Recruitment Strategies That Take a Chance

Authors: Gregory Kehne, Ariel D. Procaccia, Jingyan Wang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical evaluation of our algorithms corroborates these theoretical results. and Finally, we carry out experiments on synthetically generated data (Section 4), focusing on the linear penalty incurred by overshooting.
Researcher Affiliation Academia Gregory Kehne Harvard University, Ariel D. Procaccia Harvard University, Jingyan Wang Georgia Institute of Technology
Pseudocode Yes Algorithm 1 PGREEDY and Algorithm 2 ONESIDEDL+1
Open Source Code Yes The code to reproduce our simulation results is available at https://github.com/jingyanw/recruitment-uncertainty.
Open Datasets No We carry out experiments on synthetically generated data (Section 4), focusing on the linear penalty incurred by overshooting. In constructing instances we follow the approach of Purohit et al. [12] in their use of beta distributions to orchestrate different kinds of correlation between xi and pi. The paper states data is synthetically generated and does not provide a link to a dataset.
Dataset Splits No The paper describes experiments on synthetically generated data to evaluate algorithms, but it does not specify training, validation, or test data splits as it is not a machine learning model training task.
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 that code is available for reproduction but does not specify any software dependencies with version numbers.
Experiment Setup Yes We consider n = 50 and pmin = 0.01 throughout, and explore the greedy heuristics XGREEDY and XPGREEDY, as well as the constant-factor approximation algorithm ONESIDEDL+1 (Algorithm 2), for a range of M and λ. In constructing instances we follow the approach of Purohit et al. [12] in their use of beta distributions to orchestrate different kinds of correlation between xi and pi. We therefore first draw xi Unif[0, 1], and then produce three types of correlation as follows: Negative correlation: pi pmin + (1 xi), 10xi). Positive correlation: pi pmin + (1 pmin) Beta(10xi, 10(1 xi)). No correlation: pi Unif[pmin, 1].