Non-Gaussian Discriminative Factor Models via the Max-Margin Rank-Likelihood

Authors: Xin Yuan, Ricardo Henao, Ephraim Tsalik, Raymond Langley, Lawrence Carin

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on benchmark and real data demonstrate superior performance of the proposed model and its potential for applications in computational biology. ... We present numerical results on two benchmark (USPS and MNIST) and two real (gene expression and RNA sequencing) datasets, using part of or all methods summarized in Table 1; inference is performed via VB.
Researcher Affiliation Academia Xin Yuan EIEXYUAN@GMAIL.COM Ricardo Henao R.HENAO@DUKE.EDU Ephraim L. Tsalik E.T@DUKE.EDU Duke University, Durham, NC, 27708, USA Raymond J. Langley RLANGLEY@LRRI.ORG Department of Immunology, Lovelace Respiratory Research Institute, Albuquerque, NM 87108, USA Lawrence Carin LCARIN@DUKE.EDU Duke University, Durham, NC, 27708, USA
Pseudocode No The paper describes inference procedures and conditional posterior distributions, but it does not include a dedicated pseudocode block or an algorithm section labeled as such.
Open Source Code No All code used in the experiments was written in Matlab and executed on a 3.3GHz desktop with 16Gb RAM.
Open Datasets Yes USPS and MNIST are well-known public datasets. For the gene expression data, the paper states: "newly published tuberculosis study from Anderson et. al. (2014), consisting of gene expression intensities for 47323 genes and 334 subjects (GEO accession series GSE39941)." For RNA sequencing: "new RNA sequencing (RNAseq) sepsis study (Langley et al., 2013)." These include specific citations and accession numbers for public data.
Dataset Splits Yes USPS: "Here we use the resampled version, which is 767 images for model fitting and the remaining 773 for testing." MNIST: "The dataset is composed by 11552 training images and 1902 test images." For gene expression and RNA sequencing: "10-fold cross-validation".
Hardware Specification Yes All code used in the experiments was written in Matlab and executed on a 3.3GHz desktop with 16Gb RAM.
Software Dependencies No The paper states: "All code used in the experiments was written in Matlab". However, it does not provide a specific version number for Matlab or any other key software dependencies used in the experiments.
Experiment Setup Yes Everywhere we set K = 20, T = 5 and performance measures were averaged over 5 repetitions (standard deviations are also presented). ... In our experiments we set ϵ = 0.05... and set ψs = 1.1 and ψr = 0.001 (i.e., a non-informative prior).