Practical MLOps

Chapter 5. AutoML and KaizenML

By Noah Gift

Held down too hard by rules, partial thoughts cannot blossom. Rules without ideas is prison. Ideas without rules is chaos. Bonsai teaches us balance. Balancing rules against innovation is a pervasive problem in all of life. I once saw a play entitled “The Game of Life.” The message was that one is often asked to play the game for high stakes before anybody has explained the rules. Moreover, it’s not so easy to tell if you are winning. It often seems that beginners (and young people generally) need rules or broad theories for guidance. Then, as experience accumulates, the many exceptions and variations gradually invalidate the rules at the same time that the rules become less needed. A great advantage of bonsai over Life is that one can learn from fatal mistakes.

Dr. Joseph Bogen

It is an exciting time to be involved in build machine learning systems. Machine learning, i.e., learning from data, has a clear value to humanity in solving problems from autonomous vehicles to more effective cancer screening and treatment. At the same time, automation plays a critical role in enabling this advancement in the automation of model creation, AutoML, and the rest of the tasks surrounding machine learning, something I call KaizenML.

While AutoML is focused strictly on creating a model from clean data, KaizenML is about automating everything about the machine learning process and improving it. Let’s dive into both topics starting with the reason ...