Amortised Inference: Why Training a Network to Infer Beats Running Optimisation Every Time

Variational inference has a scaling problem. For every new data point, it asks you to solve a fresh...