Public Kaggle Competition "IceCube -- Neutrinos in Deep Ice"

TL;DR

This paper discusses a Kaggle competition focused on improving neutrino event reconstruction in IceCube using machine learning, highlighting innovative solutions and key results from hundreds of participating teams.

Contribution

It presents an overview of the competition organization, methods, and main findings, advancing the application of machine learning in neutrino physics.

Findings

Improved accuracy in neutrino direction prediction

Diverse machine learning models outperform traditional methods

Insights into effective features and training strategies

Abstract

The reconstruction of neutrino events in the IceCube experiment is crucial for many scientific analyses, including searches for cosmic neutrino sources. The Kaggle competition "IceCube -- Neutrinos in Deep ice" was a public machine learning challenge designed to encourage the development of innovative solutions to improve the accuracy and efficiency of neutrino event reconstruction. Participants worked with a dataset of simulated neutrino events and were tasked with creating a suitable model to predict the direction vector of incoming neutrinos. From January to April 2023, hundreds of teams competed for a total of $50k prize money, which was awarded to the best performing few out of the many thousand submissions. In this contribution I will present some insights into the organization of this large outreach project, and summarize some of the main findings, results and takeaways.