9.2 Differentiable Learning of QCBM
The output of a QCBM circuit is a bitstring that represents a sample from the probability distribution encoded in the quantum state. The circuit itself is, essentially, a mechanism of transforming an initial state |0⟩⊗n into a final state from which a sample is generated by means of measuring the qubits in the computational basis.
Different configurations of one-qubit and multi-qubit gates encode different probability distributions – the training of a QCBM consists of finding an optimal circuit configuration (ansatz) and an optimal set of adjustable parameters that minimise the distance between the probability distribution encoded in the final quantum state (before measurement, or "before sampling") and the probability distribution of the training dataset.
Following the structure we adopted in Chapter 8, we start with the differentiable learning approach, before moving to the non-differentiable learning method based...
