AI Paper Atlas

Research map / Probabilistic and scientific modeling / Treatment effects

Spiked-CFR: Causal Representation Learning from LLMs via Wasserstein Projection Pursuit

Fan Wang, Hengyu Yue, Yu Bowen, Weiming Liu, Zongxin Yang, Xuyun Zhang, Xiaolin Zheng, Chaochao Chen, Shuiguang Deng

ICML 2026 · Probabilistic and scientific modeling › Treatment effects · approach: Proximal and deep causal effect estimation

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 full-space Wasserstein balancing with projection-pursuit balancingLlama-CFR-Wass · ACIC-postepsilon_PEHE: 2.32 -> 2.18Improved
replace full-space Wasserstein balancing with projection-pursuit balancingLlama-CFR-Wass · IHDP-postepsilon_PEHE: 2.65 -> 2.47Improved
replace full-space Wasserstein balancing with projection-pursuit balancingLlama-CFR-Wass · Retail HeroRMSE(ATE): 2.15 -> 0.72Improved
replace full-space Wasserstein balancing with projection-pursuit balancingLlama-CFR-Wass · Sema vs. Lira (clinical)RMSE(ATE): 6.17 -> 0.56Improved
replace full-space Wasserstein balancing with projection-pursuit balancingLlama-CFR-Wass · Onab vs. Topi (clinical)RMSE(ATE): 8.14 -> 1.38Improved
use learned WPP subspace instead of random projection or PCA of the same dimension k (ablation)SPIKED-CFR · semi-synthetic and real-world datasetsrelative errorImproved
learn projection U with WPP rather than fixed projection (ablation)SPIKED-CFR · ACIC, Eren vs. Topiseed varianceImproved
extract features from the L-4 layer rather than other layers or last-token hidden state (ablation)SPIKED-CFR · semi-synthetic and clinicalnormalized errorImproved
use Llama-3-8B embeddings instead of BERT featuresCFR/forest estimators · all benchmarkseffect errorImproved
full-space Wasserstein instead of MMD in CFR on Llama embeddingsLlama-CFR · IHDP-postepsilon_PEHE: 2.51 -> 2.65Hurt
apply SPIKED-CFR balancing to raw LLM embeddingsLLM embeddings · ACIC-posttreatment classification AUCImproved
add NATURAL-style LLM probability IPW instead of SPIKED-CFRNATURAL · Eren vs. Topi (clinical)RMSE(ATE): 1.68 -> 3.99Hurt

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