Unlabeled Principal Component Analysis

Authors: Yunzhen Yao, Liangzu Peng, Manolis Tsakiris

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We assess our algorithmic pipeline on synthetic data, face images, educational and medical records, with encouraging results.
Researcher Affiliation Academia Yunzhen Yao, Liangzu Peng, Manolis C. Tsakiris School of Information Science and Technology Shanghai Tech University yaoyzh,penglz,mtsakiris@shanghaitech.edu.cn
Pseudocode Yes Algorithm 1 Two-stage Algorithmic Pipeline for UPCA
Open Source Code No The paper does not provide an explicit statement or link for the open-source code for the methodology described.
Open Datasets Yes We use the well-known database Extended Yale B [14]... The first dataset consists of the test scores of m = 707 high-school students... The second dataset consists of all the benign cases in Breast Cancer Wisconsin (Diagnostic) [12].
Dataset Splits No The paper describes data generation parameters and outlier ratios, but does not specify train/validation/test dataset splits in the context of model training for reproduction.
Hardware Specification Yes Experiments are run on an Intel(R) i7-8700K, 3.7 GHz, 16GB machine.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., programming languages, libraries, or frameworks).
Experiment Setup Yes We fix m = 50, n = 500. With dim S = r = 1 : 1 : 49, we sample S at random from the Grassmannian Gr(r, m). Then n points x j are sampled at random from the intersection of S with the unit sphere of Rm to yield X . Let nin be the number of inliers Xin and nout the number of outliers Xout, with nin + nout = n. We consider outlier ratios nout/n = 0.1 : 0.1 : 0.9.