AI Paper Atlas

Research map / Optimization / Distributed and compressed

GIANT: Globally Improved Approximate Newton Method for Distributed Optimization

Shusen Wang, Fred Roosta, Peng Xu, Michael W. Mahoney

NeurIPS 2018 · Optimization › Distributed and compressed · approach: Decentralized optimization algorithms · 30 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use GIANT instead of AGD, DANE, or L-BFGSDistributed logistic regression · MNIST8M, Epsilon, and Covtypetraining objective and test classification errorImproved
Increase the node count from 75 to 375GIANT · Covtypeper-iteration time increase: 19%Hurt
Increase the node count from 75 to 375AGD · Covtypeper-iteration time increase: 367%Hurt
Increase the node count from 75 to 375L-BFGS · Covtypeper-iteration time increase: 112%Hurt
Scale the dataset while increasing compute nodes to keep local computation per iteration the sameGIANT and DANE · Augmented datasets (n, 5n, and 25n)objective convergenceNo clear effect
Tune the fixed line-search parametersGIANT, L-BFGS, and DANE · Logistic regression experimentsconvergenceNo clear effect
Allow an inexact local linear-system solution with error on the order of the sampling errorGIANT · Smooth strongly convex objectivesconvergence rateNo clear effect
Use Nesterov acceleration in DANE to form AIDEDANE · Off-line small-scale simulationsconvergenceHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy