AI Paper Atlas

Research map / Learning theory and algorithms / Optimal transport

Optimal Transport Group Counterfactual Explanations

Enrique Valero-Leal, Bernd Bischl, Pedro Larrañaga, Concha Bielza, Giuseppe Casalicchio

ICML 2026 · Learning theory and algorithms › Optimal transport · approach: Neural optimal transport · 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
replace pointwise group counterfactuals with an explicit affine/Gaussian OT map under strict bi-Lipschitz (K=1.01)OT group counterfactuals vs Group w/ bi-Lipschitz · 15 tabular datasets, K=1.01convergence: baseline fails to converge in 50% of experimentsImproved
use OT maps instead of baseline across K valuesOT group counterfactuals · 15 tabular datasets, all Kconvergence: baseline fails in 35%Improved
enforce bi-Lipschitz via convex SDP/QCQP maps instead of unconstrained WachterPSD affine and others · W2 transport costW2: 30% higher W2 than unconstrained WachterHurt
use larger K (looser bi-Lipschitz) (ablation)baseline vs OT maps · W2 gapW2Mixed
use dense maps (PSD affine, Gaussian) instead of sparse maps (ablation)OT maps · W2 transport costW2: 90% within factor 1.7 vs 80% within 2.3Improved
use sparse maps for efficiency (W2 x log params) (ablation)OT maps · efficiencyefficiency: baseline factor 2.2 vs ours 1.1 to 1.9Improved
use sparse maps in multiobjective NSGA-II optimization (ablation)OT maps · hypervolumehypervolume: diag affine ~1.01, scaled Gaussian ~1.4 ratioImproved
use scaled Gaussian transform (ablation)Gaussian scaled · counterfactual validityvalidity: validity near 1 for all KImproved
add the Lipschitz upper bound constraint (ablation)Group counterfactual · W2 and distortionupper boundNo clear effect
use OT maps instead of baseline for runtimeOT group counterfactuals · runtimeruntime: baseline ~1250x (Lipschitz) and ~60000x (bi-Lipschitz) slower than fastest proposalImproved

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