Automated Synthesizing of Quantum Programs

1 . A method comprising:

obtaining quantum state information computed from quantum processor output data generated by a quantum resource executing an initial version of a quantum program;

providing neural network input data to a neural network, the neural network input data comprising the quantum state information and a representation of a problem to be solved by the quantum program;

obtaining neural network output data generated by the neural network processing the neural network input data;

selecting a quantum logic gate based on the neural network output data; and

generating an updated version of the quantum program that includes the selected quantum logic gate.

2 . The method of claim 1 , wherein the neural network input data comprise a state and a reward based on the quantum processor output data.

3 . The method of claim 2 , wherein the state comprises the quantum state information and the representation of the problem to be solved by the quantum program.

4 . The method of claim 3 , wherein the state comprises:

a binary array containing qubit measurements from the quantum resource executing multiple shots of the initial version of the quantum program; and

an array containing weights of a graph representation of the problem.

5 . The method of claim 3 , comprising encoding the problem.

6 . The method of claim 3 , wherein the problem to be solved comprises a combinatorial optimization problem or finding a ground state of a molecule.

7 . (canceled)

8 . The method of claim 3 , wherein the reward comprises a Hamiltonian expectation value based on the problem to be solved.

9 . The method of claim 1 , wherein the neural network output data comprise a set of values associated with a set of quantum logic gates, and the value associated with each quantum logic gate represents a prediction of a degree to which the quantum logic gate improves the quantum program.

10 . The method of claim 9 , wherein selecting the quantum logic gate comprises:

identifying a maximum value in the set of values; and

identifying the quantum logic gate associated with the maximum value.

11 . The method of claim 9 , wherein the set of quantum logic gates comprises an action space comprising:

a set of discrete-angle single-qubit rotation gates for each of a plurality of qubits; and

a set of two-qubit entangling gates for each distinct pair of qubits in the plurality of qubits.

12 . The method of claim 1 , wherein the initial version comprises a quantum logic circuit comprising a series of quantum logic gates, and generating the updated version comprises appending the selected quantum logic gate to the end of the series.

13 . The method of claim 12 , wherein appending the selected quantum logic gate to the series improves the quantum program according to a reward defined by the problem to be solved by the quantum program.

14 . (canceled)

15 . The method of claim 1 , comprising:

obtaining additional quantum processor output data generated by the quantum resource executing the updated version of a quantum program; and

selecting a value of a variable parameter of the quantum logic gate based on the additional quantum processor output data;

wherein the quantum logic gate comprises a parametric gate.

16 . (canceled)

17 . The method of claim 1 , comprising modifying the neural network based on reward data computed from the quantum processor output data.

18 . The method of claim 17 , wherein the reward data comprises a cost function based on a Hamiltonian.

19 . The method of claim 17 , wherein the neural network is modified according to a deep reinforcement learning process.

20 . The method of claim 1 , comprising executing an iterative process, where each iteration of the iterative process includes:

compiling an initial version of the quantum program for the iteration;

generating quantum processor output data for the iteration by executing the quantum program compiled for the iteration;

computing quantum state information for the iteration based on the quantum processor output data for the iteration;

operating the neural network to produce neural network output data for the iteration based on the quantum state information for the iteration;

selecting a quantum logic gate for the iteration based on the neural network output data for the iteration; and

generating an updated version of the quantum program that includes the selected quantum logic gate for the iteration.

21 - 22 . (canceled)

23 . The method of claim 1 , wherein the quantum resource comprises a quantum processor unit, multiple quantum processor units configured to operate in parallel, a quantum virtual machine, or multiple quantum virtual machines configured to operate in parallel.

24 - 27 . (canceled)

28 . The method of claim 1 , wherein the quantum processor output data are generated by the quantum resource executing multiple shots of the initial version of the quantum program, the quantum state information comprises a plurality of bitstrings, and each bitstring represents a measurement of qubit states generated by a respective one of the multiple shots.

29 . (canceled)

30 . A method comprising:

computing a reward from quantum processor output data generated by a quantum resource executing an initial version of a quantum program, wherein the reward is computed according to a problem to be solved by a policy;

modifying the policy based on the reward;

obtaining policy output data generated by the modified policy processing the reward and a representation of the problem to be solved;

selecting a quantum logic gate based on the policy output data; and

generating an updated version of the quantum program that includes the selected quantum logic gate.

31 . The method of claim 30 , further comprising initializing the policy based on a classical solution to the problem.

32 . (canceled)

33 . The method of claim 30 , wherein the problem to be solved comprises a combinatorial optimization problem, a MAXCUT problem instance, a MAXQP problem instance or a QUBO problem instance.

34 . (canceled)

35 . The method of claim 30 , wherein the policy comprises a neural network comprising a plurality of layers and a plurality of trainable weights, and modifying the policy comprises modifying the trainable weights of the neural network.

36 . The method of claim 35 , comprising:

initializing the trainable weights; and

generating the initial version of a quantum program based on the neural network comprising the initialized trainable weights.

37 . The method of claim 36 , comprising initializing the trainable weights to random values or initializing the trainable weights based on a one-step reward associated with a set of actions and observations.

38 - 39 . (canceled)

40 . The method of claim 20 , wherein a Hamiltonian of the representation of the problem admits a diagonal form with respect to a computational basis of the representation of the problem, and the method comprises, after generating an updated version of the quantum program:

determining angles of rotation gates present in the updated version of the quantum program; and

terminating the iterative process if any angles of X rotation gates deviate from π by more than π/2 radians.