Member-only story
The Wheel of Data
Why a data flywheel is a good idea for MLOps.
· The Known Unknown Matrix
∘ DevOps Known Unknowns
∘ MLOps Known Unknowns
· The Data Flywheel
· Data Orchestration and Lineage
· Conclusion
· References
This is the third and final article in the series about data-centric MLOps. In the previous articles[1][2], we looked at how dataset quality can be achieved in six stages of MLOps. The conclusion, however, was that the model and the data could not be frozen. As the world evolves, it needs to be updated continuously. How the model can be improved continuously through a cycle of dataset updates is the focus of the current article.
The current series of articles on MLOps started with an analogy with DevOps. DevOps has two cycles, Development and Operations, tightly coupled together like the following diagram:
To explain how software updates happen, I am introducing the Known Unknown matrix.
