Residual Unfairness in Fair Machine Learning from Prejudiced Data

Authors: Nathan Kallus, Angela Zhou

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate how systematic inclusion can affect fairness adjustments on an empirical example with data from the application of the Stop, Question, and Frisk (SQF) policy of the City of New York Police Department (NYPD). ... In Table 2, we report the estimated true positive and false positive rates achieved by these classifiers both in the training and in the target population.
Researcher Affiliation Academia 1Cornell University, and Cornell Tech, New York 2Cornell University, Ithaca, New York.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes We consider learning a predictor of criminal possession of a weapon from this dataset... (Further details about SQF and the dataset are provided in the supplemental Sec. D.) ... We estimate the former from the SQF data and the latter from the 2010 American Community Survey data by matching census blocks to precincts and using Laplace smoothing. ... Keefe, J. Nypd stop-and-frisk data for you. Technical report, WNYC, 2011. ... NYCLU. Stop and frisk data, 2017. URL https://www.nyclu.org/en/stop-and-frisk-data.
Dataset Splits No The paper mentions training data and target population for evaluation but does not provide specific train/validation/test dataset splits (e.g., percentages or sample counts).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types, or memory amounts) used for its experiments.
Software Dependencies No The paper mentions using logistic regression but does not specify any software components with version numbers (e.g., Python, library versions, or specific solvers).
Experiment Setup No The paper mentions fitting a logistic regression and setting an exchange rate of 25:1 for minimizing false negatives and false positives, but it does not provide specific hyperparameters (e.g., learning rate, batch size, number of epochs) or detailed training configurations.