AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Learning Debiased Representation via Disentangled Feature Augmentation

Jungsoo Lee, Eungyeup Kim, Juyoung Lee, Jihyeon Lee, Jaegul Choo

NeurIPS 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Data augmentation and generalization bias · 46 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 disentangled feature swapping augmentation (ablation)Debiased classifier · Colored MNIST, 1.0% bias-conflictingaccuracy (%): 74.03±2.40 -> 72.29±3.82Improved
Add disentangled feature swapping augmentation (ablation)Debiased classifier · Corrupted CIFAR-10, 1.0% bias-conflictingaccuracy (%): 27.73±1.02 -> 32.81±2.47Improved
Add disentangled feature swapping augmentation (ablation)Debiased classifier · BFFHQ, 0.5% bias-conflictingaccuracy (%): 59.4±2.46 -> 61.27±3.26Improved
Schedule feature augmentation after 10K iterations instead of augmenting from the beginning (ablation)Disentangled feature augmentation · Colored MNIST, 1.0% bias-conflictingaccuracy (%): 72.29±3.82 -> 81.73±2.34Improved
Schedule feature augmentation after 10K iterations instead of augmenting from the beginning (ablation)Disentangled feature augmentation · Corrupted CIFAR-10, 1.0% bias-conflictingaccuracy (%): 32.81±2.47 -> 52.31±1.00Improved
Schedule feature augmentation after 10K iterations instead of augmenting from the beginning (ablation)Disentangled feature augmentation · BFFHQ, 0.5% bias-conflictingaccuracy (%): 61.27±3.26 -> 63.87±0.31Improved
Increase bias-conflicting sample diversity from 1% to 5%, sampling 50% per minibatchTraining-data sampling · Colored MNIST unbiased testaccuracy (%): 67.19±1.99 -> 83.77±2.03Improved
Use diverse samples from the 5% pool at 1% sampling instead of limited-diversity samples from the 1% pool at 50% samplingTraining-data sampling · Colored MNIST unbiased testaccuracy (%): 67.19±1.99 -> 77.97±6.00Improved
Increase bias-conflicting sample diversity from 1% to 5%, sampling 50% per minibatchTraining-data sampling · Corrupted CIFAR-10 unbiased testaccuracy (%): 33.08±0.80 -> 46.99±0.82Improved
Use diverse samples from the 5% pool at 1% sampling instead of limited-diversity samples from the 1% pool at 50% samplingTraining-data sampling · Corrupted CIFAR-10 unbiased testaccuracy (%): 33.08±0.80 -> 36.66±0.55Improved
Use feature augmentation from the beginning rather than scheduling it (ablation)Disentangled feature augmentation · All three evaluation datasetsaccuracy (%)Hurt
Use the proposed method rather than the strongest baselineColored MNIST, 0.5%–5.0% bias-conflicting · Colored MNIST unbiased testaccuracy (%): 70.47±1.84 vs 65.22±4.41; 87.4±0.78 vs 81.73±2.34; 92.91±0.15 vs 84.79±0.95; 96.96±0.04 vs 89.66±1.09Mixed

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