by Markus Stoye

This week the first Inter-experimental LHC Machine Learning IML workshop took place at CERN. I showed my results on using deep learning for hadronic particle labeling (flavour tagging), a method that offers significant improvements in the labeling of heavy flavour jets for the CMS experiment (which I am member of). Despite deep learning as a topic is all over the media, the big CERN experiments have not used it a lot this far. In fact my application is, to my knowledge, the very first deep-learning application in CMS reconstruction.

The workshop featured several presentations on deep learning using simulated data. However, still ahead of those promising studies in simulation is the application of the tools to real data. Simulation is idealized and simplifies the real data sometimes by quite a bit, thus approaches that seemed promising in simulation did not always show the same gain in real data.

What can be tricky as well is the task to convince collaborations with thousands of members to switch from established and working methods to something new. This is especially true as people running their established methods have often no interest in switching quickly to other methods. Showing in a simulation study that a new technique might help is thus only the first step, convincing a collaboration to use this and test it in real data must be the next. Having my result out, using real data and officially approved by the CMS collaborations for reconstruction, hopefully also other deep-learners to succeed getting their collaborations to move more towards this direction.

What I especially enjoyed about the workshop was that many participants in the audience and presenters were young and enthusiastic. There were plenty of good ideas and very lively discussions over coffee, and it all gave a sense that there is still a lot of uncharted ground to explore and exciting work ahead. The vibrant positive atmosphere was somewhat different to the one of pure high energy physics workshops I attended to last year. As with the LHC we made no new big discoveries since the Higgs boson, and since no funding agency seems anxious to spend several billion Euros in a more powerful accelerator, there is a sense that the pace of progress in high-energy frontier is slowing down, rather than accelerating like in the field of machine learning.


(Markus Stoye is a CERN researcher and one of the supervisors in the CERN node of AMVA4NewPhysics. He will soon be added to the “Authors” page of the blog – NDR).