Back propagation for keras layer

PennyLane Help
2

Hi @kareem_essafty,

  1. Yes, you should be able to use backpropagation for a KerasLayer consisting of tensorflow tensors for the classical part and qulacs device for the quantum part.
  2. Depends more specifically what you mean here. How does your question compare to this one for TorchLayer? (which should have similar restrictions as KerasLayer)
  3. This is hard to debug without further context (e.g., a minimal (non-)working example). The error seems to be thrown by Keras, and you say it doesn’t occur in earlier versions, so on a superficial level I would guess it is a bug in TF/Keras, but hard to know for sure without full code here