Exacerbating Algorithmic Bias through Fairness Attacks

Authors: Ninareh Mehrabi, Muhammad Naveed, Fred Morstatter, Aram Galstyan8930-8938

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments that indicate the effectiveness of our proposed attacks. Through experimentation on three different datasets with different fairness measures and definitions, we show the effectiveness of our attacks in achieving the desired goal of affecting fairness.
Researcher Affiliation Academia 1University of Southern California 2Information Sciences Institute {ninarehm, mnaveed}@usc.edu, {fredmors, galstyan}@isi.edu
Pseudocode Yes Algorithm 1: Influence Attack on Fairness; Algorithm 2: Anchoring Attack
Open Source Code Yes 1https://github.com/Ninarehm/attack
Open Datasets Yes German Credit Dataset. This dataset comes from UCI machine learning repository (Dua and Graff 2017). COMPAS Dataset. Propublica s COMPAS dataset contains information about defendants from Broward County 2. Drug Consumption Dataset. This dataset comes from the UCI machine learning repository (Dua and Graff 2017).
Dataset Splits No The data was split into an 80-20 train and test split. No explicit validation set or split information is provided.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used in the experiments.
Experiment Setup Yes In our experiments we set λ = 1. In our experiments we set τ = 0. Hinge loss was used to control for accuracy for all the methods in our experiments as in (Koh, Steinhardt, and Liang 2018).