Making the Cut: A Bandit-based Approach to Tiered Interviewing

Authors: Candice Schumann, Zhi Lang, Jeffrey Foster, John Dickerson

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show via simulations on real data from one of the largest US-based computer science graduate programs that our algorithms make better hiring decisions or use less budget than the status quo.
Researcher Affiliation Academia University of Maryland Tufts University {schumann,zlang}@cs.umd.edu, jfoster@cs.tufts.edu, john@cs.umd.edu
Pseudocode Yes Algorithm 1 provides pseudocode for CACO. Algorithm 2 provides pseudocode for BRUTAS.
Open Source Code No The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described.
Open Datasets No The dataset used is described as "real admissions data from one of the largest US-based graduate computer science programs" and "Our dataset consists of three years (2014 16) worth of graduate applications." but no information regarding its public availability or access is provided.
Dataset Splits Yes Using information from 2014 and 2015, we used a random forest classifier [Pedregosa et al., 2011], trained in the standard way on features extracted from the applications, to predict probability of acceptance. In the testing phase, the classifier was run on the set of applicants A from 2016...
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions using a "random forest classifier [Pedregosa et al., 2011]" and "Latent Dirichlet Allocation (LDA)", but does not specify any version numbers for these software components or libraries.
Experiment Setup Yes We determined that the time taken for a Skype interview is roughly 6 times as long as a packet review, and therefore we set the cost multiplier for the second stage j2 = 6. We ran over a variety of s2 values, and we determined σ by looking at the distribution of review scores from past years. When an arm a 2 A is pulled with information gain s and cost j, a reward is randomly pulled from the arm s review scores (when s1 = 1 and j1 = 1, as in the first stage), or a reward is pulled from a Gaussian distribution with mean P(a) and a standard deviation of σ ps.