Pushed by regular progress in deep generative modeling, simulation-based inference (SBI) has emerged because the workhorse for inferring the parameters of stochastic simulators. Nonetheless, current work has demonstrated that mannequin misspecification can compromise the reliability of SBI, stopping its adoption in vital purposes the place solely misspecified simulators can be found. This work introduces strong posterior estimation~(RoPE), a framework that overcomes mannequin misspecification with a small real-world calibration set of ground-truth parameter measurements. We formalize the misspecification hole as the answer of an optimum transport~(OT) drawback between realized representations of real-world and simulated observations, permitting RoPE to be taught a mannequin of the misspecification with out inserting further assumptions on its nature. RoPE demonstrates how OT and a calibration set present a controllable steadiness between calibrated uncertainty and informative inference, even underneath severely misspecified simulators. Outcomes on 4 artificial duties and two real-world issues with ground-truth labels display that RoPE outperforms baselines and persistently returns informative and calibrated credible intervals.
- * Equal contribution
- ** Work achieved whereas at Apple
- † ETH Zürich
