Source – https://www.kotatv.com/
RAPID CITY, S.D. (KOTA) – There are 9.2 quintillion possible outcomes of the March Madness brackets. That is 18 zeros after the nine!
No one has ever verifiably picked a perfect bracket, but some students at the South Dakota School of Mines are hoping to be the first, with the use of an artificial intelligence technique.
“Guessing that is essentially impossible, you’ve got to use some sort of assisted educated guess,” said Trevor Bormann, a senior metallurgy and computer sciences student at SDSMT.
They are using a technique called neural networking that is similar to the way the nerves in the retina of our eyes process transmit information to our brains. Their bracket model utilizes basketball statistics and deep analytics, called KenPom, plus, attributes of all the teams to pick a winner.
“What we’re not doing, is we’re not picking individual games, per say,” said Dr. Kyle Caudle, a math professor at SDSMT. “The output of the neural network will predict how far a team will make it in the bracket.”
Not only are they using this season’s Ken Pom data, they are using information from years past.
“But we also use historical data,” s aid Dr. Randy Hoover, a computer science and engineering professor at SDSMT. “So, we use all the past brackets to train our neural network model. And so, we can look at teams from the historical data and see how far they’ve made it in the brackets, based on all of the statistics that they use.”
Past massive upsets actually help the modeling– take the #16 seed UMBC upset over #1 Virginia in 2018. Even though the outcome was so unlikely, the data point was added to the model.
“It basically helps it not fall in a pitfall because if you have our 20 years of data, and 1 always beats 16, it’s always going to prioritize that 1 seed over the other,” said Jackson Cates, a junior computer science student at SDSMT. “So it helps it not make a guarantee to pick that.”
“What are you guys doing with this model? Are you just using it to beat everyone in office bracket pools for the rest of your lives?” asked Anderley Penwell.
The four laugh and Bormann said they are competing.
“We’re using it to compete in a competition, called the MinneMUDAC Competition, where there’s a whole bunch of teams from around the Midwest area that are creating their own artificial intelligence and machine learning algorithms and competing to see who can get the best bracket,” said Bormann.
There are 9.2 quintillion possible brackets.
“If you took every single human on Earth, and conscripted all of them to fill out brackets, 24/7, and everyone completed a bracket a minute, it would take 2,500 years or so to finish all of them,” said Bormann.