How artificial intelligence can spur the search for new particles

While hunting for new fundamental particles, physicists have always had to speculate how particles can behave. New machine learning algorithms do not need this.

In a collision that took place at the Large Hadron Collider this April, individual charged particles (orange lines) and large jets of particles (yellow cones) were found.

The Large Hadron Collider (LHC) pushes billions of proton pairs every second. Sometimes this car manages to sway the reality a bit, and create in these collisions something unseen before. But since such events are unexpected by definition, physicists do not know what exactly they need to look for. They worry that by sifting through the data on the billions of these collisions, and by sampling some more feasible amount, they may inadvertently remove the evidence of some new physics. “We are always worried that we can throw out the baby with the water,” says Kyle Cranmer , a specialist in particle physics from New York University, working as part of the ATLAS experiment at the LHC.

Facing the task of reasonably reducing the amount of data, some physicists are trying to use machine learning technology such as “deep neural networks” in order to dredge a sea of ​​familiar events in search of new physical phenomena.
')
In the typical case of using this technology, the deep neural network learns to distinguish cats from dogs by studying a pile of photographs labeled "cat" and another pile, marked "dog." But this approach will not work in the search for new particles, because physicists cannot feed the machine images of something that they have never seen. Therefore, they have to do “learning with little supervision,” when machines start learning from known particles, and then look for rare events with less detailed information — for example, how often such events can occur in general.

In a paper published in May on the arxiv.org website of the preprints, three researchers suggested using a similar strategy for expanding the “bump hunting” technique, a classic particle search technique, which found the Higgs boson. The general idea, as one of the authors of the work, Ben Nachman , a researcher at the Lawrence Berkeley National Laboratory, writes, is to train the machine to search for rare variations in the data set.

Consider such a simple example, in the spirit of the mentioned cats and dogs, as an attempt to discover a new species of animals in a data set filled with observations of the forests of North America. If we assume that new animals will cluster in certain geographic areas (this idea corresponds to the fact that new particles are clustered around a certain mass), the algorithm should be able to choose them by systematically comparing the neighboring regions. If there are 113 caribou [North American reindeer of North America] in British Columbia, and 19 in Washington state (despite the presence of millions of squirrels there as well), the program will learn to distinguish caribou from squirrels without studying them directly. “It's not magic, but it looks like it,” said Tim Cohen , an expert in theoretical particle physics from the University of Oregon, who also studies weak supervision.

For traditional searches in particle physics, in contrast to what has been described, researchers have to make assumptions about what a new phenomenon might look like. They create a model of how new particles behave - for example, a new particle can cause disintegration into a certain set of known particles. And only after they determine what they are looking for, they can create a special search strategy. A graduate student usually takes a year of work on this task, but Nachman believes that it could be done faster and more thoroughly.

The proposed CWoLa algorithm, which means “unlabeled classification” (KBM), is able to search in existing data for any unknown particles that decay either into two lighter unknown particles of the same type or into two known particles of the same or different types. With the help of the usual search methods, teams working at the LHC should have spent at least 20 years trying to sift through all the possibilities that coincide with the second option, and for the first option today there are no search strategies at all. Nachman, who works on the ATLAS project, says that KBM is able to conduct all these searches at once.

Other experts in experimental particle physics agree that the game can be worth the candle. “We’ve already searched in different predictable places, so it’s quite important for us to go the other way and fill those voids we’ve not yet searched for,” said Kate Pachal , a physicist seeking collisions of new particles in the ATLAS project. He and his colleagues played with the idea of ​​developing flexible software capable of coping with a large range of particle masses, but none of them had any qualifications in machine learning. “I think it's time to try it,” she said.

It is hoped that neural networks will be able to determine underlying data correlations that are inaccessible to current models. Other machine learning technologies have already successfully accelerated the effectiveness of certain tasks at the LHC, for example, the definition of jets emitted by lower quarks. In that work, it was quite clear that physicists were missing some signals. “They missed some information, and if you paid $ 10 billion for an aggregate, then no information should be missed,” said Daniel Whitson , a specialist in particle physics from the University of California, Irvine.

And yet, the field of machine learning is full of cautionary stories about programs that have tangled up their hands with dumbbells (or worse ). Some at the LHC worry that all these shortcuts will reflect the work of the gremlins in the machine itself, which the experimenters so deliberately try to deliberately overlook. “When you find an anomaly, it’s not immediately clear - is this a new physics, or is it just something wrong with the detector?” Said Till Eifert , a physicist working in the ATLAS project.