Appendix 1—figure 4. Full posterior for Gabor GLM receptive field model.

SNPE posterior estimate (blue lines) compared to reference posterior (MCMC, histograms). Ground-truth parameters used to simulate the data in green. We depict the distributions over the original receptive field parameters, whereas we estimate the posterior as a Gaussian mixture over transformed parameters, see Materials and methods for details. We find that a (back-transformed) Gaussian mixture with four components approximates the posterior well in this case.