On Making Stochastic Classifiers Deterministic

Authors: Andrew Cotter, Maya Gupta, Harikrishna Narasimhan

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conclude, in Section 5, with experiments on six datasets comparing these strategies on different problems where stochastic classifiers arise. 5 Experiments We experimentally evaluate the different strategies described above for approximating a stochastic classifier with a deterministic classifier.
Researcher Affiliation Industry Google Research 1600 Amphitheatre Pkwy, Mountain View, CA 94043 {acotter,hnarasimhan,mayagupta}@google.com
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Code made available at: https://github.com/google-research/google-research/tree/master/stochastic_to_deterministic
Open Datasets Yes We use use a variety of fairness datasets with binary protected attributes: (1) COMPAS [24], where the goal is the predict recidivism with gender as the protected attribute; (2) Communities & Crime [25]... (3) Law School [27]... (4) UCI Adult [25]... (5) Wiki Toxicity [28]...
Dataset Splits Yes Table 1: Comparison of de-randomization approaches on ROC matching tasks...Train Test...Train Test... and Figure 1: Test set ROC curves for the Black group and overall population in the Law School dataset.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependency details, such as library or solver names with version numbers.
Experiment Setup Yes For hashing, we first map the input features to 2128 clusters (using a 128-bit cryptographic hash function), and apply a pairwise independent hash function to map it to 232 buckets... For Var Bin, we choose a direction β uniformly at random from the unit 2 sphere, project instances onto this direction, and have the cluster mapping divide the projected values into k = 25 contiguous bins... Additionally, we find that adding the random numbers r1, . . . , r|C| was unnecessary and take rc = 0 for all c...