Two Moons

A two-dimensional task with a posterior that exhibits both global (bimodality) and local (crescent shape) structure (Greenberg et al., 2019) to illustrate how algorithms deal with multimodality.

Reference: Greenberg et al., 2019: Automatic Posterior Transformation for Likelihood-free Inference.
Description from Lueckmann et. al, 2021: Benchmarking Simulation-Based Inference