Robust Gaussian Graphical Model Estimation with Arbitrary Corruption

Authors: Lingxiao Wang, Quanquan Gu

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our method through extensive experiments on both synthetic data and real-world genomic data. and Section 5 provides numerical results, for our method and a number of other methods, of some simulated datasets and a real example on gene expression data.
Researcher Affiliation Academia 1Department of Computer Science, University of Virginia, Charlottesville, Virginia, USA.
Pseudocode No No pseudocode or clearly labeled algorithm block was found.
Open Source Code No The paper mentions that 'The implementation of other baseline algorithms is based on R package huge1' and provides a link to an R package, but does not state that their own code is open-source or provide a link to it.
Open Datasets Yes In Section 5.2, 'Gene Expression Data', the paper states: 'we use the gene expression data of Arabidopsis thaliana, which was analyzed by Wille et al. (2004) and later on by Finegold & Drton (2011); Hirose & Fujisawa (2015)'. It also preprocesses it 'through R package limma3' with a footnote 'Available on http://bioconductor.org/packages/limma'.
Dataset Splits Yes We generate a dataset as the training sample, and an independent dataset from the same distribution as the test set. and For Robust CLIME, we set n2 = 10 and adopt 5-fold crossvalidation to choose the tuning parameter λ.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper mentions software packages like 'R package huge' and 'R package limma', but it does not specify their version numbers or any other software dependencies with specific version information.
Experiment Setup Yes For the arbitrary corruption model...we let np coordinates be arbitrarily corrupted, where we consider the corruption rate p = 0.1 for small number of corruptions and p = 0.2 for large number of corruptions. In addition, each corrupted coordinate is generated by normal distributions N(µ, σ) as follows: MA1 : µ = 1, σ = 1, MA2 : µ = 2, σ = 1. For point estimation: We set n2/n = 0.9 for t GLasso, β = 0.01 for RLL, and n2 = np for Robust CLIME. We also choose the tuning parameter λ by grid search based on its performance on the training sample and evaluate those estimators on the test set.