Efficient Decentralized Stochastic Gradient Descent Method for Nonconvex Finite-Sum Optimization Problems

Authors: Wenkang Zhan, Gang Wu, Hongchang Gao9006-9013

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we have conducted extensive experiments and the experimental results have confirmed the superior performance of our proposed method.
Researcher Affiliation Collaboration Wenkang Zhan1, Gang Wu2, Hongchang Gao1 1Department of Computer and Information Sciences, Temple University, PA, USA 2Adobe Research, CA, USA wenkang.zhan@temple.edu, gawu@adobe.com, hongchang.gao@temple.edu
Pseudocode Yes Algorithm 1: Efficient Decentralized Stochastic Gradient Descent Method (EDSGD)
Open Source Code No The paper does not provide an explicit statement or link to open-source code for the described methodology.
Open Datasets Yes In our experiments, we use six classification datasets, which are available at LIBSVM1 and Open ML2. 1https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/ 2https://www.openml.org
Dataset Splits No The paper mentions distributing samples to workers and each worker using its own dataset, but it does not specify explicit training/validation/test splits, percentages, or a cross-validation setup.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup Yes The regularization coefficient γ is set to 0.001. ... we set the batch size of the first three methods to 256 and DGET to n. As for GT-SARAH and our method, we set it to sqrt(n/K). Similar to (Sun, Lu, and Hong 2020), we set the learning rate to 0.001 for all methods.