AI Paper Atlas

Research map / Learning theory and algorithms / Fairness

Achieving Fairness Without Harm via Selective Demographic Experts

Xuwei Tan, Yuanlong Wang, Thai-Hoang Pham, Ping Zhang, Xueru Zhang

AAAI 2026 · Learning theory and algorithms › Fairness · approach: Post-processing and transport for fairness

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
decouple only the classifiers (shared ERM representation) per sensitive groupERM on Ham10000 · Ham10000 (gender)AUC: 84.07 -> 83.60Hurt
decouple representations as well as classifiers with no-harm selection (FairSDE vs Decoupled)Decoupled classifiers · Ham10000 (age)AUC: 83.91 -> 84.62Improved
add FairSDE demographic experts with no-harm selectionERM · Ham10000 (age)worst-group AUC: 75.13 -> 76.03Improved
add FairSDEERM · UTKFace (race)accuracy: 81.65 -> 84.17Improved
add FairSDEERM · MIMIC-CXR (gender)AUC: 82.28 -> 82.60No clear effect
add FairSDEERM · Harvard-GF (gender)AUC gap: 2.36 -> 1.81Improved
add FairSDE (no clear fairness gain on race gap vs ERM beyond small change)ERM · Ham10000 (gender)AUC gap: 2.76 -> 2.65No clear effect
use integer programming selection instead of greedy (ablation)FairSDE on Harvard-GF (gender) · Harvard-GF (gender)AUC gap: 2.34 -> 1.81Improved
use integer programming selection instead of greedy (ablation)FairSDE on Harvard-GF (gender) · Harvard-GF (gender)overall AUC: 82.31 -> 81.98Hurt
add fairness-regularization baseline FSCL+ERM · Ham10000 (gender)AUC: 84.07 -> 82.18Hurt
add adversarial debiasingERM · UTKFace (gender)accuracy: 78.06 -> 73.40Hurt
use greedy no-harm selection (ablation)FairSDE · Harvard-GF (gender)worst-group AUC: 80.54 -> 81.20Improved

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