The best footballers are not necessarily the ones with the best physical abilities. The difference between success and failure in football often lies in the ability to make the right decisions in seconds, where to run, and when to attack, go through, or shoot.
How can clubs help players to train their brains and their bodies?
My colleagues and I are working with the Chelsea FC Academy to develop a system to measure these artificial intelligence (AI) decision-making capabilities.
We do this by analyzing several seasons of data that track the players and the ball in each game and develop a computer model with different game positions.
The computer model provides a benchmark for comparing the performance of different players. In this way, we can measure the performance of individual players regardless of the actions of other players.
Then we can imagine what could happen if the players had definitely made a different decision. TV commentators always criticize player actions and say they should have done something different without really testing the theory. However, our computer model can show how realistic these suggestions are.
If a critic states that a player has drummed instead of passing, our system can look at the alternative outcome, taking into account factors such as how tired the player was at that point in the game.
We hope coaches and support staff use the system to help players reflect on their actions after a game and, over time, improve their decision-making abilities.
Modeling Decision- Making
Measuring these skills is extremely difficult for several reasons. First, a human being can not track all the events that take place during a game. Second, it is difficult to separate a player's actions from those of another.
For example, if a player passes the ball and a few seconds later the team loses the ball, does the player at the wrong time hand over the wrong player or person Was it someone else's fault?
To address this issue, we use a special branch of the AI known as Imitation Learning. This technology can learn behavioral computer models, such as the behavior of footballers in the field, by analyzing huge amounts of historical data.
In simple terms, the computer model learns to imitate human experts.
Most Decision Systems In AI, such as those used for board games like Go, they are based on reinforcement learning. Here, a computer learns to make decisions by repeatedly performing traits until it receives feedback that it has done the right thing, much as we train a dog to do something by giving it rewards.
But most scenarios in the real world do not attract. There is a certain reward like winning a board game.
Mimicking learning attempts to understand the underlying decision-to develop a policy by looking at it as an expert does a task and then tries to imitate the expert.
Modeling football experts (players) is very difficult because they make decisions with advanced skills that are difficult to program into a computer. For example, choosing the item to notice, choosing the right reaction, and anticipating what other players will do.
For the computer model to be realistic, the historical data on which it is based must reflect the real world as much as possible. It should not only show how players move in relation to each other and the ball, but also how tired they are and how the game situation is.
For example, players want to attack or try to defend or even defend if they want to win or lose. (In some tournaments, a team might want to lose a match, making it easier to position in the next round.)
Change analysis after the game
We have already developed a system to build a model the movements of the players to each other and the ball, with which the performance can be examined.
We now intend to make the model more realistic by adding details of the players' body positions, heart rate (to show tiredness), and playing conditions. We will then develop the system to measure the abilities of current players and hope to have a fully functional system within two years.
We expect the way players and coaches will analyze the games, especially after post-game analysis. This helps players to be more reflective by seeing how their actions could have made a difference. Scouts and clubs could select players and identify talents based on data on those crucial decision-making abilities.
The extension of AI from controlled environments with board games to complex applications in practice remains a daunting challenge. But people are very good at adapting to complex, changing environments and making decisions.
By learning how to mimic human decision-making, AI will be able to cope with all kinds of unfamiliar environments where people do not always follow the rules
This article was written by Varuna De Silva, Lecturer at Loughborough University's Institute for Digital Technologies re-released under a Creative Commons license. Read the original article.