Causal EpiNets: Neural Bounds for Individual Treatment Effects

A new arXiv paper (2605.07065v1) introduces Causal EpiNets, a neural framework for estimating individual treatment effects using Probability of Necessity and Sufficiency (PNS). Standard plug-in estimators fail due to structural probability violations and extremum bias from max-min operators, producing narrow intervals. The proposed method uses an anchored neural architecture to guarantee constraint satisfaction and precision-corrected intersection-bound inference with Epistemic Neural Networks for uncertainty quantification. Empirical evaluations show the approach maintains correct coverage.