AI Paper Atlas

Research map / Learning theory and algorithms / Fairness

RECAST: Model Reconstruction via Counterfactual-Aware Wasserstein Geometry under Limited Data

Xuan Zhao, Lena Krieger, Zhuo Cao, Arya Bangun, Hanno Scharr, Ira Assent

ICML 2026 · Learning theory and algorithms › Fairness · approach: Fairness auditing and regularization · 0 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
add RECAST (Wasserstein barycentric prototypes with soft CFs) over SAMPLES (CFs as full labeled samples)SAMPLES / MLP target · COMPAS (query size 100)fidelity: 0.415 -> 0.855Improved
use RECAST instead of CCACCA / DT target · COMPAS (query size 100)fidelity: 0.657 -> 0.869Improved
use RECAST instead of SAMPLES on logistic regression targetsSAMPLES / LR target · Adult Income (query size 100)fidelity: 0.909 -> 0.913No clear effect
use RECAST instead of CCA on a Housing MLP targetCCA / MLP target · California Housing (query size 100)fidelity: 0.716 -> 0.712No clear effect
use RECAST instead of TRA (which assumes a decision tree target)TRA / DT target · Adult Income (query size 100)fidelity: 0.842 -> 0.894Improved
use RECAST instead of SAMPLES when CFs come from C-CHVAESAMPLES / MLP target · Adult Income (CF generator C-CHVAE)fidelity: 0.357 -> 0.923Improved
use RECAST instead of SAMPLES when CFs come from DiCESAMPLES / MLP target · Adult Income (CF generator DiCE)fidelity: 0.933 -> 0.923Hurt
use RECAST instead of baselines under CF generator variationSAMPLES and CCA · Adult Income (varying CF generators)fidelity variabilityImproved
use RECAST under additive test noise (tau up to 0.4)SAMPLES and CCA · Adult Income, full and near-threshold setsfidelity dropImproved
use RECAST when the query budget is small (25 to 400)SAMPLES and CCA · Adult/COMPAS/HELOC/HousingfidelityImproved
increase query budget until supervised no-CF reconstruction catches upRECAST vs no-CF baseline · query-budget sweepfidelity advantageMixed
use RECAST fairness diagnostics (TIDP/TIEO) on reconstructed modelSAMPLES and CCA · fairness diagnostic error under noiseabsolute diagnostic differenceImproved

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