AI Paper Atlas

Research map / Learning theory and algorithms / Fairness

From Gradient Volume to Shapley Fairness: Towards Fair Multi-Task Learning

Xiao Wang, Yuying Han, Dazi Li, Fei Zhang, Min Tang

ICLR 2026 · Learning theory and algorithms › Fairness · approach: Fairness-constrained classification

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 Shapley-value weights from VolDet to loss-based RLW (ablation)RLW · NYU-v2Delta m%: 7.78 -> 7.60Improved
add Shapley-value weights from VolDet to DWA (ablation)DWA · NYU-v2Delta m%: 3.57 -> 2.17Improved
add Shapley-value weights from VolDetPro to UW (ablation)UW · NYU-v2Delta m%: 4.05 -> -1.50Improved
replace FairGrad weights with Shapley-VolDet weightsSVFair · NYU-v2Delta m%: -4.66 -> -8.81Improved
use SVFair instead of PIVRGSVFair · CityscapesMean Rank: MR 2.50 (PIVRG) vs 6.88 (VolDet); Delta m% -0.45 vs -2.08Mixed
use VolDetPro instead of VolDet (ablation)SVFair · Office-31Delta m%: -1.53 -> -1.62Improved
use VolDetPro instead of VolDet (ablation)SVFair · CelebADelta m%: -0.63 -> 0.11Hurt
increase temperature tau from 1.0 to 5.0 (ablation)SVFair(VolDet) · CityscapesDelta m%: 0.10 -> -2.16Improved
increase temperature tau from 0.5 to 5.0 (ablation)SVFair(VolDet) · Office-31Delta m%: -1.49 -> -0.97Hurt
add Monte Carlo Shapley estimation with K=1k to SVFair on CelebASVFair · CelebAsingle-epoch runtime (s): 810 -> 870 (K=1k); 1086 (K=5k)Hurt

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