AI Paper Atlas

Research map / Optimization / Optimizers

[Re] AdaBelief Optimizer: Adapting Stepsizes by the Belief in Observed Gradients

Anirudh Buvanesh, Madhur Panwar

NeurIPS 2022 · Optimization › Optimizers · approach: Second-order and structured preconditioning

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 AdaBelief instead of AdamResNet18 · ImageNet valtop-1 accuracy: 63.79 -> 70.08Improved
use AdaBelief instead of SGDResNet18 · ImageNet valtop-1 accuracy: 70.23 -> 70.08No clear effect
use AdaBelief instead of AdaBoundResNet18 · ImageNet valtop-1 accuracy: 68.13 -> 70.08Improved
use AdaBelief instead of AdamSN-GAN · CIFAR-10FID: 13.05 -> 12.52Improved
use AdaBelief instead of RAdamSN-GAN · CIFAR-10FID: 12.70 -> 12.52No clear effect
use AdaBelief instead of SGDSN-GAN · CIFAR-10FID: 49.70 -> 12.52Improved
use AdaBelief instead of Adam (Padam p=1/2)small WGAN · CIFAR-10FID: 96.6 -> 83.0Improved
use AdaBelief instead of Adam (Padam p=1/2)small WGAN-GP · CIFAR-10FID: 73.5 -> 61.8Improved
decrease Padam p from 1/2 toward 0 (more SGD-like) (ablation)Padam · small WGANFID: 96.6 -> 469.3Hurt
use AdaBelief instead of Padam p=1/8ResNet18 · ImageNet valtop-1 accuracy: 70.07 -> 70.08No clear effect
use AdaBelief instead of AdamTransformer · IWSLT14 BLEUBLEU: 35.60 -> 35.85Improved
use AdaBelief instead of AdaHessianTransformer · IWSLT14 BLEUBLEU: 35.90 -> 35.85Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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