Modern approaches for simulation-based inference build upon deep learning surrogates to enable approximate inference with computer simulators, as is often the case in the physical sciences and in particle physics. In practice, the faithfulness of the inference
results are, however, rarely guaranteed. For example, Hermans et al., 2021 have shown that current simulation-based inference algorithms can produce Bayesian posteriors that are overconfident, hence risking false inferences.
In this talk, we will review simulation-based inference, together with the main inference algorithms, and present Balanced Neural Ratio Estimation (BNRE), a variation of the NRE algorithm designed to produce posterior approximations that tend to be more conservative, hence improving their reliability. Examples from the physical sciences
and from particle physics will be discussed throughout.
