RWS outperforms current methods in learning SCFMs.
problem Learning models with stochastic control flow is challenging.
method Revisited reweighted wake-sleep algorithm for SCFMs.
result RWS learns better models and inference networks with more particles.