A Quantum Diaries Survivor

Yes, this is supposedly a particle physics blog, not a machine learning one - and yet, I have been finding myself blogging a lot more about machine learning than particle physics as of late. Why is that? Well, of course the topic of algorithms that may dramatically improve our statistical inference from collider data is of course dear to my heart, and has been so since at least two decades (my first invention, the "inverse bagging" algorithm, is dated 1992, when nobody even knew what bagging was). But the more incidental reason is that now _everybody_ is interested in the topic, and that means all of my particle physics and astroparticle physics colleagues. 

For the tenth anniversary of this blog being hosted by Science 2.0, which is coming in a few days, I decided to reinstall the habit I once had of weekly picking and commenting on a result from high-energy physics research, a series I called "The Plot Of The Week". These days I am busier than I used to be when this blog started being published here, so I am not sure I will be able to keep a weekly pace for this series; on the other hand I want to make an attempt, and the first step in that direction is this article.

Last Monday and Tuesday I gave a few lectures on Machine Learning at a Data Science school (IDPASC) in Braga, Portugal. I think that this topic has received so much attention in the last few years, with heaps of excellent resources now freely available online, that it is very difficult to be original and provide useful information to any student who is proactive enough to google "auto-encoders" by herself.

Update: a reader points out that a similar idea was already proposed and implemented in a commercial program. I'm glad to know this! (I would certainly not try to push my own implementation against a commercial product). I was however disappointed to see that the implementation, while perfectly acceptable from the point of view of quantum mechanics, is lacking in a few important ways from the chess logic point of view (some comments are in the thread below). Anyway, this is an example of a good idea coming too late...

What is dark matter (DM)? This is one of the most pressing questions in fundamental science nowadays. We have observed that only one fifth of the matter that exists in the Universe clusters into stars and emits light - the rest appears to only interact gravitationally, producing phenomena we can study through the dynamics of galaxy rotation or by observing the deflection of light passing through it.

Nima Arkani-Hamed needs no introduction - he's a superstar theoretical physicist, and whenever he speaks, his colleagues listen - so much so that his seminars regularly overrun twice past their scheduled duration, without anybody blinking. And today it's your lucky day (and mine), as you get to listen to a clear thinker explaining what really is the status of research in fundamental physics, and why it is actually extremely exciting, much to the discomfort of those who would prefer that public money were spent to reduce taxes (if you don't get the pun, please leave).

Today's news is that five months after Alessandro Strumia's controversial talk at a conference on "Theory and Gender", CERN decided to terminate the Italian theorist's status of "guest professor", effectively cutting its ties with him. The decision certainly affects the ability of Strumia to further his research in particle phenomenology, which centered on models of physics beyond the Standard Model, and is rather unprecedented. 

As the regulars here already know, I am an employee of the INFN. This is the "Istituto Nazionale di Fisica Nucleare", which translates as "National institute for nuclear physics", a slight misnomer of historical origin, as the institute today actually centers its activities on SUB-nuclear physics - i.e. study of elementary particles (but nuclei are still one of the targets!). 

The ATLAS and CMS collaborations released yesterday a joint document where they discuss the combination of their measurements of the rate of production of single top quarks in proton-proton collisions delivered by the LHC collider. The exercise is not an idle one, as the physics behind the production processes is interesting, and its study as well as the precise comparison of experimental results and theory predictions improves our ability to predict other reactions, wherein we might find deviations from the currently accepted theory, the Standard Model.

I am very happy to report today that the CMS experiment just confirmed to be an excellent spectrometer - as good as they get, I would say - by discovering two new excited B hadrons. The field of heavy meson spectroscopy proves once again to be rich with new gems ready to be unearthed, as we collect more data and dig deeper. For such discoveries to be made, collecting as many proton-proton collisions as possible is in fact the decisive factor, along with following up good ideas and preserving our will to not leave any stone unturned.

On March 25 to 27 will be held the school titled "Data Science in (astro)particle physics and cosmology", in Braga (Portugal). The lecturers are prof. Glen Cowan (RHUL), who will cover Statistics, and myself, who will cover topics in Machine Learning. I thought I would mention this here, as for me it is a novelty - in the past years I have often given lectures in advanced statistics topics at various Ph.D. schools around the world, but I never focused explicitly and solely on ML.

In the previous post I discussed, among other things, a purely empirical observation on the mass spectrum of elementary particles, which I summarized in a graph where on the vertical scale I put the year of discovery, and where I only cared to plot particles with a mass above a keV - in fact, we know that neutrinos have non-zero masses, but we have not measured them and they are of the order of an eV or below. Okay, for simplicity I will re-publish the graph below.