www-ai.cs.tu-dortmund.de/LEHRE/SEMINARE/WS2021/TrustworthyAIMachineLearning/literature/zafar2019.pdf
= 0, y = 1) = N ([2, 0], [5, 1; 1, 5])
p(x|z = 1, y = 1) = N ([2, 3], [5, 1; 1, 5])
p(x|z = 0, y = −1) = N ([−1,−3], [5, 1; 1, 5])
p(x|z = 1, y = −1) = N ([−1, 0], [5, 1; 1, 5])
An accuracy-maximizing logistic [...] = 0, y = 1) = N ([1, 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])
We then train a accurac [...] decision time.
34
Fairness Constraints: A Flexible Approach for Fair Classification
1 W
5 M
1 2 3 4 5-1-2-3-4-5
5 W 5 M
x = 0
Figure 12: Covariance constraints may perform unfavorably in the presence of outliers …
