AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

Weak Adaptation Learning: Addressing Cross-Domain Data Insufficiency With Weak Annotator

Shichao Xu, Lixu Wang, Yixuan Wang, Qi Zhu

ICCV 2021 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Conditional distribution alignment · 9 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
train with weak annotator labels plus WAL stages instead of target data onlyB_t · CIFAR-10accuracy: 30.46 -> 61.71Improved
apply WAL instead of fine-tuning all parametersB_f2 · M->S digitsaccuracy: 77.92 -> 80.00Improved
apply WAL instead of target-only trainingB_t · VisDA-Caccuracy: 32.86 -> 40.83Improved
apply WAL instead of best SSDA method FANFAN · VisDA-Caccuracy: 32.99 -> 40.83Improved
fine-tune only last three layers on weak-label pretrained modelB_f1 · S->M digitsaccuracy: 94.79 -> 80.41Hurt
fine-tune all layers instead of last three (ablation)B_f1 · VisDA-Caccuracy: 27.67 -> 35.03Improved
use a much weaker annotator (23.79 accuracy) (ablation)WAL · CIFAR-10accuracy: 61.71 -> 42.29Hurt
increase number of labeled target samples (ablation)WAL · CIFAR-10accuracyImproved
increase accuracy of weak annotator (ablation)WAL · CIFAR-10accuracyImproved
add unlabeled target samples via SSDA methods (FAN, MME, ENT)SSDA baselines · VisDA-Caccuracy: 32.99No 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