AI Paper Atlas

Research map / Optimization / Optimizers

Towards Understanding Why Lookahead Generalizes Better Than SGD and Beyond

Pan Zhou, Hanshu YAN, Xiaotong Yuan, Jiashi Feng, Shuicheng YAN

NeurIPS 2021 · Optimization › Optimizers · approach: Adam convergence and adaptive learning rates · 1 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
add local regularizer beta_q/2||theta-theta_{q-1}||^2 to stagewise lookahead (SLRLA vs SLA)SLA · CIFAR10 ResNet18accuracy (%): 95.27 -> 95.47Improved
add local regularizer to stagewise lookahead (SLRLA vs SLA)SLA · CIFAR100 VGG16accuracy (%): 70.20 -> 70.63Improved
add local regularizer to stagewise lookahead (SLRLA vs SLA)SLA · ImageNet ResNet18accuracy (%): 70.30 -> 70.47Improved
add local regularizer to stagewise lookahead (SLRLA vs SLA)SLA · CIFAR10 WRN-16-10accuracy (%): 95.73 -> 96.08Improved
add lookahead (SLA) on top of stagewise SGDStagewise SGD · CIFAR100 WRN-16-10accuracy (%): 78.95 -> 79.54Improved
use SLRLA instead of adaptive optimizer AdamAdam · CIFAR100 ResNet18accuracy (%): 76.88 -> 78.58Improved
use SLRLA instead of adaptive optimizer AdamWAdamW · CIFAR10 VGG16accuracy (%): 90.75 -> 92.63Improved
use SLRLA instead of AdaBeliefAdaBelief · ImageNet ResNet18accuracy (%): 70.08 -> 70.47Improved
increase lookahead alphalookahead and SLA · regularized softmax MNISTtraining loss and test lossMixed
increase alpha for SLRLASLRLA · regularized softmax MNISTtest accuracyMixed
tune regularization constant beta_q over a large rangeSLRLA · ImageNetaccuracyNo clear effect
use stagewise LR decay with lookahead rather than linear decay (theory)lookahead · regularized softmax MNISTexcess riskImproved

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