Neural network-based quantum error correction decoding method and apparatus, and chip
View Patent ↗This application discloses a neural network-based QEC decoding method. The method includes: obtaining error syndrome information of a quantum circuit; performing block feature extraction on the error syndrome information by using a neural network decoder, to obtain feature information; and performing fusion decoding processing on the feature information by using the neural network decoder, to obtain error result information, the error result information being used for determining a data qubit in which an error occurs in the quantum circuit and a corresponding error type. In this application, a block feature extraction manner is used, a quantity of channels of feature information obtained by each feature extraction is reduced, and inputted data of next feature extraction is reduced, which reduces a quantity of feature extraction layers in a neural network decoder. Therefore, a decoding time used by the neural network decoder is reduced, thereby achieving real-time error correction.
1. A neural network-based quantum error correction decoding method performed at a computer device, the method comprising:
obtaining error syndrome information of a quantum circuit, the error syndrome information being a data array formed by eigenvalues of a stabilizer generator of a quantum error correction code;
performing block feature extraction on the error syndrome information by using a neural network decoder, to obtain feature information, a feature extraction layer of the neural network decoder being configured to perform the block feature extraction on inputted data, the block feature extraction being configured to, after the inputted data is partitioned into at least two blocks, perform parallel feature extraction on the at least two blocks by using at least two feature extraction units; and
performing fusion decoding processing on the feature information by using the neural network decoder, to obtain error result information, the error result information being used for determining a data qubit in which an error occurs in the quantum circuit and a corresponding error type.
2. The method according to claim 1 , wherein the neural network decoder comprises m cascaded feature extraction layers, m being a positive integer; and
the performing block feature extraction on the error syndrome information by using a neural network decoder, to obtain feature information comprises:
performing block feature extraction on the error syndrome information by using the m feature extraction layers, to obtain the feature information,
the first feature extraction layer being configured to perform block feature extraction on the error syndrome information, the i th feature extraction layer being configured to perform block feature extraction on a feature extraction result of a previous feature extraction layer, i being an integer greater than 1 and less than or equal to m.
3. The method according to claim 1 , wherein the error syndrome information comprises T data arrays, and each data array is obtained by performing one error syndrome measurement on the quantum circuit by using the quantum error correction code, T being an integer greater than 1; and
after the obtaining error syndrome information of a quantum circuit, the method further comprises:
classifying the error syndrome information into at least two data units,
one data unit comprising T array units at a same position in the T data arrays.
4. The method according to claim 1 , wherein the error syndrome information is real error syndrome information obtained by performing error syndrome measurement with noise on the quantum circuit by using the quantum error correction code; and the neural network decoder comprises a first decoder and a second decoder;
the first decoder is configured to decode the real error syndrome information, to obtain a logic error class corresponding to the real error syndrome information, the logic error class being a class obtained after an error occurred in the quantum circuit is mapped;
the second decoder is configured to decode the real error syndrome information, to obtain perfect error syndrome information corresponding to the real error syndrome information, the perfect error syndrome information being information obtained by performing error syndrome measurement without noise on the quantum circuit; and
after the performing fusion decoding processing on the feature information by using the neural network decoder, to obtain error result information, the method further comprises:
determining, according to the logic error class and the perfect error syndrome information, the data qubit in which the error occurs in the quantum circuit and the corresponding error type.
5. The method according to claim 4 , wherein the determining, according to the logic error class and the perfect error syndrome information, the data qubit in which the error occurs in the quantum circuit and the corresponding error type comprises:
obtaining a first error result corresponding to the logic error class;
obtaining a second error result corresponding to the perfect error syndrome information; and
determining, according to the first error result and the second error result, the data qubit in which the error occurs in the quantum circuit and the corresponding error type.
6. The method according to claim 4 , wherein the obtaining a first error result corresponding to the logic error class comprises:
selecting any element from elements comprised in the logic error class as the first error result,
the logic error class comprising at least one equivalent error element.
7. The method according to claim 4 , wherein the obtaining a second error result corresponding to the perfect error syndrome information comprises:
looking up a mapping table to obtain simple errors respectively corresponding to error syndrome points in the perfect error syndrome information, the mapping table comprising a mapping relationship between at least one group of error syndrome points and simple errors; and
multiplying the simple errors respectively corresponding to the error syndrome points, to obtain the second error result.
8. The method according to claim 4 , wherein the determining, according to the first error result and the second error result, the data qubit in which the error occurs in the quantum circuit and the corresponding error type comprises:
calculating a product of the first error result and the second error result, to obtain the data qubit in which the error occurs in the quantum circuit and the corresponding error type.
9. The method according to claim 4 , wherein there are k second decoders, k being a positive integer and k being related to a length of the quantum error correction code;
the k second decoders are configured to respectively decode the real error syndrome information, to obtain k perfect error syndrome bits; and
the k perfect error syndrome bits are used for performing merging to obtain the perfect error syndrome information.
10. The method according to claim 1 , wherein training data of the neural network decoder is generated by:
probabilistically generating an error on a physical qubit comprised in a sample quantum circuit;
probabilistically generating an error on an auxiliary qubit corresponding to the sample quantum circuit, the auxiliary qubit being used for performing measurement to obtain error syndrome information of the sample quantum circuit;
probabilistically generating an error on a controlled NOT gate comprised in an eigenvalue measurement circuit corresponding to the sample quantum circuit, the eigenvalue measurement circuit being configured to measure an eigenvalue of a stabilizer generator;
probabilistically generating a measurement error in a case that error syndrome measurement is performed on the sample quantum circuit by using a quantum error correction code; and
obtaining the error syndrome information and error result information of the sample quantum circuit, and generating the training data.
11. The method according to claim 1 , wherein training data of the neural network decoder is generated by:
performing quantum process tomography on a sample quantum circuit, and extracting a noise model of the sample quantum circuit;
simulating an evolution of a quantum state of the sample quantum circuit under the action of noise based on the noise model; and
obtaining the error syndrome information and error result information of the sample quantum circuit, and generating the training data.
12. The method according to claim 1 , further comprising:
generating an error-correction control signal according to the error result information, the error-correction control signal being used for correcting an error generated by the quantum circuit; and
transmitting the error-correction control signal to the quantum circuit.
13. A computer device, comprising a processor and a memory, the memory storing one or more instructions, the one or more instructions being loaded and executed by the processor to perform a plurality of operations including:
obtaining error syndrome information of a quantum circuit, the error syndrome information being a data array formed by eigenvalues of a stabilizer generator of a quantum error correction code;
performing block feature extraction on the error syndrome information by using a neural network decoder, to obtain feature information, a feature extraction layer of the neural network decoder being configured to perform the block feature extraction on inputted data, the block feature extraction being configured to, after the inputted data is partitioned into at least two blocks, perform parallel feature extraction on the at least two blocks by using at least two feature extraction units; and
performing fusion decoding processing on the feature information by using the neural network decoder, to obtain error result information, the error result information being used for determining a data qubit in which an error occurs in the quantum circuit and a corresponding error type.
14. The computer device according to claim 13 , wherein the neural network decoder comprises m cascaded feature extraction layers, m being a positive integer; and
the performing block feature extraction on the error syndrome information by using a neural network decoder, to obtain feature information comprises:
performing block feature extraction on the error syndrome information by using the m feature extraction layers, to obtain the feature information,
the first feature extraction layer being configured to perform block feature extraction on the error syndrome information, the i th feature extraction layer being configured to perform block feature extraction on a feature extraction result of a previous feature extraction layer, i being an integer greater than 1 and less than or equal to m.
15. The computer device according to claim 13 , wherein the error syndrome information comprises T data arrays, and each data array is obtained by performing one error syndrome measurement on the quantum circuit by using the quantum error correction code, T being an integer greater than 1; and the plurality of operations further comprise:
after obtaining error syndrome information of a quantum circuit;
classifying the error syndrome information into at least two data units,
one data unit comprising T array units at a same position in the T data arrays.
16. The computer device according to claim 13 , wherein the error syndrome information is real error syndrome information obtained by performing error syndrome measurement with noise on the quantum circuit by using the quantum error correction code; and the neural network decoder comprises a first decoder and a second decoder;
the first decoder is configured to decode the real error syndrome information, to obtain a logic error class corresponding to the real error syndrome information, the logic error class being a class obtained after an error occurred in the quantum circuit is mapped;
the second decoder is configured to decode the real error syndrome information, to obtain perfect error syndrome information corresponding to the real error syndrome information, the perfect error syndrome information being information obtained by performing error syndrome measurement without noise on the quantum circuit; and
the plurality of operations further comprise:
after performing fusion decoding processing on the feature information by using the neural network decoder, determining, according to the logic error class and the perfect error syndrome information, the data qubit in which the error occurs in the quantum circuit and the corresponding error type.
17. A non-transitory computer-readable storage medium, storing one or more instructions, the one or more instructions being loaded and executed by a processor of a computer device to perform a plurality of operations including:
obtaining error syndrome information of a quantum circuit, the error syndrome information being a data array formed by eigenvalues of a stabilizer generator of a quantum error correction code;
performing block feature extraction on the error syndrome information by using a neural network decoder, to obtain feature information, a feature extraction layer of the neural network decoder being configured to perform the block feature extraction on inputted data, the block feature extraction being configured to, after the inputted data is partitioned into at least two blocks, perform parallel feature extraction on the at least two blocks by using at least two feature extraction units; and
performing fusion decoding processing on the feature information by using the neural network decoder, to obtain error result information, the error result information being used for determining a data qubit in which an error occurs in the quantum circuit and a corresponding error type.
18. The non-transitory computer-readable storage medium according to claim 17 , wherein the neural network decoder comprises m cascaded feature extraction layers, m being a positive integer; and
the performing block feature extraction on the error syndrome information by using a neural network decoder, to obtain feature information comprises:
performing block feature extraction on the error syndrome information by using the m feature extraction layers, to obtain the feature information,
the first feature extraction layer being configured to perform block feature extraction on the error syndrome information, the i th feature extraction layer being configured to perform block feature extraction on a feature extraction result of a previous feature extraction layer, i being an integer greater than 1 and less than or equal to m.
19. The non-transitory computer-readable storage medium according to claim 17 , wherein the error syndrome information comprises T data arrays, and each data array is obtained by performing one error syndrome measurement on the quantum circuit by using the quantum error correction code, T being an integer greater than 1; and the plurality of operations further comprise:
after obtaining error syndrome information of a quantum circuit;
classifying the error syndrome information into at least two data units,
one data unit comprising T array units at a same position in the T data arrays.
20. The non-transitory computer-readable storage medium according to claim 17 , wherein the error syndrome information is real error syndrome information obtained by performing error syndrome measurement with noise on the quantum circuit by using the quantum error correction code; and the neural network decoder comprises a first decoder and a second decoder;
the first decoder is configured to decode the real error syndrome information, to obtain a logic error class corresponding to the real error syndrome information, the logic error class being a class obtained after an error occurred in the quantum circuit is mapped;
the second decoder is configured to decode the real error syndrome information, to obtain perfect error syndrome information corresponding to the real error syndrome information, the perfect error syndrome information being information obtained by performing error syndrome measurement without noise on the quantum circuit; and
the plurality of operations further comprise:
after performing fusion decoding processing on the feature information by using the neural network decoder, determining, according to the logic error class and the perfect error syndrome information, the data qubit in which the error occurs in the quantum circuit and the corresponding error type.