CN113159303B - Quantum circuit-based artificial neuron construction method - Google Patents
Quantum circuit-based artificial neuron construction method Download PDFInfo
- Publication number
- CN113159303B CN113159303B CN202110231652.6A CN202110231652A CN113159303B CN 113159303 B CN113159303 B CN 113159303B CN 202110231652 A CN202110231652 A CN 202110231652A CN 113159303 B CN113159303 B CN 113159303B
- Authority
- CN
- China
- Prior art keywords
- quantum
- weight
- neuron
- state
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N10/00—Quantum computing, i.e. information processing based on quantum-mechanical phenomena
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- Neurology (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses an artificial neuron construction method based on a quantum circuit, which belongs to the field of quantum machine learning, and comprises the steps of firstly encoding the input and the weight of a neuron onto a quantum computing ground state, then acting a controlled unitary gate containing the weight value of the neuron on the input quantum state, and finally obtaining the output of the neuron through quantum phase estimation. The quantum neuron model for realizing the scheme mainly comprises three parts of quantum circuits: the first part is an input and weight interaction quantum circuit which can well realize the function that the neuron receives input values from different connection strengths; the second part is a phase estimation quantum circuit, and the circuit realizes the function of a neuron activation function; the third part is a weight update quantum circuit that enables the conversion of an update to a weight value into an update to each qubit state representing a weight quantum state. The invention has the advantage of quantum information processing.
Description
Technical Field
The invention belongs to the field of quantum machine learning, in particular to the field of quantum neural networks, and particularly relates to an artificial neuron construction method based on a quantum circuit.
Background
During the last twenty years, machine learning has grown up rapidly and has become a technological base in the big data age, and in the field of quantum information science, the combination of machine learning and quantum computing technology has also received attention from more and more researchers. One main idea of combining machine learning and quantum computing is to utilize superposition of quantum states and acceleration of quantum algorithm to solve the problems of huge data volume and slow training process in the current data science, and the research in the field of quantum machine learning originates from the research on a quantum neural network at the earliest.
The quantum neural network (Quantum Neural Network, QNN) is a novel model combining the characteristics of an artificial neural network (Artificial Neural Network, ANN) and quantum computing (Quantum Computation), and theoretically the quantum neural network is more intelligent than a classical neural network, and has more effective learning and generalization capabilities. Many studies have attempted to quantize nonlinear activation functions such as sigmoid of classical neurons and build more general quantum neurons and their network models. These quantum neural networks offer advantages over classical neural networks in certain applications. However, the problem is that these schemes all introduce nonlinear operators, and although nonlinear quantum mechanics has been studied for a long time, whether nonlinear algorithms can realize quantum implementation is highly controversial. From this, it can be seen that the general quantum neural network research is still in the exploration stage, and how to more reasonably construct a quantum neural network integrating nerve computation and quantum computation is also a non-trivial open subject.
Various quantum neurons and their network models have been proposed by scholars around the world, and many models have been widely used. For example, in the medical field, a quantum-excited neural network has been used for breast cancer prediction; in the financial field, a quantum excitation mixture based on quantum multilayer perceptrons helps to overcome the random work dilemma of financial time series prediction; in the engineering field, a quantum neural network based on a rough set is suitable for transformer fault diagnosis and is very useful in processing uncertain data; in the function approximation problem, compared with a classical back propagation neural network, the quantum back propagation neural network has the advantages of faster learning speed and stronger fitting capability. Quantum neural networks are used in different fields of real life and exhibit certain advantages over conventional neural networks.
Disclosure of Invention
The present invention is directed to solving the above problems of the prior art. An artificial neuron construction method based on quantum circuits is provided. The technical scheme of the invention is as follows:
the artificial neuron construction method based on the quantum circuit is characterized by comprising the following steps of:
the classical input and weight information of the quantum neuron model are respectively encoded on a quantum computing ground state, the quantum neuron model processes the quantum information, and the classical information is encoded on the quantum computing ground state and can be processed as the quantum information;
acting a controlled unitary gate containing neuron weight values on classical input information encoded to a quantum computing ground state;
and obtaining neuron output through quantum phase estimation, wherein the quantum neurons are constructed to be used for constructing a quantum neural network. The quantum neural network is a novel model combining the artificial neural network with quantum computing characteristics, and theoretically, the quantum neural network is more intelligent compared with the traditional neural network, and has more effective learning and generalization capability. Many proposed models have been applied to function approximation, big data processing, joint storage and automation control systems, and the like. The neuron model provided by the invention can be used as a more effective machine learning model after constructing a quantum neural network, and can be used for pattern recognition, information processing, signal detection and analysis and the like.
Further, the quantum neuron model is composed of three parts of quantum circuits, and specifically comprises:
1) Input and weight interaction quantum circuit: the first stage process of quantum neuron calculation is realized by the partial quantum circuit, and is mainly through the interaction of a quantum state of element input and a controlled weight unitary gate, the stage is used for realizing that the neuron receives input values from different connection strengths, and writing a result into the phase of an eigenvalue of a system state formed by the input, the weight quantum state and auxiliary quantum bits;
2) Output estimation quantum circuit: the part of quantum circuits realize the second stage process of quantum neuron calculation, quantum inverse Fourier transform is applied to a count quantum bit register, the quantum inverse Fourier transform realizes the function of classical neuron activation, and then the state of the count quantum bit register is read out through measurement in a calculation ground state, so that the corresponding output of the quantum neuron is obtained;
3) Weight updating quantum circuit: the part of quantum circuits realize the updating of the weight, are used for adjusting the weight after comparing the actual output value with the expected output value, and convert the updating of the weight value into the updating of each quantum bit state representing the quantum state of the weight.
Further, in the 1) input-weight interaction quantum circuit, if the neuron input vector after the neuron threshold value θ has been introduced is X' = (X) 1 ；x 2 ；...；x n ；-1)，x i E { -1,1}, and the corresponding weight vector is W' = (W) 1 ；w 2 ；...；w n ；θ)，w i θ∈ (-1, 1), where i=1, …, n, n is the number of elements in the input vector that does not introduce the neuron threshold, and encoding the input vector X 'and the weight vector W' separately onto the quantum computing ground state has:
the values of the weight vectors are then written as parameters to the controlled unitary gateIn the middle, there arek represents the state value of the count qubit register, and after the count qubit register passes through the Hadamard gate, the unitary gate U controlled by the count qubit register and the weight quantum state k (W ') acts on the input quantum state |X ' '>Before inverse quantum Fourier transform, the state of the whole system is as follows:
h represents a single-quantum bit gate Hadamard gate in quantum computing, andrepresentation ofThe Hadamard gates respectively act on t quantum bits of the count quantum bit register, the stage mainly realizes the function that the neuron receives input values from different connection strengths, and the result is written into the phase of the eigenvalue of the system state formed by the input, the weight quantum state and the auxiliary quantum bit.
Further, controlled unitary doorsK-w in index i Representing a state value of k and a weight quantum state value of w of the count qubit register i When then operator U k (w i ) Applied to input qubit |x i >On the input qubit |x i >Unitary door U (w i ) The definition is as follows:
wherein the method comprises the steps ofThen act on the entire input quantum state |X ''>Can be written as a controlled unitary gate U (W')
Further, the 2) applying an inverse Fourier transform to the count qubit register in the output estimation quantum circuit to cause Is a hypothetical variable, used to represent +.>Is a value of (2); delta is defined unitary gate U (w i ) One variable of (a) is +.>The following results were obtained:
the inverse quantum Fourier transform is used to realize the action of the activation function, and the measured quantum register state is |l>＝|l 1 l 2 …l t >，l t Representing the state value of the t-th qubit in the count qubit register, an estimated value is obtainedSo the neuron outputs are +.>y E [0, 1), and when the output of the perceptron model is y E { 1,1}, only the estimated value is concerned with whether the estimated value is larger than +.>I.e. whether the first qubit of the count qubit register is |1>Then let |l 1 >＝|0>Y= -1 when and l when 1 >＝|1>Y=1.
Further, whenPeak, if->Being an integer, the measurement is performed on a calculation basis to obtain:
if it isIt is known from the theory of quantum phase estimation algorithm that, not an integer, the phase value is +.>To the m bits with a success probability of at least 1-epsilon, only +.>In a simple sorting task, it is verified whether the output is greater than +.>Thus sorting, only 2 count qubits are needed.
Further, in the 3) weight updating quantum circuit, referring to the quantum circuit for implementing the SUM function SUM and the CARRY function CARRY in the existing quantum adder circuit, for the weight value quantum state |w i >The j-th qubit |w i_j >Given |w i_j >，|Δw i_j >And carry c j Calculation andnew carry c j-1 The weight update quantum circuit thus constructed realizes the following procedure:
|Δw i ,w i >→|Δw i ,w i +Δw i ＞
the process shows that the updating of the weight value can be converted into updating of each qubit state representing the weight quantum state and can be realized on a quantum circuit.
The invention has the advantages and beneficial effects as follows:
the invention provides a quantum neuron model based on quantum circuit realization, wherein the calculation step shown in claim 1 meets the basic function of classical neuron calculation. In claim 2, the first part of input and weight interact quantum circuit, realizing the function that the neuron receives input values from different connection strengths, and can exponentially save storage resources because the input and the weight are represented by quantum states, and can greatly improve the calculation efficiency by utilizing quantum parallelism. The second part outputs an estimated quantum wire, the inverse quantum fourier transform of which performs the function of classical neuron activation functions, and does not involve nonlinear operators, so it can be implemented entirely by quantum wires. And the third part of the weight updating quantum circuit is used for updating and converting the weight value to update each quantum bit state representing the weight quantum state, and updating the weight can be realized through the constructed quantum circuit. While the calculation of the lines of the sections as claimed in claims 3, 5 and 7 strictly adhere to the basic principles of quantum calculations.
Many of the quantum neuron and network models proposed at present introduce nonlinear operators for realizing the function of classical neuron activation functions, but only nonlinear operators in quantum computing theory have measurement operations, so that the proposed models cannot be completely realized by quantum circuits, and are difficult to run on quantum computers. The nonlinear operator in the model provided by the invention only has strategy operation, so the model can be completely realized by a quantum circuit, and the model algorithm provided by the invention can be operated on a quantum computer in the future.
Drawings
FIG. 1 is a block diagram of a quantum neuron in accordance with a preferred embodiment of the present invention;
FIG. 2 is a quantum circuit diagram of a quantum neuron according to the present invention;
FIG. 3 is a graph of quantum circuits for interaction of inputs and weights in the present invention;
FIG. 4 is a graph of an output estimation quantum circuit in the present invention;
FIG. 5 is a quantum circuit diagram for implementing SUM function SUM and CARRY function CARRY;
FIG. 6 is a graph of a weight update quantum circuit in accordance with the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and specifically described below with reference to the drawings in the embodiments of the present invention. The described embodiments are only a few embodiments of the present invention.
The technical scheme for solving the technical problems is as follows:
the invention combines quantum computing and artificial neurons, and provides a new neuron model which can be realized by a quantum circuit, and a quantum neuron structure diagram of the model is shown in figure 1. Inputs and weights of neurons are first encoded onto quantum computing ground states separately, and then controlled unitary gates containing the weights of neurons are applied to these input quantum states (neuron thresholds are introduced by auxiliary quantum bits). The result of the input and weight interactions is the phase of the eigenvalue of the system state consisting of input and weight quantum states and auxiliary quantum bits, and finally the neuron output is obtained by quantum phase estimation. From the quantum neuron structure diagram we constructed a quantum circuit as shown in fig. 2 for implementing the quantum neuron.
If the neuron input vector after the neuron threshold value θ has been introduced is X' = (X) 1 ；x 2 ；...；x n ；-1)，x i E { -1,1}, encoding the input vector X' onto the quantum computing ground state (median-1 from |0 in the quantum model>The expression) is as follows:
and the corresponding weight vector is W' = (W) 1 ；w 2 ；...；w n ；θ)，w i θ ε (-1, 1). Writing the value of vector W' into binary form W i ＝(w i_0 w i_1 w i_2 …w i_s ) b (w i_0 For sign bits), the encoding of the weight vector W' onto the quantum computing ground state has:
the values of the weight vectors are then written as parameters to the controlled unitary gateIn the middle, there areThe unitary gate acting on the input qubits can be defined here as +.>To achieve the unitary gate we define U first j The method comprises the following steps:
wherein the method comprises the steps ofAnd j=1, …, s. Here delta is the variable value determined by the input and weight vector, which is introduced in order to enable the result of the first stage to fall into a stable interval, since the second stage phase estimate has periodicity. Whereby the controlled unitary doors can be broken down into a series of controlled doors
In which the received qubit |w i_0 >The controlled gate X represents a Pauli-X gate. Unitary gating for input qubits
In the first stage, mainly discussing the interaction of the neuron input quantum state and the controlled weight unitary gate, the quantum circuit for achieving the interaction is shown in fig. 3. For convenience of description, we will represent the state of the counting qubit bundle as |k>The weight quantum state W is represented by the quantum state |w of each weight value thereof i >Representation (including neuron threshold θ) and input quantum state |X ''>The quantum state |x of each input value thereof i >And (3) representing. To be clearerTo the corresponding weighted unitary gates of the input quantum states, we use U k (w n ) Applied to |x n >For example, there are the following procedures:
the controlled unitary door is also detailed in fig. 3Combined unitary door->To each qubit |x n >Quantum wires of the above process. Acting on input qubit |x at a series of unitary gates n >After the last, a CNOT gate is introduced again for the purpose of enabling |x n >Restoring its original calculated ground state.
According to the result, the initial state of the system is as follows:
then, after applying the corresponding controlled weight unitary gate U to each input qubit, it is:
from the above results, it can be seen that the result of the interaction of the input and the weight has been written into the phase of the eigenvalue of the system state consisting of the input and the weight quantum states and the auxiliary quantum bits, and then the corresponding output can be obtained by using the quantum phase estimation algorithm.
In the second stage, an inverse quantum fourier transform is applied to the count qubit register, and its state is then read out by a measurement operation. Output estimation quantum circuit as shown in fig. 4, the quantum circuit achieves the function of classical neuron activation function and finally obtains neuron output. In the last stage, the input quantum line interacting with the weights uses phase kick back to write the phase (fourier basis) of unitary gate U (W') acting on the input quantum state into t qubits in the count register, and the state of the whole system is:
next, an inverse fourier transform is applied to the count qubit register, for simplicity purposes, to giveThe following results were obtained:
the above expression is given inReaching a peak. If->Being an integer, there is a high probability that the measurement is performed on the calculation basis:
if it isIt is known from the theory of quantum phase estimation algorithm that, not an integer, the phase value is +.>To the m bits with a success probability of at least 1-epsilon, only +.>In a simple sorting task, it is verified whether the output is greater than +.>Therefore, the sorting is performed, so that the task can be well completed by only 2 count qubits.
The final measured count quantum register state is |l>＝|l 1 l 2 …l t >Then an estimated value can be obtainedWhereby the output of the neuron is +.>y.epsilon.0, 1). The output of the known perceptron model is y e-1, only if the estimate is greater than +.>I.e. whether the first qubit of the count qubit register is |1>Then let |l 1 >＝|0>Y= -1 when and l when 1 >＝|1>Y=1.
To implement the weight update of the quantum neuron model, we refer to the existing quantum adder circuit, and respectively give the quantum circuit implementing the SUM function SUM as shown in fig. 5 (a), and the quantum circuit implementing the new CARRY function CARRY as shown in fig. 5 (b). For the weight value quantum state |w i >The j-th qubit |w i_j >Let the |w be i_j >，|Δw i_j >And carry c j We need to calculate andnew carry c j-1 . In the calculation process, we neglect sign bits temporarily, and only need to compare their respective sign bits before calculation and then calculate according to a binary algorithm. Realization rightsThe re-update quantum wire requires 3 registers: the 1 st one requires s qubits, the input and input states of which are |Deltaw i_1 ,Δw i_2 ,…,Δw i_s >The method comprises the steps of carrying out a first treatment on the surface of the The 2 nd one requires s+1 qubits with input state of |0,w i_1 ,w i_2 ,…,w i_s >The output state is | (w i +Δw i ) _+ ,(w i +Δw i ) _1 ,…,(w i +Δw i ) _s >The method comprises the steps of carrying out a first treatment on the surface of the The 3 rd is represented by s initial states of |0>Is used to record the carry and reset |0 at the end>. The resulting weight update quantum wire diagram is shown in fig. 6. The first s CARRY in the quantum wire, each will be |c j ,Δw i_j ,w i_j ,0>Become->The final carry gives the high order number (w i +Δw i ) _+ . Then use a CNOT door handle +.>Becomes (Deltaw) i_1 ,w i_1 ) And then operating the handle (c) by one SUM 1 ,Δw i_1 ,w i_1 ) Becomes (c) 1 ,Δw i_1 ,(w i +Δw i ) _1 ). And then performing the carrier and SUM operations s-1 times. As a result, for each qubit |w i_j >Are summed and each auxiliary qubit is restored to its initial state |0>。
The quantum circuit realizes the following processes:
|Δw i ,w i >→|Δw i ,w i +Δw i >
the process shows that the updating of the weight value can be converted into updating of each qubit state representing the weight quantum state and can be realized on a quantum circuit.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
The above examples should be understood as illustrative only and not limiting the scope of the invention. Various changes and modifications to the present invention may be made by one skilled in the art after reading the teachings herein, and such equivalent changes and modifications are intended to fall within the scope of the invention as defined in the appended claims.
Claims (4)
1. The artificial neuron construction method based on the quantum circuit is characterized by comprising the following steps of:
the classical input and weight information of the quantum neuron model are respectively encoded on a quantum computing ground state, the quantum neuron model processes the quantum information, and the classical information is encoded on the quantum computing ground state and can be processed as the quantum information;
acting a controlled unitary gate containing neuron weight values on classical input information encoded to a quantum computing ground state;
obtaining neuron output through quantum phase estimation, wherein the quantum neurons are constructed to be used for constructing a quantum neural network;
the quantum neuron model consists of three parts of quantum circuits, and specifically comprises the following components:
1) Input and weight interaction quantum circuit: the first stage process of quantum neuron calculation is realized by the partial quantum circuit, and is mainly through the interaction of a quantum state of element input and a controlled weight unitary gate, the stage is used for realizing that the neuron receives input values from different connection strengths, and writing a result into the phase of an eigenvalue of a system state formed by the input, the weight quantum state and auxiliary quantum bits;
2) Output estimation quantum circuit: the part of quantum circuits realize the second stage process of quantum neuron calculation, quantum inverse Fourier transform is applied to a count quantum bit register, the quantum inverse Fourier transform realizes the function of classical neuron activation, and then the state of the count quantum bit register is read out through measurement in a calculation ground state, so that the corresponding output of the quantum neuron is obtained;
3) Weight updating quantum circuit: the part of quantum circuits realize the update of the weight, are used for adjusting the weight after comparing the actual output value with the expected output value, and convert the update of the weight value into the update of each quantum bit state representing the quantum state of the weight;
the 1) the neuron input vector after the neuron threshold value θ has been introduced is X' = (X) in the input-weight interaction quantum circuit 1 ；x 2 ；...；x n ；-1)，x i E { -1,1}, and the corresponding weight vector is W' = (W) 1 ；w 2 ；...；w n ；θ)，w i θ∈ (-1, 1), where i=1,..n, n is the number of elements in the input vector that do not introduce neuron threshold, encoding the input vector X 'and the weight vector W' separately onto the quantum computing ground state has:
the values of the weight vectors are then written as parameters to the controlled unitary gateIn the middle, there arek represents the state value of the count qubit register, and after the count qubit register passes through the Hadamard gate, the unitary gate U controlled by the count qubit register and the weight quantum state k (W ') acts on the input quantum state |X ' '>Before inverse quantum Fourier transform, the state of the whole system is as follows：
H represents a single-quantum bit gate Hadamard gate in quantum computing, andthe Hadamard gate which respectively acts on t quantum bits of the count quantum bit register is shown, the stage mainly realizes the function that the neuron receives input values from different connection strengths, and the result is written into the phase of the eigenvalue of the system state formed by the input, the weight quantum state and the auxiliary quantum bit;
controlled unitary doorK-w in index i Representing a state value of k and a weight quantum state value of w of the count qubit register i When then operator U k (w i ) Applied to input qubit |x i >On the input qubit |x i >Unitary door U (w i ) The definition is as follows:
wherein the method comprises the steps ofThen act on the entire input quantum state |X ''>Controlled unitary gate U (W') write
2. The method of quantum wire-based artificial neuron construction according to claim 1, wherein the 2) output estimation quantum wireIn (2) applying an inverse Fourier transform to a count qubit register to cause Is a hypothetical variable, used to represent +.>Is a value of (2); there is->The following results were obtained:
the inverse quantum Fourier transform is used to realize the action of the activation function, and the measured quantum register state is |l>＝|l 1 l 2 …l t >，l t Representing the state value of the t-th qubit in the count qubit register, an estimated value is obtainedSo the neuron outputs are +.>y E [0, 1), and when the output of the perceptron model is y E { 1,1}, only the estimated value is concerned with whether the estimated value is larger than +.>I.e. whether the first qubit of the count qubit register is |1>Let |l 1 >＝|0>Y= -1 when and l when 1 >＝|1>Y=1.
3. The method for constructing artificial neurons based on quantum wires according to claim 2, wherein whenPeak, if->Being an integer, the measurement is performed on a calculation basis to obtain:
if it isNot integers, it is known from the theory of quantum phase estimation algorithm that +.>To the m bits with a success probability of at least 1-epsilon, only +.>In a simple sorting task, it is verified whether the output is greater than +.>Thus sorting, only 2 count qubits are needed.
4. The method as claimed in claim 1, wherein 3) the quantum circuit for weight update refers to a quantum circuit for realizing SUM function SUM and CARRY function CARRY in existing quantum adder circuit, and the quantum state |w for weight value i >The j-th qubit |w i_j >For givingFix |w i_j >，|Δw i_j >And carry c j Calculation andnew carry c j-1 The weight update quantum circuit thus constructed realizes the following procedure:
|Δw i ,w i >→|Δw i ,w i +Δw i >
the process can be used for converting the update of the weight value into the update of each qubit state representing the weight quantum state, and can be realized on a quantum circuit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110231652.6A CN113159303B (en) | 2021-03-02 | 2021-03-02 | Quantum circuit-based artificial neuron construction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110231652.6A CN113159303B (en) | 2021-03-02 | 2021-03-02 | Quantum circuit-based artificial neuron construction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113159303A CN113159303A (en) | 2021-07-23 |
| CN113159303B true CN113159303B (en) | 2023-07-21 |
Family
ID=76884038
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110231652.6A Active CN113159303B (en) | 2021-03-02 | 2021-03-02 | Quantum circuit-based artificial neuron construction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113159303B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114446414B (en) * | 2022-01-24 | 2023-05-23 | 电子科技大学 | Reverse synthetic analysis method based on quantum circulation neural network |
| CN114358319B (en) * | 2022-03-22 | 2022-06-21 | 合肥本源量子计算科技有限责任公司 | Machine learning framework-based classification method and related device |
| CN115293254A (en) * | 2022-07-29 | 2022-11-04 | 合肥本源量子计算科技有限责任公司 | Quantum multilayer perceptron-based classification method and related equipment |
| CN117713385B (en) * | 2024-02-05 | 2024-05-03 | 威海双城电气有限公司 | Electric power detection system and method based on combination of quantum computation and artificial intelligence |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019055847A1 (en) * | 2017-09-15 | 2019-03-21 | President And Fellows Of Harvard College | Quantum artificial neural networks |
| CN110674921A (en) * | 2019-07-11 | 2020-01-10 | 中国科学技术大学 | Method for constructing quantum feedforward neural network based on classical training |
| WO2020245013A1 (en) * | 2019-06-04 | 2020-12-10 | Universita' Degli Studi Di Pavia | Artificial neural network on quantum computing hardware |
| CN112101518A (en) * | 2020-08-05 | 2020-12-18 | 华南理工大学 | Quantum system capable of simulating any nonlinear activation function |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103018660B (en) * | 2012-12-25 | 2015-04-22 | 重庆邮电大学 | Multi-fault intelligent diagnosing method for artificial circuit utilizing quantum Hopfield neural network |
| CN105976408A (en) * | 2016-04-28 | 2016-09-28 | 北京大学 | Digital holographic compression transmission method of quantum backward propagation nerve network |
| JP6963277B2 (en) * | 2017-10-24 | 2021-11-05 | 日本電信電話株式会社 | Conversion device, judgment device, and calculation device |
| CN111368920B (en) * | 2020-03-05 | 2024-03-05 | 中南大学 | Quantum twin neural network-based classification method and face recognition method thereof |
-
2021
- 2021-03-02 CN CN202110231652.6A patent/CN113159303B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019055847A1 (en) * | 2017-09-15 | 2019-03-21 | President And Fellows Of Harvard College | Quantum artificial neural networks |
| WO2020245013A1 (en) * | 2019-06-04 | 2020-12-10 | Universita' Degli Studi Di Pavia | Artificial neural network on quantum computing hardware |
| CN110674921A (en) * | 2019-07-11 | 2020-01-10 | 中国科学技术大学 | Method for constructing quantum feedforward neural network based on classical training |
| CN112101518A (en) * | 2020-08-05 | 2020-12-18 | 华南理工大学 | Quantum system capable of simulating any nonlinear activation function |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113159303A (en) | 2021-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113159303B (en) | Quantum circuit-based artificial neuron construction method | |
| Ravi et al. | Financial time series prediction using hybrids of chaos theory, multi-layer perceptron and multi-objective evolutionary algorithms | |
| Li et al. | Deep feature learning architectures for daily reservoir inflow forecasting | |
| Kulkarni et al. | Quantum computing methods for supervised learning | |
| Zhou et al. | Shallow and deep neural network training by water wave optimization | |
| Adhikari et al. | A homogeneous ensemble of artificial neural networks for time series forecasting | |
| Dash et al. | Prediction of financial time series data using hybrid evolutionary Legendre neural network: Evolutionary LENN | |
| Khan et al. | A derivative-free method for quantum perceptron training in multi-layered neural networks | |
| CN114519430A (en) | Soft quantum neural network system and mode identification method | |
| Bar et al. | An approach based on quantum reinforcement learning for navigation problems | |
| Patel et al. | Quantum inspired binary neural network algorithm | |
| CN111144500A (en) | Differential privacy deep learning classification method based on analytic Gaussian mechanism | |
| Kumar et al. | Neural-network techniques for software-quality evaluation | |
| Laokondee et al. | Quantum Neural Network model for Token allocation for Course Bidding | |
| Altares-López et al. | AutoQML: Automatic generation and training of robust quantum-inspired classifiers by using evolutionary algorithms on grayscale images | |
| CN114511092B (en) | Graph attention machine mechanism implementation method based on quantum circuit | |
| Alam et al. | Novel deep generative simultaneous recurrent model for efficient representation learning | |
| Wang et al. | Stochastic adaptive CL-BFGS algorithms for fully complex-valued dendritic neuron model | |
| Rajabi et al. | Long-term prediction in Tehran stock market using a new architecture of Deep neural networks | |
| Shafie et al. | Backpropagation neural network with new improved error function and activation function for classification problem | |
| Guo et al. | A Deep Neural Network Based Model For Jane Street Market Prediction | |
| Huang et al. | Post-variational quantum neural networks | |
| Das et al. | A harmony search-based artificial neural network for stock market prediction | |
| Soni et al. | In depth analysis, applications and future issues of artificial neural network | |
| Nayak et al. | ACRRFLN: Artificial chemical reaction of recurrent functional link networks for improved stock market prediction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |