Title: Analyzing positional data from soccer games with deep reinforcement learning
Authors: Ulf Brefeld - Leuphana University of Luneburg (Germany) [presenting]
Abstract: Positional data are regularly recorded in many soccer leagues, such as Bundesliga. The data are captured at 25fps and contains $(x,y)$ coordinates of all players and the ball. Although that data are available for some years, they are mainly used for computing heat maps showing the whereabouts of players or descriptive statistics like the covered kilometers by a player. By contrast, we will show that the data can actually be used for more sophisticated analyses using state-of-the-art deep architectures. We will present examples dealing with player movement models, controlled zones, and predictions of player actions.