AI Paper Atlas

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

Direct Distillation between Different Domains

Jialiang Tang, Shuo Chen, Gang Niu, Hongyuan Zhu, Joey Tianyi Zhou, Chen Gong, Masashi Sugiyama

ECCV 2024 · Label-efficient and robust learning › Knowledge distillation · approach: Distillation with causal and adaptive guidance · 15 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 domain-invariant feature transfer loss (ablation)4Ds · Office-Homeaccuracy: 80.25 -> 82.77Improved
add categorical knowledge transfer loss (ablation)4Ds · Office-Homeaccuracy: 81.80 -> 82.77Improved
use fusion-activation weighting instead of no weights (ablation)4Ds · Office-Homeaccuracy: 80.07 -> 82.77Improved
use fusion-activation instead of block-wise plus channel-wise weights (ablation)4Ds · Office-Homeaccuracy: 81.83 -> 82.77Improved
use adapters on the teacher instead of fixed or fully learnable teacher (ablation)4Ds · Office-Homeaccuracy: 79.64 / 78.55 -> 82.77Improved
train adapted teacher with cross-entropy and knowledge transfer losses (ablation)4Ds · Office-Homeaccuracy: 81.22 / 81.86 -> 82.77Improved
use one-stage 4Ds instead of two-stage DAA-SHOTstudent ResNet18 · Average of four datasetsaccuracy: 85.61 -> 87.80Improved
apply two-stage NRC-based adaptation plus distillationstudent ResNet18 · Average of four datasetsaccuracy: 84.60 -> 82.77 (DAA-NRC)Hurt
apply traditional KD methods (VKD, Fitnets, VID, SPKD, SemCKD, SRRL) from a source-trained teacherstudent ResNet18 · Average of four datasetsaccuracy: 84.60 -> 80.48 to 83.10Hurt
use 4Ds instead of training the compact student on target datastudent ResNet18 · PACSaccuracy: 94.67 -> 97.65Improved

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