AI Paper Atlas

Research map / Optimization / Optimizers

GGD: Grafting Gradient Descent

Yanjing Feng, Yongdao Zhou

JMLR 2024 · Optimization › Optimizers · approach: SGD momentum and step-size schedules

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 grafting gradient instead of Adam's sampled gradient (GAdam)Adam · MNISTmin squared L2 norm of full gradient: 5.1588 -> 5.2779 (x1e-3)Hurt
use grafting gradient instead of mini-batch SGD gradient (GGD)MBSGD · MNISTmin squared L2 norm of full gradient: 5.3185 -> 5.8986 (x1e-3)Hurt
use grafting gradient instead of SVRG gradient (GSVRG)SVRG · MNISTmin squared L2 norm of full gradient: 0.13407 -> 0.15203 (x1e-3)Hurt
use grafting gradient instead of Adam's sampled gradient (GAdam)Adam · CIFAR-10min squared L2 norm of full gradient: 1.6565 -> 1.4659 (x1e-4)Improved
use grafting gradient instead of mini-batch SGD gradient (GGD)MBSGD · CIFAR-10min squared L2 norm of full gradient: 634.79 -> 2681.0 (x1e-4)Hurt
use grafting gradient instead of SVRG gradient (GSVRG)SVRG · CIFAR-10min squared L2 norm of full gradient: 2.0849 -> 3.7647 (x1e-4)Hurt
use GGD instead of vanilla SGDvanilla SGD · ijcnn1, a9a, covtype, rcv1 (logistic regression)train lossImproved
use GGD with diminishing stepsize instead of SGD-IS or MBSGDSGD-IS / MBSGD · ijcnn1, a9a, covtype, rcv1 (logistic regression)train lossMixed
use GGD-WR-Adam instead of AdamAdam · ijcnn1, a9a, covtype, rcv1 (logistic regression)train lossNo clear effect
use GGD-WR-SVRG instead of SVRGSVRG · covtype (logistic regression)train lossNo clear effect
use GGD-WR-SVRG instead of SVRGSVRG · ijcnn1 (logistic regression)train lossImproved
use GGD, GAdam, GSVRG instead of MBSGD, Adam, SVRGMBSGD / Adam / SVRG · MNIST (LeNet-5)train lossImproved

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