Neural network establishing method for quantum readout, method for mitigating quantum readout error using neural network and apparatus thereof

Patent metadata

Field	Value
Publication number	US-12488272-B2
Application number	US-202218074828-A
Country	US
Kind code	B2
Filing date	Dec 5, 2022
Priority date	Dec 3, 2021
Publication date	Dec 2, 2025
Grant date	Dec 2, 2025

Abstract

The present disclosure relates to a quantum computer technology and a method in which a learning apparatus a neural network for quantum readout acquires a plurality of actual measurement results including noise in quantum readout using a quantum circuit, acquires an ideal measurement result of the quantum circuit in correspondence to each of the plurality of actual measurement results including noise, creates training data from a set of the actual measurement results including noise and the ideal measurement results, and trains a neural network for mitigating errors, which are generated in quantum readout, using the created training data.

First claim

What is claimed is:

1 . A method in which a learning apparatus using at least one processor constructs a neural network for quantum readout, the method comprising:
acquiring a plurality of actual measurement results including noise in quantum readout using a quantum circuit by means of the learning apparatus; acquiring an ideal measurement result of the quantum circuit in correspondence to each of the plurality of actual measurement results including noise by means of the learning apparatus; creating training data from a set of the actual measurement results including noise and the ideal measurement results by means of the learning apparatus; and training a neural network for mitigating errors, which are generated in quantum readout, using the created training data by means of the learning apparatus.

2 . The method of claim 1 , wherein the acquiring of actual measurement results creates a noise probability distribution showing a plurality of actual measurement results including noise in quantum readout by applying certain single qubit rotation to a qubit using a quantum circuit composed of single qubit gates.

3 . The method of claim 2 , wherein the quantum circuit creates the noise probability distribution by applying an angle randomly and independently created to all qubits of a qubit system for a gate corresponding to rotation about one axis of Bloch sphere.

4 . The method of claim 1 , wherein the acquiring of ideal measurement results calculates each ideal probability distribution showing the ideal measurement results of the quantum circuit by measuring qubits on the basis of computation using a rotation angle of the quantum circuit in correspondence to the plurality of actual measurement results including noise.

5 . The method of claim 1 , wherein the creating of training data
sets a noise probability distribution showing the actual measurement results including noise as input of the neural network, sets an ideal probability distribution showing the ideal measurement results as output of the neural network, and creates training data composed of each set by matching the noise probability distribution and the ideal probability distribution.

6 . The method of claim 1 , wherein the training of a neural network uses a deep learning model composed of an input layer showing a probability of measuring a computational base state, a hidden layer composed of a fully connected layer, and an output layer showing a probability of measuring computational base state in an ideal case, and
a noise probability distribution showing actual measurement results and an ideal probability distribution showing the ideal measurement results of the training data are set as input of the input layer and output of the output layer, respectively, and an ideal measurement result is inferred from an actual measurement result, thereby training the deep learning model for mitigating errors that are generated in quantum readout.

7 . The method of claim 6 , wherein the deep learning model uses a Rectified Linear Unit (ReLU) as an activation function of each hidden node,
the output layer uses a softmax function such that output shows a probability distribution, a loss function optimizes a weight and bias of a neural network using categorical cross entropy, and a free parameter is updated by an optimizer in which a hyperparmeter such as a learning rate is heuristically selected.

8 . One or more non-transitory computer-readable medium storing one or more instructions, wherein the one or more instructions that can be executed by one or more processors
acquires a plurality of actual measurement results including noise in quantum readout using a quantum circuit, acquires an ideal measurement result of the quantum circuit in correspondence to each of the plurality of actual measurement results including noise, creates training data from a set of the actual measurement results including noise and the ideal measurement results, and trains a neural network for mitigating errors, which are generated in quantum readout, using the created training data.

9 . A method of reducing quantum readout errors, the method comprising:
acquiring a measurement result of a readout object by performing quantum readout by means of a quantum computer; inputting the acquired measurement result of the readout object into a neural network previously constructed in relation to errors, which are generated in quantum readout, by means of the quantum computer; and inferring an ideal measurement result corresponding to the measurement result of the readout object using the neural network by means of the quantum computer, the neural network is constructed by acquiring a plurality of actual measurement results including noise in quantum readout using a quantum circuit, acquiring an ideal measurement result of the quantum circuit in correspondence to each of the plurality of actual measurement results including noise, creating training data from a set of the actual measurement results including noise and the ideal measurement results, and training a neural network for mitigating errors that are generated in quantum readout using the created training data.

10 . The method of claim 9 , wherein the acquiring a measurement result of a readout object acquires a probability distribution, in which a final state of a qubit is reduced in a quantum state, by performing quantum readout through computation-based projective measurement.

11 . The method of claim 9 , wherein the neural network creates a noise probability distribution showing a plurality of actual measurement results including noise in quantum readout by applying certain single qubit rotation to a qubit using a quantum circuit composed of single qubit gates.

12 . The method of claim 11 , wherein the quantum circuit creates the noise probability distribution by applying an angle randomly and independently created to all qubits of a qubit system for a gate corresponding to rotation about one axis of Bloch sphere.

13 . The method of claim 9 , wherein the neural network calculates each ideal probability distribution showing the ideal measurement results of the quantum circuit by measuring qubits on the basis of computation using a rotation angle of the quantum circuit in correspondence to the plurality of actual measurement results including noise.

14 . The method of claim 9 , wherein the neural network uses a deep learning model composed of an input layer showing a probability of measuring a computational base state in actual measurement, a hidden layer composed of a fully connected layer, and an output layer showing a probability of measuring computational base state in an ideal case; the noise probability distribution showing an actual measurement result and the ideal probability distribution showing an ideal measurement result of the training data are set as an input of the input layer and output of the output layer, respectively; and an ideal measurement result is inferred from an actual measurement result, thereby training the deep learning model for mitigating errors that are generated in quantum readout.

15 . An apparatus for reducing quantum readout errors, the apparatus comprising:
a neural network previously constructed in relation to errors that are generated in quantum readout; and a quantum computer having a qubit controller, acquiring a measurement result of a readout object by performing quantum readout, inputting the acquired measurement result of the readout object into the neural network, and inferring an ideal measurement result corresponding to the measurement result of the readou

Assignees

• Research & Business Found Sungkyunkwan Univ

Inventors

• Oh Byung Du
• Kim Jihye
• Park Kyungdeock
• Chong Yonuk
• Hwang Euyheon

Classifications

• G06N10/70 Primary

  Quantum error correction, detection or prevention, e.g. surface codes or magic state distillation · CPC title

• B82Y10/00

  Nanotechnology for information processing, storage or transmission, e.g. quantum computing or single electron logic · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title