In research, machine learning (ML) is data science. In production, it’s software, which needs to be managed just like every other component in the software development lifecycle.
Managing functions created by data is very different from managing those created by human hands. ML code must be versioned and coordinated with other versions of a solution—and version changes have side effects, sometimes requiring pull back. Importantly, changes in incoming data can indicate not just when an algorithm needs updating, but also the ML function.
In this presentation, Kris addresses all of the issues that impact the ML development lifecycle, frequently called "ML ops.” He covers the native mechanisms in AWS for managing ML ops as well as the most popular commercial and open source alternatives. In fact, ML ops is a fast-evolving field. There’s probably no single solution that's perfect for your situation, but with a broad view of the challenges and alternatives, you'll be able to make informed choices.
Presenter: Kris Skrinak, Global Machine Learning Technical Lead, AWS