Testing Group Fairness via Optimal Transport Projections

Authors: Nian Si, Karthyek Murthy, Jose Blanchet, Viet Anh Nguyen

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 6. Numerical Experiments Our experiments use the following three datasets: Arrhythmia (Dua & Graff, 2017), COMPAS (Multi Media LLC, 2016) and Drug (Fehrman et al., 2017). The details of the datasets are provided in Appendix B.2. In the first experiment, we test the fairness of the Tikhonovregularized logistic and SVM classifiers by the equal opportunity criterion. We randomly split 70%-30% of the data as a train-test set. Figure 1 reports the test statistics, fairness rejection threshold, and the accuracy of the classifier.
Researcher Affiliation Collaboration 1Department of Management Science & Engineering, Stanford University 2Engineering Systems and Design pillar, Singapore University of Technology and Design 3Vin AI Research, Vietnam.
Pseudocode Yes Algorithm 1 Computing P(ˆPN) for one-dimensional φ( ) 1: Input: Data {d(xi), C(xi), φ(ui, EˆPN [U])}N i=1. 2: Output: the optimal value P(ˆPN). 3: Let s P i [N] C(xi)φ(ui, EˆPN [U]); 4: For i [N], compute ti d(xi) 1(1 2C(xi))φ(ui, EˆPN [U])sgn(s);
Open Source Code No The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper.
Open Datasets Yes Our experiments use the following three datasets: Arrhythmia (Dua & Graff, 2017), COMPAS (Multi Media LLC, 2016) and Drug (Fehrman et al., 2017). The details of the datasets are provided in Appendix B.2.
Dataset Splits No We randomly split 70%-30% of the data as a train-test set.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment.
Experiment Setup No The paper mentions 'Tikhonov-regularized logistic and SVM classifiers' and plots results against a 'regularization parameter λ', but it does not specify concrete experimental setup details such as learning rates, batch sizes, optimizers, or other hyperparameters needed for replication.