AI Paper Atlas

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

Similarity-Preserving Knowledge Distillation

Frederick Tung, Greg Mori

ICCV 2019 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 1,227 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 similarity-preserving loss to student trainingWideResNet-16-1 student, WRN-16-2 teacher · CIFAR-10test error (%): 8.74 -> 7.52Improved
add SP loss versus attention transferWideResNet-40-2 student, WRN-16-8 teacher · CIFAR-10test error (%): AT 4.47 vs SP 4.55Hurt
use activations from the end of each block instead of only the last conv layer (ablation)SP distillation · CIFAR-10test errorNo clear effect
use post-softmax scores instead of activations to compute similarity (ablation)SP distillation · CIFAR-10test errorHurt
add SP loss to fine-tuning of a pretrained studentMobileNet-0.25 (teacher MobileNet-0.5) · Describable Texturestest error (%): 42.45 -> 41.30Improved
add SP loss to fine-tuning of a pretrained studentMobileNetV2-0.35 (teacher MobileNetV2-1.0) · Describable Texturestest error (%): 41.25 -> 40.29Improved
use SP instead of attention transferMobileNet/MobileNetV2 students · Describable Texturestest error (%): SP won 19 of 30 MobileNet splits and 24 of 30 MobileNetV2 splitsImproved
add SP loss to student trainingShuffleNetV2-0.5 (teacher ShuffleNetV2-1.0) · CINIC-10test error (%): 20.09 -> 18.56Improved
combine SP with attention transferShuffleNetV2-0.5 (teacher ShuffleNetV2-1.0) · CINIC-10test error (%): 18.50 -> 18.20Improved
combine SP with traditional KDShuffleNetV2-0.5 (teacher ShuffleNetV2-1.0) · CINIC-10test error (%): KD 18.62 -> KD+SP 18.35Mixed
combine SP with attention transferMobileNetV2-0.35 (teacher MobileNetV2-1.0) · CINIC-10test error (%): 16.10 -> 15.66Improved
vary SP weight gamma from 10 to 64,000 (ablation)SP distillation · CINIC-10top-1 errorNo 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