Time Series Forecasting Based on Variational Recurrent Model

In this paper, we use variational recurrent model to investigate the time series forecasting problem. Combining recurrent neural network (RNN) and variational inference (VI), this model has both deterministic hidden states and stochastic latent variables while previous RNN methods only consider deterministic states. Based on comprehensive experiments, we show that the proposed methods significantly improves the state-of-art performance of chaotic time series benchmark and has better performance on real-worl data. Both single-output and multiple-output predictions are investigated.