Monoculture in Matching Markets

Authors: Kenny Peng, Nikhil Garg

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate empirically that our theoretical results hold in markets with heterogeneous firms, as well as when firms use ML models to rank applicants. We provide an overview of our computational experiments; in particular, we test our theoretical predictions in an experiment where firms use ML models (trained on the ACSIncome dataset [19]) to evaluate applicants.
Researcher Affiliation Academia Kenny Peng Cornell Tech kennypeng@cs.cornell.edu Nikhil Garg Cornell Tech ngarg@cornell.edu
Pseudocode No The paper describes models and processes but does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code to reproduce our experiments (and all other figures) is available at a publicly-available repository.6 https://github.com/kennylpeng/monoculture
Open Datasets Yes We use the ACSIncome dataset specifically, individuals from Texas in 2018 [19].
Dataset Splits No For each of the 10 2 = 45 pairs of features, we train a logistic regression model using a 90 percent train split. We then evaluate markets where each applicant corresponds to a point in the test split.
Hardware Specification No All experiments can be completed on a typical laptop in at most a few hours.
Software Dependencies No The models we train use the default sklearn Logistic Regression hyperparameters.
Experiment Setup No The models we train use the default sklearn Logistic Regression hyperparameters.