How machine learning gave Australia an edge in the Tokyo pool
Lots of interesting stories that have come out on the recently concluded Tokyo Olympics. But this one’s on a subject that is applicable to a whole host of disciplines outside of sport (indeed, we featured one on cricket recently). In an increasingly digital world, where there is plenty of data footprint, big data analytics can indeed be a source of competitive advantage as the experience of the Australian swimming relay team shows from this article.
“Individual swims would account for seven of the nine gold medals as the rising stars and seasoned racers of the squad rose to the occasion. Yet it would be a return of six medals across seven relays, the most of any competing nation, that underpinned the group effort from the first day until the last.
Much of that success came from the quality of the swimmers, the planning from coaches and hours spent working on split-second changeovers…But part of it arose from a world-leading collaboration between Swimming Australia’s high-performance specialists and tech giant Amazon, which used machine learning to not only suggest the best possible combinations for Australia but predict with uncanny accuracy the likely squads to be deployed by rivals.
…“[Previously] I’d have to prepare the relay data almost three months out from the international competition, there were three computer screens, the FINA website, feeding everything into Excel spreadsheets, trying to work out what’s the best strategy.
“It was a bit of a laugh really, for them, they thought we were a bit archaic in how we did things.”
The next stop was a visit to the AWS Lab in Silicon Valley in California, where the concept of using machine learning to fine tune the members and orders of relays began to take flight.
….“We talked about what was the best order for us to swim in … but could we also predict what the other teams were going to do? Because that would give us a real competitive edge. If you know someone is going to swim in the second leg or a relay, or anchor, or lead, you can find out who the best match might be, or engineer your team around that possibility,” Corones said.
“We were picking the order and the times these relay teams would swim between 0.1 of a second. It was unbelievable. That was the test, the proof of concept. Then were like ‘we’re on here’,” Corones said.
The variables at play were immense, with Corones and the AWS team considering the most successful historical order, the age of the swimmers and how many days since their last personal best. It also turned up some useful trends, including one that suggested there was a 60 per cent chance of younger and slower swimmers swimming a personal best in the second half of relays.
… “The best example was that relay, which is new in the program. We fooled, or a lot of people were surprised, with the order that we swam. We mixed that order from heats to finals. A lot of people were surprised we didn’t have Cate (Campbell) there, because she is a huge relay swimmer.
“We ran all those different scenarios, we looked at anchoring with Campbell, anchoring with Kyle Chalmers. We were able to predict the American team and were pretty close on all of the international teams across all seven relays, so that was very satisfying.”
…“Swimming Australia had a lot of systems that were collecting different types of data. By bringing this together into a central cloud-based data lake, the Amazon Machine Learning Solutions Lab and Swimming Australia developed a solution that pulled together athlete performance to support coach’s decision-making in competition, predict possible and probable competition outcomes, and influence tactical racing strategies while saving hours of manual analysis.””