High-energy neutrinos are evidence of dramatic processes in the Universe: exploding stars, gamma-ray bursts and cataclysmic processes involving black holes and neutron stars. They are the most abundant massive elementary particles, however, they have hardly any mass and are electrically neutral. This allows them to travel through space almost entirely unhindered, but it also makes them notoriously difficult to detect.
The IceCube neutrino observatory at the South Pole is the first of its kind. It consists of an underground cubic kilometer of ice in which strings of light-sensitive detectors register traces of the elusive particles. Neutrinos are not observed directly, though. When they randomly interact with the ice, they produce electrically charged secondary particles, which in turn emit bluish so-called Cherenkov light.
Over the past ten years, researchers have developed several approaches to reconstruct the original direction of the neutrinos from these sporadic luminous events in the ice. Existing solutions, however, are either fast but inaccurate, or more accurate but at the price of enormous computational time.
New ideas may soon come from the international data science community. Data scientists and machine learning aficionados can find a wide variety of data sets on the internet platform Kaggle to test their models, share codes, and learn new skills.
The goal of the "IceCube - Neutrinos in Deep Ice" Kaggle is to use machine learning routines to develop new solutions to determine the direction of a neutrino particle. It is a public competition in which participants, working alone or in teams, could ultimately help scientists develop significantly faster and more accurate algorithms for analysing neutrino events. This would significantly increase the chance of identifying cosmic neutrino sources.
Link to the competition: https://www.kaggle.com/competitions/icecube-neutrinos-in-deep-ice
Dr. Philipp Eller (TUM)