How Oregon State University AI researchers are using Nvidia GPU’s to help AI’s learn Professional Football
Professor Alan Fern: “AI will revolutionise sports,” a computer science professor who is using videos and artificial intelligence to teach computers how to understand football, and to perhaps provide strategic insights that many human coaches might not see using Deep Learning techniques.
The NBA is already using AI and neural networks to track players and strategic plays, at present mainly for tracking player performance but eventually may provide new and improved plays. Football is far more strategic and involves just over twice as many players for the computer to monitor, and the potential strategies and possibilities increase exponentially.
“Football is the most strategic sport out there.” Fern says; “Every play is a different collection of actions of 11 people. That gives rise to lots of complex strategies.”
The neural network must first learn the game as well as a human before it can even attempt to suggest new strategies. Fern has been training the OSU with hundreds of hours of high school football, and Oregon State plays. It is now able to detect the snap, and the difference between kicking, passing, and running plays and distinguishing between defence or offence. Automatic Planning is a branch of AI that Fern is hoping to use to plan more efficient strategies and action plans to achieve goals.
The NVIDIA Tesla K80 GPU Accelerator, and CUDA programming model is making it possible for the neural network to keep track of all 22 players at once, begin to understand the game, track individual players and understand what each player is doing. Using NFL’s “all 22” videos which are shot from above so that the computer can clearly see the strategies the same way that human coaches analyse the plots.
CUDA is being used to simulate blood flow in arteries, to help find hidden clots and plaque, to analyse Air Traffic Flow to help alleviate congestion, and Nanoscale Molecular Dynamics — simulating chemistry and the behaviours of molecules.
Once the technology becomes more well known and wide spread, there “will be an AI arms race” — Fern
When the neural network has learned enough, it will be able to assist training as well, and possibly point out when a player is in danger of being injured or over exerting themselves. Underperforming players will be noticed right away, and possibly given better training regimes to strengthen the aspects where they are lacking.
To read more of my work on Apple News please visit stirlospace