AI Paper Atlas

Research map / Learning theory and algorithms / Fairness

A Fair Bayesian Inference through Matched Gibbs Posterior

Jihu Lee, Kunwoong Kim, Sehyun Park, Insung Kong, Dongyoon Yang, Yongdai Kim

ICLR 2026 · Learning theory and algorithms › Fairness · approach: Fairness auditing and regularization

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 matched deviation penalty with learnable matching T (matched Gibbs posterior) vs deterministic fair baselinesfair ReLU MLP · CRIME consistency score at Delta_W^1/2 ~ 0.09Con: 0.767 -> 0.837Improved
use matched Gibbs posterior instead of MMD-based Gibbs/variational posteriorsfair Bayesian inference · CRIMEaccuracy: 0.736 vs 0.696 (gibbs mmd), 0.729 (variational mmd)Improved
use matched Gibbs posterior instead of adv, reduction, gapregfair classifier · ADULT consistency scoreCon: 0.927 -> 0.955Improved
use matched Gibbs posterior instead of baselinesfair classifier · DUTCH consistency scoreCon: 0.939 -> 0.965Improved
reduce prior temperature tau of matching prior from 1.0 to 0.1 and 0.01 (ablation)matched Gibbs posterior · ADULTAcc (x10^2): 82.888 -> 82.892No clear effect
vary number of pretrain epochs (10, 20, 30, 40) for MCMC initialization (ablation)matched Gibbs posterior · ADULTPareto trade-offNo clear effect
use matched Gibbs posterior with strict fairness instead of constrained random-walk MHposterior sampling under fairness constraint · ADULTacceptance ratio: 0.02 vs 0.897Improved
tighten constraint threshold in constrained random-walk MHrandom-walk Metropolis-Hastings · ADULTacceptance ratio: 0.58 -> 0.02Hurt
tighten DP level delta for samples from standard mean-field Gaussian VImean-field Gaussian VI · CRIME logistic regressionproportion of fair samples: 0.014 -> 0.000Hurt
use gibbs matched for Ece under strict fairnessmatched Gibbs posterior · CRIME and DUTCHEceMixed

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