AI Paper Atlas

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

On Feature Learning in the Presence of Spurious Correlations

Pavel Izmailov, Polina Kirichenko, Nate Gruver, Andrew Gordon Wilson

NeurIPS 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Data augmentation and generalization bias · 22 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
apply DFR (retrain last layer on held-out balanced data) to ERM featuresERM, ResNet-50 · WaterbirdsWGA (%): 68.9 -> 91.1Improved
use group DRO / RWG / RWY training instead of ERM before applying DFRfeature extractor · Waterbirds, CelebA, FMOW, textDFR WGA: ERM+DFR 91.1 vs GDRO-ES+DFR 89.9 on WaterbirdsNo clear effect
early stopping for group-robust methods before DFR (ablation)group DRO, RWG, RWY · Waterbirds, CelebADFR WGA: GDRO 88.2 vs GDRO-ES 89.9 (Waterbirds)No clear effect
use stronger pretrained architecture (ConvNext Large IN22k) instead of ResNet-50 (ablation)ERM + DFR · WaterbirdsDFR WGA: 91.1 -> 97.2Improved
stronger base model on MultiNLI (DeBERTa-Large vs BERT-Base) (ablation)ERM + DFR · MultiNLIDFR WGA: 73.8 -> 84.8Improved
finetune feature extractor on Waterbirds vs only pretrained features (ablation)ResNet-50 IN1k + DFR · WaterbirdsDFR WGA: 88.2 -> 92.9Improved
finetune ConvNext Large IN22k on Waterbirds (ablation)ConvNext Large + DFR · WaterbirdsDFR WGA: 94 -> 94.3No clear effect
finetune feature extractor on target data (ablation)models on CelebA/FMOW · CelebADFR WGA: 88.3 -> 92.2Improved
use supervised ImageNet pretraining vs random init (ablation)ResNet-50 · Waterbirds, FMOW, CelebADFR WGAImproved
add weight decay (non-zero) before DFR (ablation)ResNet-50 IN1k · CelebAbase WGA vs DFR WGA: base WGA +~5%; DFR WGA best at zero weight decayMixed
MixUp augmentation (ablation)ResNet-50 · FMOWDFR WGA: 42% -> 30%Hurt
Random Erasing augmentation (ablation)ResNet-50 · WaterbirdsDFR WGAHurt

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