Hi! Would it ever make sense to have multiple embedding layers in a variational quantum circuit? In all the tutorials I have seen, there is one embedding layer followed by some number of entangling layers. Could we for example have a new embedding layer before each entangling layer? Why/why not? Thanks!
Hey @Eino! Welcome to the forum 
You definitely can! There’s no reason why you can’t. So long as you logically process the data being embedded, it all seems fine to me. Do you have an application in mind? Or are you just curious? 
Hi @isaacdevlugt, thanks!
I’m generating polymers using a variational quantum circuit and I’m just trying out different circuits. I tried to have a new embedding layer before each entangling layer, but the result was worse than the result with one embedding layer. I also didn’t find any example where multiple embedding layers would have been used. Because of this I’m guessing that one embedding layer is usually preferred over multiple ones. I’m just wondering if there are any reasons why it would in some cases be helpful to use multiple embedding layers.
You might want to check out the Data-reuploading models, which propose to do multiple embedding layers. The original paper gives some motivation related to neural networks for which this might be beneficial, but personally my intuition is that having multiple embedding layers for the same data allows you to access different frequency degrees of freedom for your quantum circuit than you would otherwise have.