AI Paper Atlas

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

Efficient Knowledge Distillation from Model Checkpoints

Chaofei Wang, Qisen Yang, Rui Huang, Shiji Song, Gao Huang

NeurIPS 2022 · Label-efficient and robust learning › Knowledge distillation · approach: Distillation with causal and adaptive guidance · 9 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
use half-way intermediate teacher instead of fully converged teacherKD · CIFAR-100 WRN-40-2 -> MobileNetV2top-1 accuracy %: 68.03 -> 68.99Improved
use half-way intermediate teacher instead of fully converged teacherKD · CIFAR-100 ResNet-110 -> ResNet-32top-1 accuracy %: 72.48 -> 72.49No clear effect
use half-way intermediate teacher instead of fully converged teacherKD · ImageNet ResNet-34 -> ResNet-18top-1 accuracy %: 70.66 -> 70.86Improved
use best intermediate checkpoint instead of final teacherKD · CIFAR-100 WRN-40-2 -> WRN-40-1top-1 accuracy %: 72.68 -> 73.26Improved
use Snapshot Ensemble (full + own intermediate) instead of Full Ensemble teacherEnsemble KD · Average over CIFAR-100 and Tiny-ImageNettop-1 accuracy %: 66.11 -> 66.63Improved
use Snapshot Ensemble instead of Full Ensemble as teacherteacher ensemble · Average ensemble accuracytop-1 accuracy %: 73.06 -> 69.95Hurt
select teacher checkpoint by maximizing I(X;F)+I(Y;F) (ablation)KD · CIFAR-100 average of four pairstop-1 accuracy %: 70.46 -> 70.87Improved
use half-way checkpoint T0.5 instead of full teacher (ablation)KD · CIFAR-100 average of four pairstop-1 accuracy %: 70.46 -> 70.76Improved
use early checkpoint T0.3 (60th epoch) instead of full teacher (ablation)KD · CIFAR-100 average of four pairstop-1 accuracy %: 70.46 -> 69.78Hurt
increase Snapshot Ensemble size k at same size as Full Ensemble (ablation)Ensemble KD · CIFAR-100 WRN-40-2 -> WRN-40-1top-1 accuracy %Improved

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