��ݮ��Ƶ

“There would be plenty of games where I thought I had a really good game and helped the team win, but it didn't show up in the box score,” says David.

It wasn’t that he never scored–he recorded 10 points in 14 playoff games while leading the Thunderbirds to win the Northern Ontario Junior Hockey League and Dudley Hewitt Cup championships in 2015–but it wasn’t his primary responsibility on the team.

“I would pride myself on making a good first pass out of the defensive zone or breaking up plays,” says David. “I always wondered: how can these things be measured?”

Now a PhD candidate in computer science, David is using his expertise in multi-agent systems, an area of artificial intelligence that looks at situations involving numerous decision-makers, to answer this question.

It all started with a hackathon in the Fall of 2020. Hosted by Rogers, Sportsnet and the University of ��ݮ��Ƶ, the competition challenged participants to use new puck and player-tracking data (first introduced in the 2020 Stanley Cup Playoffs) to enrich the fan experience.

David, who was already working on his PhD at the time, leapt at the opportunity to combine his passion for hockey with his expertise in computer science. His team, Game², won first prize for an application that gave fans new ways to play fantasy hockey and bet on in-game events. In the process, David realized that this new tracking data could transform the way we measure performance in hockey.

After the hackathon, David, along with his brother Daniel Radke, a master’s student in computer science in Germany at the time, University of ��ݮ��Ƶ computer science professor Tim Brecht and Masters student Alex Pawelczyk, published that used puck and player-tracking data from the NHL to introduce several innovative new hockey metrics. These metrics assessed how players pass, respond to defensive pressure and move on the ice. A , which refined the team’s passing model by accounting for player movement and the possibility of indirect passes off the boards, won the best research paper at , the Linköping Hockey Analytics Conference.

“Now that we have tracking data, we can measure things like passing lanes and defensive pressure and learn a lot about a player's tendencies and situational awareness,” says David. “We can assess how risky a player’s passes are, for instance, and we can look at how they behave in certain contexts, like specific parts of the ice.”

These advanced metrics have a multitude of applications: they can be used to improve players’ on-ice decision-making, help coaches strategize for opponents, and enhance the way teams evaluate talent and build their rosters.

And the NHL is taking notice. He was recently hired as a consultant by the Chicago Blackhawks, who are eager to put his insights into practice. “They are very keen embrace an evidence-based way of building a team,” says David. “There was a lot of buy-in from ownership all the way down to the coaching staff. It was a perfect fit.”

But it’s not only hockey teams that can benefit from this work. David envisions applying his research to other “invasion games,” such as basketball and soccer, in which teams succeed by invading an opponent’s territory. This is the topic of he is publishing at the Autonomous Agents and Multiagent Systems conference in June.

“I believe that what statistics did for baseball, multi-agent systems can do for hockey, basketball and soccer,” says David.