AI Paper Atlas

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

Don’t blame Dataset Shift! Shortcut Learning due to Gradients and Cross Entropy

Aahlad Manas Puli, Lily H Zhang, Yoav Wald, Rajesh Ranganath

NeurIPS 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 0 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 confidence control (SD, sigma-damp, norm, sigma-stitch) to ERMERM · Waterbirds, CelebA, CivilCommentstest worst-group accuracy: 3-10% on Waterbirds, 4-9% on CelebA, 7-10% on CivilCommentsImproved
Use spectral decoupling instead of ERMERM · Waterbirdstest worst-group accuracy: 70.8 -> 80.5Improved
Use confidence control instead of CnCCnC · Waterbirdstest worst-group accuracy: 68.0 -> 80.5Improved
Use SD instead of CnCCnC · CivilCommentstest worst-group accuracy: 68.8 -> 70.1Improved
Use confidence control instead of JTTJTT · CelebAtest worst-group accuracy: 75.2 -> 81.4Improved
Use large learning rate and weight decay in ERM (ablation)ERM log-loss · CelebAtest worst-group accuracy: 56% -> 72.8%Improved
Shrink validation set to 99 samples (1/200) (ablation)ERM · CelebAtest worst-group accuracy: more than 20%Hurt
Use sigma-stitch with only 99 validation samples (ablation)sigma-stitch · CelebAtest worst-group accuracy: 76.2%Improved
Use confidence control instead of ERM in pure setting (no group labels)ERM · CelebA and Waterbirdstest worst-group accuracy: 3-8%Improved
Use SD instead of ERM in pure settingERM · Waterbirds pure settingtest worst-group accuracy: 70.9 -> 78.7Improved
Use SD_log instead of ERM in pure settingERM · CivilComments pure settingtest worst-group accuracy: 60.8 -> 67.4Improved
Use SD instead of ERM in pure setting on CivilCommentsERM · CivilComments pure settingtest worst-group accuracy: 60.8 -> 60.4No 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