Addressing Bias in Online Selection with Limited Budget of Comparisons

Authors: Ziyad Benomar, Evgenii Chzhen, Nicolas Schreuder, Vianney Perchet

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we confirm our theoretical findings via numerical experiments, and we give further insight regarding the behavior of the algorithms we presented and how they compare to each other. In all the empirical experiments of this section, each point is computed over 106 independent trials.
Researcher Affiliation Collaboration Ziyad Benomar ENSAE, Ecole Polytechnique, Fair Play joint team ziyad.benomar@ensae.fr Evgenii Chzhen CNRS, LMO, Université Paris-Saclay evgenii.chzhen@universite-paris-saclay.fr Nicolas Schreuder CNRS, Laboratoire d informatique Gaspard Monge (LIGM/UMR 8049) nicolas.schreuder@cnrs.fr Vianney Perchet CREST, ENSAE, Criteo AI LAB Fairplay joint team vianney.perchet@normalesup.org
Pseudocode Yes A formal description is given in Algorithm 1, and a visual representation for the case of three groups is provided in Figure 1. Algorithm 1: Dynamic-Threshold algorithm AB (αk,b)k [K],b B
Open Source Code Yes The code used for the experiments is available at github.com/Ziyad-Benomar/Addressing-bias-in-online-selection-with-limited-budget-of-comparisons.
Open Datasets No The paper does not use external public datasets. It simulates a theoretical problem with parameters (N, λk) without relying on pre-existing datasets.
Dataset Splits No The paper describes numerical experiments for theoretical models and does not involve typical dataset splits (training, validation, test) found in empirical machine learning studies.
Hardware Specification No The paper does not specify the type of hardware (e.g., CPU, GPU models, memory, or cloud instances) used for running the numerical experiments.
Software Dependencies No The paper provides a link to the code repository but does not explicitly list specific software dependencies with version numbers (e.g., programming language, libraries, or solvers).
Experiment Setup Yes In this section, we confirm our theoretical findings via numerical experiments, and we give further insight regarding the behavior of the algorithms we presented and how they compare to each other. In all the empirical experiments of this section, each point is computed over 106 independent trials. ... for N = 500, λ = 0.7, for B {0, 1, 2} ...