AI Paper Atlas

Research map / Probabilistic and scientific modeling / Causal discovery

Transporting the Past: An Optimal Transport View of Backtracking Counterfactuals

Zhangyi Liu, Sayedpouria Fatemi, Suvrit Sra

NeurIPS 2026 · Probabilistic and scientific modeling › Causal discovery · approach: Counterfactual identification and bounds · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Frames backtracking counterfactual inference in SCMs as optimal transport, transporting the posterior over exogenous variables onto the set of exogenous configurations that yield a desired counterfactual outcome.

Problem

Backtracking counterfactual inference in structural causal models: a factual observation induces a posterior over exogenous variables, so counterfactual inference is distributional rather than point-valued.

Approach

Two complementary optimal-transport frameworks move the exogenous posterior onto the set of exogenous configurations satisfying the desired counterfactual condition, giving a distributional treatment of backtracking counterfactuals.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
allow dependence among counterfactual exogenous variables (OT-CF) instead of enforcing independence (ablation)OT-CF · 5D nonlinearD_U: 0.240670 vs 0.250959 (Hard)Mixed
use Hard constrained solver instead of soft solversIOT-CF · synthetic SCMsCF residual: exact CF satisfactionImproved
use Hard constrained solver instead of OT-CFIOT-CF · 10D nonlinearD_U: 0.795014 (OT-CF) vs 0.798030 (Hard)No clear effect
use Penalty instead of Marginal solver (ablation)IOT-CF · 6D polynomialD_U: 0.783031 -> 0.384916Improved
use OT-CF instead of interventional counterfactualOT-CF · 5D nonlinearD_U: 0.4216 -> 0.2407Improved
increase Gaussian transfer kernel parameter from 0.5 to 50 (ablation)probabilistic backtracking · Example 1D_K: 2.566089 -> 0.151979Improved
increase Gaussian transfer kernel parameter from 0.5 to 50 (ablation)probabilistic backtracking · Example 2D_K: 2.212211 -> 0.070130Improved
add independence constraint (IOT-CF) to OT-CF (ablation)IOT-CF · illustrative Examplecounterfactual outcome uniquenessImproved

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