This repository was created during my master thesis about fairness in machine learning. It consists of two modules:
- Metrics: Some metrics that are typically used to evaluate fairness
- Models: Models that can be used to enforce algorithmic fairness
A proposal how one can enforce individual fairness, given a distance metric that captures the similarity between individuals/samples. The proposal includes a WGAN-like architecture, where the distance a-la-Wasserstein is computed between the generator output and the labels of the similar samples. The latter are provided by the so-called Sampler, which incoroprates the already mentioned distance metric. By default, the sampler uses the euclidean distance.
