IFIC study looks at Artificial Intelligence as a compass for new physics

Artificial Intelligence is not unknown to high-energy physics. On the contrary, it has been used for more than 30 years in the processing of data produced by large colliders. Machines trained by machine learning techniques, capable of analyzing and finding patterns in huge amounts of data, are a fundamental tool in the arsenal of high-energy physics. Now, a group of researchers from the Institute of Corpuscular Physics (IFIC), a research centre at the University of Valencia Science Park (PCUV), proposes to go one step further and use Artificial Intelligence to explore the most promising physical models. 

In the study, the team of Martin Hirsch, Luca Mantani and Veronica Sanz proposes a new strategy for analyzing LHC data using genetic algorithms, an artificial intelligence technique inspired by natural selection. In practice, the algorithm starts with a broad set of candidate models, each representing a possible extension of the Standard Model. From there, it applies small variations ("mutations" and "combinations" between them) to generate new alternative models. That is, the algorithm creates new theoretical models from those already known and each resulting model is evaluated by comparing its predictions with the available experimental data. Models that fit better with the data, or show greater ability to improve future searches, are considered more promising. These models "survive" and are used as a basis for the next generation, while the less compatible ones are discarded. Thus, the algorithm automatically and efficiently identifies which theoretical directions have more potential to reveal new physics.

This approach is particularly powerful because it allows for the efficient exploration of a gigantic, multidimensional theoretical space which cannot be systematically traversed by traditional methods. Instead of analyzing each model separately, something unapproachable due to the large number of possible combinations, the genetic algorithm goes through the options "jumping" among the most promising ones. By selecting, combining and modifying only the best-performing models, the search naturally concentrates on those areas of space where it is most likely to find signals of new physics.

By about 2040, the LHC will cease operation and its successor will most likely be an even larger collider capable of entering uncharted territory. Thus AI could be used as a compass to guide the search for new physics in the next generation of colliders

The goal is not only to better explain current data, but something more ambitious: identify which scenarios have more "hope of discovery" in the near future. That is, it aims to help decide which of the current models have more options for making truly novel contributions.

This study is particularly relevant at a time when particle physics will make several qualitative leaps in the medium and long term, thanks to the improvement of the world’s largest colliders. In particular, the Large Hadron Collider (LHC) will see a significant increase in the intensity of its particle beams over the next few years, entering into a phase known as high luminosity. On the other hand, by about 2040, the LHC will cease its activity and its successor will most likely be an even larger collider capable of entering uncharted territory. Thus, AI could be used as a compass to guide the search for new physics in the next generation of colliders.

The results show that this approach is able to identify signals that traditional analyses might overlook, opening up a new way of combining particle physics, advanced statistics and artificial intelligence. All this would serve to address one of the greatest problems of modern physics: that the Standard Model of Particle Physics is unfinished. In order to complete it, or perhaps replace it, it is necessary to find new phenomena which will enable us to discern between the various theoretical proposals. Tools with Artificial Intelligence can help us, precisely, to explore this unknown territory much more effectively and to guide the search for new fundamental theories.