Quantum Continuous-Variable Recurrent Neural Network model for PennyLane QML framework
Please read a guide about quantum neural network on strawberryfields first.
To create an analog of Elman RNN we need to have followng building blocks:
-
Data encoding procedure: Displacement encoding is used.
-
Quantum linear layer: Can be created from the quantum layer from the guide without non-Gaussian activation gates at the end.
-
Inverse to quantum linear layer operator to reset a group of qumodes to the vacuum state after each step. Because all quantum operators
-
Unitary operator acting on a one group of qumodes while beeing controlled by another group of qumode: A sequence of controlled phase operations is used to transfer information from between groups of qumodes.
-
Activation function: It is just a layer of non-Gaussian gates.
On a picture bellow you can see a block of the proposed quantum continuous-variable RNN acting on some input sequence at time t.
After performing Quantum CV RNN classification on the hidden qumodes state can be performed.