Machine learning mapping for quantum processing units

1. A non-transitory, computer-readable medium storing instructions that when executed effectuate operations comprising:

obtaining, with a classical computer system, an objective function to be solved by a quantum computing system, wherein:

the objective function involves more variables than any of one or more quantum computers of the quantum computing system have logical qubits;

decomposing, with the classical computer system, the objective function into a plurality of sub-problems, each of the sub-problems involving as many or fewer variables than the one or more quantum computers have logical qubits;

causing, with the classical computer system, the quantum computing system to solve each of the sub-problems, wherein solving each of the sub-problems comprises:

obtaining a set of raw outputs from a plurality of shots run by the quantum computing system processing the given sub-problem, wherein obtaining a raw output from a shot run by the quantum computing system processing the given sub-problem comprises:

initializing initial states of the logical qubits of the quantum computing system, the initial states of the logical qubits corresponding to values of at least some variables of the objective function;

applying a given set of perturbations to the logical qubits; and

measuring the perturbed states of the logical qubits; and

wherein the measured perturbed states of the logical qubits represent raw output from the shot run; and

determining a solution to the given sub-problem based on an expectation value determined from the set of raw outputs, wherein determining the expectation value comprises:

determining a subset of the raw outputs having a greater than a threshold frequency of occurrence and determining the expectation value based on a measure of central tendency of the subset;

aggregating solutions to the sub-problems to determine a solution to the objective function, wherein aggregating solutions to the sub-problems to determine a solution to the objective function comprises determining the solution to the objective function based on the measured perturbed states of the logical qubits; and

storing, with the classical computer system, the solution to the objective function in memory.

2. The medium of claim 1 , wherein decomposing the objective function into the plurality of sub-problems comprises:

decomposing the objective function based on a type of qubit with which the quantum computing system is implemented.

3. The medium of claim 1 , wherein:

the variables are unknown variables of the objective function to be determined by solving the objective function.

4. The medium of claim 1 , wherein the operations comprise:

iteratively, through a plurality of iterations, adjusting initial state values of the variables by, for each iteration, causing qubits of the quantum computing system to be initialized to current initial state values, obtaining outputs of the quantum computing system responsive to the current initial state values, determining adjustments to the current initial state values for a next iteration based on results of applying the outputs of the quantum computing system responsive to the current initial state values to the objective function.

5. The medium of claim 4 , wherein the objective function is differentiable, and wherein the adjustments are determined by computing partial derivatives of the initial state values with respect to the objective function and determining, based on the partial derivatives, directions in which to adjust the initial state values between iterations.

6. The medium of claim 1 , wherein:

at least some of the sub-problems are solved by different quantum computers of the quantum computing system; and

at least some of the sub-problems are solved by the same quantum computer of the quantum computing system at different times.

7. The medium of claim 1 , wherein:

the solution to the objective function comprises finding values of the variables that produce a local or global optimum of the objective function; and

decomposing the objective function comprises decomposing a first quantum circuit into a plurality of smaller quantum circuits.

8. The medium of claim 1 , wherein:

decomposing the objective function into the plurality of sub-problems comprises steps for decomposing the objective function; and

the quantum computing system comprises means for quantum computing.

9. The medium of claim 1 , wherein initializing initial states of the logical qubits comprises initializing initial states of the logical qubits for a plurality of shots;

wherein applying a given set of perturbations to the logical qubits comprises applying the given set of perturbations to the logical qubits for the plurality of shots;

wherein measuring the perturbed states of the logical qubits comprises measuring the perturbed states of the logical qubits for the plurality of shots; and

wherein determining the solution to the objective function based on the measured perturbed states of the logical qubits comprises:

determining an expectation value based on the measured perturbed states of the logical qubits for the plurality of shots; and

determining the solution to the objective function based on the expectation value.

10. The medium of claim 9 , wherein determining the expectation value comprises determining the expectation value based on a probability distribution of the measured perturbed states of the logical qubits for the plurality of shots.

11. The medium of claim 9 , wherein determining the expectation value comprises determining the expectation value based on a subset of measured perturbed states which have a frequency of occurrence greater than a threshold.

12. The medium of claim 9 , further comprising training a machine learning model, based on a subset of the plurality of shots, to determine the given set of perturbations to apply to the logical qubits based on the initial states of the logical qubits.

13. The medium of claim 9 , further comprising training a machine learning model, based on a subset of the plurality of shots, to determine the initial states of the logical qubits based on the applied given set of perturbations.

14. A non-transitory, computer-readable medium storing instructions that when executed effectuate operations comprising:

obtaining, with a classical computer system, an objective function to be solved by a quantum computing system, wherein:

the objective function involves more variables than any of one or more quantum computers of the quantum computing system have logical qubits;

decomposing, with the classical computer system, the objective function into a plurality of sub-problems, each of the sub-problems involving as many as or fewer variables than the one or more quantum computers have logical qubits;

causing, with the classical computer system, the quantum computing system to solve each of the sub-problems, wherein solving each of the sub-problems comprises:

obtaining a set of raw outputs from a plurality of shots run by the quantum computing system processing the given sub-problem, wherein obtaining a raw output from a shot run by the quantum computing system processing the given sub-problem comprises:

initializing initial states of the logical qubits of the quantum computing system, the initial states of the logical qubits corresponding to values of at least some variables of the objective function;

applying a given set of perturbations to the logical qubits; and

measuring the perturbed states of the logical qubits; and

wherein the measured perturbed states of the logical qubits represent raw output from the shot run;

statistically aggregating solutions to the sub-problems to determine a solution to the objective function, the solution to the objective function comprise an expectation value solving the objective function wherein:

statistically aggregating comprises determining the expectation value based on a probability distribution of at least some of the raw outputs of the plurality of sub-problems; and

determining the expectation value comprises determining a subset of the raw outputs of each of the sub-problems having greater than a threshold of occurrence in the raw outputs and determining a central tendency based on the subset; and

storing, with the classical computer system, the solution to the objective function in memory.

15. The medium of claim 14 , wherein

aggregating solutions to the sub-problems to determine a solution to the objective function comprises determining the solution to the objective function based on the measured perturbed states of the logical qubits.

16. The medium of claim 14 , further comprising determining, with the classical computer system, a direction in parameter space in which to adjust parameters to further optimize the objective function and/or the plurality of sub-problems based on the expectation value.

17. A non-transitory, computer-readable medium storing instructions that when executed effectuate operations comprising:

generating a set of training data for solving a multi-variable problem on a quantum computing system, comprising:

initializing initial states of a set of qubits of the quantum computing system, wherein qubits of the set of qubits correspond to variables of the multi-variable problem;

applying a given set of multi-variable perturbations to the set of qubits;

measuring the perturbed states of the set of qubits corresponding to the initial states;

determining an expectation value of the multi-variable problem based on the measured perturbed states;

adjusting, based on the expectation value of the multi-variable problem, the initial states of the set of qubits to generate adjusted states;

initializing the adjusted states of the set of qubits of the quantum computing system;

applying the given set of multi-variable perturbations to the set of qubits;

measuring the perturbed states of the set of qubits corresponding to the adjusted states;

determine an adjusted expectation value of the multi-variable problem based on the measured perturbed states; and

generating the set of training data from the expectation values corresponding to the adjusted states and adjustment values corresponding to differences between the initial states and the adjusted states of the set of qubits;

based on the set of training data, training a machine learning model to adjust the set of qubits to optimize the value of the multi-variable problem; and

storing parameters of the trained machine learning model in memory.

18. The medium of claim 17 , wherein the multi-variable problem comprises more variables than the quantum computing system has qubits;

wherein generating the set of training data comprises;

mapping the multi-variable problem into a plurality of sub-problems comprising equal or fewer variables than the quantum computing system has qubits; and

generating the set of training data for solving each of the plurality of sub-problems; and

wherein training the machine learning model to optimize the value of the multi-variable problem comprises training a plurality of machine learning models to optimize the value of the plurality of sub-problems; and

wherein storing the parameters of the trained machine learning model in memory comprises storing parameters of the plurality of machine learning models in memory.

19. The medium of claim 17 , wherein the multi-variable problem comprises more variables than the quantum computing system has qubits and

wherein generating the set of training data comprises;

mapping the multi-variable problem into a plurality of sub-problems comprising equal or fewer variables than the quantum computing system has qubits; and

generating the set of training data for solving each of the plurality of sub-problems; and

wherein training the machine learning model to optimize the value of the multi-variable problem comprises training the machine learning model to map the multi-variable problem into the plurality of sub-problems.

20. The medium of claim 17 , wherein training the machine learning model to optimize the value of the multi-variable problem comprises training the machine learning model to determine initial states of the set of qubits.

21. The medium of claim 17 , wherein training the machine learning model to optimize the value of the multi-variable problem comprises training the machine learning model to determine a given set of multi-variable perturbations to be applied to the set of qubits.

22. A non-transitory, computer-readable medium storing instructions that when executed effectuate operations comprising:

generating a set of training data for solving a multi-variable problem on a quantum computing system, wherein the multi-variable problem comprises more variables than the quantum computing system has qubits, comprising:

mapping the multi-variable problem into a plurality of sub-problems comprising equal or fewer variables than the quantum computing system has qubits; and

generating the set of training data for solving each of the plurality of sub-problems by:

for each of a set of multi-variable perturbations:

initializing initial states of a set of qubits of the quantum computing system, wherein qubits of the set of qubits correspond to variables of the multi-variable problem;

applying a given set of multi-variable perturbations to the set of qubits by:

initializing initial states of logical qubits of the quantum computing system, the initial states of the logical qubits corresponding to values of at least some variables of the multi-variable problem;

applying a given set of perturbations to the logical qubits;

measuring the perturbed states of the set of qubits; and

determine a value of the multi-variable problem based on the measured perturbed states of the set of qubits;

based on the set of training data, training a machine learning model to optimize the value of the multi-variable problem, wherein:

training the machine learning model to optimize the value of the multi-variable problem comprises training a plurality of machine learning models to optimize the value of the plurality of sub-problems; and

training the machine learning model to optimize the value of the multi-variable problem comprises training the machine learning model to map the multi-variable problem into the plurality of sub-problems;

wherein training the machine learning model to optimize the value of the multi-variable problem comprises training the machine learning model to determine initial states of the set of qubits;

wherein training the machine learning model to optimize the value of the multi-variable problem comprises training the machine learning model to determine a given set of multi-variable perturbations to be applied to the set of qubits; and

storing parameters of the trained machine learning model in memory, wherein storing the parameters of the trained machine learning model in memory comprises storing parameters of the plurality of machine learning models in memory.