Stochastic subgradient for composite convex optimization with functional constraints

Authors: Ion Necoara, Nitesh Kumar Singh

JMLR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 5 detailed numerical simulations are provided that support the effectiveness of our method in real problems. ... Tables 1 and 2 give the results of our algorithm SSP for robust (ρ > 0) and nominal (ρ = 0) classification formulations. We choose the parameters λ = 0.1, 0.2, 0.3, β = 1.96, and stopping criterion 10 2. We consider a dataset of CT scan images... We also resize all images into 190 × 190 pixels.
Researcher Affiliation Academia Ion Necoara EMAIL Automatic Control and Systems Engineering Department, University Politehnica Bucharest, Spl. Independentei 313, 060042 Bucharest, Romania. Gheorghe Mihoc-Caius Iacob Institute of Mathematical Statistics and Applied Mathematics of the Romanian Academy, 050711 Bucharest, Romania. Nitesh Kumar Singh EMAIL Automatic Control and Systems Engineering Department, University Politehnica Bucharest, Spl. Independentei 313, 060042 Bucharest, Romania.
Pseudocode Yes Algorithm 1 (SSP): Choose x0 ∈ Y and stepsizes αk > 0 and β ∈ (0, 2) For k ≥ 0 repeat: Sample independently ζk ∼ P1 and ξk ∼ P2 and update : vk = proxαkg(·,ζk) (xk − αk ∇f(xk, ζk)) (5) zk = vk − β (h(vk, ξk))+ ∇h(vk, ξk) 2 ∇h(vk, ξk) (6) xk+1 = ΠY(zk). (7)
Open Source Code No The codes were written in Matlab and run on a PC with i7 CPU at 2.1 GHz and 16 GB memory. (No specific link or statement about open-source release is provided for the authors' own code.)
Open Datasets Yes We consider a dataset of CT scan images having two classes, covid and non-covid, available at https://www.kaggle.com/plameneduardo/ sarscov2-ctscan-dataset. ... PMU-UD, sobar-72 and divorce datasets available on https://archive-beta.ics.uci. edu/ml/datasets; and the rest from LIBSVM library https://www.csie.ntu.edu.tw/ ~cjlin/libsvmtools/datasets/.
Dataset Splits Yes To implement our algorithm we have taken 1488 data in which 751 are of Covid patients and 737 of Non-Covid patients. Then, we divide them into training data and testing data. For training data we have taken 1240 images in which 626 are of Covid and 614 are of Non-Covid. For testing data we have taken 248 images in which 125 are of Covid and 123 of Non-Covid. ... We divided each dataset into 80% for training and 20% for testing.
Hardware Specification Yes The codes were written in Matlab and run on a PC with i7 CPU at 2.1 GHz and 16 GB memory.
Software Dependencies No The codes were written in Matlab and run on a PC with i7 CPU at 2.1 GHz and 16 GB memory. (The version of Matlab is not specified, nor are specific versions for other software components like the 'Matlab solver lsqlin' or 'LIBSVM library' mentioned later.)
Experiment Setup Yes We choose the parameters λ = 0.1, 0.2, 0.3, β = 1.96, and stopping criterion 10 2. ... Then, the robust sparse classification problem can be formulated as: min w,d,u λ i=1 ui + w 1 ... We choose the parameters λ = 0.1, 0.2, 0.3, β = 1.96, and stopping criterion 10 2. ... For the constrained least-squares problem, we consider an adaptive stepsize: αk = δ AT ζkxk − bζk 2 / Aζk(AT ζkxk − bζk) 2 , where δ ∈ (0, 2). ... Both algorithms were stopped when max(Ax − b , (Cx − d)+ ) ≤ 10 −3.