Offline Contextual Bandits with High Probability Fairness Guarantees

Authors: Blossom Metevier, Stephen Giguere, Sarah Brockman, Ari Kobren, Yuriy Brun, Emma Brunskill, Philip S. Thomas

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

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our algorithm on three applications: a user study with an automated tutoring system, a loan approval setting using the Statlog German credit data set, and a criminal recidivism problem using data released by Pro Publica. To demonstrate the versatility of our approach, we use multiple well-known and custom definitions of fairness. In each setting, our algorithm is able to produce fair policies that achieve performance competitive with other offline and online contextual bandit algorithms.
Researcher Affiliation Academia 1College of Information and Computer Sciences 2Computer Science Department University of Massachusetts Amherst Stanford University
Pseudocode Yes Algorithm 1 Robin Hood (D, = {δi}k i=1, ˆZ{ˆzi j}d j=1, E = {Ei}k i=1) ... Algorithm 2 Candidate Utility(θ, Dc, , ˆZ, E)
Open Source Code No The paper does not provide a direct link to open-source code for the Robin Hood algorithm or explicitly state that the code is being released.
Open Datasets Yes We use the Statlog German Credit data set, which includes a collection of loan applicants, each one described by 20 features, and labels corresponding to whether or not each applicant was assessed to have good financial credit [32]. ... This experiment uses recidivism data released by Pro Publica as part of their investigation into the racial bias of deployed classification algorithms [3].
Dataset Splits No The paper describes partitioning data into 'candidate selection set Dc' and 'safety set Ds' within the algorithm (Algorithm 1), but it does not specify standard training, validation, and test dataset splits with percentages or sample counts for external reproducibility of the overall evaluation.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or specific cluster configurations) used to run the experiments.
Software Dependencies No The paper mentions using 'Student s t-test' and 'CMA-ES [19]' but does not provide specific version numbers for these or other software dependencies.
Experiment Setup No The paper mentions using 'CMA-ES [19] as our optimization method' but does not specify its hyperparameters or other detailed training configurations (e.g., learning rates, batch sizes, epochs) for the experiments.