A0664
Title: Pattern recognition in elite soccer with only a few labeled situations
Authors: Ulf Brefeld - Leuphana University of Luneburg (Germany) [presenting]
Abstract: The identification of strategies and tactical patterns is key to pre- and post-match analyses in team sports, and analysts usually spend a great deal of their time watching and annotating video footage. The different ways to automatically annotate patterns of interest so that the analyst can select the most relevant ones for further analyses are discussed. In general, supervised machine learning approaches suggest themselves for this task, but they often require large amounts of labelled situations to successfully learn the target concepts. Since this is often difficult in practice, we will focus on label-efficient approaches, such as self-supervised pre-training, to learn representations of the data, which allows solving pattern and event detection in soccer with only a few annotated situations.
