Quantum convolutional neural network circuit

Patent metadata

Field	Value
Publication number	US-12586367-B2
Application number	US-202318528155-A
Country	US
Kind code	B2
Filing date	Dec 4, 2023
Priority date	Dec 5, 2022
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

Aspects of the present disclosure relate generally to systems and methods for classifying an object in an image using a quantum convolutional neural network (QCNN). A method includes receiving the image; compressing, using an autoencoder, the image by extracting a plurality of features; encoding the plurality of features into quantum states by applying a quantum feature map corresponding to a quantum circuit unique to the image, where parameters of the quantum circuit depend on pixel values in the image; inputting the encoded plurality of features into the QCNN that is trained to detect the object and generate a classification of the object; and outputting the classification of the object generated by the QCNN.

First claim

What is claimed is:

1 . A method for classifying an object in an image using a quantum convolutional neural network (QCNN), comprising:
receiving the image depicting the object; compressing, using an autoencoder, the image by extracting a plurality of features; encoding the plurality of features into quantum states using a quantum feature map that corresponds to a quantum encoding circuit unique to the image, wherein parameters of the quantum encoding circuit depend on pixel values in the image; inputting the encoded plurality of features into the QCNN that is executed on a parameterized quantum circuit (PQC) and trained to detect the object and generate a classification of the object from a plurality of classes, wherein the QCNN comprises a plurality of layers including one or more convolutional layers that apply a unitary operator to neighboring qubits in the one or more convolutional layers; and outputting the classification of the object generated by the QCNN.

2 . The method of claim 1 , wherein outputting the classification of the object generated by the QCNN comprises:
receiving one or more measurements from the QCNN; and inputting the one or more measurements into a fully connected neural network that is configured to output the classification from a plurality of classes.

3 . The method of claim 1 , wherein the quantum encoding circuit includes one qubit per feature of the plurality of features.

4 . The method of claim 1 , wherein the unitary operator is a quasi-local unitary operator applied in a translationally invariant way for a finite depth.

5 . The method of claim 1 , wherein the unitary operator is a two qubit unitary operator that applies an arbitrary unitary operation using CNOT gates and single qubit rotations.

6 . The method of claim 1 , wherein the quantum encoding circuit is composed of tensor objects, and wherein a controller translates the tensor objects into native hardware gates.

7 . The method of claim 1 , wherein encoding the plurality of features comprises applying one of: product state encoding or unary encoding.

8 . The method of claim 1 , wherein the autoencoder is one of: an autoencoder executing principal component analysis, a fully connected autoencoder, or a locality preserving variational autoencoder.

9 . The method of claim 8 , wherein the autoencoder is the autoencoder executing principal component analysis when an amount of qubits in the PQC is less than a threshold amount of qubits.

10 . The method of claim 9 , wherein the autoencoder is the fully connected autoencoder or the locality preserving variational autoencoder when the amount of qubits in the PQC is not less than the threshold amount of qubits.

11 . The method of claim 8 , wherein the autoencoder is the locality preserving variational autoencoder when a latent space dimension is less than a threshold size.

12 . The method of claim 11 , wherein the autoencoder is the fully connected autoencoder when the latent space dimension is not less than the threshold size.

13 . The method of claim 8 , wherein the autoencoder is the locality preserving variational autoencoder, wherein each pixel in a latent space includes compressed information only about a local region of the image, and wherein an overall dimension of the latent space is determined by a size of a local fully connected filter.

14 . The method of claim 1 , wherein training the QCNN comprises optimizing variational parameters of the QCNN using a cost function comprising one of: mean-squared error or binary cross entropy.

15 . The method of claim 1 , further comprising prior to compressing the image:
converting the image to a grayscale image; and equalizing pixel intensities in the converted image.

16 . A quantum information processing (QIP) system for classifying an object in an image using a quantum convolutional neural network (QCNN), comprising:
at least one memory; and at least one hardware processor coupled with the at least one memory and configured, individually or in combination, to:
receive the image depicting the object;
compress, using an autoencoder, the image by extracting a plurality of features;
encode the plurality of features into quantum states using a quantum feature map that corresponds to a quantum encoding circuit unique to the image, wherein parameters of the quantum encoding circuit depend on pixel values in the image;
input the encoded plurality of features into the QCNN that is executed on a parameterized quantum circuit (PQC) and trained to detect the object and generate a classification of the object from a plurality of classes, wherein the QCNN comprises a plurality of layers including one or more convolutional layers that apply a unitary operator to neighboring qubits in the one or more convolutional layers; and
output the classification of the object generated by the QCNN.

Assignees

• Ionq Inc
• Hyundai Motor Co Ltd
• Kia Corp

Inventors

• Iaconis Jason John
• Kim Sang Hyub
• Park Soon Cheol
• Jo Hanlae
• Kim Sangtae

Classifications

• G06V10/7715

  Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods · CPC title

• B82Y10/00

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

• G06V10/82 Primary

  using neural networks · CPC title

• G06V10/454

  Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN] · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title