Stochastic variational inference

Stochastic variational inference ([10]) learns shared latent variables with stochastic optimization. It uses subsets of the data to compute noisy estimates of the gradients. The method can be used to scale the inference on large datasets, if the model has a specific structure.