Do we only have single layer neural network for each input & output in RNN ?
What exactly to you mean by “single layer neural network”?
If you look at this slide from later in that same video, you can see what is inside each of those vertical slices. There are weight matrices for each of the input a, the input x, the output a, and the output y. So it’s fairly complicated and expensive to train. Adding more complexity would have to be traded against any possible benefits.
Thanks for explaining 