Centralized Selection with Preferences in the Presence of Biases

Authors: L. Elisa Celis, Amit Kumar, Nisheeth K. Vishnoi, Andrew Xu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Further, extensive empirical validation of these results in real-world and synthetic settings, in which the distributional assumptions may not hold, are presented.
Researcher Affiliation Academia 1Yale University 2IIT Delhi.
Pseudocode Yes In this section, we provide the pseudocodes for the three algorithms we compare in Section 4, as well as extensions for relaxed bounds in Appendix G. These include a special case of the Gale-Shapley algorithm (Ast; Algorithm 1), the algorithm to satisfy group-wise proportional representation constraints (Agroup; Algorithm 2), the algorithm to satisfy institution-wise constraints (Ainst-wise; Algorithm 3), and modifications of Algorithm 2 and Algorithm 3 under relaxed bounds.
Open Source Code Yes The code and data can be found here1. 1https://github.com/sandrewxu/Centralized Selectionwith Preference Bias
Open Datasets Yes The data contains the test scores of 384,977 students from IITJEE 2009, self-reported gender and official birth category, opening and closing ranks, and capacities of major-institute pairs (JOSAA, 2009).
Dataset Splits No The paper describes simulation setups and iterations for evaluation but does not specify training, validation, or test dataset splits in the context of machine learning model training.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, or cloud resources) used for running the experiments are mentioned in the paper.
Software Dependencies No The paper states, 'We implement algorithms and run all empirical work in Python3,' but does not provide specific version numbers for Python or any other software dependencies or libraries.
Experiment Setup Yes Setup. We fix n = 1000, p = 5, and ki = 100 for i [p]. For D {DGauss, DPareto} and β { 1/4}, we vary γ [0, γmax] and calculate P(1) and U over 50 iterations.