Research map / Probabilistic and scientific modeling / Treatment effects
Causal Transformer for Estimating Counterfactual Outcomes
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use the three-subnetwork Causal Transformer | Counterfactual outcome estimation · MIMIC-III semi-synthetic, random trajectories, τ=1–10 | RMSE: 0.30 -> 0.20; 0.48 -> 0.38; 0.59 -> 0.45; 0.65 -> 0.49; 0.68 -> 0.52; 0.71 -> 0.53; 0.72 -> 0.55; 0.74 -> 0.56; 0.76 -> | Helped |
| Use the Causal Transformer | Counterfactual outcome estimation · MIMIC-III real-world, τ=1–5 | RMSE: 4.84 -> 4.59; 9.15 -> 8.99; 9.81 -> 9.59; 10.15 -> 9.91; 10.40 -> 10.14 | Helped |
| Add CDC balancing to CT (ablation) | Causal Transformer · MIMIC-III semi-synthetic, random trajectories, τ=1–10 | RMSE: 0.20 vs 0.20; 0.38 vs 0.38; 0.45 vs 0.45; 0.49 vs 0.50; 0.52 vs 0.52; 0.53 vs 0.55; 0.55 vs 0.56; 0.56 vs 0.58; 0.58 vs | Helped |
| Replace trainable relative PE with non-trainable relative PE (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: ±0.00; −0.02; +0.01; −0.03 | Mixed |
| Replace relative PE with trainable absolute PE (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: +0.04; +0.16; +0.15; +1.00 | Hurt |
| Remove attentional dropout (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: ±0.00; +0.07; +0.00; +0.09 | Hurt |
| Remove cross-attention (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: +0.03; +0.16; +0.06; +0.10 | Hurt |
| Remove EMA (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: +0.03; +0.38; +0.03; +0.33 | Hurt |
| Remove CDC balancing (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: −0.01; −0.02; ±0.00; +0.07 | Mixed |
| Replace CDC loss with gradient reversal (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: +0.02; +0.17; +0.08; +0.33 | Hurt |
| Use a single-subnetwork encoder-decoder with gradient reversal (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: +0.16; +0.08; +0.05; +0.23 | Hurt |
| Use a single-subnetwork encoder-decoder with domain confusion (ablation) | CT · Tumor growth simulator, γ=1/4 and τ=1/6 | normalized RMSE: −0.03; +0.10; −0.03; +0.23 | Mixed |
Closest papers by meaning
- Towards Characterizing Domain Counterfactuals for Invertible Latent Causal Models (ICLR 2024 · similarity 0.90)
- Causal Contrastive Learning for Counterfactual Regression Over Time (NeurIPS 2024 · similarity 0.90)
- Counterfactual Generative Modeling with Variational Causal Inference (ICLR 2025 · similarity 0.89)
- A Dual-module Framework for Counterfactual Estimation over Time (ICML 2024 · similarity 0.89)
- Adversarially Balanced Representation for Continuous Treatment Effect Estimation (AAAI 2024 · similarity 0.88)
- Counterfactual Contrastive Learning with Normalizing Flows for Robust Treatment Effect Estimation (ICML 2025 · similarity 0.88)
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.