Variational Quantization for Forecasting with State Space Models

A new forecasting model that combines discrete state space hidden Markov models with neural network architectures and variational training procedures to enable accurate predictions on large datasets with diverse time series.