AI Paper Atlas

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

Source-Free Adaptation to Measurement Shift via Bottom-Up Feature Restoration

Cian Eastwood, Ian Mason, Chris Williams, Bernhard Schölkopf

ICLR 2022 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Source-free domain adaptation · 3 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
Use bottom-up training rather than adapting all layers simultaneously (ablation)FR · EMNIST-DAaccuracy: 74.4±0.8 -> 86.1±0.1Improved
Use bottom-up training rather than adapting all layers simultaneously (ablation)FR · CIFAR-10-Caccuracy: 87.2 ± 0.7 -> 89.4 ± 0.2Improved
Use bottom-up training rather than adapting all layers simultaneously (ablation)FR · CIFAR-100-Caccuracy: 65.5 ± 0.2 -> 68.5 ± 0.2Improved
Use bottom-up training rather than adapting all layers simultaneously (ablation)FR · Camelyon17accuracy: 84.6±0.6 -> 84.5±0.8; 86.0±1.1 -> 87.0±1.2; 89.5±0.4 -> 89.7±0.5Mixed
Add marginal logit-distribution alignment (ablation)Feature Restoration · CIFAR-10-CaccuracyNo clear effect
Add marginal logit-distribution alignment (ablation)Feature Restoration · CIFAR-100-CaccuracyImproved
Use symmetric KL divergence instead of asymmetric KL divergence (ablation)Feature Restoration · CIFAR-10-CaccuracyNo clear effect
Use symmetric KL divergence instead of asymmetric KL divergence (ablation)Feature Restoration · CIFAR-100-CaccuracyImproved
Align source and target features rather than only replacing batch-normalization statisticsFeature Restoration · EMNIST-DAaccuracy: 46.2±1.1 -> 74.4±0.8Improved
Use feature-distribution restoration instead of entropy-minimizing SHOTBUFR · EMNIST-DAECE: 29.6±3.7 -> 4.7±0.2Improved
Adapt features with BUFR instead of entropy-minimizing SHOTBUFR · EMNIST-DA severeaccuracy: 55.1±23.5 -> 84.6±0.2Improved
Correct label shift before applying feature restoration (ablation)FR · VisDA-C with ImageNet pretrainingaverage accuracy: 73.7 -> 82.8Improved

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