|
The application and approval process, filtering, certification, authentication, etc. sounds like a separation problem. Since the site is about math, IIRC there's some math for separation problems?? Uh, since the site is for "math/stat" and there have to be rates of false positives, that people are already complaining about, and false negatives, I don't see people complaining about spam, then we're into statistical hypothesis testing, right? Sooo, to do better on the rates of false negatives, we want more data for a more powerful test. Sooo, we need a multi-variate test. Since no way can we justify assuming probability distributions for all the relevant data, we need a distribution-free test. So, where can we find one of those??? Disclosure: This question is just an exercise. For an answer, I published one of those. So, it's an applied probability calculation based on an algebraic group of measure preserving transformations! It may be a rationalization of resampling theory. Crudely the result is obvious, but a proof is tricky. It may be that the work is a stimulation for and or connection with approximate independence, e.g., maybe as in some work of Choquet student M. Talagrand. |