AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Guiding Data Allocation for Robust Subpopulation Generalization

Zhaoying Pan, Yipei Wang, Shenyu Lu, Xiaoqian Wang

NeurIPS 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Distributionally robust optimization · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Studies whether fully balanced subgroup proportions are the unique optimal training configuration for robust subpopulation generalization, and argues that balance is costly and not always necessary.

Problem

Models achieve high average training accuracy but perform poorly under subpopulation shift; balancing subgroup proportions is a common data-centric fix but can be costly, infeasible, and not always necessary.

Approach

Questions whether balanced training data is the unique optimal configuration for robust subpopulation generalization, and instead studies how to guide data allocation across subgroups rather than enforcing full balance.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
target Grad-derived configuration instead of balanced in under-samplingUnder-sampling · Waterbirds-syn (Example 2)adjusted accuracy: 0.777 -> 0.901Improved
target Grad-derived configuration instead of balanced in over-samplingOver-sampling · Waterbirds-syn (Example 2)worst-group accuracy: 0.366 -> 0.541Improved
target Grad-derived configuration instead of balanced in under-samplingUnder-sampling · Waterbirds-syn (Example 2)worst-group accuracy: 0.837 -> 0.848Improved
use balanced over-sampling instead of original dataWaterbirds-syn (Example 2) · Waterbirds-synworst-group accuracy: 0.376 -> 0.366No clear effect
use imbalanced Grad allocation for last-layer retraining stage twoDFR · Waterbirds/CelebAworst-group accuracy: 89.30 -> 89.10 (Waterbirds); 87.49 -> 87.36 (CelebA)No clear effect
use imbalanced Grad allocation for last-layer retraining stage twoAFR · CelebAworst-group accuracy: 80.74 -> 83.84Improved
use imbalanced Grad allocation for last-layer retraining stage twoAFR · Waterbirdsworst-group accuracy: 86.90 -> 88.27Improved
use imbalanced Grad allocation for last-layer retraining stage twoGSR · CelebAworst-group accuracy: 85.74 -> 86.67Improved
allocate data along Grad direction instead of balancing direction (ablation)Data allocation · Waterbirds-syn, UrbanCars-syn, CelebA-subset, CivilComments-subsetadjusted accuracyImproved
use non-balanced training configurations instead of balanced (ablation)Subgroup data configuration · Gaussian theory and four datasetsadjusted accuracy: balanced 0.921 vs 90th percentile 0.918No 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