FINRA Bridges Business and Technology Through Machine Learning Competitions
Michael Hawman
Innovation and collaboration are two of FINRA’s four core values. Michael Hawman, Director of Technology for Delivery Services, returned to FINRA after participating in AWS re:Invent in 2022 with a proposal that would align these two values for FINRA’s business and technology divisions.
“One of the talks I signed up for [at re:Invent] was a workshop on DeepRacer,” Michael says, referring to an AWS 3D racing simulator that utilizes “a fully autonomous 1/18th scale race car, driven by reinforcement learning.”
“I attended the workshop, trained a model and raced. Afterwards, I thought it would be great to bring back to FINRA so our staff could start getting exposed to some machine learning concepts. You don't need a technical background; you're shown how things work, and the system is built to give you a head start in some areas,” Michael says.
“You are teaching software to make decisions within its environment via rewards,” he adds. “Then you get a practical application of a car driving around a track.”
This is how FINRA has embraced an opportunity to encourage its business and technology staff to learn together, upskill in machine learning, while creating a common language between the two teams.
“Creating that common language can be difficult in traditional software development, but it's even more so within data science,” Michael says. “It’s exposing people to some of these concepts, so they don't feel as intimidated by generative AI and machine learning in the future.”
To support staff in this effort, Michael and his team have provided office hours and learning sessions.
"We have hosted several workshops, so people have seen how easy it is to participate, and it's a fun competition with a leaderboard.” Michael says. “There's motivation for people when they see their name advance, and then see somebody else get ahead.”
This is Michael’s second year hosting the machine learning-based event within FINRA.
In 2023, Michael explored bringing DeepRacer to the Createathon but opted to implement the effort separately.
“We wanted to do something on our own,” Michael says. “Then, we had a successful 2023 and, when it came to figuring out if we were going to do it again, Createathon was the first place we explored to see if they were interested.”
This year, DeepRacer is a key activity at the ninth annual Createathon, FINRA’s premiere innovation event.
“It naturally fits in with the Createathon because both efforts share a common goal; to bring business and technology together,” Michael says. “Then they created their theme "Formula: One FINRA,” and kind of doubled down on the racing concept.”
According to Michael, it’s proven to be a worthwhile effort to get both business and technology staff through the machine learning door.
"It helps us to have members of the business go through that process of training a model, using a certain approach and seeing what it does,” Michael says. “They go back and look at the data to figure out how to change the approach; tweak something in the parameters of data science or maybe the strategy around how they’re rewarding it for making turns. They’ll ask themselves, ‘Am I rewarding the right thing?’ Then go back, analyze what they’ve done and try to figure out where they can make changes relative to what they’re trying to get it to do.”
The feedback Michael and his team have received continues to be positive.
“Most of the feedback I receive from businesses is around ease of use,” Michael said. “They started off unsure about it because it sounded very technical, but once they started, they realized how simple it was. Last year, four of the top five participants were non-technical and had no AI/ML experience. Positive feedback from technology is regarding the real-world application of these skills and the opportunity to dive into machine learning a little bit, figure out the inner workings, looking at the data in terms of how it's making decisions and then being able to tweak their strategies from there.”
Tim Thompson
Tim Thompson, Senior Vice President of Enterprise Risk Management and executive sponsor of Createathon, tried his hand at reinforcement learning with Michael’s assistance.
“Diving into machine learning in this way has been a fun and rewarding undertaking for me,” Tim says. “To date, I’ve worked in different positions across FINRA’s business divisions, and I can see how the opportunity to understand and interact with machine learning could put each of these areas in a position to enhance output as well as increase effectiveness. This knowledge and experience will also enable us to partner more effectively with our technology peers to solve future business problems because we will come to the table with a better understanding of a complex and technical subject matter.”
Machine Learning has also been a recent topic point on FINRA’s podcast, FINRA Unscripted. Follow the link below to learn how FINRA is exploring the integration of these concepts into market regulation and surveillance.
Episode 98: Deep Learning: The Future of the Market Manipulation Surveillance Program
