Stochastic Iterative Hard Thresholding for Graph-structured Sparsity Optimization

Authors: Baojian Zhou, Feng Chen, Yiming Ying

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments to show the efficiency and effectiveness of the proposed algorithms. We conduct experiments on both synthetic and real datasets to evaluate the performance of GRAPHSTOIHT.
Researcher Affiliation Academia 1Department of Computer Science, SUNY at Albany, Albany, NY, USA 2Department of Mathematics and Statistics, SUNY at Albany, Albany, NY, USA.
Pseudocode Yes Algorithm 1 GRAPHSTOIHT
Open Source Code Yes The source code and datasets are accessible at: https: //github.com/baojianzhou/graph-sto-iht.
Open Datasets Yes We apply it to the breast cancer dataset in Van De Vijver et al. (2002), which contains 295 training samples including 78 positives (metastatic) and 217 negatives (non-metastatic). We use the Protein-Protein Interaction (PPI) network in Jacob et al. (2009)4.
Dataset Splits Yes All related parameters are tuned by 5-fold-cross-validation on each training dataset. We tune b and η on an additional validation dataset with 100 observations.
Hardware Specification Yes All experiments are tested on 56 CPUs of Intel Xeon(R) E5-2680 with 251GB of RAM.
Software Dependencies No All codes are written in Python and C language. No specific version numbers for software libraries, frameworks, or languages are provided.
Experiment Setup Yes All methods terminate when Axt+1 y 10 7 (corresponding to convergence ) or t/n 500 (corresponding to the maximum number of epochs allowed). For GRAPHSTOIHT, one epoch contains n iterations. We fix the sparsity s = 8 and n = 180 and try different b from set {1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 180}. We consider 16 different η from set {0.1, 0.2, . . . , 1.5, 1.6}.