When Cate Campbell touched the wall in front of the American Abbey Weitzeil on the last stage of the women’s layer relay in Tokyo, it ended a record game for the Australian team at the Olympic Games. Individual swims would account for seven of the nine gold medals as the team’s rising stars and seasoned runners lived up to this opportunity. However, it would be a return of six medals in seven seasons, most from any of the competing nations, which would keep the group performing from day one to the last.
Much of that success has been based on the quality of the swimmers, the planning of the coaches, and the hours they spent in fractions of a second, one of which was a 0.04-second piece of Campbell’s ruthlessness that almost hit the mark in his desperate battle against Wide line.
But part of it came from a world-leading collaboration between the high-performance specialists at Swimming Australia and technology giant Amazon, which not only used machine learning to suggest the best possible combinations for Australia, but also predicted squads with astonishing accuracy likely to be used by rivals.
Swimming Australia signed a contract with Amazon Web Services for cloud and data solutions in mid-2019. That didn’t mean much to Jess Corones, SA Performance Solutions Manager; When she thought of Amazon, she thought of online shopping.
But a link in an email led them to a project that AWS had been working on with an entirely different sport, Formula 1. Once the light bulb went out, there was no going back on a project that led to Silicon Valley, Las Vegas and finally the Tokyo Pool, where the rewards of years of hard work would unfold in real time.
“The link led to a little video of Amazon’s work they did with Formula One to use data to get information about the sport and understand what was going on,” Corones said. “I saw this video and thought, ‘This could be a game changer.’
“[Before] I had to prepare the relay data almost three months before the international competition, there were three computer screens, the FINA website where everything was entered into Excel spreadsheets to find out which strategy is the best. .
“They had a little fun, they thought we were a little archaic in the way we do things. The next stop was a visit to the AWS Lab in Silicon Valley, California, where the concept of using machine learning to refine the limbs and orders for squadrons began to rise, which was just the beginning as the uses became apparent.
“We talked about what is the best order to swim … but could we also predict what the other teams would do? Because that would give us a real competitive advantage. If you know someone will swim in the second leg,” or a squadron, or an anchor or a leader, you can figure out who’s the best fit, or shape your team around that opportunity, ”Corones said.
As the algorithm refined as more and more historical data was entered, it was time to test the process under race conditions. The concept was first put to the test at the finals of the International Swim League in Las Vegas in early 2020.
Corones was in the stands with Amazon employees and Australian swim coaches Michael Bohl and Ron McKeon. What they saw was proof that the system was working and confirmation that it would be an important part of the team’s strategic stepping stone to Tokyo.
“We chose the order and the times that these relays would swim within 0.1 seconds. It was amazing. That was the test, the proof of the concept. So we said, ‘We’re here,'” Corones said. The variables at stake were enormous: Corones and the AWS team took into account the most successful historic assignment, the age of the swimmers, and the number of days since their last personal best. It also turned up some useful trends, including one that suggested there was a 60 percentage probability that younger and slower swimmers would have theirs in the second half of the season achieve personal best.
But everything would have counted little if things hadn’t gone well in Tokyo. Overall, they managed to create a highly competitive and highly strategic mixed combination with bronze that banished the glory of the crown. With several combinations in the game and very little space. Accidentally, Australia played perfectly to secure a podium and keep Emma McKeon’s pursuit of seven medals afloat.
“The best example was this season, which is new to the program. We are fooling ourselves, or a lot of people were surprised in what order we swim. We mixed that order from the playoffs to the finals. Many people were surprised that we didn’t. I don’t have Cate (Campbell) with me because she’s a great relay swimmer.
“We’re doing all these different scenarios, we want to anchor with Campbell, anchor with Kyle Chalmers. We were able to predict the American team and we were pretty close to all the international teams in all seven seasons, so it was very satisfying.”
The machines haven’t come to rule the world yet because human variables are at stake, as in the case of the women’s 4 x 200 meter relay, where Australia was the gold favorite, broke its old world record, but only Bronze could hold behind China and the US, Australia led this season with Ariarne Titmus and McKeon getting a fraction of their predicted times and opening the door for the other two teams to take the lead and stand out in the final two stages.
Those intangibles will always be there, said Karl Durrance, corporate director of AWS Australia. However, he said arming training and performance personnel with intelligent and evolving data will soon be the standard for top-class racing. Many systems that have collected different types of data. By merging in a central cloud-based data lake, Amazons Machine Learning Solutions Lab and Swimming Australia developed a solution that brings together the performance of athletes to support coach decision-making in competition, predict prospects and likely competition results, and influence tactical races Strategies and saves hours of manual analysis.
