Distributionally Robust Fair Principal Components via Geodesic Descents

Authors: Hieu Vu, Toan Tran, Man-Chung Yue, Viet Anh Nguyen

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results on real-world datasets show the merits of our proposed method over state-of-the-art baselines. ... 5 NUMERICAL EXPERIMENTS
Researcher Affiliation Collaboration Hieu Vu, Toan Tran Vin AI Research, Vietnam Man-Chung Yue Hong Kong Polytechnic University Viet Anh Nguyen Vin AI Research, Vietnam
Pseudocode Yes Algorithm 1 Riemannian Subgradient Descent for (7)
Open Source Code Yes The code for all experiments is available in supplementary materials.
Open Datasets Yes We consider a wide variety of datasets with ranging sample sizes and number of features. Further details about the datatasets can be found in Appendix C. ... please refer to Samadi et al. (2018) for Default Credit and Labeled Faces in the Wild (LFW) data sets, and Olfat & Aswani (2019) for others.
Dataset Splits Yes To emphasize the generalization capacity of each algorithm, we split each dataset into a training set and a test set with ratio of 30% 70% respectively... We find the best hyper-parameters by 3-fold cross validation, and prioritize the one giving minimum value of the summation (ABDiff.+ARE.).
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments (e.g., CPU, GPU models, or cloud computing instances).
Software Dependencies No The paper does not explicitly list any software dependencies with specific version numbers (e.g., 'Python 3.8' or 'PyTorch 1.9').
Experiment Setup Yes RFPCA. We search α {0.05, 0.1, 0.15} and λ {0., 0.5, 1., 1.5, 2.0, 2.5}. ... we set the number of iteration for our subgradient descent algorithm to τ = 1000 and also repeat the Riemannian descent for 20 randomly generated initial point U0. ... Fair PCA... we set the number of iterations to 1000... We search the learning rate η of the algorithm from set of 17 values evenly spaced in [0.25, 4.25] and {0.1}. ... CFPCA... we search δ from {0., 0.1, 0.3, 0.5, 0.6, 0.7, 0.8, 0.9}, and ... we fix δ = 0 while searching µ in {0.0001, 0.001, 0.01, 0.05, 0.5}.