT in general?

  • Andy: No further suggestions for now.