Group Sparse Optimization via lp,q Regularization
Authors: Yaohua Hu, Chong Li, Kaiwen Meng, Jing Qin, Xiaoqi Yang
JMLR 2017 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In the aspect of application, we present some numerical results on both the simulated data and the real data in gene transcriptional regulation. ... The purpose of this section is to carry out the numerical experiments of the proposed PGM for the ℓp,q regularization problem. We illustrate the performance of the PGM-GSO among different types of ℓp,q regularization, in particular, when (p, q) =(2,1), (2,0), (2,1/2), (1,1/2), (2,2/3) and (1,2/3), by comparing them with several state-of-the-art algorithms for simulated data and applying them to infer gene regulatory network from gene expression data of mouse embryonic stem cell. |
| Researcher Affiliation | Academia | Yaohua Hu EMAIL College of Mathematics and Statistics Shenzhen University Shenzhen 518060, P. R. China; Chong Li EMAIL School of Mathematical Sciences Zhejiang University Hangzhou 310027, P. R. China; Kaiwen Meng EMAIL School of Economics and Management Southwest Jiaotong University Chengdu 610031, P. R. China; Jing Qin EMAIL School of Life Sciences The Chinese University of Hong Kong Shatin, New Territories, Hong Kong and Shenzhen Research Institute The Chinese University of Hong Kong Shenzhen 518057, P. R. China; Xiaoqi Yang EMAIL Department of Applied Mathematics The Hong Kong Polytechnic University Kowloon, Hong Kong |
| Pseudocode | Yes | Algorithm 1 (PGM-GSO) Select a stepsize v, start with an initial point x0 Rn, and generate a sequence {xk} Rn via the iteration |
| Open Source Code | Yes | The R package of the proximal gradient method for solving group sparse optimization, named GSpar O in CRAN, is available at https://CRAN.R-project.org/package=GSpar O |
| Open Datasets | Yes | The transcriptome data in m ESCs for gene regulatory network inference are downloaded from Gene Expression Omnibus (GEO). 245 experiments under perturbations in m ESC are collected from three papers Correa-Cerro et al. (2011); Nishiyama et al. (2009, 2013). Each experiment produced transcriptome data with or without overexpression or knockdown of a gene, in which the control and treatment have two replicates respectively. |
| Dataset Splits | No | The paper describes generating simulated data and using 'golden standards' for real data evaluation, but it does not specify explicit training/test/validation splits for model training. The real data analysis uses 'two independent golden standards' to validate inferred networks, implying an evaluation setup rather than distinct dataset splits for model learning stages. |
| Hardware Specification | Yes | All numerical experiments are implemented in Matlab R2013b and executed on a personal desktop (Intel Core Duo E8500, 3.16 GHz, 4.00 GB of RAM). |
| Software Dependencies | Yes | All numerical experiments are implemented in Matlab R2013b |
| Experiment Setup | Yes | We choose the stepsize v = 1/2 in the tests of the PGM-GSO. Two key criteria to characterize the performance are the relative error x x 2/ x 2 and the successful recovery rate, where the recovery is defined as success when the relative error between the recovered data and the true data is smaller than 0.5%; otherwise, it is regarded as failure. We carry out six experiments with the initial point x0 = 0 (unless otherwise specified). |