Auditing and Enforcing Conditional Fairness via Optimal Transport

Authors: Mohsen Ghassemi, Alan Mishler, Niccolo Dalmasso, Luhao Zhang, Vamsi K. Potluru, Tucker Balch, Manuela Veloso

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we compare the effectiveness of the methods described in Sections 4 and 5 on four datasets commonly used in the fairness literature (Fabris et al. 2022). We include two classification tasks on the Drug (Fehrman, Egan, and Mirkes 2016) and Adult (Becker and Kohavi 1996) datasets; and two regression tasks on the Law School (Ramsey, Wightman, and Council 1998) and Communities and Crime (Redmond 2009) datasets.
Researcher Affiliation Collaboration 1J.P.Morgan AI Research 2Department of Applied Mathematics and Statistics, Johns Hopkins University EMAIL, EMAIL
Pseudocode Yes (See Appendix D for pseudocode for our actual algorithms.)
Open Source Code No The paper does not provide a concrete statement about releasing its own source code or a direct link to a repository for the methodology described.
Open Datasets Yes We include two classification tasks on the Drug (Fehrman, Egan, and Mirkes 2016) and Adult (Becker and Kohavi 1996) datasets; and two regression tasks on the Law School (Ramsey, Wightman, and Council 1998) and Communities and Crime (Redmond 2009) datasets.
Dataset Splits No The paper mentions running experiments multiple times ("We report the mean over 10 runs for every method-hyperparameter combination") but does not specify the exact percentages or methodology for training/validation/test splits of the datasets used.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU models, or cloud computing specifications used for running its experiments.
Software Dependencies No The paper mentions using a multi-layer perceptron (MLP) and specific loss functions but does not list any specific software libraries or their version numbers, such as Python, PyTorch, TensorFlow, or scikit-learn versions.
Experiment Setup Yes In all experiments, we use a multi-layer perceptron (MLP) with two hidden layers containing 50 and 20 nodes and a rectified linear unit (Re Lu) activation function. The loss function is set to be cross-entropy for classification (after a softmax activation) and mean squared error (MSE) for regression.