Fast OSCAR and OWL Regression via Safe Screening Rules

Authors: Runxue Bao, Bin Gu, Heng Huang

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on a variety of datasets show that our screening rule leads to a significant computational gain without any loss of accuracy, compared to existing competitive algorithms.
Researcher Affiliation Collaboration 1Department of Electrical and Computer Engineering, University of Pittsburgh, Pittsburgh, USA 2JD Finance America Corporation.
Pseudocode Yes Algorithm 1 Safe Screening Rule for OWL Regression with Iterative Strategy; Algorithm 2 Accelerated Proximal Gradient Descent Algorithm with Safe Screening Rules; Algorithm 3 Stochastic Proximal Gradient Descent Algorithm with Safe Screening Rules.
Open Source Code No The paper states that algorithms are implemented in MATLAB but does not provide a link or explicit statement about code availability.
Open Datasets Yes Duke Breast Cancer and Colon Cancer datasets are from the LIBSVM repository, which is available at https: //www.csie.ntu.edu.tw/ cjlin/libsvmtools/datasets/. Indoor Loc Longitude and Slice Localization datasets are from the UCI benchmark repository (Dua & Graff, 2017), which is available at https://archive.ics.uci.edu/ml/datasets.php.
Dataset Splits No To compare the prediction error of different algorithms, we randomly divide the dataset into the training and testing set in proportion to 4 : 1 and use root mean squared error (RMSE) as the performance criterion of the linear regression tasks. The paper specifies a training and testing split, but no explicit validation split.
Hardware Specification Yes Our experiments were performed on a 4-core Intel i7-6820 machine.
Software Dependencies No We implement all the algorithms in MATLAB. The paper mentions MATLAB but does not provide a specific version number or other software dependencies with versions.
Experiment Setup Yes tolerance error ϵ of duality gap and dual infeasibility in our experiments are set as 10 6. ... the hyperparameters of the size of mini-batch, the number of inner loop and step size η, which range from 5 to 100, 5 to 80 and 10 6 to 10 3 respectively for different datasets, are selected by grid search. ... we set pi = i e τ, i = 1, 2, 3, τ = 2 for Duke Breast Cancer, Indoor Loc Longitude and Slice Localization datasets and τ = 3 for Colon Cancer, Cardiac Left and Cardiac Right datasets.