Discover Financial Services applies machine learning through a Robocar event powered by Amazon SageMaker

The Discover Financial Services (DFS) team members who attended AWS re:Invent agreed that the Robocar Rally was an extremely impactful experience. By participating in this hackathon, six members of Discover’s core team received hands-on experience using machine learning (ML) and deep learning on AWS. They had a blast and created lasting memories! Discover’s Cloud Center of Excellence (CCoE) later proposed an idea: Let’s find a way to replicate a similar event onsite at the Discover Headquarters. The Discover CCoE had a goal to increase awareness of AWS AI/ML services within their leadership and data science teams in order to impact the landscape of machine learning experimentation. The event came to life in just under 2.5 months.

Team composition

During the event, Discover created six teams, with six individuals per team, to participate in the hackathon and ultimately compete head-to-head. Discover had diverse representation from across many business functions. This created a mixed environment from developers to security experts, and it included a few machine learning practitioners. The team size and number of teams was a result of coming to a happy medium between customer interest and lessons learned passed down from previous Robocar rally events.

Robocar roles

AWS used a prescriptive approach to managing the event and strategically assigning roles to each team-member:

• Driver: Controls the car while training the model, ensuring clean data is collected, and is responsible for controlling the Robocar during the event runs.
• Robo Chairperson: Voice of team and the scribe. The point of contact to discuss all modifications and deviations from original instructions, overall team information, training data, and model information (such as laps driven per data set, model naming conventions, and performance).
• Hardware Engineer: Keeps track of troubleshooting Robocar issues such as wrong cabling, servo failures, battery life, etc. It is very important to ensure that Robocar batteries are constantly charged.
• Software Engineer: Manages the Robocar directories that were flashed on the SD card. Understands each directory and becomes familiar with the structure and intent of the files. Handles model deployment and script execution.
• AWS Architect: Drives the AWS Management Console. The AWS Architect is in charge of verifying that the training data is on Amazon S3, the team repo is created in Amazon ECR, and the Amazon SageMaker training is completed. The Architect references model artifacts while ensuring that the correct IAM policies are in place to do so.
• Pit Boss: Guides the team through their experience running through the event by using their past experience to help explain concepts and troubleshoot. This is a strategic enablement role that also acts as a proxy between team members and AWS account team.

Along with the 40 plus people who made up the competing teams and the hundreds of spectators, there was also a live stream for Discover’s internal employees to view.

Conclusion

Discover believes that the machine learning education experience can be enhanced and accelerated with activities that bring a sense of fun, are hands-on, and provide a sense of community. The event was an innovative way to get Discover’s developers excited about what’s possible and try out new AWS services to rapidly deliver business value in the domains discussed in this blog post. Discover team members, including those with no ML experience, were able to use behavioral cloning with Convolutional Neural Networks (CNNs) and build models within just a couple hours after the Robocar assembly. Teams thought of the Robocar Rally as an easy and fun way to get started and tap into capabilities Discover is actively building in-house. The goals were achieved and the feedback from Discover was overwhelmingly positive. One executive noted, “This on-campus event provided an engaging way for employees to collaborate and get hands-on experience working on machine learning.”

We encourage all of you to see how easy it is to build, train, tune, and deploy machine learning models using Amazon SageMaker. Share your efforts with our growing community of projects. Let us know what you build, and we hope to see you at this year’s re:Invent!

About the Author

As a Solutions Architect, Todd Escalona spends his time evangelizing the AWS Cloud globally for his customers and partners, while listening to understand their goals and working backwards from there. He defines requirements, provides architectural guidance around specific use cases, and assists in designing applications and services that are scalable, reliable, and performant. His interests spread across various technologies such as Artificial Intelligence, Machine Learning and serverless event driven architectures.