Chapter 10. Trends in Production ML
So far in this book, we have looked at computer vision as a problem to be solved by data scientists. Because machine learning is used to solve real-world business problems, however, there are other roles that interface with data scientists to carry out machine learning—for example:
- ML engineers
-
ML models built by data scientists are put into production by ML engineers, who tie together all the steps of a typical machine learning workflow, from dataset creation to deployment for predictions, into a machine learning pipeline. You will often hear this being described as MLOps.
- End users
-
People who make decisions based on ML models tend to not trust black-box AI approaches. This is especially true in domains such as medicine, where end users are highly trained specialists. They will often require that your AI models are explainable—explainability is widely considered a prerequisite for carrying out AI responsibly.
- Domain experts
-
Domain experts can develop ML models using code-free frameworks. As such, they often help with data collection, validation, and problem viability assessment. You may hear this being described as ML being “democratized” through no-code or low-code tools.
In this chapter, we’ll look at how the needs and skills of people in these adjacent roles increasingly affect the ML workflow in production settings.
Tip
The code for this chapter is in the 10_mlops folder of the book’s GitHub repository. We will provide file names ...
Get Practical Machine Learning for Computer Vision now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.