AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

MetaDistiller: Network Self-Boosting via Meta-Learned Top-Down Distillation

Benlin Liu, Yongming Rao, Jiwen Lu, Jie Zhou, Cho-Jui Hsieh

ECCV 2020 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 37 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 meta-learned top-down soft targets for intermediate outputsResNet18 self-boosting · CIFAR-100top-1 accuracy: 78.25 -> 79.05Improved
add meta-learned top-down soft targetsResNet50 · CIFAR-100top-1 accuracy: 79.71 -> 80.85Improved
add meta-learned soft targetsVGG19(BN) · CIFAR-100top-1 accuracy: 74.27 -> 75.12Improved
add meta-learned soft targetsResNet18 · ImageNettop-1 accuracy: 69.84 -> 70.52Improved
add meta-learned soft targetsVGG19(BN) · ImageNettop-1 accuracy: 72.45 -> 73.40Improved
add meta-learned soft targetsResNet50 · ImageNettop-1 accuracy: 75.24 -> 76.11Improved
use MetaDistiller instead of a ResNet152 teacher with FitNetResNet50 student · CIFAR-100top-1 accuracy: 80.20 -> 80.85Improved
use MetaDistiller instead of a ResNet152 teacher with ATResNet18 student · CIFAR-100top-1 accuracy: 78.81 -> 79.05Improved
ensemble all intermediate outputs at inferenceMetaDistiller ResNet18 · CIFAR-100top-1 accuracy: 79.05 -> 80.05Improved
add generated soft targets on top of hard-label supervision (vs DSN) (ablation)ResNet18 · CIFAR-100top-1 accuracy: 78.01 -> 79.05Improved
add generated soft targets on top of hard-label supervision (vs DSN) (ablation)ResNet50 · CIFAR-100top-1 accuracy: 79.33 -> 80.85Improved
add generated soft targets on top of hard-label supervision (vs DSN) (ablation)ResNet101 · CIFAR-100top-1 accuracy: 80.54 -> 81.39Improved

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