Matrix Completion with Noisy Entries and Outliers

Authors: Raymond K. W. Wong, Thomas C. M. Lee

JMLR 2017 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Its promising empirical performance is demonstrated via a sequence of simulation experiments, including image inpainting. [...] Two sets of numerical experiments and a real data application were conducted to evaluate the practical performances of the proposed methodology.
Researcher Affiliation Academia Raymond K. W. Wong EMAIL Department of Statistics Texas A&M University College Station, TX 77843 USA; Thomas C. M. Lee EMAIL Department of Statistics University of California Davis, CA 95616, USA
Pseudocode Yes Details of this algorithm based on pseudo data matrix are given in Algorithm 1. [...] Algorithm 1 The General Robust Algorithm; Algorithm 2 Robust-Impute
Open Source Code No The paper describes two algorithms (Algorithm 1 and Algorithm 2) and evaluates their performance, but does not provide any explicit statement about releasing the source code or a link to a code repository.
Open Datasets Yes In this experiment the target matrix is the so-called Lena image that has been used by many authors in the image processing literature. [...] In this application the target matrix is an image from a Landsat Thematic Mapper data set publicly available at http://ternauscover.science.uq.edu.au/.
Dataset Splits Yes To evaluate the recovered matrix, the observed pixels were split into training, validation and testing sets consisting 80%, 10% and 10% of the observed (nonzero) entries respectively.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments. It only describes the algorithms and their empirical performance.
Software Dependencies No The paper mentions implementing algorithms (Algorithm 1 and Algorithm 2) and comparing them to Soft-Impute, but does not specify any programming languages, libraries, or software dependencies with version numbers.
Experiment Setup Yes For each simulated data set, the target matrix was generated as X0 = UV , where U and V are random matrices of size 100 r with independent standard normal Gaussian entries. Then each entry of X0 is contaminated by additional independent Gaussian noise with standard deviation σ, which is set to a value such that the signal-to-noise ratio (SNR) is 1. [...] In this study, we used two values for r (5, 10), three values for p (0, 0.05, 0.1) and three values for q (0.25, 0.5, 0.75). [...] The average training and testing errors of the recovered images of matrix ranks 50, 75, 100 and 125 are reported in Table 1.