The offer of Ph.D. positions in Physics at the University of Padova has opened just a few days ago, and I wish to advertise it here, giving some background on the matter for those of you who are interested in the call or know somebody who could be.

The Ph.D. in Italy In Italy, Ph. D.s last three years (the duration can be extended but this is not recommended). Courses start with the first academic semester, so the calls typically open in the spring, and admission tests are run in the summer. The system is not too different from that of other countries, but there are peculiarities of which you might want to be aware.

First of all, the courses are funded with a monthly salary for the three years of duration (it is not a large sum; it is roughly equivalent to the starting salary of school teachers). This means that depending on the available funding, there are a fixed number of open positions - variable per University and per course of study. I am not aware of the typical ratio between positions and applications, but I suspect it is of the order of six to one for Physics in Padova. There are this year a total of seventeen positions (2 with no salary, 3 with fixed theme).

Still speaking about money (it is, I reckon, and important parameter in the equation, for a student who must decide where to apply), the research groups that the Ph.D. candidates join usually have funds that allow the students to travel to the research centres, to conferences, and to schools; in addition, the course itself grants some funds to the students for these activities. This adds some slack to the financial aspects. Indeed, I recall that when I was a Ph.D. student (22+ years ago) I traveled four to five times a year to Fermilab, where my experiment (CDF) was located, and as I did I received a substantial per-diem (in fact it is the same that researchers and professors receive). All in all, the two months abroad allowed me to save about a couple thousand extra dollars.

Of course, there is an admission test. It used to be a combination of a written test (exercises in physics plus a composition on some research topic of interest of the candidates) and an oral colloquium with the selection committee, but recently the written test was removed. So you basically only have to convince the committee that you are a bright student and that you have an interesting research plan. And that is where I can suggest something.

When you visit the web page of the Physics Department, the information on the doctoral offer is complemented with data on what are the open lines of research. There are links to the research groups that give an overall idea, so that if you already know that you want to do a Ph.D. in particle physics to search for new phyiscs with CMS data (say), you can verify that there is a very active group of scientists collaborating with the CMS experiment in Padova. On the other hand, there is no specific indication of what could be the topic of your Ph.D. thesis if you passed the selection.

While I do believe strongly that a Ph.D. candidate should have the complete freedom to choose the specific topic of his or her research, and the supervisor should accept the choice and find ways to make it doable, it is in my opinion a good idea to provide prospective students with some indications of what could be appealing research topics which would be supported by seniors in the institution. So that is the purpose of this post...

I can tell you that as far as Physics is concerned Padova is ranked very high among Italian universities, and for a good reason - there are a high number of outstanding researchers, whose production is really excellent overall. I cannot go through a list as I am sure I would forget to mention important elements, but e.g. we do have a great theory group, as well as very active experimental groups both in particle physics, in astroparticle physics, in neutrino physics, and in nuclear physics.

Machine Learning for HEP In addition to all of that, if you are interested in machine learning applied to physics, there I am. I have been a pioneer in these techniques for data analysis in CDF and CMS, and I am still actively working on a few lines of research which might inspire an outstanding Ph.D. thesis. Let me mention two things I have been working on lately, with two Ph.D. students that participate in the INSIGHTS ITN (an European Community-funded network of research institutes for training of young researchers), before I venture to describe a research plan for a possible new Ph.D. student.

One line of research concerns ways to incorporate the effect of systematic uncertainties in the optimization of a data analysis flow. Say you want to discriminate a small signal from backgrounds using a neural network: you can do a great job training your tool, but if the NN does not know (in its loss function) what nuisances affect the final extraction of inference on the parameter of interest (the cross section of the sought for process), you are not going to really optimize anything. An algorithm developed by Pablo de Castro (a former Ph.D student of mine), INFERNO (see the relative 2018 Computer Physics Communications article here), is now in the process of being applied to CMS searches by one of the INSIGHTS ESR working in Padova. As this is a very new and promising research area in machine learning, I am very excited by this plan.

With another student there is an ongoing effort to search for a rare decay of the neutral B meson, using a neural network to operate a regression of the reconstructed mass of the hadron, thereby improving its observability. Again, we are going to use cutting-edge tools for this task.

Two specific ideas for groundbreaking theses And now, about the two research directions that would benefit from the work of a bright new graduate student: the first is the development of unsupervised learning tools for new physics searches in CMS data. As the future of machine learning is unsupervised -something recognized by the biggest experts in the field- there is a lot of interest in this direction. I have developed an anomaly detection algorithm, RanBox, which is however still in its R&D phase. Currently a undergraduate student in Statistics is working on it, porting the code in python (I wrote the original algorithm in C++). The abstract of a research on this topic for a Ph.D student could be the following:

The CMS experiment has collected over 150 inverse femtobarns of proton-proton collisions from Run 2 of the Large Hadron Collider. While these data have been used for a variety of new physics searches inspired by theoretical models, it is important to search also for unknown new processes without a theoretical bias. The interest of unsupervised learning techniques to help us look for anomalies in the feature space of the observed events has therefore risen in the recent past. The proposed research plan includes the development of a novel anomaly detection tool suitable for LHC searches, and its application to Run 2 CMS data.

A second area of research involves the optimization of detector design using machine learning tools. This is a very ambitious and original project, as detectors for particle physics applications are multi-purpose instruments, and the construction of an optimization measure is highly non-trivial; not to mention the fact that the intrinsic stochasticity of subnuclear phenomena are ill-suited for differential programming tools which are the key players for end-to-end optimization strategies. Yet we cannot be blind to the fact that in the future, particle reconstruction will be left in the hands of artificial intelligence. For an AI tool, the redundancy we build in our instruments may not be as important as other factors.

It then transpires that we cannot build detectors the way we have been for the past 50 years, thinking that we are making the correct construction choices because these have served us well in the past. A true Pareto optimality can and should be investigated for large future projects in a disciplined way. A thesis looking for some simple applications of this concept would be groundbreaking in my opinion.

By the way, I have recently published a study of the geometry optimization of a simple detector that will study muon-electron scatterings to reduce the uncertainty of the anomalous magnetic moment of the muon; that study was not done using machine learning tools, but in my mind it is a first step in the direction I would like to take.

So that's it - if you are a bright student who wishes to make a dent with a Ph.D. in Physics, and you want to come and work with me on one of these topics, drop me a line! My e-mail is (first name dot last name at gmail dot com).