The objective of this talk is to introduce a fairness interpretability framework for measuring and explaining the bias in classification and regression models at the level of a regressor distribution. In our work, we measure the model bias across sub-population distributions in the model output using the Wasserstein metric. To properly quantify the contributions of predictors, we take into account the favorability of both the model and predictors with respect to the non-protected class. The quantification is accomplished by the use of transport theory, which gives rise to the decomposition of the model bias and bias explanations to positive and negative contributions (flows). This information is useful when utilizing the approach for bias mitigation and access to the protected attribute is prohibited. To gain more insight into the role of favorability and allow for additivity of bias explanations, we adapt techniques from cooperative game theory.

Relevant papers:
https://arxiv.org/abs/2011.03156
https://arxiv.org/abs/2111.11259

Zoom link:  https://ncsu.zoom.us/j/97638681103?pwd=dDJrRkE3d3NQZEhrRlhOMDc4T0pRUT09
passcode: NAseminar