Contextual bandits with concave rewards, and an application to fair ranking

Authors: Virginie Do, Elvis Dohmatob, Matteo Pirotta, Alessandro Lazaric, Nicolas Usunier

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare the empirical performance of our algorithm to relevant baselines on a music recommendation task. We present two experimental evaluations of our approach, which are fully detailed in App. B.
Researcher Affiliation Collaboration Virginie Do PSL University & Meta AI virginiedo@meta.com Elvis Dohmatob, Matteo Pirotta, Alessandro Lazaric, Nicolas Usunier Meta AI {dohmatob,pirotta,lazaric,usunier}@meta.com
Pseudocode Yes Algorithm 1: FW-Lin UCBRank: linear contextual bandits for fair ranking. Algorithm 2: Generic Frank-Wolfe algorithm for CBCR. Algorithm 3: FW-lin UCB: linear CBCR with K arms. Algorithm 4: FW-Square CB: contextual bandits with concave rewards and regression oracles Algorithm 5: FW-lin UCBRank: linear contextual bandits for fair ranking.
Open Source Code No The paper does not contain an explicit statement or link to open-source code for the methodology described.
Open Datasets Yes Following (Patro et al., 2020), we use the Last.fm music dataset from (Cantador et al., 2011)
Dataset Splits No The paper mentions generating contexts and rewards from a dataset, and sets parameters like 'k = 10' for ranking slots, but does not provide explicit train/validation/test dataset splits, percentages, or methodology for splitting the data used in experiments.
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments.
Software Dependencies No Our experiments are fully implemented in Python 3.9. Using the Python library Implicit, MIT License: https://implicit.readthedocs.io/.
Experiment Setup Yes For all algorithms, the regularization parameter of the Ridge regression is set to λ = 0.1. We set β = 0.5 for all objectives and for welf, we set α = 0.5. We use β0 = 0.01. We choose d = 10 in the data generation and λ = 0.1 in the Ridge regression.