Member-only story
Any product has its own lifecycle. Same goes to a machine learning system. It does not just work all the way until the end of the world. It’s a loop, looping the steps again and again until the end of the application.
I will mostly use a visual inspection process in manufacturing process as an example in this article.
A simplified machine learning lifecycle would look like this:
Data collection
Its all start with data (well, it should start with problem definition, but we are talking about a machine learning system, so we assume the problem was identified, and solvable by machine learning techniques).
Data collection is a generalized term. It involves a few tasks here:
Data source identification
Where does the data come from? What are the factors affecting the data quality?
In a product inspection process, for example, the data source can be from an equipment in a factory, and the data is the images captured by a camera. The setting of the camera will affect the data collected. Most of the time, we would want the camera of different machines in the same process to have the same specifications and setting to ensure similar data features can be captured. E.g: area of focus, brightness, image…
