WO2023173878A1

WO2023173878A1 - Quantum neural network training method and apparatus

Info

Publication number: WO2023173878A1
Application number: PCT/CN2022/141665
Authority: WO
Inventors: 姜金哲; 张新; 李辰; 李红珍
Original assignee: 苏州浪潮智能科技有限公司
Priority date: 2022-03-17
Filing date: 2022-12-23
Publication date: 2023-09-21
Also published as: CN114358258B; CN114358258A

Abstract

The present application relates to the technical field of quantum computing. Provided are a quantum neural network training method and apparatus. The quantum neural network training method comprises: constructing an expected-value solving circuit according to a quantum neural network algorithm circuit; sequentially updating the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit; after the current parameters corresponding to all the quantum gates in the quantum neural network algorithm circuit have been updated, determining whether the current parameters corresponding to all the quantum gates in the quantum neural network algorithm circuit all converge; and if all the current parameters converge, completing the quantum neural network training method. By means of the implementation of the method, an operation is directly performed on a quantum state in a Hilbert space, and then a quantum neural network is trained. The effect of executing quantum neural network training in a quantum computer is achieved, and the training efficiency of a quantum neural network algorithm is greatly improved.

Description

A kind of quantum neural network training method and device

Cross-references to related applications

This application claims priority to the Chinese patent application submitted to the China Patent Office on March 17, 2022, with the application number 202210263335.7 and the application title "A quantum neural network training method and device", the entire content of which is incorporated herein by reference. Applying.

Technical field

This application relates to the field of quantum computing technology, and in particular to a quantum neural network training method and device.

Background technique

Neural network is an algorithmic mathematical model of distributed parallel information processing. It was first an algorithm or model in the field of artificial intelligence. At present, neural network has developed into a multi-disciplinary subject field, and with the progress of deep learning Be valued and respected again.

With the development of quantum computing science, neural networks are combined with it to develop new quantum neural network algorithms. A series of special operations are performed on qubits through a combination of quantum gates. In quantum neural networks, qubits correspond to the concept of bits in traditional computers. There are two existing states |0> and |1>, which correspond to the switch state of traditional bits. Quantum gate is an important tool for implementing quantum machine learning algorithms and is a control operation on quantum states. As an operation that can carry parameters, it plays an important role in quantum neural networks. It mainly completes the control operations and entanglement operations between qubits. By combining various gates in an orderly manner, we can get what we want in the final output. information to implement specific algorithms.

Quantum neural network is based on quantum computing, and the machine learning algorithm implemented on a quantum computer has good generalization performance compared with traditional computing. At present, there is no method to train quantum neural network algorithms on quantum computers. Training quantum neural networks is still done through the machine learning framework of traditional computers, which cannot give full play to the advantages of quantum computing. Under the current hardware development conditions, this is a compromise and compromise method. However, with the further development of quantum computers, there is an urgent need for a training process that can be performed directly on quantum computers.

Contents of the invention

In order to solve the problems of the existing technology, embodiments of the present application provide a quantum neural network training method and device to overcome the problem in the existing technology that the quantum neural network cannot be directly trained through a quantum computer.

In order to solve one or more of the above technical problems, the technical solutions adopted in this application are as follows:

Embodiments of the present application provide a quantum neural network training method for neural network training on a quantum computer, including:

Construct an expected value solving circuit based on the quantum neural network algorithm circuit;

Update the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit in sequence;

After completing the update of the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit, determine whether the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit have all converged;

If all current parameters converge, the quantum neural network training method is completed;

Among them, the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit are sequentially updated, including:

Get the current parameters corresponding to the quantum gate being updated as parameters to be processed;

Use the expected value solving circuit to act on the operation results of the parameters to be processed and the preset values to obtain the first expected value and the second expected value;

Calculate the parameter gradient of the parameter to be processed according to the first expected value and the second expected value;

Operate the parameters to be processed according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated.

Furthermore, a quantum neural network training method also includes: if the current parameters corresponding to at least one quantum gate do not converge, executing the quantum neural network training method in a loop.

Furthermore, the method also includes:

Construct a quantum neural network algorithm circuit. The quantum neural network algorithm circuit at least includes a quantum gate with parameters, and the parameters of the quantum gate are the current parameters.

Furthermore, the expected value solution circuit is isomorphic to the quantum neural network algorithm circuit;

The expected value calculated by the expected value solving circuit is obtained by multiplying the expected value calculated by the quantum neural network algorithm circuit by the preset scaling factor.

Further, using the expected value solving circuit to act on the operation result of the parameter to be processed and the preset value, obtaining the first expected value and the second expected value includes:

Use the expected value solving circuit to act on the sum of the parameter to be processed and the preset value to obtain the first expected value;

Use the expected value solving circuit to act on the difference between the parameter to be processed and the preset value to obtain the second expected value.

Further, calculating the parameter gradient of the parameter to be processed based on the first expected value and the second expected value includes:

In the Hilbert space, perform a rotation operation on the preset quantum state in the first preset direction with an angle of the first expected value to obtain the first intermediate quantum state;

In the Hilbert space, perform a rotation operation on the first intermediate quantum state in the second preset direction with an angle of the second expected value to obtain the second intermediate quantum state;

Input the second intermediate quantum state into the parameter gradient calculation circuit to calculate the parameter gradient;

The above parameter gradient design circuit is designed based on the quantum phase estimation algorithm.

Further, the parameters to be processed are operated according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated, including:

In the Hilbert space, perform a rotation operation on the preset quantum state in the third preset direction with the angle of the current parameter to obtain the third intermediate quantum state;

In the Hilbert space, perform a rotation operation on the third intermediate quantum state in the fourth preset direction with an angle of parameter gradient to obtain the fourth intermediate quantum state;

Input the fourth intermediate quantum state into the gradient update calculation circuit to calculate the current parameters;

The above gradient update calculation circuit is designed based on the quantum phase estimation algorithm.

An embodiment of the present application provides a quantum neural network training device, which includes:

Expected value circuit construction module, current parameter update module, parameter convergence judgment module;

The expected value circuit building module is used to construct the expected value solving circuit based on the quantum neural network algorithm circuit;

The current parameter update module is used to sequentially update the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit:

Operate the parameters to be processed according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated;

The parameter convergence judgment module is used to determine whether the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit have all converged after updating the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit;

If all current parameters converge, the quantum neural network training method is completed.

Further, the device also includes: algorithm building module;

An algorithm building module is used to construct a quantum neural network algorithm circuit. The quantum neural network algorithm circuit includes at least one quantum gate with parameters, and the parameters of the quantum gate are current parameters.

Further, the current parameter update module includes:

Parameter acquisition sub-module, expected value solution sub-module, gradient calculation sub-module, parameter update sub-module;

The parameter acquisition submodule is used to obtain the current parameters corresponding to the quantum gate being updated as parameters to be processed;

The expected value solving submodule is used to use the expected value solving circuit to act on the operation results of the parameters to be processed and the preset values to obtain the first expected value and the second expected value;

The gradient calculation submodule is used to calculate the parameter gradient of the parameter to be processed based on the first expected value and the second expected value;

The parameter update submodule is used to operate the parameters to be processed according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated.

The beneficial effects brought by the technical solutions provided by the embodiments of this application are:

By constructing a quantum neural network circuit and an expectation value solving circuit, direct manipulation of quantum states in Hilbert space is achieved, and then the quantum neural network is trained. It plays the role of performing quantum neural network training on quantum computers. Greatly improves the training efficiency of quantum neural network algorithms.

Description of the drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 is a schematic diagram of a quantum neural network training method provided by an embodiment of the present application;

Figure 2 is a schematic diagram of another quantum neural network training method provided by an embodiment of the present application;

Figure 3 is a schematic diagram of a quantum gate circuit module provided by an embodiment of the present application;

Figure 4 is a schematic diagram of a quantum circuit module for calculating the expected value _da provided by an embodiment of the present application;

Figure 5 is a schematic diagram of a quantum circuit module for calculating the expected value d _b provided by the embodiment of the present application;

Figure 6 is a schematic flowchart of a gradient solution provided by an embodiment of the present application;

Figure 7 is a schematic flowchart of updating quantum gate parameters provided by an embodiment of the present application;

Figure 8 is a schematic diagram of a quantum neural network training device provided by an embodiment of the present application:

Figure 9 is a schematic diagram of another quantum neural network training device provided by an embodiment of the present application:

Figure 10 is a schematic sub-module diagram of a current update module provided by an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only Some of the embodiments of this application are provided, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

Unless otherwise defined, technical terms or scientific terms used in this disclosure shall have the usual meaning understood by a person with ordinary skill in the art to which this disclosure belongs. "First", "second" and similar words used in this disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. Likewise, similar words such as "a", "an" or "the" do not indicate a quantitative limitation but rather indicate the presence of at least one. The numbers in the drawings of the description only indicate the distinction between various functional components or modules, and do not indicate the logical relationship between the components or modules. Words such as "include" or "comprising" mean that the elements or things appearing before the word include the elements or things listed after the word and their equivalents, without excluding other elements or things. Words such as "connected" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right", etc. are only used to express relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

Hereinafter, various embodiments according to the present disclosure will be described in detail with reference to the accompanying drawings. It should be noted that, in the drawings, the same reference numerals are given to components that substantially have the same or similar structures and functions, and repeated descriptions about them will be omitted.

In the existing technology, there is a lack of methods to directly train quantum neural network algorithms on quantum computers. Instead, a machine learning framework is used to simulate the training of quantum neural networks on traditional computers within a limited number of bits. It is impossible to fully utilize the powerful parallel computing capabilities of quantum computers with high qubit numbers. Therefore, there is an urgent need for a quantum neural network training method and device that can directly train quantum neural networks on quantum computers. The embodiment of this application discloses a quantum neural network training method based on the characteristics of quantum computing. The specific technical solution is as follows:

In one embodiment, as shown in Figure 1, a quantum neural network training method includes:

Step S1: Construct an expected value solving circuit based on the quantum neural network algorithm circuit;

Step S2: Update the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit in sequence;

Step S3: After completing the update of the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit, determine whether the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit have all converged;

In another embodiment, step S3 of a quantum neural network training method also includes:

If the current parameters corresponding to at least one quantum gate do not converge, return to step S1 to execute the quantum neural network training method in a loop.

After the judgment of step S3, if the current parameters corresponding to a quantum gate do not converge, then return to step S1 and execute the above-mentioned quantum neural network training method in a loop. The execution method of quantum neural network training when the parameters do not converge is further elaborated.

Below, the detailed implementation steps of a quantum neural network training method will be explained.

In another embodiment, as shown in Figure 2, a quantum neural network training method includes:

Step S0: Construct a quantum neural network algorithm circuit. The quantum neural network algorithm circuit includes at least one quantum gate with parameters, and the parameters of the quantum gate are the current parameters;

The circuit of the quantum neural network algorithm is composed of quantum gates with parameters. According to the Parameter Shift (parameter shift) theory, the expected value of a quantum circuit output is expressed by the following formula:

Among them, U _G (θ) = ^-iaθG , α is a real parameter, A is a measurement operator,

is a vector state,

is a vector state

The complex conjugate of .

The gradient of parameter θ can be expressed as the following formula, where quantum gate G has only two eigenvalues e ₀ and e ₁ , so:

in,

According to the above formula, if you want to find the gradient of the parameter, you need to know

and

value. Therefore, two quantum circuits are needed

and

Solve for d _a and d _b . Because of quantum circuits

and

Algorithm circuit with quantum neural network

isomorphic, and only differs by one scaling factor r, according to

Easy to build quantum circuits

and

As shown in Figure 3-Figure 5. pass

and

The corresponding expected values d _a and d _b can be found.

According to the relationship between the expected value solving circuit and the quantum neural network algorithm circuit, which is isomorphic and has a phase difference scaling coefficient r, the expected value solving circuit is constructed based on the quantum neural network algorithm circuit.

Step S2: Update the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit in sequence.

Specifically include:

Step S21: Obtain the current parameters corresponding to the quantum gate being updated as parameters to be processed;

Step S22: Use the expected value solving circuit to act on the operation result of the parameter to be processed and the preset value to obtain the first expected value and the second expected value;

Specifically include:

Step S221: Use the expected value solving circuit to act on the sum of the parameter to be processed and the preset value to obtain the first expected value;

Step S222: Use the expected value solving circuit to act on the difference between the parameter to be processed and the preset value to obtain the second expected value;

Among them, the default value is

use

Acts on the parameter θ to be processed and the preset value

The sum of gets the first expected value d _a , that is

use

Acts on the parameter θ to be processed and the preset value

The difference gets the second expected value d _b , that is

Step S23: Calculate the parameter gradient of the parameter to be processed based on the first expected value and the second expected value;

Specifically include:

Step S231: In the Hilbert space, perform a rotation operation on the preset quantum state in the first preset direction with an angle of the first expected value to obtain the first intermediate quantum state;

Step S232: In the Hilbert space, perform a rotation operation on the first intermediate quantum state in the second preset direction with an angle of the second expected value to obtain the second intermediate quantum state;

Step S233: Input the second intermediate quantum state into the parameter gradient calculation circuit, and calculate the parameter gradient.

Among them, the above parameter gradient design circuit is designed based on the quantum phase estimation algorithm (Quantum Phase Estimation, QPE).

In one embodiment, as shown in Figure 6. The mathematical expression of the parameter gradient according to the parameter θ to be processed:

For the quantum state _| 0>, perform a rotation operation of angle da in the first preset direction in the Hilbert space, and then perform a rotation operation of angle d _b in the second preset direction, and then through the quantum phase The parameter gradient designed by the estimation algorithm (QPE) is used to design the circuit to obtain the parameter gradient of the parameter θ

Wherein, the first preset direction and the second preset direction have opposite rotation directions in Hilbert space.

Step S24: Operate the parameters to be processed according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated.

Specifically include:

Step S241: In the Hilbert space, perform a rotation operation on the preset quantum state in the third preset direction with the angle of the current parameter to obtain the third intermediate quantum state;

Step S242: In the Hilbert space, perform a rotation operation on the third intermediate quantum state in the fourth preset direction with an angle of parameter gradient to obtain the fourth intermediate quantum state;

Step S243: Input the fourth intermediate quantum state into the gradient update calculation circuit, and calculate the current parameters;

Among them, the above parameter gradient update calculation circuit is designed based on the quantum phase estimation algorithm (Quantum Phase Estimation, QPE).

In one embodiment, as shown in Figure 7, the parameter gradient calculated according to step S233

Update parameters.

For the quantum state |0> in the Hilbert space, perform a rotation operation with the angle being the parameter θ to be processed in the third preset direction, and then perform the rotation operation with the angle being the gradient in the fourth preset direction.

The rotation operation is performed, and then the updated parameter θ′ is obtained through the parameter gradient design circuit designed by the quantum phase estimation algorithm (QPE). Wherein, the third preset direction and the fourth preset direction have opposite rotation directions in the Hilbert space.

Step S3: After completing the update of the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit, determine whether the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit have all converged.

Perform the following operations based on whether all current parameters converge:

In another embodiment, as shown in Figure 8, a quantum neural network training device includes: an expected value circuit construction module 1, a current parameter update module 2, and a parameter convergence judgment module 3;

Expected value circuit building module 1, used to construct an expected value solving circuit based on the quantum neural network algorithm circuit;

The current parameter update module 2 is used to sequentially update the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit;

Parameter convergence judgment module 3 is used to judge whether the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit have all converged after updating the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit.

In another embodiment, as shown in Figure 9, a quantum neural network training device further includes: algorithm building module 0.

Algorithm building module 0 is used to construct a quantum neural network algorithm circuit. The quantum neural network algorithm circuit includes at least one quantum gate with parameters, and the parameters of the quantum gate are used as current parameters.

In one embodiment, as shown in Figure 10, the current parameter update module 2 includes: a parameter acquisition sub-module 21, an expected value solution sub-module 22, a gradient calculation sub-module 23, and a parameter update sub-module 24.

Parameter acquisition sub-module 21 is used to acquire the current parameters corresponding to the quantum gate being updated as parameters to be processed;

The expected value solving sub-module 22 is used to use the expected value solving circuit to act on the operation results of the parameters to be processed and the preset values to obtain the first expected value and the second expected value;

The gradient calculation sub-module 23 is used to calculate the parameter gradient of the parameter to be processed according to the first expected value and the second expected value;

The parameter update submodule 24 is used to operate the parameters to be processed according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated.

In some embodiments:

according to

Easy to build quantum neural network algorithm circuit

and

As shown in Figure 3-5.

Specifically include:

According to quantum circuits

and

And find the corresponding first expected value _da and second expected value d _b .

Among them, the default value is

use

Acts on the parameter θ to be processed and the preset value

The sum of gets the first expected value d _a , that is

use

Acts on the parameter θ to be processed and the preset value

The difference gets the second expected value d _b , that is

Step S23: Calculate the parameter gradient of the parameter to be processed according to the first expected value d _a and the second expected value d _b ;

Specifically include:

Among them, the above parameter gradient design circuit is designed based on the quantum phase estimation algorithm (Quantum Phase Estimation, QPE). The process is shown in Figure 6.

Specifically include:

Among them, the above parameter gradient update calculation circuit is designed based on the quantum phase estimation algorithm (Quantum Phase Estimation, QPE). The process is shown in Figure 7.

In some embodiments:

The expected value of the output of a quantum circuit is expressed by:

Among them, U _G (θ)=e ^-iaθG , where quantum gate G has only two eigenvalues e ₀ and e ₁ , and the gradient of θ is expressed as:

in,

Steps S1 to S2 have been described in detail in the above embodiments and will not be described again here.

In some embodiments:

The following describes a quantum neural network training device with reference to Figures 8-10.

The neural network training device includes: an expected value circuit construction module 1, a current parameter update module 2, and a parameter convergence judgment module 3;

The front parameter update module 2 is shown in Figure 10 and includes: parameter acquisition sub-module 21, expected value solution sub-module 22, gradient calculation sub-module 23, and parameter update sub-module 24.

In some embodiments:

As shown in Figure 9, a quantum neural network training device also includes: algorithm building module 0.

Among them, the expected value circuit construction module 1, the current parameter update module 2, and the parameter convergence judgment module 3 have been described in detail in the third embodiment and will not be described again here.

In particular, according to embodiments of the present application, the process described above with reference to the flowchart may be implemented as a computer software program. For example, embodiments of the present application include a computer program product including a computer program loaded on a computer-readable medium, the computer program containing program code for performing the method illustrated in the flowchart. In such embodiments, the computer program may be downloaded and installed from the network through the communication device, or from memory, or from ROM. When the computer program is executed by an external processor, the above-mentioned functions defined in the method of the embodiment of the present application are performed.

It should be noted that the computer-readable medium in the embodiments of the present application may be a computer-readable signal medium or a computer-readable non-volatile readable storage medium, or any combination of the above two. The computer-readable non-volatile readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer readable non-volatile readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard drives, random access memory (RAM), read only memory (ROM) ), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In embodiments of the present application, a computer-readable non-volatile readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with an instruction execution system, apparatus, or device. In embodiments of the present application, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, in which computer-readable program codes are carried. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable non-volatile readable storage medium that can be sent, propagated, or transmitted for use by an instruction execution system, apparatus, or device or programs used in conjunction with it. Program code contained on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency, radio frequency), etc., or any suitable combination of the above.

The above-mentioned computer-readable medium may be included in the above-mentioned server; it may also exist separately without being assembled into the server. The computer-readable medium carries one or more programs. When the one or more programs are executed by the server, the server: in response to detecting that the peripheral mode of the terminal is not activated, obtains the frame rate of the application on the terminal. ; When the frame rate meets the screen off condition, determine whether the user is obtaining screen information of the terminal; in response to the determination result that the user is not obtaining screen information of the terminal, control the screen to enter the immediate dimming mode.

Computer program code for performing operations of embodiments of the present application may be written in one or more programming languages, including object-oriented programming languages—such as Java, Smalltalk, or a combination thereof. ), C++ (language), and also includes conventional procedural programming languages—such as the "C (language)" language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In situations involving remote computers, the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (such as an Internet service provider through Internet connection).

Each embodiment in this specification is described in a progressive manner. The same and similar parts between the various embodiments can be referred to each other. Each embodiment focuses on its differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For relevant details, please refer to the partial description of the method embodiment. The systems and system embodiments described above are only illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in One location, or it can be distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Persons of ordinary skill in the art can understand and implement the method without any creative effort.

The technical solutions provided by this application have been introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; At the same time, for those of ordinary skill in the art, there will be changes in the specific implementation and application scope based on the ideas of this application. In summary, the contents of this specification should not be construed as limiting this application.

The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application. Inside.

Claims

A quantum neural network training method for neural network training on a quantum computer, characterized in that the method includes:

Construct an expected value solving circuit based on the quantum neural network algorithm circuit;

Sequentially update the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit;

After completing the update of the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit, determine whether the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit have all converged;

If all current parameters converge, the quantum neural network training method is completed;

Among them, updating the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit in sequence includes:

Get the current parameters corresponding to the quantum gate being updated as parameters to be processed;

Use the expected value solving circuit to act on the operation result of the parameter to be processed and the preset value to obtain the first expected value and the second expected value;

Calculate the parameter gradient of the parameter to be processed according to the first expected value and the second expected value;

The parameters to be processed are operated according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated.
A quantum neural network training method according to claim 1, characterized in that the method further includes: if the current parameters corresponding to at least one quantum gate do not converge, executing the quantum neural network training method in a loop.
A quantum neural network training method according to claim 2, characterized in that the method also includes:

Construct a quantum neural network algorithm circuit, which includes at least one quantum gate with parameters, and the parameters of the quantum gate are current parameters.
A quantum neural network training method according to any one of claims 1-3, characterized in that the expected value solving circuit is isomorphic with the quantum neural network algorithm circuit;

The expected value calculated by the expected value solving circuit is obtained by multiplying the expected value calculated by the quantum neural network algorithm circuit by a preset scaling factor.
A quantum neural network training method according to claim 4, characterized in that the quantum gate includes a first eigenvalue and a second eigenvalue.
A quantum neural network training method according to claim 5, characterized in that the preset scaling coefficient is determined by the first eigenvalue, the second eigenvalue and preset real parameters.
A quantum neural network training method according to claim 6, characterized in that the preset scaling coefficient is determined in the following manner:

Calculate the difference between the first characteristic value and the second characteristic value;

Calculate the half value of the preset real parameter;

The product value of the difference value and the half value is calculated, and the product value is determined as the preset scaling factor.
A quantum neural network training method according to any one of claims 1 to 3, characterized in that the use of the expected value solving circuit acts on the operation results of the parameters to be processed and the preset values to obtain The first expected value and the second expected value include:

Use the expected value solving circuit to act on the sum of the parameter to be processed and the preset value to obtain a first expected value;

The second expected value is obtained by using the expected value solving circuit to act on the difference between the parameter to be processed and the preset value.
A quantum neural network training method according to claim 8, characterized in that the expected value solving circuit includes a first expected value solving circuit and a second expected value solving circuit.
A quantum neural network training method according to claim 9, characterized in that, using the expected value solving circuit to act on the sum of the parameter to be processed and the preset value, obtaining the first expected value includes:

The first expected value solving circuit is used to act on the sum of the parameter to be processed and the preset value to obtain a first expected value.
A quantum neural network training method according to claim 9, characterized in that said using the expected value solving circuit to act on the difference between the parameter to be processed and the preset value to obtain the second expected value includes:

The second expected value solving circuit is used to act on the difference between the parameter to be processed and the preset value to obtain a second expected value.
A quantum neural network training method according to claim 8, characterized in that the preset value is determined by a preset scaling coefficient.
A quantum neural network training method according to claim 12, characterized in that the preset value is the value of pi divided by four times of the preset scaling coefficient.
A quantum neural network training method according to any one of claims 1 to 3, characterized in that, calculating the parameter gradient of the parameter to be processed according to the first expected value and the second expected value includes: :

In the Hilbert space, perform a rotation operation on the preset quantum state in the first preset direction with an angle of the first expected value to obtain the first intermediate quantum state;

In the Hilbert space, perform a rotation operation on the first intermediate quantum state in a second preset direction with an angle of a second expected value to obtain a second intermediate quantum state;

Input the second intermediate quantum state into the parameter gradient calculation circuit to calculate the parameter gradient;

The parameter gradient design circuit is designed based on the quantum phase estimation algorithm.
A quantum neural network training method according to claim 14, characterized in that the first preset direction and the second preset direction are opposite to the rotation directions in the Hilbert space.
A quantum neural network training method according to claim 14, characterized in that the preset quantum state is a vacuum state of a free field.
A quantum neural network training method according to any one of claims 1 to 3, characterized in that, the parameters to be processed are operated according to the parameter gradient to obtain the current value corresponding to the quantum gate being updated. Parameters include:

In the Hilbert space, perform a rotation operation on the preset quantum state in the third preset direction with the angle of the current parameter to obtain the third intermediate quantum state;

In the Hilbert space, perform a rotation operation on the third intermediate quantum state in a fourth preset direction with an angle equal to the parameter gradient to obtain a fourth intermediate quantum state;

Input the fourth intermediate quantum state into the gradient update calculation circuit to calculate the current parameters;

The gradient update calculation circuit is designed based on the quantum phase estimation algorithm.
A quantum neural network training method according to claim 17, characterized in that the third preset direction and the fourth preset direction are opposite to the rotation directions in the Hilbert space.
A quantum neural network training method according to claim 17, characterized in that the preset quantum state is a vacuum state of a free field.
A quantum neural network training device, characterized in that the device includes:

Expected value circuit construction module, current parameter update module, parameter convergence judgment module;

The expected value circuit building module is used to construct an expected value solving circuit based on the quantum neural network algorithm circuit;

The current parameter update module is used to sequentially update the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit:

Get the current parameters corresponding to the quantum gate being updated as parameters to be processed;

Use the expected value solving circuit to act on the operation result of the parameter to be processed and the preset value to obtain the first expected value and the second expected value;

Calculate the parameter gradient of the parameter to be processed according to the first expected value and the second expected value;

Operate the parameter to be processed according to the parameter gradient to obtain the current parameters corresponding to the quantum gate being updated;

The parameter convergence judgment module is used to judge whether the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit have all converged after updating the current parameters corresponding to all quantum gates in the quantum neural network algorithm circuit;

If all current parameters converge, the quantum neural network training method is completed.