ADOM: Accelerated Decentralized Optimization Method for Time-Varying Networks

Authors: Dmitry Kovalev, Egor Shulgin, Peter Richtarik, Alexander V Rogozin, Alexander Gasnikov

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we perform experiments with logistic regression for binary classification with ℓ2 regularization. That is, our loss function has the form j=1 log(1 + exp( bija ijx)) + r 2 x 2, (19) where aij Rd and bij { 1, +1} are data points and labels, r > 0 is a regularization parameter, and m is the number of data points stored on each node. In our experiments we use function sklearn.datasets.make classification from scikit-learn library for dataset generation. We generate a number of datasets consisting of 10,000 samples, distributed to the n = 100 nodes of the network with m = 100 samples on each node. We vary r to obtain different values of the condition number κ. We also vary the number of features d.
Researcher Affiliation Academia 1King Abdullah University of Science and Technology, Thuwal, Saudi Arabia 2Moscow Institute of Physics and Technology, Dolgoprudny, Russia 3Higher School of Economics, Moscow, Russia.
Pseudocode Yes Algorithm 1 PNGD: Projected Nesterov Gradient Descent; Algorithm 2 ADOM: Accelerated Decentralized Optimization Method
Open Source Code No The paper does not provide any statement or link indicating that the source code for the methodology described is publicly available.
Open Datasets No In our experiments we use function sklearn.datasets.make classification from scikit-learn library for dataset generation. We generate a number of datasets consisting of 10,000 samples, distributed to the n = 100 nodes of the network with m = 100 samples on each node.
Dataset Splits No The paper mentions generating datasets and distributing samples but does not specify any train/validation/test splits or percentages.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments.
Software Dependencies No The paper mentions using "sklearn.datasets.make classification from scikit-learn library" and cites "LIBSVM" but does not provide specific version numbers for these software components.
Experiment Setup Yes We vary r to obtain different values of the condition number κ. We also vary the number of features d. ... To calculate the dual gradient F (z) we use T = 3 steps of AGD in ADOM and T = 30 steps of AGD in DNM. ... We switch between 2 networks every t iterations, where t {50, 20, 10, 5}. ... We use T = 1 iterations of GD to calculate F (zk g) in ADOM.