Fig. 1.

Download original image
Workflow of a typical SBI with neural posterior estimation (NPE), i.e., directly estimating the posterior. The forward modeling process transforms parameters to be inferred to data vector, and therefore includes the process of measuring summary statistics from simulation results in our case. The density estimator takes parameter-data pair {θ,d} as input, and after the training procedure returns an estimation of posterior. Note: for the neural likelihood estimation (NLE) and neural ratio estimation (NRE), the target to be trained (shown in green block) is the likelihood and the ratio of likelihood and prior, respectively. Combined with the prior, this allows us to easily build the posterior using Bayes law.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.