ML deployments vary largely from regular software deployments. Any ML system evolves with data. So, along with code, models and datasets both need to be version-controlled. So, the regular CI/CD workflows might not work right off the bat for maintaining ML reliably.
In this talk, we’ll discuss different flavors of incorporating CI/CD into an ML system with varying degrees of automation, technical complexity, tooling, and stage of development. By the end of the session, the participants will have a conceptual framework of how CI/CD can be effectively approached for an ML project development.