AI Paper Atlas

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

Lipschitz Continuity Guided Knowledge Distillation

Yuzhang Shang, Bin Duan, Ziliang Zong, Liqiang Nie, Yan Yan

ICCV 2021 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 33 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 Lipschitz continuity loss (lambda 0 -> 3.2) (ablation)KD, setup (a) · CIFAR-100 setup (a)top-1 error (%): 21.69 -> 20.33Improved
add Lipschitz continuity loss (lambda 0 -> 3.2) (ablation)KD, setup (b) · CIFAR-100 setup (b)top-1 error (%): 23.43 -> 21.48Improved
add Lipschitz continuity loss (lambda 0 -> 1.6) (ablation)KD, setup (c) · CIFAR-100 setup (c)top-1 error (%): 26.47 -> 23.56Improved
add Lipschitz continuity loss (lambda 0 -> 3.2) (ablation)KD, setup (d) · CIFAR-100 setup (d)top-1 error (%): 27.68 -> 23.78Improved
raise lambda from 3.2 to 6.4 (ablation)LONDON, setup (a) · CIFAR-100 setup (a)top-1 error (%): 20.33 -> 21.87Hurt
raise lambda from 3.2 to 6.4 (ablation)LONDON, setup (d) · CIFAR-100 setup (d)top-1 error (%): 23.78 -> 25.22Hurt
turn off Lipschitz continuity loss on setup (b) (ablation)LONDON · CIFAR-100 setup (b)validation accuracyImproved
use LONDON instead of OFDOFD, setup (a) · CIFAR-100 setup (a)top-1 error (%): 20.89 -> 20.33Improved
use LONDON instead of OFDOFD, setup (d) · CIFAR-100 setup (d)top-1 error (%): 24.44 -> 23.78Improved
use LONDON instead of OFDOFD · ImageNet val (ResNet152 -> ResNet50)top-1 error (%): 21.65 -> 21.12Improved
use LONDON instead of OFDOFD · ImageNet val (ResNet50 -> MobileNet)top-1 error (%): 28.75 -> 27.64Improved
use LONDON instead of OFDOFD · PASCAL VOC2007 test (SSD, ResNet18)mAP: 73.08 -> 73.82Improved

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