Algorithmic Fairness Verification with Graphical Models

Authors: Bishwamittra Ghosh, Debabrota Basu, Kuldeep S Meel9539-9548

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we empirically evaluate the performance of FVGM. We first present the experimental setup and the objective of our experiments, followed by experimental results.
Researcher Affiliation Academia 1 School of Computing, National University of Singapore, Singapore 2 Equipe Scool, Univ. Lille, Inria, UMR 9189 CRISt AL, CNRS, Centrale Lille, France
Pseudocode No The paper describes algorithmic steps using recursive definitions and dynamic programming but does not provide structured pseudocode or an algorithm block.
Open Source Code Yes *Source code: https://github.com/meelgroup/justicia
Open Datasets Yes We perform the scalability analysis on five real-world datasets studied in fair ML literature: UCI Adult, German-credit (Doshi-Velez and Kim 2017), COMPAS (Angwin et al. 2016), Ricci (Mc Ginley 2010), and Titanic (https://www.kaggle.com/c/titanic).
Dataset Splits Yes We perform five-fold cross-validation on a dataset.
Hardware Specification No The computational work for this paper was performed on resources of Max Planck Institute for Software Systems, Germany and the National Supercomputing Centre, Singapore (https://www.nscc.sg).
Software Dependencies No We implement a prototype of FVGM in Python (version 3.8). We deploy the Scikit-learn library for learning linear classifiers such as Logistic Regression (LR) and Support Vector Machine (SVM) with linear kernels. For learning a Bayesian network on the converted Boolean data, we deploy the PGMPY library (Ankan and Panda 2015).
Experiment Setup Yes During discretization, we apply a gird-search to estimate the best bin-size within a maximum bin of 10. To convert the coefficients of features into integers, we employ another grid-search to choose the best multiplier within {1, 2, . . . , 100}.