At a large financial services institution like ours, multi-tenant software architecture for machine learning at scale is non-negotiable due to regulatory and compliance requirements and the nature of our business working with consumer financial data. However, the number and complexity of moving parts required to establish multi-tenancy in the enterprise — from security to bugging, maintenance, governance, and other challenges — renders this an incredibly challenging task. In essence, it requires two systems to work simultaneously, with multiple grounds of users, who all need competently segregated user experiences). I will discuss how Capital One is solving the unique challenges, creating custom solutions, and making open source technology work for us in order to become an industry pioneer in transforming core machine learning environments from single-node to multi-tenant systems. The audience will come away from my talk with a broader understanding of some of the technical learnings and best practices for the challenges that arise, and potential approaches to solve them, when scaling ML across large, complex organizations like ours.
Session Summary
Learnings from the Frontier: Building Multi-Tenant Compute Systems in the Enterprise
MLconf 2022 San Francisco
Zachary Hanif
Capital One
Vice President, Machine Learning
Learn more »
Capital One’s investments in building out a large engineering organization, fully moving to the cloud, re-architecting our applications and data platforms, and embracing machine learning at scale have made us a pioneer in the ability to scale ML across the enterprise in our industry.