Using race cars to teach machine learning to kids
Teaching real world lessons with the AWS DeepRacer
December 1, 2020
Localized shutdowns are still affecting a large swath of the United States with many schools in full or partial shutdown. This summer was particularly challenging for parents who faced months without virtual school or camps to help their kids engaged. That’s when a small group of Capital One associates came together to create a fun event for the children of their colleagues across the company. Over the course of a few planning sessions, the first ever AWS DeepRacer Kids Cup League was born. In its original form, the AWS DeepRacer is a competition designed to enable the discovery of machine learning through hands-on application, with a fully autonomous 1/18th scale race car driven by reinforcement learning (a type of machine learning) and a global racing league.
At the start of 2020, the DeepRacer planning committee expected to run an event similar to their event in 2019 -- completely in person and limited to our associates. In 2019, more than 350 participants took part in the Capital One AWS DeepRacer League. But like much of this year, plans changed. After the COVID-19 pandemic hit, Capital One closed its offices and the team had to rethink the event in a virtual environment. The core planning committee of Lisa Tovar (Senior Manager, Card), Cadence Weber (Manager, Tech College), and Denisa Karfikova (Principal Project Manager, Technology) reworked the associate competition, and saw the potential to expand the event. A simple way to learn machine learning by teaching an autonomous vehicle to quickly race around a track could appeal to adults and children alike. The Kids Cup was born.
“It was a whole mindshift change where we wanted to benefit not just our associates but our children who were out of classrooms as well,” Lisa recalled. “We realized that with the agile mindset that Capital One sowed in its culture, there was no reason why we couldn’t pull this off and do the hard work to make it happen.”
The event turned out to be an entertaining and “out of the ordinary” lesson in reinforcement learning and real world implications. For Lisa’s daughter Izzie, she learned how to create a reward function with her AWS DeepRacer vehicle so that it could move around the track. As she quickly learned, Izzie needed to use the reward function to navigate the vehicle to stay within the tracks to reach its destination. The function's purpose is to provide numerical rewards for doing the right thing, in this case making the vehicle head around the track in the fastest time possible. Lisa sought to make reward functions easier to understand by relating it to the real world concept of rewards and consequences relevant to children.
“I used an analogy for lying to explain the reward function as a way to teach Izzie about what happens when people lie,” Lisa explained. “When a person tells a lie, it could spiral out of control, the consequence. When that person tells the truth, they could work out those issues with their parents, the reward. Analogous to DeepRacer, the model uses a reward system to stay on the track, the higher the reward the more likely for the car to stay on the track.
Getting a chance to work with machine learning through the use of a race car was something that intrigued Alexandra Restrepo, a Senior Associate in Client Services who handles various channels for the New Jersey region on behalf of commercial clients. Over the course of the competition, the team provided mentors that students could engage with during office hours. Mentors were there to help the kids understand how their models worked and provided suggestions on how they could improve.
“It was so nice seeing the trajectory of these children training their models to prevent erratic driving,” Alexandra said. “They knew that they wouldn’t get it right the first time. But after going to our mentor sessions, they learned to have patience and iterate over and over again. AWS provided a great platform that showed ways to achieve proper coding.”
Alexandra’s son Christopher enjoyed the experience as a participant in the Kids Cup League. The at-home experience from the COVID-19 pandemic had left him wanting something different to do since he missed school. He noted, “The Kids Cup made me see what I could do in the future as a career path.”
The event also drew in Jeff Gill, a Business Director who’s worked at Capital One for nine years, and his daughter Sophia who enjoys watching Formula One car races with him. What Jeff enjoyed most about the Kids Cup League was the upfront training the children received.
“The training helps spark interest in technology,” Jeff explained. “By learning about the structure of the reward function and thinking through the different weights on outcomes for your model to reward, Sophia began taking on a different outlook. Teaching her about tech through this manner not only cleared up what used to seem mysterious about data and coding, but made her realize that coding isn’t as hard as it may seem.”
All three Capital One parents agreed that the Kids Cup was an unique experience not just for their children, but a thoughtful way for parents to connect with their children.
“As far as the COVID-19 pandemic affecting the work-life balance, Capital One has been there as an advocate to do what’s right for our associates,” Alexandra explained. “The company understands that the external environment impacts an education gap that many of us parents feel unable to fill on our own. In this case, they gave us resources through the DeepRacer Kids Cup League to make us feel less overwhelmed.”
For Jeff, the Kids Cup felt exactly like something that the company would do because of the Mission and Values we live and strive for every day at work. “We say we’re a tech company that a bank would build,” Jeff noted. “Events like these true back to how we fundamentally see everything from a tech lens. People are seeing how everything we do is infused with leveraging data and streamlining things to get to better decisions. This ultimately helps the customer experience as well.”
As for Lisa, helping associates get to spend time with their kids during the pandemic was a no-brainer.
“I’m not aware of any other companies who go all in on events like these, Izzie’s first interaction with technology was at a Capital One Take your Daughter to work day event, it was an immersive experience that started her on a path to want to learn more”” Lisa said. “DeepRacer was a natural extension of that and very Capital One to want to give working parents a break, while continuing to spark an interest in technology for their kids.”