AI Paper Atlas

Research map / Learning theory and algorithms / Sample complexity

On-Demand Sampling: Learning Optimally from Multiple Distributions

Nika Haghtalab, Michael Jordan, Eric Zhao

NeurIPS 2022 · Learning theory and algorithms › Sample complexity · approach: Sample complexity of ranking and selection · 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
replace GDRO with R-MDLResNet-50 · CelebA, standard regularizationworst-group accuracy (%): 41.7 -> 88.9Improved
replace GDRO with R-MDLResNet-50 · Waterbirds, standard regularizationworst-group accuracy (%): 76.9 -> 86.4Improved
replace GDRO with R-MDLBERT · MultiNLI, standard regularizationworst-group accuracy (%): 66.6 -> 70.3Improved
replace GDRO with R-MDLResNet-50 · Waterbirds, strong regularizationworst-group accuracy (%): 84.6 -> 89.4Improved
replace GDRO with R-MDLResNet-50 · CelebA, strong regularizationworst-group accuracy (%): 86.7 -> 88.8Improved
replace GDRO with R-MDLBERT · MultiNLI, early stoppingworst-group accuracy (%): 77.7 -> 43.1Hurt
replace GDRO with R-MDLResNet-50 · CelebA, early stoppingworst-group accuracy (%): 88.3 -> 90.6Improved
remove strong weight-decay regularizationGDRO · Waterbirdsworst-group accuracy (%): 84.6 -> 76.9Hurt
remove strong weight-decay regularizationGDRO · CelebAworst-group accuracy (%): 86.7 -> 41.7Hurt
remove strong weight-decay regularizationR-MDL · Waterbirdsworst-group accuracy (%): 89.4 -> 86.4No clear effect
remove strong weight-decay regularizationR-MDL · CelebAworst-group accuracy (%): 88.8 -> 88.9No clear effect
replace GDRO with R-MDLResNet-50 · CelebA, standard regularizationgap between average and worst-group accuracy (%): 53 -> 3.4Improved

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