Oblivious Data for Fairness with Kernels

Authors: Steffen Grünewälder, Azadeh Khaleghi

JMLR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our key contributions of this work, briefly summarized below, are theoretical; we also provide an evaluation of the proposed approach through experiments in the context of classification and regression1. ... 7. Empirical evaluation In this section we report our experimental results for classification and regression.
Researcher Affiliation Academia Steffen Grünewälder EMAIL Azadeh Khaleghi EMAIL Department of Mathematics and Statistics Lancaster University Lancaster, UK
Pseudocode Yes Algorithm 1 Generating the oblivious kernel matrix; the sum over an empty index set is treated as 0
Open Source Code Yes Our implementations are available at https://github.com/azalk/Oblivious.git.
Open Datasets Yes We evaluated the performance of our method on the so-called Adult dataset, which is a benchmark dataset publicly available on the UCI Machine Learning Repository (Dua and Graff, 2017).
Dataset Splits Yes The smaller option was used in data extraction, giving a total of 1628 training and 12661 testing data-points respectively. ... For each value of γ we generate 500 data points for ORR and M-ORR to infer the conditional expectations and further 500 data points are used by all three methods to calculate the ridge regression solution. For simplicity, we fixed a partition for the conditional expectation: the set S = [ 5, 5] is split into a dyadic partition consisting of 16 sets. Each method uses a validation set of 100 data points (which are different from the 500 training data points) to select the regularization parameter λ from 2 5, 2 4, . . . , 25. A test set of size 100 is used to calculate the mean squared error (MSE).
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments. It mentions training models but lacks any hardware specifications.
Software Dependencies No The paper provides a link to a GitHub repository, implying software is used, but does not explicitly list any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions, etc.).
Experiment Setup Yes The hyperparameters were selected using a 5-fold cross-validation. The SVM regularization parameter C was varied between 2 4, 2 3, . . . , 24, and γ was selected from {0.001, 0.01, 0.1, 1}. ... We use an RBF kernel with σ = 1. ... Each method uses a validation set of 100 data points (which are different from the 500 training data points) to select the regularization parameter λ from 2 5, 2 4, . . . , 25.