TOP
Search the Dagstuhl Website
Looking for information on the websites of the individual seminars? - Then please:
Not found what you are looking for? - Some of our services have separate websites, each with its own search option. Please check the following list:
Schloss Dagstuhl - LZI - Logo
Schloss Dagstuhl Services
Seminars
Within this website:
External resources:
  • DOOR (for registering your stay at Dagstuhl)
  • DOSA (for proposing future Dagstuhl Seminars or Dagstuhl Perspectives Workshops)
Publishing
Within this website:
External resources:
dblp
Within this website:
External resources:
  • the dblp Computer Science Bibliography


Please select your role
to obtain appropriate quick-links.
As a LIPIcs/OASIcs author, you will be invited by e-mail to register at the Dagstuhl Submission Server. The server will guide you through the publication workflow. Preliminary information can be found here:
Quick Links

Dagstuhl Seminar 24081

Computational Approaches to Strategy and Tactics in Sports

( Feb 18 – Feb 23, 2024 )
(Click in the middle of the image to enlarge)
Permalink
Please use the following short url to reference this page: https://www.dagstuhl.de/24081
Organizers
Contact
Shared Documents
Publications
Schedule
Summary
Show Summary

The rapid growth in spatio-temporal data in sport over the past decade has generated numerous methodological developments from the statistical and machine learning communities. The richness of modern sports data is enabling sports researchers to analyze every action and decision during a competitive event in increasing detail. Two central topics that have emerged from this new phase of methodological research in sport are data-driven approaches to strategy & tactics. In a nutshell, strategy & tactics allow weaker teams or athletes to win over stronger ones. Therefore, they are one of the most interesting and challenging aspects in sports.

Although both terms describe similar aspects and are even sometimes used interchangeably, they range on different time scales. A strategy serves as an overarching umbrella to reach long-term goals. Hence, strategic decisions involve long-term training plans, signing players and coaches, as well as deciding on team formation, pacing, equipment, rotations, or playing philosophy. On a shorter time scale a match/race strategy is the plan made by coaches and athletes before the start of the match or race.

Tactics, on the other hand, is rather short-term. Tactics are the execution and adaptations to the planned strategy to have an edge over the opponent during the match or race. Tactics are therefore often broken down into building blocks or patterns that can be easily communicated to athletes. Note that communication is of utmost importance as tactics are invented by the coaching staff while their implementation is the task of the players/athletes. A tactical pattern may thus involve only subgroups of athletes, or subsections of a race and assign concrete tasks for predefined, context-sensitive situations.

The goal of this Dagstuhl Seminar was to bring together a diverse set of researchers from both academia and industry working on these topics. The seminar drew from people with various backgrounds in terms of area of specialization (Artificial Intelligence, Operations Research, Sport Science, Statistics), role (Academic, Data Provider, Federation, Sports Club) and sport (Australia Rules Football, Baseball, Basketball, Darts, Ice Hockey, Soccer, Speed Skating, Tennis, Wheelchair Rugby).

The seminar was structured around three themes:

Discovery

The goal of this theme was to discuss different methods that can automatically identify tactical and strategic patterns from spatio-temporal data. Examples were given for problems such as detecting formations, identifying commonly occurring sequences of actions (e.g., passing sequences), discovering player movement trajectories, and deciding where players should aim a tennis serve.

Evaluation

This theme focused on the challenges and pitfalls associated with trying to evaluate the finding of computational approaches to identifying strategies and tactics. This theme focused on highlighting a number of methodological issues and describing ways to assess the validity of discoveries. There were a number of interesting examples given about how causal analysis could be used to evaluate the efficacy of certain tactics. Finally, the potential and risks for using large language models in sports were also discussed.

Communication

This theme tackled the problem of how to communicate the findings from tactical studies to an interdisciplinary audience. The emphasis was on how to marry finding from the research literature to things that could be translated into practice. A key point that was made is that it is crucial to think about what types of information will be useful and actionable for practitioners.

The first three days of the seminar focused on one theme, which was introduced with a longer tutorial and then shorter presentations. The final full day of the seminar was open to all topics under the themes and there was a greater focus on presentations from early-career researchers in attendance. The seminar also featured two panels and (small) group discussions about five different topics.

Results

During the seminar, we identified and agreed upon the following action points aimed at trying to continue integrating the various different communities (Sports Science, Operations Research, Statistics, Artificial Intelligence) working on computational approaches to tactics in sports:

  1. We will collect a list of venues where computational approaches to tactics in sports are often published. We will host this on the web: https://dtai.cs.kuleuven.be/sports/venues/
  2. We will explore setting up a slack or discord channel to facilitate more continuous interaction and the ability to quickly get answers to questions. Joris Bekkers and Jan Van Haaren will take the lead on this point.
  3. We have setup a document that contains the biographies, contact details, and topics of interest for all seminar participants that are willing to share their information. That will help people stay in touch.
  4. We will strive to setup some basic tutorials that illustrate how to implement standard, concepts that reoccur across sports. For example, many team sports have variants of plus-minus, expected possession value metrics, and expected statistics such as expected goals (soccer, ice hockey) or expected rush yard gained (American Football).
  5. We will continue to promote the mailing list for disseminating computational sports-related information (job ads, conference call for papers, etc.) and we will use this list to distribute the report on the seminar to reinforce our thanks to the attendees and excitement about the seminar’s outcomes: ml-ai-4sports@googlegroups.com
Copyright Ulf Brefeld, Jesse Davis, Laura de Jong, and Stephanie Kovalchik
Motivation
Show Motivation

The past decade has seen a rapid growth in the ability to collect large-scale spatiotemporal data sets about sports. Ideally, such data should be used to inform strategic and tactical decision making. On the one hand, strategy is the long-term planning of training sessions, signing of coaches and athletes, rotation and the plan made before a match/race. On the other hand, tactics are short-term and involve the execution and adaptation to the match/race plan. Having insights into the efficacy and feasibility of strategies and tactics is particularly important and challenging within sports because effective and novel strategy & tactics allow weaker teams or athletes to win against stronger ones. Unfortunately, the size, richness, and complexity of modern spatiotemporal sports data means that automated analysis is essential. Alas, the nature of the data has posed a number of challenges for classic analysis techniques. This has spurred the development of novel statistical and machine learning techniques in order to perform more fine-grained analysis of every action and decision during a competitive event.

In this Dagstuhl Seminar, we aim to bring together sports researchers in academia and industry to understand how they are using machine learning and statistical techniques to analyze strategy and tactics. Understanding, formalizing, discovering, and analyzing strategy & tactics pose a number of key challenges from a computational perspective, which include the following aspects:

  1. From a modeling perspective, what is the best way to represent discovered strategic and tactical concepts?
  2. How can we evaluate the efficacy of different strategies and tactics?
  3. How do you communicate the findings from tactical studies to an interdisciplinary audience?
  4. How can computational approaches be used to support decision making for coaching?
  5. What kind of data and domain knowledge are needed to conduct strategic and tactical studies?

The goal of the seminar is to explore and discuss how we can and should answer these questions.

Copyright Ulf Brefeld, Jesse Davis, Laura de Jong, and Stephanie Kovalchik
Participants
Show Participants
Related Seminars
  • Dagstuhl Seminar 06381: Computer Science in Sport (2006-09-17 - 2006-09-20) (Details)
  • Dagstuhl Seminar 08372: Computer Science in Sport - Mission and Methods (2008-09-07 - 2008-09-10) (Details)
  • Dagstuhl Seminar 11271: Computer Science in Sport - Special emphasis: Football (2011-07-03 - 2011-07-06) (Details)
  • Dagstuhl Seminar 13272: Computer Science in High Performance Sport - Applications and Implications for Professional Coaching (2013-06-30 - 2013-07-03) (Details)
  • Dagstuhl Seminar 15382: Modeling and Simulation of Sport Games, Sport Movements, and Adaptations to Training (2015-09-13 - 2015-09-16) (Details)
  • Dagstuhl Seminar 21411: Machine Learning in Sports (2021-10-10 - 2021-10-15) (Details)
Classification
  • Artificial Intelligence
Keywords
  • sports
  • machine learning
  • spatio-temporal data
  • explainability
Seminar 24081
© 2024 Schloss Dagstuhl – Leibniz-Zentrum für Informatik GmbH
ImprintContactAccessibilityPrivacy