Fair GLASSO: Estimating Fair Graphical Models with Unbiased Statistical Behavior

Authors: Madeline Navarro, Samuel Rey, Andrei Buciulea, Antonio G. Marques, Santiago Segarra

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical validation includes synthetic and real-world simulations that illustrate the value and effectiveness of our proposed optimization problem and iterative algorithm.4 Experiments We illustrate the ability of Fair GLASSO to reliably estimate both synthetic and real-world graphs from data while promoting unbiased connections.
Researcher Affiliation Academia Madeline Navarro Rice University nav@rice.edu Samuel Rey King Juan Carlos University samuel.rey.escudero@urjc.es Andrei Buciulea King Juan Carlos University andrei.buciulea@urjc.es Antonio G. Marques King Juan Carlos University antonio.garcia.marques@urjc.es Santiago Segarra Rice University segarra@rice.edu
Pseudocode Yes Algorithm 1: Fair GLASSO from Gaussian observations.
Open Source Code Yes Moreover, the code, which is included in the submission for completeness, will be made available on Git Hub if the draft is accepted.
Open Datasets Yes The Movie Lens dataset, a common benchmark for fair graph machine learning, exemplifies our ability to form unbiased models from networks used for recommendation systems. ... (links provided in Appendix G: https://grouplens.org/, https://dl.acm.org/, http://www.sociopatterns.org/datasets/high-school-dynamic-contact-networks/, http://www.sociopatterns.org/datasets/high-school-contact-and-friendship-networks/)
Dataset Splits No Hyperparameters for optimization methods are either chosen to showcase specific scenarios as in Sections 4.2 and 4.4 or chosen via classical hyperparameter tuning methods as stated in Appendix G.
Hardware Specification Yes The experiments are run on a computer with AMD Ryzen Threadripper 3970X 32-Core Processor, two Nvidia Titan RTX GPU, and 188GB of RAM.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes Extensive experimental details including our performance metrics, the baselines with which we compare, and the real-world datasets are provided in Appendix G; these details are summarized here. We include additional experiments on the effect of varying the hyperparameters µ1 and µ2 and violating assumptions (AS1)-(AS4) of Theorem 1 on Appendix H.