Fast Stochastic Composite Minimization and an Accelerated Frank-Wolfe Algorithm under Parallelization

Authors: Benjamin Dubois-Taine, Francis Bach, Quentin Berthet, Adrien Taylor

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide theoretical convergence rates and also present numerical results to demonstrate the performance of our proposed algorithms. The numerical experiments include different variants of our proposed algorithms and compare them with existing ones on both synthetic and real-world datasets.
Researcher Affiliation Academia Department of Computer Science and Engineering, The Chinese University of Hong Kong, Shenzhen, P.R. China
Pseudocode Yes Algorithm 1: FSMC Algorithm 2: Accelerate Frank-Wolfe Algorithm (AFW) for SCM
Open Source Code No No explicit statement or link providing access to the source code for the methodology described in this paper.
Open Datasets Yes We use the a9a dataset and the w8a dataset which are commonly used benchmark datasets.
Dataset Splits Yes We randomly choose 80% of the training data from both a9a and w8a datasets as the training set and the remaining 20% as the test set.
Hardware Specification Yes All numerical experiments are performed on a desktop PC with an Intel Core i7-3210M CPU, 2.50GHz and 8GB RAM.
Software Dependencies No The algorithms are implemented in Python 3.8. This only provides a programming language version, not specific library or solver versions as required.
Experiment Setup Yes We choose the initial step size α0 = 10−1 and the parameter γ = 100 for Algorithm 1 and Algorithm 2 respectively. The batch size for SGD and FSMC-VR are set to 500 in this experiment. The initial point is set to be x0 = 0.