AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Enhancing Sharpness-Aware Optimization Through Variance Suppression

Bingcong Li, Georgios B. Giannakis

NeurIPS 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Confidence calibration objectives · 3 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
replace SAM with VaSSO (variance-suppressed adversary)SAM · CIFAR10 VGG-11-BNtest accuracy (%): 93.82 -> 94.10Improved
replace SAM with VaSSOSAM · CIFAR10 PyramidNet-110test accuracy (%): 97.85 -> 97.93Improved
replace SAM with VaSSOSAM · CIFAR100 ResNet-18test accuracy (%): 80.96 -> 81.30Improved
replace SAM with VaSSOSAM · CIFAR100 PyramidNet-110test accuracy (%): 85.60 -> 85.85Improved
replace SAM with VaSSOSAM · ImageNet ResNet-50test accuracy (%): 77.16 -> 77.42Improved
replace SAM with VaSSOSAM · ImageNet ViT-S/32test accuracy (%): 68.98 -> 69.54Improved
add VaSSO variance suppression to GSAM (V+G)GSAM · ImageNet ResNet-50test accuracy (%): 77.20 -> 77.48Improved
add VaSSO variance suppression to GSAM (V+G)GSAM · ImageNet ViT-S/32test accuracy (%): 69.42 -> 69.61Improved
replace SAM with VaSSO (theta=0.9)SAM · IWSLT-14 De-En Transformervalidation perplexity: 5.00 -> 5.00No clear effect
replace AdamW with VaSSO (theta=0.4)AdamW · IWSLT-14 De-En TransformerBLEU: 34.66 -> 34.88Improved
use smaller theta (0.4 instead of 0.9) in VaSSO (ablation)VaSSO · IWSLT-14 De-En TransformerBLEU: 34.81 -> 34.88Improved
replace SAM with VaSSO (theta=0.2) under label noiseSAM · CIFAR10 25% label noisetest accuracy (%): 96.39 -> 96.48Improved

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