Ah — do write weeknotes :) They are a great way to keep that “internal dialogue” going, beyond anything else.

I’m referring to the traditional definition of supervised/unsupervised learning (e.g. https://en.wikipedia.org/wiki/Supervised_learning): in this context, the issue is that problems are intrinsically one or the other. For example, a “yes/no” diagnosis is a supervised learning problem, while unsupervised learning problems are for example related to clustering, as in the case of “can this algorithm unearth hidden patterns in this dataset”. A problem can only ever been one or the other; but, of course, the same data can potentially be used for problems that are either.

By encouraging our partners to frame their problem the right supervised/unsupervised way, we can achieve two things together: enable them to explore different classes of problems once they understanding the difference, and allow them to work on increasing their data quality and completeness (for example, if the labelling necessary to attack a supervised learning problem is somewhat wanting).