Am not clear on two of the three supervised methods for monitoring/detecting data drift and shift: `statistical process control`

and `error distribution monitoring`

.

**Statistical Process Control**

In the final video of the week called `Continuous Evaluation and Monitoring`

, from 4min24sec, it is stated that `statistical process control`

expects “errors” to have a binomial distribution. What “errors” is it referring to and why binomial ? Are the errors the numbers of wrong predictions made with serving raw input datasets …?

Furthermore, two equations are provided on the slide, but no explanation is given on what the symbols refer to. In the final quiz, there is a question (number 2) that seems a bit pointless - as it doesn’t test our knowledge of these equations either. Anyone can answer it correctly without knowing what the symbols in the equation actually refer to.

**Error Distribution Monitoring**

I gathered that the overall technique is called `error distribution monitoring`

, but that it is implemented using an algorithm like “adaptive windowing”, and you calculate the mean “error” rate over a these windows of data (presumably windows mean batches of serving/training examples which are streaming into the ML pipeline…?). As with statistical process control, I don’t know what is meant by “error” here ?

(Finally, I suspect I would benefit from any advice on my googling skills to find papers/resources, because my simple searching for “statistical process control” or “error distribution monitoring” hasn’t led me to resources that answer my questions. Is there a neat/efficient/alternative way to find relevant papers in arXiv, via google or otherwise ?)