Hands-On Artificial Intelligence for Beginners

Basic structure

RNNs differ from other networks in the fact that they have a recursive structure; they are recurring over time. RNNs utilize recursive loops, which allow information to persist within the network. We can think of them as multiple copies of the same network, with information being passed between each successive iteration. Without recursion, an RNN tasked with learning a sentence of 10 words would need 10 connected copies of the same layer, one for each word. RNNs also share parameters across the network. Remember in the past few chapters how the number of parameters we had in our network could get unreasonably large with complex layers? With recursion and parameter sharing, we are able to more effectively learn increasingly ...