A leading data scientist at artificial intelligence computing company NVIDIA foresees a future of sports scouting and recruiting that is not only more objective but also more remote.
Joshua Patterson, NVIDIA’s director of applied solutions engineering, told SportTechie that the deep learning and full-motion video analytics enabled by graphics processing units (GPUs) could overhaul the way colleges and professional teams evaluate players through the rapid processing of huge reams of data and video.
“That’s kind of the revolutionary thing of GPUs in sports,” Patterson said. “Things that used to not be possible are now becoming trivial.”
Much of the work for, say, a college football assistant coach is combining review of prospect-submitted game films and extensive travel to see players in person. On video, they might pause, rewind and circle key movements, Patterson noted, such as “This elbow motion is really great” or “This route was run in this specific way” and “When they turned on this post route, they really shook the defender.”
“They talk about these things almost scientifically, but they’re still humans eyeballing things,” Patterson said, adding: “Using full-motion video analytics, you can start to quantify how fast or strong or nimble a player is without ever having to go watch games.”
Instead, he envisions more teams running videos through neural networks and algorithms to discern such nuances, helping — though not eliminating — the work coaches already do.
Patterson’s data bonafides include his scientific advisory position to the U.S. Census Bureau and his work as a data science principal at Accenture Technology Labs. There, he collaborated on a project that analyzed every shot taken during the 2012-2013 NBA season in a stunning visual project posted at HotShotCharts.com. (Though the site is no longer operational, Wired magazine did a write-up.)
Get The Latest Sports Tech News In Your Inbox!
Similarly, a researcher at MapD recently used NVIDIA’s GPU to compile 13 years of NBA shot data, showing just how profound the league’s outward trend of shot selection is. The volume of three-pointers shot in March 2005 had nearly doubled by March 2017. Interestingly the conversion rate of those shots had not changed appreciably, but the number of players comfortable taking long-range shots had. Mid-range shots, correspondingly, declined precipitously, and the data indicated a roughly 50 percent better efficiency in three-point shot over medium-distance two-pointers. While these are not new revelations, they are presented in a format that’s easily digested by the reader.
While sensors and wearables have their place, Patterson said, he sees such investment as a larger barrier than what can be gleaned non-invasively through an optical-tracking system installed with cameras around the stadium.
“I think deep learning is really going to change the recruiting space heavily,” he said.