Google DeepMind has jointly developed soccer tactical AI 'TacticAI' with Premier League powerhouse Liverpool, which can increase the chances of scoring in corner kicks and reduce the probability of conceding goals.
The prestigious English soccer club Liverpool, to which Japanese national soccer player Wataru Endo belongs, is also known as a club that actively utilizes
TacticAI: an AI assistant for football tactics - Google DeepMind
https://deepmind.google/discover/blog/tacticai-ai-assistant-for-football-tactics/
DeepMind and Liverpool FC develop AI to advise on football tactics | New Scientist
https://www.newscientist.com/article/2422562-deepmind-and-liverpool-fc-develop-ai-to-advise-on-football-tactics/
Google's AI is now suggesting new football tactics
https://www.zmescience.com/future/googles-ai-is-now-suggesting-new-football-tactics/
Google DeepMind unveils AI football tactics coach honed with Liverpool
https://www.ft.com/content/e5a64dd3-7fe0-4db4-9f65-6f7517c2c573
Liverpool, a powerful English soccer club, has shown off a number of miraculous comebacks. One of the most well-known upsets by Liverpool is the comeback play against Barcelona in the semi-finals of the Champions League in the 2018-2019 season, also known as the ' Miracle at Anfield .' In the final tournament, which was played at home and away, Liverpool lost the first away game 3-0. As a result, in order to advance to the semi-finals, they now need to score at least three goals in their second home match against Barcelona, whose team includes Lionel Messi, who is considered the best soccer player in the world. However, Liverpool went on the offensive from the beginning of the match, and in the end, a corner kick taken quickly by local player Trent Alexander-Arnold resulted in the fourth goal that took them through to the semi-finals, against one of the world's best clubs, Barcelona. succeeded in a come-from-behind victory.
Reds' miraculous comeback against Barça: Liverpool 4-0 Barcelona | Champions League - YouTube
A corner kick is one of the plays that has a high probability of leading to a goal, like this dramatic goal. In order to generate more goals from such corner kicks, it is necessary to identify the characteristics of the players participating in the match and the tactics of the opponent, and to respond flexibly to the situation.
In order to generate more points from such corner kicks, Google DeepMind researchers and Liverpool staff collaborated to develop the soccer tactical AI ``TacticAI''. It uses predictive and generative AI to provide tactical insights regarding corner kicks, and was published in the academic journal Nature Communications on March 19, 2024 local time.
TacticAI: an AI assistant for football tactics | Nature Communications
https://www.nature.com/articles/s41467-024-45965-x
Google DeepMind and Liverpool have been collaborating since 2021 to advance sports analysis AI, and have published multiple papers as research results so far. The first paper published was about `` game planning ,'' focusing on analysis of penalty kicks (PK) and considering the usefulness of using AI to support soccer tactics. Later, in 2022, a prototype of a prediction system for analyzing soccer data called the `` Graph Imputer '' was devised. As a new AI following these, TacticAI was developed by Google DeepMind and Liverpool.
In machine learning, input data for which prediction results are known is called 'gold standard data.' In soccer, where plays are constantly flowing, it is said to be extremely difficult to obtain such gold standard data, but TacticAI uses geometric technology that supports the creation of generalizable AI models. By adopting a deep learning approach, we succeeded in achieving excellent performance.
Google DeepMind researchers and Liverpool experts are working together to evaluate TacticAI's performance. The results revealed that the tactics proposed by TacticAI were preferred by experts 90% of the time over the tactics actually implemented in soccer matches.
TacticAI is a complete AI system that combines predictive models and generative models. First, the predictive model predicts what will happen, and then it calculates what happened in past plays from a dataset. Search and suggest 'How can I achieve a specific result?'
TacticAI allows you to answer three key questions:
1: What happens in a particular corner kick tactical setup?
Such as which players are most likely to receive the ball or which players are most likely to attempt a shot.
2: What happens when you play a setup (assembly of a pre-prepared attack)?
Have similar tactics worked well in the past?
3: How should tactics be adjusted to achieve specific results?
For example, how to change the positioning of players on the defensive side to reduce the chance of being hit with a shot.
TacticAI treats each player as a node in the graph in order to convert the corner kick situation into a graph display. Google DeepMind said, ``By representing the corner kick setup as a graph, the implicit relationships between players can be directly modeled.'' Each node has characteristics such as the player's position, speed, and height. In addition, it seems that the latest status of each node will be updated by using message passing on the neural network.
The diagram below summarizes how TacticAI handles corner kicks. Four possible plays are generated and fed into the core AI model, the TacticAI model, as predictive information. In corner kicks, questions such as ``Who will touch the ball first?'', ``Who is likely to take a shot?'', and ``How can we increase the possibility of scoring a goal or lower the possibility of conceding a goal?'' It is possible to output information such as ” and contribute to the team's tactical suggestions.
By leveraging predictive and generative models, TacticAI can assist coaches by searching for similar corner kick situations and simulating different tactics. Traditionally, to develop tactics, analysts would replay video of many games over and over again, looking for similar situations or studying rival teams.
On the other hand, TacticAI can automatically calculate the player's numerical expression, allowing experts to easily and efficiently search for similar situations in the past. Google DeepMind further validates intuitive observations through extensive qualitative research with soccer experts. The results showed that TacticAI's top 1 search was successful in finding relevance 63% of the time. This far exceeds the 33% chance of finding a relationship found with existing approaches based on directly analyzing the similarity of players' positions.
By using TacticAI, it is possible to modify the corner kick tactics used by human coaches and reduce the probability of being hit by a defensive shot. TacticAI can also provide tactical recommendations, such as adjusting the positions of all players on a particular team, and from this suggestion “coaches will not only be able to identify important patterns, but also the key to tactical success or failure.” Google DeepMind claims that players can be identified more quickly.
The graph below is an example of the data that TacticAI can actually output. (A) is an example of a corner kick where a shot was attempted, (B) is a proposal to reduce the probability of a shot by adjusting the defender's positioning and movement speed in the data in (A), ( C) is the data that suggests (B) reduces the probability that an attacking player will receive the ball, and (D) is the data that suggests that the series of proposals up to this point can be generated to suggest various tactical options to the coach. thing.
Google DeepMind says of TacticAI, ``Our quantitative analysis shows that TacticAI accurately predicts ball receiver and shot situations in corner kicks, and that player positioning is as important as the actual development of the play.'' We also qualitatively evaluated TacticAI's proposals in a blind case study in which evaluators did not know which tactics were proposed and by whom.Experts affiliated with Liverpool , found that TacticAI's suggestions were indistinguishable from actual corner kick tactics, and were favored over the original tactic 90% of the time.This is not only because TacticAI's predictions are accurate; It shows that it's useful.'
Sports like soccer are also dynamic areas for AI development because they feature real-world multi-agent interactions using multimodal data. As such, advances in AI for sports 'could have applications in a variety of fields, from computer games to robotics,' says Google DeepMind.
Related Posts:
in Software, Posted by logu_ii