www-ai.cs.tu-dortmund.de/LEHRE/SEMINARE/SS21/TrustworthyAIMachineLearning/zafar2017a.pdf
Bulge Prox. Crime C1 C2 C3
Male 1 1 1 3 1 1 1 C1 7 3 3
Male 2 1 0 3 1 1 0 Male 3 0 1 7 1 0 1
C2 3 7 3 Female 1 1 1 3 1 0 1 Female 2 1 0 7 1 1 1
C3 3 7 7 Female 3 0 0 3 0 1 0
Figure 1: Decisions of three [...] isfy Eqs. (1, 3-5).
Some recent works [7, 18] have investigated the impossi- bility of simultaneously satisfying multiple notions of fair- ness. Chouldechova [7] and Kleinberg et al. [18], show that, when [...] 2], [5, 2; 2, 5])
p(x|z = 1, y = 1) = N ([2, 3], [10, 1; 1, 4])
p(x|z = 0, y = −1) = N ([0,−1], [7, 1; 1, 7])
p(x|z = 1, y = −1) = N ([−5, 0], [5, 1; 1, 5])
Then, we train an unconstrained logistic regression …