Causal Foundation Models
A new paper frames causal inference as a foundation-model task: pretrain once, estimate effects on new datasets in context.
The authors define causal foundation models as neural networks that estimate quantities such as average treatment effect without updating their weights for each new dataset. The paper positions them against the usual custom causal pipeline of mechanism choice, estimator selection, and model training. It is a practical introduction, with causal-inference and machine-learning background plus example code and Jupyter notebooks. HF Daily Papers' note
The authors define causal foundation models as neural networks that estimate quantities such as average treatment effect without updating their weights for each new dataset. The paper positions them against the usual custom causal pipeline of mechanism choice, estimator selection, and model training. It is a practical introduction, with causal-inference and machine-learning background plus example code and Jupyter notebooks. HF Daily Papers' note
score 5