<ul><li>Optimal transport-based approaches are being used for deriving counterfactuals to quantify algorithmic discrimination.</li><li>Alternative methodologies have been proposed to address challenges in interpreting these methods, such as using causal graphs and iterative quantile regressions.</li><li>Transporting categorical variables has been a challenge, which led to the proposal of a novel approach involving converting them into compositional data and transporting within the probabilistic simplex of R^d.</li><li>The effectiveness of this approach was demonstrated through an illustration on real-world data, along with discussions on limitations.</li></ul>

Optimal Transport on Categorical Data for Counterfactuals using Compositional Data and Dirichlet Transport

Discover more