SLCP

A challenging inference task designed to have a simple likelihood and a complex posterior (Papamakarios et al., 2019b; Greenberg et al., 2019): The prior is uniform over five parameters $\theta$ and the data are a set of four two-dimensional points sampled from a Gaussian likelihood whose mean and variance are nonlinear functions of $\theta$. This induces a complex posterior with four symmetrical modes and vertical cut-offs.

Reference: Greenberg et al., 2019: Automatic Posterior Transformation for Likelihood-free Inference.
Papamakarios et al., 2019: Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flows.
Description from Lueckmann et. al, 2021: Benchmarking Simulation-Based Inference