AI Paper Atlas

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

Multi-Label Knowledge Distillation

Penghui Yang, Ming-Kun Xie, Chen-Chen Zong, Lei Feng, Gang Niu, Masashi Sugiyama, Sheng-Jun Huang

ICCV 2023 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 29 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 multi-label logits distillation (MLD) to the student (ablation)ResNet-34 student · MS-COCOmAP: 70.31 -> 70.68Improved
add class-aware label-wise embedding distillation (CD) on top of MLD (ablation)L2D · MS-COCOmAP: 70.68 -> 71.91Improved
add instance-aware label-wise embedding distillation (ID) on top of MLD (ablation)L2D · MS-COCOmAP: 70.68 -> 71.79Improved
combine MLD, CD and ID (ablation)L2D · MS-COCOmAP: 71.91 -> 72.87Improved
use conventional feature-based KD (RKD, PKT, ReviewKD) for multi-labelResNet-34 student · MS-COCOmAP: 70.31 -> 70.39 (ReviewKD)No clear effect
use L2D instead of the best baseline PS on different architecturesSwin-T teacher, MobileNet v2 student · MS-COCOmAP: 72.42 -> 74.21Improved
distill with L2D on same-architecture WRN pairWRN-50 student · MS-COCOmAP: 74.45 -> 76.61Improved
apply L2D when the vanilla student outperforms the teacher (reversed KD)ResNet-101 student · MS-COCOmAP: 73.98 -> 75.51Improved
apply L2D on VOC with fewer labels per imageResNet-18 student · VOC2007mAP: 84.01 -> 85.71Improved
L2D balancing parameters variedL2D · MS-COCOmAPNo 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