by Giles Strong

So, it’s been a while since my last post, apologies for that, but the summer has been both busy and eventful, so let me summarise what’s been happening.

LIP Summer Internships

Over the summer, LIP, my research institute, hosted about 30 masters students from local universities. For about two months, the students worked on a variety of internship projects organised by the experimental groups. The culmination being a day long series of presentations by the students.

My supervisor and I organised a project for two of these students, Antonio and Gançalo, who worked on a search for di-Higgs production via the decay of a hypothetical heavy Higgs. This is a similar search to what I work on, except that instead of the di-Higgs mass being a distribution, it is now a resonance located at some unknown value. The search still suffers from a very low production cross-section, meaning that any visible signal is easily swamped by other background processes; a perfect scenario for a machine learning classifier.

The two students worked well together and implemented a complex series of neural networks to optimise the analysis. The last step being to perform a hypothesis test to determine the expected performance on the classifier is applied to actual collider data, and the results were very promising! Their final presentation may be found here, and the full schedule here.

Visiting ESRs

During September, Alessia, the network ESR based at Université Catholique de Louvain, came to work on secondment at LIP. Alessia is one of the authors of the MoMEMta package, which provides a highly customisable way of implementing the Matrix Element Method. Unlike the neural network classifiers I normally work with, MEM uses analytical knowledge of particle theory to compute weights for events under certain hypotheses. The difficulty being the fact that we reconstruct particles in the event via a detector which is not completely accurate, so one has to account for the difference the reconstructed particle and the particles the MEM expects to receive.

It took a bit of work, but eventually we were able to get the method up and running for the hh→bb𝜏𝜏 search. Unfortunately, the discrimination performance on its own was not as good as I was able to get with a NN classifier. However, the raw weights computed by the tool proved to be useful input features for a new classifier, which was able to outperform the old one.

Towards the end of September, Pablo, the ESR based at INFN Padova, who had been attending a physics school in nearby Evora, joined us for a few days to get up to speed on what we’d done, and to help clarify the direction of and upcoming publication we’re working on.

The Remote Operations Centre

The CMS experiment at CERN uses a control room containing many displays to monitor the operation of the detector. These diagnostic displays are available online (some even public) and so similar, remote rooms can be used to show the running of CMS and the LHC. LIP recently finished installing one such room at one of the local universities (IST, Lisbon).

DSCN1919

The main purpose of the room for CMS is scientific outreach, and students are invited to come and find out about the experiment and high-energy physics, and get an insight into just how vast and complicated the LHC machinery is. So for one day a week I now work in a different office (which has air conditioning!). The room is also used by another experiment called Auger, to monitor their detector based in Argentina.

Preparation for the next deliverable

Deadlines in the network are pretty tight, and for me the next one is coming up at the end of this month. For the previous deliverable last February I had performed initial tests to see whether or not neural networks were applicable to my analysis. The next step was then to improve and refine them. So far this was all being done on some very roughly simulated data, and for this deliverable I need to move to a more accurate simulation.

This took some time, since a lot of new data needed to be produced, and the regressors and classifiers reimplemented. The use of the MEM was also a requirement for the deliverable, so Alessia’s visit was well timed. Luckily, though, the work seems to be coming together and all results have been obtained. All that remains is to finish writing them up and then go through a few drafts.

It’s also good to have obtained all the results, since tomorrow I’m off to Louvain in Belgium for the mid-term review of the network. There we’ll be meeting with our project officer from the European Commission, presenting updates of our work, and discussing how to best proceed with the various directions of the network during the second half. Hopefully there’ll also be time to sample some of the local beers and chocolates as well.