AI RESEARCH

Conservative neural posterior estimation via distributionally robust training

arXiv CS.LG

ArXi:2605.28516v1 Announce Type: cross Simulation-based inference with neural posterior estimation (NPE) often yields overconfident and unreliable posteriors under limited simulation budgets. To address this, we propose DRO-NPE, a distributionally robust approach that replaces the standard NPE objective with a worst-case loss over a Wasserstein ambiguity set. We