Using player tracking data from 7,176 corners filmed in the Premier League between 2020 and 2021, the researchers first created a graphical representation of player placement and determined player position, movement, height, and weight. Encoded as nodes on the graph. , and relationships between players as boundaries between players. They then used an approach called geometric deep learning, which exploits the symmetry of a soccer field to reduce the amount of processing required by a neural network. (This is not a new strategy; a similar approach was used in DeepMind's seminal AlphaGo research.)
The resulting model led to the creation of a number of tools that may be useful to soccer coaches. Based on the player's positioning at the moment the kick is taken, TacticAI can predict which player is most likely to make first contact with the ball and whether a shot will be taken as a result. Then, recommendations on how best to adjust player position and movement to maximize (for attacking teams) or minimize (for defending teams) the chances of a shot being taken. can be generated. This means moving the defender to cover the near post. For example, placing a man at the edge of an area.
Liverpool's soccer experts particularly liked how TacticAI's recommendations could pinpoint attackers or defenders who are “falling asleep at the wheel”, which are critical to the success of a particular tactic, Beličković said. . Analysts spend hours watching video footage, trying to find weaknesses in an opponent's defense to target, or holes in their own team's performance to double down on in training. “But in different situations he is very difficult to track 22 people,” says Veličković. “With a tool like this, you can quickly see which players are not making the right moves and which players should be doing something different.”
TacticAI can also be used to find other corners that feature similar patterns of players and movements, again saving analysts time. According to DeepMind, the suggestions made by this model are twice as useful as the current technique, which is based solely on the player's body coordinates and does not take movements or physical characteristics into account, and is said to be useful by Liverpool's coach. It was evaluated. (The two corners may look the same, but if one corner has a tall striker on the edge of the box and the other corner runs towards the near post, that's probably important.)
Part of this effort is to address the lack of appropriate language to describe the vast range of events that can occur on a street corner, said DeepMind's Zhe Wang, another lead contributor to the paper. It is said that Unlike American football, which has a time-honored and deep nomenclature for various plays and running routes, this level of detailed set-piece choreography in soccer is a relatively new phenomenon. “Each coach may have their own way of describing the corner kick patterns they observe,” Wang said. “So at TacticAI, we want to harness the power of deep learning to establish a common language for describing corner kick patterns.”
In the future, the researchers hope to incorporate TacticAI into a natural language interface, allowing coaches to run queries in text to get answers to the problems they are trying to solve in the field, according to the paper. He says he is thinking about it. Velichkovic said the model could also be used during games for coaches to refine corner routines on the fly, but it's most likely useful in the days before games to free up coaches' time. He said that. “We don't want to build AI systems to replace experts,” he says Veličković. “We want to build AI systems that enhance the capabilities of experts, allowing them to do their jobs more efficiently and spend more time on the creative part of coaching. Masu.”
