Back to search

FRINATEK-Fri prosj.st. mat.,naturv.,tek

Use Artificial Intelligence to pinpoint Dark Matter at the LHC

Alternative title: Bruk av kunstig intelligens til å finne mørk materie i data fra LHC

Awarded: NOK 9.8 mill.

What is the nature of dark matter? This is one of the most pertinent questions in modern physics. A number of observations have confirmed the existence of this non-luminous component of matter, referred to as dark matter. Calculations show that the dark matter makes up 84% of the total matter content in the universe. Yet we know very little about its nature. The particle accelerator "The Large Hadron Collider" at CERN (European Council for Nuclear Research) recreates the conditions of the early universe by accelerating, then colliding protons at a speed close to the speed of light. If dark matter consists of particles, it could be produced in these collisions. This project uses data from the ATLAS experiment, which is one of the experiments that collects data from these collisions. The goal is to shed light upon the dark matter. For decades, particle physicists have focused on searching for specific models that can explain dark matter - so far without success. This has been the state-of-the-art method, because the data from the experiments are enormous, both in size and complexity. In this project, we have a different approach: instead of searching for something specific, we concentrate the search on what differs from what we refer to as known physics. Using new methods based on artificial intelligence and machine learning techniques, we search for dark matter in data from the ATLAS experiment in a new and more model-independent way than has previously been done. It is important to note that although machine learning algorithms are powerful, they are also opaque. That is: We do not have access to understand the underlying decisions made by these so-called black-box algorithms. This gives cause for concern and should be handled with care. While using machine learning techniques in our search for dark matter, we are also developing reliable and interpretable methods for artificial intelligence models. Two PhD candidates and one postdoctoral fellow have been engaged to work on this project. Together we work on developing exciting and novel methods for analyzing the data from high energy physics collisions: can methods from e.g. computer vision help to bring more profound understanding of what actually happens in the collision point and thereby bring new insight into the nature of dark matter?

We are living in an extremely exciting time: The current paradigm change in computer science based on artificial intelligence (AI), combined with the highest statistics ever data from the upcoming Run3 of the Large Hadron Collider at CERN, open up unprecedented opportunities to shed light upon the long standing mystery of Dark Matter. This "invisible" component of matter matter is estimated to make up 84% of the matter content of the universe, and without it galaxies and galaxy clusters would not have formed. Yet, we know very little about its particle nature! With this project we will develop novel search strategies for Dark Matter candidates in data from the ATLAS experiment at the Large Hadron Collider using AI and Machine Learning techniques, targeting a yet inaccessible and thus unexplored part of the phase space. The bold aim is to pin-point Dark Matter. With a model-independent search strategy guided by astrophysical measurements of the amount of Dark Matter in the universe today, this project represents novel search methods that may lead to a break-through for revealing the nature of Dark Matter. Although powerful, Machine Learning algorithms are opaque, and the concern that we do not fully understand the underlying decisions made by these so-called black-box algorithms is very much justified. The huge and complex data analysis setting of the ATLAS experiment enables questioning and interpreting AI decision taking. Furthermore, it allow us to analyse how the use of AI is influencing our traditional critical and analytical thinking mindsets. Along with applying supervised, weakly-supervised and unsupervised Machine Learning techniques in the search for Dark Matter, this project will seek to develop reliable and interpretable AI models. An important part will also be to focus on the analytic reasoning itself; how does the use of black-box algorithms affect the way we think analytically?

Funding scheme:

FRINATEK-Fri prosj.st. mat.,naturv.,tek