AI Paper Atlas

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

Distilling Holistic Knowledge With Graph Neural Networks

Sheng Zhou, Yucheng Wang, Defang Chen, Jiawei Chen, Xin Wang, Can Wang, Jiajun Bu

ICCV 2021 · Label-efficient and robust learning › Knowledge distillation · approach: Distillation with causal and adaptive guidance · 72 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 HKD holistic distillation on top of KDKD, ResNet32x4 -> ResNet8x4 · CIFAR100top-1 accuracy (%): 73.55 -> 76.13Improved
add HKD holistic distillation on top of KDKD, ResNet50 -> MobileNetV2 · CIFAR100top-1 accuracy (%): 67.83 -> 70.72Improved
add the KD loss to HKDHKD, ResNet32x4 -> ResNet8x4 · CIFAR100top-1 accuracy (%): 75.63 -> 76.13Improved
add the KD loss to HKDHKD, ResNet50 -> VGG8 · CIFAR100top-1 accuracy (%): 74.86 -> 74.88No clear effect
replace CRD+KD with HKD+KDstudent network, ResNet32x4 -> ShuffleNetV2 · TinyImageNettop-1 accuracy (%): 58.95 -> 59.31Improved
replace CRD with HKDstudent network, ResNet34 teacher to ResNet18 student · ImageNettop-1 accuracy (%): 53.76 -> 54.07Improved
replace CRD+KD with HKD+KDMobileNetV2 student transferred from CIFAR100 · TinyImageNet linear probeaccuracy (%): 28.98 -> 30.55Improved
replace CRD+KD with HKD+KDMobileNetV2 student transferred from CIFAR100 · STL-10 linear probeaccuracy (%): 65.87 -> 67.28Improved
add vanilla KD to the studentMobileNetV2 student transferred from CIFAR100 · TinyImageNet linear probeaccuracy (%): 23.01 -> 22.92No clear effect
use a two-layer GNN (L=2) instead of one layer (ablation)HKD+KD, ResNet32x4 -> ResNet8x4 · CIFAR100top-1 accuracy (%): 76.13 -> 76.05No clear effect
use a two-layer GNN (L=2) instead of one layer (ablation)HKD+KD, VGG13 -> MobileNetV2 · CIFAR100top-1 accuracy (%): 70.48 -> 70.28No clear effect
use a two-layer GNN (L=2) instead of one layer (ablation)HKD+KD, ResNet50 -> VGG8 · CIFAR100top-1 accuracy (%): 74.85 -> 74.82No clear effect

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