DevOps has helped improve the regular software lifecycle - can’t we just apply it to Machine Learning without having to call it something else?
You can find plenty of articles on the problems of putting machine learning models in production (a 2020 report claims only 22% of companies surveyed have successfully deployed a model into production). Why is that - it’s just software right?
After working in a DevOps role for a while, I’ve spent the last two years working on a Machine Learning team at Ibotta. I’ll talk about how ML really is different in some ways, but how it can still benefit from DevOps principles. We’ll cover cultural and technical differences, what other teams and disciplines are important for success, how we (Ibotta) get models into production and what I’d like to do next.