Program

Monday, June 26

2 pm - 5 pm (lecture hall S3 043)

Welcome and introduction (François Rioult, Albrecht Zimmermann)

Detecting compensations in running biomechanics: from the lab towards optimizing performance and rehabilitation (Benedicte Vanwanseele, Tim op de Beeck) slides

Recreational distance running is one of the most popular forms of physical activity. Running confers health benefits ranging from the prevention of chronic diseases to improvements in mental health. Unfortunately, running is still characterized by a high (re)occurrence of musculoskeletal overuse injuries. Attempts to identify risk factors for running related injuries (RRIs) have mainly yielded poor results. The main reasons for these poor results are that they mainly use a reductionist approach that disregards the complex multifactorial nature of RRIs, are retrospective laboratory studies, and do not monitor musculoskeletal load during training sessions.

The combination of wearable technology, artificial intelligence (AI) and biomechanical knowledge to model data collected during training sessions and rehabilitation has contributed to a better understanding of the complex interactions between environment, fatigue and injuries. We will discuss the data challenges that arise when analyzing these kind of data and the importance of hardware, sensor attachment and correct feature extraction methods to obtain good data, useful models, and actionable interpretations. 

Data analysis for individual sports: Different types of Markovian models and applications in swimming and climbing (Nicolas Vergne) slides (w/video, w/o video)

Markovian models allow to deal with dependent data, by conditioning the probability of a state at the present time by the state of the previous time. After a brief description of classical Markov models, two extensions of these models will be presented: the drifting Markov models and the semi-Markov models. The very recent construction of drifting semi-Markov models will be presented. Finally, drifting Markov models will be applied to data from swimming and climbing.

Tuesday, June 27

9:30 am - 11 pm (lecture hall S3 043)

Performance Optimisation in Speed Skating and Road Cycling (Arno Knobbe) slides speed-skating cycling

In elite sports, reaching the podium requires attention to detail. Many of these details require optimisation of a wide range of parameters, and careful analysis of the available data is necessary in order to achieve or approach this optimum. For example, when preparing an athlete for an upcoming race, many aspects of the training distribution over the weeks prior to the race will have a modest, but potentially decisive impact on the outcome. Careful analysis of the historical data might give you clues about how to optimise this preparation. In this talk, I will discuss endeavours in two sports disciplines where substantial data science efforts allowed us to dig up detailed insights that could potentially lead to a competitive edge. In the first sport, speed skating, we analysed a large collection of training schedules in order to look for opportunities and pitfalls for future races. In road cycling, we model the non-trivial physiological response of selected athletes to different exertions, through data collected from the power meter on the bike. In both cases, we focus on the challenge of using detailed personal data in order to produce an athlete-specific model that captures the individual characteristics of the elite athlete.

11 pm - 12 pm (lecture hall S3 043)

Interior geo-localization in the training facility of the Vikings (Caen Handball) François Rioult

2 pm - 5 pm (Computer room S3 160 [S3 159, if necessary])

Practical session: Contextualizing physical data in handball: working with raw data to "perceive" handball (John Komar) slides (560 MB, .pptx)

This session proposes to start from raw data, directly collected in match, in order to give them more meaning as to the sporting activity practiced. Giving meaning to raw data comes down to contextualizing this data, taking an interest in specific events, identifiable game phases, and more generally specificities related to the activity that is practiced.
We will use LPS data collected during handball matches by the team's physical trainer, which meets all the criteria for raw, unstructured data. In this sense, data cleaning (synchronization, calibration) is a preliminary and essential step to analysis. Once the data is “cleaned”, we will try to perceive “handball” in it: define game phases (e.g., counter attack), identify types of defense, identify types of runs (e.g., lateral, rear). This work then makes it possible to contextualize the physical data generally collected by the physical trainer (e.g., number of accelerations, distance traveled at different running intensities), and thus to really measure the cost of the strategy/tactic used. 'a team defends using a staggered defense, it will increase its chances of intercepting passes, but it also increases the energy cost by more distance traveled and accelerations achieved - the analysis of physical data must therefore to be interpreted in the light of the tactics/strategy employed.

Wednesday, June 28

9:30 am - 12 pm (Computer room S3 160 [S3 159, if necessary])

Practical session: Basketball data science (Ambra Macis) slides

This short course consists of two parts. The first one provides the understanding of the basic concepts of basketball data science. The second one concerns basketball data and the R package BasketballAnalyzeR. Starting from a description of the various sources and forms of Basketball data, this part introduces the R package BasketballAnalyzeR and suggests how to use it with some basic and advanced methods of statistical data analysis.

2 pm - 3:30 pm (lecture hall S3 043)

Optimal coaching: how the tools of statistical modelling and artificial intelligence can help (Mathieu Rosenbaum) slides

In this lecture, we show through examples how some mathematical tools combined with artificial intelligence techniques can help staffs improve their decisions. We focus on football teams and consider various issues that need to be addressed before, during and after games. Our aim is to illustrate the power of science driven tools to quantify decisions made by staffs and players.

3:30 pm - 17 pm (lecture hall S3 043)

Using machine learning to assess player quality in team sports (Albrecht Zimmermann) slides

Thursday, June 29

9:30 am - 12 pm (lecture hall S3 043)

Spatio-temporal challenges in analyzing Soccer data (Ulf Brefeld) slides

Data originating from sports often requires non-standard ways of processing. Particularly data tracking data from team sports is often difficult as there are different types of agents (players team A, players team B, ball) where each agent's position is measured automatically at regular and discrete intervals. Tracking data is often enriched with match events that are often annotated manually. Hence, timestamps in event data are irregularly distributed across the game and may differ from the timestamps in the tracking data. In the first part of this talk, I will present different ways to  convert tracking and event data from football (soccer) into usable formats. The second part uses these as input representations to models that compute interesting results. We will focus on the models and discuss applications and outcomes (e.g., pattern detection, availability,…).

Data-Informed Pragmatism: Parma Calcio’s Use of Analytics in Football (presentation will be in French, with the slides in English) (Sébastien Coustou) slides

Discover how Parma Calcio leverages data analytics in football with a pragmatic approach. We collaborate with leading providers like Opta, Transfermarkt, WyScout, and Wimu, utilizing terabytes of data through our big data architecture. In-house machine learning capabilities further enhance our analytical capabilities. This presentation showcases the integration of data insights across scouts, senior leadership, and staffs, illustrating Parma Calcio's data-informed pragmatism at the forefront of football analytics.