Deep Learning in Time Series Analysis

11 Recurrent Neural Networks (RNN)

11.1 Introduction

The neural networks discussed in the preceding chapters, are purely static learning models that implement mapping from the input vector to the outputs. However, they cannot deal with time related information in a dynamic system. Consider the industrial process of gas furnace as an example, the CO2 concentration is not only dependent on the current air flow rate but also affected by the process conditions in the preceding steps. Another example can be seen from data analysis with EEG signal in which the temporal property of the data plays a key role in the pattern recognition of the signal. Evidently, handling the dynamic nature of data sequences requires deep networks other than those traditional ...