Protecting the Protected Group: Circumventing Harmful Fairness

Authors: Omer Ben-Porat, Fedor Sandomirskiy, Moshe Tennenholtz5176-5184

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper, we uncover another mechanism underlying harmful fairness even in static settings: Imposing a fairness constraint can make the disadvantaged group worse off if the fairness constraint and the utilities of the population mismatch. [...] Finally, we suggest additional ways to deal with the mismatch if the underlying utilities can be approximated from data. [...] In this section, we provide evidence for the applicability of WE-fairness by developing tools for computing bank-optimal WE classifiers. Our goal is to show how the bank can use the assumed utility proposed by the regulator to compute approximately optimal classifiers. [...] Proposition 8. Fix a small δ > 0 and assume that the bank has access to a sample of (X, A, Y, α , v) and to estimators u and ˆr such that E [|u v|] ηu and E [|ˆr r|] ηr for small enough ηu and ηr. Then, a (ε, ε) bank-optimal v-WE classifier with 6 1 P(A = 0) + 1 P(A = 1) max{ηu, ηr} can be computed with probability 1 δ on a sample of size O 1 max{ηu, ηr} δ + log log 1 max{ηu, ηr}
Researcher Affiliation Academia Omer Ben-Porat1, Fedor Sandomirskiy2,3, Moshe Tennenholtz2 1Tel-Aviv University 2Technion Israel Institute of Technology 3Higher School of Economics, St. Petersburg, Russia
Pseudocode No The paper does not contain any pseudocode blocks or clearly labeled algorithm sections.
Open Source Code No The paper does not provide any explicit statements about open-sourcing code or links to a code repository for the methodology described.
Open Datasets No The paper mentions working with "historical data" and presents a numerical "Example 3" with hypothetical probabilities and revenue/loss values. However, it does not provide access information (link, DOI, citation to a public source) for any specific dataset used in an empirical study.
Dataset Splits No The paper is theoretical and uses a numerical example (Example 3) for illustration. It does not describe experiments that would involve training, validation, or test dataset splits. Therefore, no information on these splits is provided.
Hardware Specification No The paper does not specify any hardware used for the computations or analysis described. It focuses on theoretical concepts and mathematical models.
Software Dependencies No The paper does not list any specific software or library dependencies with version numbers. It primarily discusses theoretical models and mathematical properties.
Experiment Setup No The paper does not describe an experimental setup with specific hyperparameters, training configurations, or system-level settings, as it is primarily a theoretical paper with a numerical example.