The Nerds have Come to Play
Sport is in the throes of a radical shift. Raw strength and technique can get you far enough, but a powerful computer and a brilliant analytical mind can do so much more. How do you pick the right players, how do you win more games, how do you get fans more involved? These have in the past been answered by gut and experience. Now, increasingly, coaches and managers are turning to machine learning and advanced computing on big data.
Big data, big applications
John Guttag, previously the head of MIT’s Electrical Engineering and Computer Science Department, who currently co-heads the Computer Science and Artificial Intelligence Laboratory’s Networks and Mobile Systems Group, has been doing extensive research along with his team on the application of advanced computational techniques in medicine, finance and, more recently, the use of big data and analytics in sports.
“Data-driven medicine is still our primary job – to help people have better cardiac health, avoiding infections, reducing brain damage…” Guttag said in a talk entitled “Tracking the Action” that he delivered recently in Doha. “But about two years ago we decided to work on more important problems like making more money (we did a lot of work on financial analytics) and sports,” he joked. “The interesting thing is that it’s the same underlying mathematics for all three. We build predictive models using machine learning, data mining, algorithms, signal processing, computer vision and graphics. Even though it looks like we are working on different things, deep down we are doing the same thing in different domains.”
Talking further on sports analytics, Guttag said there are two aspects to it – business analytics around sports, like pricing of tickets, and performance analytics. “When it comes to performance analytics, we are concerned about how to win more matches and also what techniques to use to make the viewing experience more interesting for fans watching it on their TVs. There is a strong overlap between the two. The same kind of information that’s valuable to managers in preparing a team is of great interest to people watching the game.”
"MIT researchers will complement the QCRI team, especially in the area of image processing and computer vision. MIT’s vast experience will help in jump-starting the project and accelerating its progress” Dr Mohamed Hefeeda, Principal Scientist, Qatar Computing Research Institute.
The writing on the wall
His team has already been doing extensive work to apply these techniques in baseball and basketball, and Guttag said that while the evidence for their success in basketball is accumulating, it is “crystal-clear in baseball that analytical techniques for player selection have radically revised the way teams bring on players”. Football players are some of the very best paid athletes in the world, baseball players even more so. Statisticians, however, cost much less to hire. “So investing in a statistician gives clubs and teams much more benefit for their money. In fact, that seems to be the trend in a lot of sports,” Guttag said. “The two teams competing for the baseball championship this year, Boston and St Louis, both have invested heavily in analytics and it has paid off for them.” In fact, Major League Baseball will be incorporating analytics in its broadcasting from next season onwards, he said. While groundbreaking and proven research has been done on player selection, there hasn’t been much work done with regard to game preparation and in-game decision-making. But Guttag assures that analytics will come to play a pivotal role in game strategies very soon.
A fruitful partnership
“The area is becoming so big and important, and we are very excited to extend this idea to soccer, or football as you call it.” While it is surprising that Europe, which is the epicentre of all things football, is not yet convinced of the advantages of incorporating analytics in game strategy and planning, player selection and viewer experience, Qatar is taking the first steps in this direction. A dedicated team at Qatar Computing Research Institute, headed by Dr Mohamed Hefeeda, will be conducting collaborative research with Guttag’s team at MIT to develop this technology for football and introduce it to the world. “We are aspiring to create a dedicated research lab, probably the first in the world for football. Other sports will be considered in the longer term. We will develop efficient and distributed machine learning algorithms that can process the vast amount of data generated during sports matches,” Dr Hefeeda said. Guttag also mentioned that MIT would benefit enormously from this cooperation given the growing importance of and intellectual excitement in this region. “The agreement is a milestone for us too. This is our first real, successful attempt to collaborate with people in this part of world.”
Guttag believes that with the help of QCRI, much of what has been done in other areas can be successfully applied to football. “We have done a lot of advanced analytics in basketball and there is a strong relation between these two sports. They have similar objectives: they are both very free-flowing in the way the players move in relation to the other players. A lot of the work can be ported successfully,” he said. But in many ways, it’s also more challenging. “Football is more complex to analyse: more players, larger fields, fewer goals, and fewer games between the same teams,” said Dr Hefeeda. “QCRI has researchers with expertise in data mining, cloud computing and video processing. MIT researchers will complement the QCRI team, especially in the area of image processing and computer vision. MIT’s vast experience in the domain will help in jump-starting the project and accelerating its progress. The collaboration with MIT will also help QCRI in recruiting top researchers.”
Sports analytics 101
So how would sports analytics work? “The important step is finding interesting questions,” Guttag says. “Like, for example, what is the relationship between ball control and winning? Does a team win because they control the ball or is that just incidental?
“The next step is to formalise that question,” he continues. “The hardest part is to control for other variables like the players, which is the biggest and most important variable – how good are the players and what are they good at? To use machine learning, you have to work very hard to understand all the variables and control for those that are not pertinent to the actual question you are answering. That’s probably what we spend more time on than anything else.
“Another thing to worry about is finding a proxy for the outcome of interest. The outcome here is winning but the problem is that there aren’t many games or goals. So to get enough examples to arrive at statistically valid conclusions, you need to find ways to look at proxies. Maybe corner kick differentials or energy budget (are players expending energy efficiently?). And then we have to track down data that sheds light on the question, choosing (or inventing) appropriate analytics techniques to work on them.”
Acquiring data is probably the easier part of the job. “Appropriate cameras installed in the arena film games and give you the (x,y) coordinates of players and officials and (x,y,z) coordinates of the ball at the rate of 25 times per second. Such data exists for about 2,000 football matches. This is better, more high-resolution than we’ll ever need.” The real challenge is sifting through this noise to recognise meaningful signals, watching out for patterns and causalities and testing them out.
“The project is in its very early stages,” Dr Hefeeda says. “It has been approved only a few months ago. QCRI will be hiring several researchers for this project in different research areas. In the first year, I am expecting three to five researchers to join the project, with the number set to grow as the project progresses. There are at least three main research areas that serve the sports analytics project: (i) image and video analysis, (ii) machine learning and data mining, and (iii) large-scale system design. This is, of course, in addition to the domain knowledge about various sports, which we will get from collaboration with professionals and experts (coaches, players, and managers) at different sports organisations and teams such as the Qatar Football Association, and Qatar Olympics Committee.”
The future of sports analytics
A video picks up more information than you’d realise. The visualisation analytics work done at MIT has proved that information about the human body (like heart rate) can be measured, with accuracy comparable to hospital-grade monitors, through motion magnification.The change in colour of your skin with every burst of blood pumped through the veins. The small but sure bobbing of your head with the force of the high-speed injection of blood. These are already finding applications in medicine. The technology is still nascent and doesn’t work with fast-moving objects. But Guttag believes these tools, when developed, can enhance the vision of television viewers and coaches.
To read the original article, visit the website.
In the Media
Cyber security experts in Qatar and the region have advised caution and not to access any emails or other information from unknown sources, following the recent wave of cyber attacks across the world...
As the world ends, will you lock arms and sing “Kumbayah” or embark on a path of law-breaking, anti-social behavior? A new study, based upon the virtual actions of more than 80,000 players of the ...
Researchers from MIT and the Qatar Computing Research Institute have developed a novel new facility in the current rush of interest towards computer vision – an algorithm that can identify overweight...
The Qatar Computing Research Institute’s new Creative Space, which conducts fun activities to teach children computing skills, has successfully held its first Open House event. About 100 children ...
The QCRI – MIT CSAIL Annual Research Project Review is open to the public on Monday, March 27, 2017, at the HBKU Research Complex Multipurpose Room. The annual meeting is a highlight of a ...
Machine Learning and Data Analytics Symposium - MLDAS 2017 Building on the success of the three previous events , Boeing and QCRI will hold the Fourth Machine Learning and Data Analytics Symposium (...