Contrastive Examples for Addressing the Tyranny of the Majority
This addresses fairness issues in AI for underrepresented groups, offering a method to mitigate bias in applications like face recognition, though it is incremental as it builds on existing generative techniques.
The paper tackles algorithmic bias in computer vision, where models favor majority groups in training data, by proposing contrastive examples that intervene on group memberships to balance datasets. They show that generative adversarial networks can create these examples, reducing bias as measured by equalized odds on tabular and image datasets like CelebA and Diversity in Faces.
Computer vision algorithms, e.g. for face recognition, favour groups of individuals that are better represented in the training data. This happens because of the generalization that classifiers have to make. It is simpler to fit the majority groups as this fit is more important to overall error. We propose to create a balanced training dataset, consisting of the original dataset plus new data points in which the group memberships are intervened, minorities become majorities and vice versa. We show that current generative adversarial networks are a powerful tool for learning these data points, called contrastive examples. We experiment with the equalized odds bias measure on tabular data as well as image data (CelebA and Diversity in Faces datasets). Contrastive examples allow us to expose correlations between group membership and other seemingly neutral features. Whenever a causal graph is available, we can put those contrastive examples in the perspective of counterfactuals.