AI Paper Atlas

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

A Theory of Label Propagation for Subpopulation Shift

Tianle Cai, Ruiqi Gao, Jason Lee, Qi Lei

ICML 2021 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Conditional distribution alignment · 8 citations (OpenAlex)

Read the paper

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
Replace distribution matching (MDD) with consistency-based FixMatch label propagation on ENTITY-30 subpopulation shiftSwAV-finetuned ResNet50 on ENTITY-30 · ENTITY-30 UDA targetTarget Acc: +8.65 over MDD (72.60 vs 63.95)Improved
Add FixMatch consistency-based adaptation on top of source-only trainingSwAV-finetuned ResNet50 on ENTITY-30 · ENTITY-30 UDA targetTarget Acc: +15.87 over source-only (72.60 vs 56.73)Improved
Add FixMatch finetuning on top of MDD on Office-31MDD · Office-31 (6 transfer tasks)average accuracy: +0.68 (89.84 vs 89.16)Improved
Add FixMatch finetuning on top of MDD on Office-HomeMDD · Office-Home (12 transfer tasks)average accuracy: +1.6 (69.6 vs 68.0)Improved
Add FixMatch on top of source-trained model: source accuracy under FixMatch decreases slightlyFixMatch on ENTITY-30 · ENTITY-30 UDA sourceSource Acc: -1.04 vs source-only (90.87 vs 91.91)Hurt
Add DANN domain-adversarial training on ENTITY-30SwAV-finetuned ResNet50 · ENTITY-30 UDA targetTarget Acc: +4.3 over source-only (61.03 vs 56.73), with high variance (±4.63)Improved
Add MDD distribution matching on ENTITY-30SwAV-finetuned ResNet50 · ENTITY-30 UDA targetTarget Acc: +7.22 over source-only (63.95 vs 56.73)Improved

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