Variational analog quantum oracle learning

1 . A method for training a quantum machine to provide candidate parameters for optimization operations based on oracle outputs comprising:

configuring a quantum annealer based on a set of coupling parameters of a Hamiltonian;

performing quantum annealing using the configured quantum annealer to obtain annealer output samples, wherein each sample of the annealer output samples indicates state values of qubits of the quantum annealer; and

for each pairwise combination of indices of the annealer output samples:

determining a first subset of the annealer output samples, wherein a product of state values indexed by the pairwise combination of indices is positive for each sample of the first subset;

determining a second subset of the annealer output samples, wherein a product of state values indexed by the pairwise combination of indices is negative for each sample of the second subset;

providing the first subset of annealer output samples to an oracle executing on a classical computing system to obtain a first set of oracle outputs;

providing the second subset of annealer output samples to the oracle to obtain a second set of oracle outputs;

determining a first expectation based on the first set of oracle outputs and a second expectation based on the second set of oracle outputs;

determining a comparison value between the first and second expectations; and

updating a coupling parameter indexed by the pairwise combination of indices based on the comparison value and a learning rate parameter.

2 . The method of claim 1 , further comprising:

obtaining a relaxation factor between zero and one;

determining whether a sample count associated with a first pair of indices is less than a sample count threshold, wherein determining the first expectation comprises determining a sample count of the first subset; and

in response to a determination that the sample count is less than the sample count threshold, updating a first coupling parameter associated with the first pair of indices by multiplying the first coupling parameter by the relaxation factor.

3 . The method of claim 2 , further comprising setting the sample count threshold to a value less than m/2, where m is a total count of the annealer output samples.

4 . The method of claim 1 , further comprising:

determining that a first set of variables are correlated based on a determination that the first set of variables satisfy a correlation threshold; and

mapping a first qubit and a second qubit of the quantum annealer with the first set of variables based on a determination that the first and second qubits are coupled qubits.

5 . The method of claim 1 , wherein the learning rate parameter is a first learning rate parameter, further comprising:

obtaining a set of anneal offsets; and

updating the set of anneal offsets based on a second learning rate parameter, wherein updating the coupling parameter comprises updating the coupling parameter based on the set of anneal offsets.

6 . One or more tangible, non-transitory, machine-readable media storing instructions that, when executed by one or more processors, effectuate operations comprising:

configuring a quantum annealer based on a set of coupling parameters of a Hamiltonian;

performing a quantum annealing using the configured quantum annealer to obtain annealer output samples, wherein each sample of the annealer output samples indicates state values of qubits of the quantum annealer for a set of indices of the annealer output samples;

determining a first subset of the annealer output samples, wherein a product of state values indexed by the set of indices is positive for each sample of the first subset;

determining a second subset of the annealer output samples, wherein a product of state values indexed by the set of indices is negative for each sample of the second subset;

providing, to an oracle, the first and second subsets of annealer output samples to obtain a first set of oracle outputs and a second set of oracle outputs;

determining a first expectation based on the first set of oracle outputs and a second expectation based on the second set of oracle outputs;

determining a comparison value between the first and second expectations; and

updating a coupling parameter indexed by the set of indices based on the comparison value and a learning rate parameter.

7 . The media of claim 6 , wherein configuring the quantum annealer comprises randomly mapping a first qubit of the quantum annealer to a first variable of the Hamiltonian.

8 . The media of claim 6 , wherein configuring the quantum annealer comprises:

determining a correlation value between a first variable of the Hamiltonian and a second variable of the Hamiltonian;

determining whether the correlation value satisfies a correlation criterion;

in response to a determination that the correlation value satisfies the correlation criterion, selecting a first qubit and a second qubit based on a determination that the first and second qubits within a proximity threshold of each other; and

mapping the first qubit to the first variable and the second qubit to the second variable.

9 . The media of claim 8 , wherein determining whether the correlation value satisfies the correlation criterion comprises determining whether the correlation value is greater than a correlation threshold.

10 . The media of claim 6 , the operations further comprising:

obtaining an initial set of state values representing a local minimum energy state, wherein:

configuring the quantum annealer comprises configuring the quantum annealer based on the initial set of state values;

performing the quantum annealing comprises performing a set of reverse annealing operations to determine a set of reverse-annealed states; and

performing the set of reverse annealing operations comprises reducing a field strength of the quantum annealer based on a reversal distance;

determining a third set of oracle outputs based on the set of reverse-annealed states; and

updating the reversal distance based on the third set of oracle outputs.

11 . The media of claim 10 , wherein reducing the field strength of the quantum annealer comprises reducing the field strength linearly with respect to time.

12 . The media of claim 6 , wherein:

the quantum annealing is a first quantum annealing;

updating the coupling parameter comprises updating the coupling parameter without updating a field value parameter of the Hamiltonian;

the annealer output samples is a first set of annealer output samples;

the comparison value is a first comparison value;

the operations further comprising:

performing a second quantum annealing after configuring the quantum annealer based on the updated coupling parameter to obtain a second set of annealer output samples, wherein each sample of the annealer output samples is associated with a first index value;

providing, to the oracle, the second set of annealer output samples to obtain a third set of oracle outputs; and

updating a field value parameter indexed by the first index value based on the third set of oracle outputs and the learning rate parameter.

13 . The media of claim 12 , wherein the learning rate parameter is a first learning rate parameter, further comprising:

performing a third quantum annealing after configuring the quantum annealer based on the updated field value parameter to obtain a third set of annealer output samples, wherein each sample of the third set of annealer output samples indicates is associated with the first index value;

providing, to the oracle, the third set of annealer output samples to obtain a fourth set of oracle outputs; and

updating an anneal offset indexed by the first index value based on the fourth set of oracle outputs and a second learning rate parameter.

14 . A system comprising:

one or more processors; and

memory storing computer program instructions that, when executed by the one or more processors, cause the one or more processors to effectuate operations comprising:

configuring a quantum annealer based on a parameter of a Hamiltonian;

performing annealing using the configured quantum annealer to obtain output samples, wherein each sample of the output samples indicates state values of elements of the quantum annealer;

providing, to an oracle, the output samples to obtain a set of oracle outputs;

determining an expectation based on the set of oracle outputs; and

updating the parameter based on the expectation and a learning rate parameter.

15 . The system of claim 14 , wherein updating the parameter comprises updating the parameter based on a relaxation factor, wherein the relaxation factor is between zero and one.

16 . The system of claim 14 , the operations further comprising:

obtaining a maximum change threshold; and

determining whether a change in energy between a first epoch and a second epoch satisfies the maximum change threshold, wherein:

the energy indicates a Hamiltonian energy of a set of qubits of the quantum annealer; and

updating the parameter comprises updating the parameter in response to a determination that the maximum change threshold is not satisfied.

17 . The system of claim 14 , the operations further comprising:

obtaining a configuration weight;

determining a first value based on the configuration weight;

determining a second value based on the configuration weight, wherein the second value is greater than the first value; and

determining a ratio between the first value and the second value, wherein updating the parameter comprises updating the parameter based on the ratio.

18 . The system of claim 14 , wherein configuring the quantum annealer comprises:

mapping a first subset of qubits of the quantum annealer to variables of the Hamiltonian;

mapping a second subset of qubits of the quantum annealer to a set of hidden units representing higher-order interactions between the variables, wherein the second subset of qubits are all initialized to the same state value; and

setting an upper limit to anneal offsets of the second subset of qubits.

19 . The system of claim 14 , the operations further comprising:

determining an expectation for a state of a qubit based on a state values of the output samples of the quantum annealer associated with the qubit, wherein a variable is mapped to the qubit; and

determining a value of the variable based on the expectation.

20 . The system of claim 14 , the operations further comprising:

obtaining an anneal offset for a qubit of the quantum annealer;

selecting an anneal offset difference based on a probability distribution;

wherein configuring the quantum annealer comprises configuring the quantum annealer to update a field applied on the qubit based on the anneal offset;

evolving the state of the quantum annealer to obtain a first set of output samples;

re-configuring the quantum annealer based on a sum of the anneal offset and the anneal offset difference;

evolving the re-configured quantum annealer to obtain a second set of output samples;

determining a first energy value based on the first set of output samples and the parameter of the Hamiltonian;

determining a second energy value based on the second set of output samples and the parameter of the Hamiltonian; and

updating the anneal offset for the qubit based on a difference between the first energy value and the second energy value.