Probabilistic Counterfactual Inference for Discrete Outcomes in Gaussian-Process Causal Models
The paper says discrete counterfactuals in GP causal models need the right noise coupling, not just a good observational fit.
The authors extend GP-SCM counterfactual inference to binary, nominal, and ordinal outcomes with explicit noise-abduction mechanisms. They test the framework on synthetic SCMs where ground-truth counterfactuals are known. One reported failure mode is structural: treating ordinal data as merely categorical roughly triples counterfactual error, and more data does not fix it. Forcing order onto nominal data instead damages the fitted structural equation itself. ArXiv · AI/CL/LG's note
The authors extend GP-SCM counterfactual inference to binary, nominal, and ordinal outcomes with explicit noise-abduction mechanisms. They test the framework on synthetic SCMs where ground-truth counterfactuals are known. One reported failure mode is structural: treating ordinal data as merely categorical roughly triples counterfactual error, and more data does not fix it. Forcing order onto nominal data instead damages the fitted structural equation itself. ArXiv · AI/CL/LG's note
score 4