@katcorr
This is the link to a textbook Phil Thomas shared on fairness and machine learning: https://fairmlbook.org. It's publicly available online.
Looks like the chapter on classification could potentially be useful down the road. Just skimming through, there's some discussion on independence, separation, sufficiency, and some conflicts between different measures with application to a model with racial discrepancies in predicting credit scores.
I'm not sure if it makes more sense to read this now or later, but I just wanted to include it here.
P.S. I think the way this may fit in with the work on the Seldonian Algorithm is that once we have a better understanding of that and the code, we can observe behavior when we attempt to satisfy multiple constraints or we can evaluate solutions that satisfy different constraints etc. Basically, we could play around with the fairness definitions using the Seldonian Algorithm.
@katcorr This is the link to a textbook Phil Thomas shared on fairness and machine learning: https://fairmlbook.org. It's publicly available online.
Looks like the chapter on classification could potentially be useful down the road. Just skimming through, there's some discussion on independence, separation, sufficiency, and some conflicts between different measures with application to a model with racial discrepancies in predicting credit scores.
I'm not sure if it makes more sense to read this now or later, but I just wanted to include it here.
P.S. I think the way this may fit in with the work on the Seldonian Algorithm is that once we have a better understanding of that and the code, we can observe behavior when we attempt to satisfy multiple constraints or we can evaluate solutions that satisfy different constraints etc. Basically, we could play around with the fairness definitions using the Seldonian Algorithm.