AI Paper Atlas

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

Online Ensemble Model Compression using Knowledge Distillation

Devesh Walawalkar, Zhiqiang Shen, Marios Savvides

ECCV 2020 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 68 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
train students with the online ensemble framework instead of individual baseline trainingcompressed students · CIFAR100 ResNet110, fifth studentaccuracy: 42.26 -> 46.76Improved
train students with the ensemble frameworkcompressed students · CIFAR10 ResNet110, fifth studentaccuracy: 92.27 -> 93.28Improved
train students with the ensemble frameworkcompressed students · ImageNet ResNet18, fifth studenttop-1 accuracy: 55.23 -> 58.52Improved
train students with the ensemble frameworkcompressed students · CIFAR100 ResNet32, fourth studentaccuracy: 61.85 -> 61.17Hurt
train students with the ensemble frameworkcompressed students · CIFAR10 and CIFAR100accuracy: about 1% better on averageImproved
train with ensemble framework on SVHN and ImageNetfourth and fifth students · SVHN and ImageNetaccuracy: around 3% better on averageImproved
use ensemble teacher versus KD-ONEensemble teacher · CIFAR100 ResNet-32error rate: 26.61 -> 26.09Improved
use ensemble teacher versus KD-ONEensemble teacher · SVHN ResNet-32error rate: 1.83 -> 1.97Hurt
raise cross-entropy ratio alpha from 0.3 to 0.7 (ablation)ensemble loss · CIFAR100 ResNet110weighted average accuracy: 161.71 -> 179.31Improved
raise alpha from 0.7 to 0.9 (ablation)ensemble loss · CIFAR100 ResNet110weighted average accuracy: 179.31 -> 171.18Hurt
set beta=gamma=0.15 instead of beta=0.2, gamma=0.1 (ablation)ensemble loss · CIFAR100 ResNet110weighted average accuracy: 164.37 -> 177.65Improved
set temperature T=2 instead of T=1 (ablation)distillation · CIFAR100 ResNet110mean student accuracy: 61.78 -> 63.32Improved

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