R Deep Learning Essentials - Second Edition

Recurrent neural network model

The deep learning networks we have seen so far have no concept of memory. Every new piece of information is treated as atomic and has no relation to what has already occurred. But, sequences are important in time series and text classification, especially sentiment analysis. In the previous section, we saw how word structure and order matters, and we used CNNs to resolve this. While this approach worked, it does not resolve the problem completely as we still must pick a filter size, which limits the range of the layer. Recurrent neural networks are deep learning layers which are used to solve this problem. They are networks with feedback loops that allow information to flow and therefore are able to remember ...