Optimizing Generalized Rate Metrics with Three Players

Authors: Harikrishna Narasimhan, Andrew Cotter, Maya Gupta

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on different fairness tasks confirm the efficacy of our approach.
Researcher Affiliation Industry Google Research 1600 Amphitheatre Pkwy, Mountain View, CA 94043 {hnarasimhan, acotter, mayagupta}@google.com
Pseudocode Yes Algorithm 1 Oracle-based Optimizer Initialize: λ0 for t = 0 to T 1 do...
Open Source Code Yes All implementations are in Tensorflow.2 See Appendix G for additional details. 2https://github.com/google-research/google-research/tree/master/generalized_rates
Open Datasets Yes We use five datasets: (1) COMPAS, where the goal is to predict recidivism with gender as the protected attribute [44]; (2) Communities & Crime, where the goal is to predict if a community in the US has a crime rate above the 70th percentile [45], and we consider communities having a black population above the 50th percentile as protected [27]; (3) Law School, where the task is to predict whether a law school student will pass the bar exam, with race (black or other) as the protected attribute [46]; (4) Adult, where the task is to predict if a person s income exceeds 50K/year, with gender as the protected attribute [45]; (5) Wiki Toxicity, where the goal is to predict if a comment posted on a Wikipedia talk page contains non-toxic/acceptable content, with the comments containing the term gay considered as a protected group [47].
Dataset Splits No The paper mentions using datasets but does not explicitly provide training, validation, and test split percentages or counts in the main text.
Hardware Specification No The paper does not provide any specific details regarding the hardware used for running the experiments.
Software Dependencies No The paper mentions 'All implementations are in Tensorflow' but does not specify a version number or other software dependencies with versions.
Experiment Setup No The paper mentions using linear models and hinge losses as surrogates, and specifies the objectives and constraints for the fairness tasks, but does not provide specific hyperparameters or detailed training configurations like learning rates, batch sizes, or optimizers.