A Hybrid Variance-Reduced Method for Decentralized Stochastic Non-Convex Optimization

Authors: Ran Xin, Usman Khan, Soummya Kar

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments are provided to illustrate our main technical results. In this section, we illustrate our theoretical results on the convergence of the proposed GT-HSGD algorithm with the help of numerical experiments.
Researcher Affiliation Academia 1Department of Electrical and Computer Engineering, Carnegie Mellon University, Pittsburgh, PA, USA 2Department of Electrical and Computer Engineering, Tufts University, Medford, MA, USA.
Pseudocode Yes Algorithm 1 GT-HSGD at each node i
Open Source Code No The paper does not provide an explicit statement or link for the open-sourcing of the code for the described methodology.
Open Datasets Yes Data. To test the performance of the applicable decentralized algorithms, we distribute the a9a, covertype, KDD98, Mini Boo NE datasets uniformly over the nodes and normalize the feature vectors such that θi,j = 1, i, j. The statistics of these datasets are provided in Table 2. Table 2. Datasets used in numerical experiments, all available at https://www.openml.org/.
Dataset Splits No The paper describes using datasets but does not specify explicit train, validation, or test splits, nor does it refer to a standard splitting methodology.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. It only refers to a "network of n nodes" and "machines or edge devices".
Software Dependencies No The paper does not list specific software dependencies with version numbers, such as programming languages, libraries, or frameworks used for implementation.
Experiment Setup No The paper discusses parameter tuning procedures (e.g., finding step-size and minibatch size candidate sets) but does not provide the specific hyperparameter values or training configurations used for the final experiments presented in the results.