Optimal Mean Robust Principal Component Analysis

Authors: Feiping Nie, Jianjun Yuan, Heng Huang

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Both theoretical analysis and empirical studies demonstrate our new methods can more effectively reduce data dimensionality than previous robust PCA methods. 5. Experimental Results The main goal of PCA is to reduce the dimensionality such that the reduced features represent and reconstruct the original data as good as possible. In the experiments, we show how well the reconstruction of the proposed new optimal mean robust PCA methods compared to the previous PCA and robust PCA methods.
Researcher Affiliation Academia Feiping Nie FEIPINGNIE@GMAIL.COM Jianjun Yuan WRIYJJ@GMAIL.COM Heng Huang HENG@UTA.EDU Computer Science and Engineering Department, University of Texas at Arlington, Arlington, TX, 76019
Pseudocode Yes Algorithm 1 Algorithm to solve the problem (13). Initialize D as an identity matrix while not converge do..., Algorithm 2 Algorithm to solve the problem (27). Initialize x C while not converge do..., Algorithm 3 Algorithm to solve the problem (30). Let 1 < ρ < 2. Initialize µ = 0.1, E = 0, Λ = 0 while not converge do
Open Source Code No The paper does not contain any explicit statement about releasing source code for the described methodology or a link to a code repository.
Open Datasets Yes In the experiments, we use 12 benchmark face image datasets, including AT&T (Samaria & Harter, 1994), UMIST (Graham & Allinson, 1998), Yale (face data), Yale B (Georghiades et al., 2001), Palm (Hou et al., 2009), CMU-PIE (Sim & Baker, 2003), FERET (Philips et al., 1998), MSRA, Coil (Nene et al., 1996), JAFFE, MNIST, and AR. We downloaded the image data from different websites.
Dataset Splits No The paper mentions randomly selecting 20% images and applying occlusions, but it does not provide specific details on how the dataset was partitioned into training, validation, and test sets, or specify any cross-validation setup.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments.
Software Dependencies No The paper does not provide specific software details, such as library names with version numbers, that would be needed to replicate the experiments.
Experiment Setup Yes In each dataset, we randomly select 20% images and randomly place a 1/4 size of occlusion in the selected images. The reconstruction error is calculated as P i xr i xo i 2,... We choose the range of γ based on the suggestion from (Wright et al., 2009), in which the γ is suggested with the scale of m 1 2 (m is the dimension of matrix Z). Considering the size of images used in our experiments, we select the range of γ from 30 to 90.