Research map / Probabilistic and scientific modeling / Treatment effects

CausalPFN: Amortized Causal Effect Estimation via In-Context Learning

NeurIPS 2025 · Probabilistic and scientific modeling › Treatment effects · approach: Proximal and deep causal effect estimation · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 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.

Trains a single transformer (CausalPFN) on large libraries of simulated causal data-generating processes so it amortizes causal effect estimation via in-context learning on new observational datasets.

Problem

Selecting and tuning an appropriate causal effect estimator from dozens of specialized methods requires substantial manual effort and domain expertise, for estimating CATE and ATE from observational data.

Approach

Amortize the causal estimator-selection workflow into one prior-data-fitted transformer trained once on simulated DGPs satisfying ignorability; at inference it takes an observational dataset in context and outputs individual and average treatment effect estimates with uncertainty, without per-dataset training.

Key results

• Trained once on simulated ignorable DGPs, CausalPFN infers CATE/ATE on new datasets without per-dataset training (exact numeric results not given in provided text).

Measured effects reported in the paper

Limitations

• Relies on the ignorability (unconfoundedness) assumption holding in the simulated prior and in real data
• Results depend on the coverage of the simulated DGP prior

Experimental setup

Datasets: simulated data-generating processes satisfying ignorability

Benchmarks: CATE and ATE benchmark datasets (names not stated in provided text)

Metrics: CATE error ATE error calibration

Baselines: TabPFN fine-tuned causal forest DragonNet X-learner R-learner T-learner S-learner

Builds on: TabPFN prior-data fitted networks DragonNet

Closest papers by meaning

• Do-PFN: In-Context Learning for Causal Effect Estimation (NeurIPS 2025 · similarity 0.89)
• Generalized Bayes for Causal Inference (NeurIPS 2026 · similarity 0.88)
• Structural Causal Bottleneck Models (NeurIPS 2026 · similarity 0.88)
• Estimating Interventional Distributions with Uncertain Causal Graphs through Meta-Learning (NeurIPS 2025 · similarity 0.87)
• Towards Learning and Explaining Indirect Causal Effects in Neural Networks (AAAI 2024 · similarity 0.87)
• NESTER: An Adaptive Neurosymbolic Method for Causal Effect Estimation (AAAI 2024 · similarity 0.87)