GUEDJ Benjamin, University College London (Centre for Artificial Intelligence, Department of Computer Science) et Inria Lille – Nord Europe.
Résumé : Massive data collection in sports has revived the old scientific utopia of better understanding the world by automatically converting data into knowledge. We present clustering and prediction algorithms (called Magma and MagmaClust) based on multi-task Gaussian processes to deal with inhomogeneous time series, and illustrate how both outperform the state-of-the-art techniques. We apply those algorithms to predict performance profiles for professional swimmers and help early detection of promising athletes.
Joint work with Arthur Leroy, Pierre Latouche, Servane Gey (all with Université de Paris)
References: https://arxiv.org/abs/2007.10731 and https://arxiv.org/abs/2011.07866
