As the twenty-three regular readers of this blog know [(c) A.Manzoni], lately I have shifted the main focus of my research to advanced applications of deep learning to basic science. This does not mean that I do not continue to participate in the CMS experiment at the Large Hadron Collider at CERN – this remains the main focus of my research; but that means what remains of my brain functionality is mostly invested in thinking about the future applications of today’s and tomorrow’s computing innovations.
In a sense, the type of CMS work I am currently involved in (co-chair of the thesis allocation committee; member of the CMS statistics committee; supervisor of a doctoral thesis and member of an analysis group) n is not as demanding in terms of brain processing. Or, to be more precise, let’s say that it doesn’t force me to “think outside the box” too much, whereas this other job does.
The INFN, the institute I work for, gives me research freedom so I don’t have to ask permission to use my time one way or the other. However, the INFN requires me to declare, once a year, what fraction of my time is spent on each experiment in which I participate. These numbers are then used for general statistics as well as to provide funding for travel to conferences and to do collaborations. experimentation; they are also used to decide if I have copyright in these experiences (you have to pass a certain threshold to earn this right).
It’s a bit of a shame that the MODE collaboration, of which I am the scientific coordinator, is not yet an entity recognized by the INFN (largely my fault, but it’s very hard to be put on the map when your activity does not specifically target one of the five research areas that constitute the subject of INFN research, but rather all of them at the same time). Thus, MODE currently relies only on the goodwill of its members, as it has no funding to operate as a group. In other words, with the exception of a few sponsors. The first is IRIS-HEP, an American organization funded by the NSF (“Institute for Research and Innovation in Software for High-Energy Physics”). The other is JENAS, an Appec – NuPecc – ECFA consortium which supports a few initiatives of interest to the whole particle physics / nuclear physics and neutrino physics / astrophysics community.
I take the paragraph above to mean that if you have extra money, you might consider funding a Ph.D. student to work with us for the future of basic science! If not, well, there are other ways to support our work. In fact, we recently announced the second edition of a workshop on Differentiable Programming for Experimental Design, the focus of MODE research.
The workshop follows the first edition, which took place in Louvain-la-Neuve last September and was very well attended, with 30 participants on site and more than 70 participants online. It is funded by a large contributuon from the sponsors mentioned above and will take place in the beautiful city of Kolymbari, on the northwest coast of Crete, from September 12th to 16th. So if you are interested or working on learning applications to basic science research, you will certainly be happy to hear about the latest developments at our workshop. Even better, you can submit an abstract and come and give a talk… The Covid-19 is not over yet, but on-site participation has indeed resumed.
Above: the fantastic beach of Balos, 30 km west of the workshop site, accessible by car or boat.
If you decide to come, I strongly advise you to take a few extra days before the conference, to enjoy a visit to the wonderful beaches of western Crete: Falasarna, Balos, Elafonissi. The region can be reached more comfortably by landing in Chania (served by Aegean but also by several other airlines), or in Heraklion, or by boat to the port of Chania. Believe me, you won’t be disappointed…and the level of food quality you’ll get from restaurants like this will not be forgotten.
As for the scientific program: we will have fully plenary sessions devoted to recent advances in applications to HEP, astro-HEP, neutrino physics, nuclear physics, as well as some invited talks and a data challenge. Although the workshop focuses on the true cutting edge of differentiable programming applications, such as the one below, you should expect general coverage as well of more “standard” uses of deep learning, such as classification of images for jet marking, track reconstruction, deep regression, anomaly detection, etc.
Well, I hope you’ll join us! The regular rate for 4 days includes accommodation, three meals and use of all facilities, and is at a very low 580 eur, so you won’t break the bank…
As for the “cutting edge” apps MODE focuses on: take the gif below as an example. In this case, the software is a differentiable pipeline that models the problem of optimizing a detector to find out if someone has hidden a block of uranium inside a cubic meter of scrap metal. – an interesting use case for border control and smelter security. To this end, cosmic rays striking the volume are tracked by detection panels above and below, and inference about material within the volume is made based on the amount of scatter the rays experience when they cross it.
The loss function of the problem includes the efficiency of locating volumes that hide the uranium within, as well as a penalty in the cost of detector panels. The system “learns” which configuration of sensing elements is most effective, minimizing loss. Of course, the above is only a very quick and rough explanation, but it is enough to give an idea of the complexity of the task – the software has to perform several steps to perform the task:
1) generate a cosmic ray stream with a simulator
2) to simulate the trajectories of muons in the material and their interaction with the detection elements (hits)
3) carry out a reconstruction of the trajectories using the detected impacts
4) deduce a map of the density of matter in the unknown volume
5) construct a test statistic to discriminate volumes containing hidden uranium from normal volumes
6) evaluate the performance of discrimination, and calculate a loss function
7) change the configuration of the detection elements according to the gradient of the loss function
8) iterate until convergence.
Here is the animated GIF: at the top you can see the loss function decreasing as the system “learns” the best configuration; below you can see the detector panels above and below the unknown volume (four above and four below) as they move, viewed from the side and from above. The problem considered here is very “symmetric” (the blocks of uranium are randomly simulated in the volume) so one could imagine that there is not much to discover in terms of optimal arrangement, but the system always includes non-trivial details about the most efficient layout. Explaining your choices is then easy by looking at the loss profile and working it back to the inference performance.
Although this is only a “demonstration” application of differentiable programming to detector design, it contains all the ingredients that can be deployed to tackle more difficult and scientifically interesting problems. But muon tomography itself is a growing field of applied science, so we plan to release this software and share it with a large community![The gif has been produced by G.C. Strong who is leading the development effort for this software package].
Tommaso Dorigo (see his personal webpage here) is an experimental particle physicist who works for INFN and the University of Padua, and collaborates with the CMS experiment at CERN LHC. He coordinates the MODE collaboration, a group of physicists and computer scientists from eight institutions in Europe and the United States that aim to enable end-to-end optimization of detector design with differentiable programming. Dorigo is editor of Reviews in Physics and Physics Open. In 2016 Dorigo published the book “Anomaly! Collider Physics and the Quest for New Phenomena at Fermilab”, an insider’s view on the sociology of large particle physics experiments. You can get a copy of the book on Amazon, or contact them for a free pdf copy if you have limited financial means.