CN116310473A - Quantum neural network image classification method based on error mitigation - Google Patents
Quantum neural network image classification method based on error mitigation Download PDFInfo
- Publication number
- CN116310473A CN116310473A CN202211410414.2A CN202211410414A CN116310473A CN 116310473 A CN116310473 A CN 116310473A CN 202211410414 A CN202211410414 A CN 202211410414A CN 116310473 A CN116310473 A CN 116310473A
- Authority
- CN
- China
- Prior art keywords
- quantum
- error
- state
- representing
- neural network
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 230000000116 mitigating effect Effects 0.000 title claims abstract description 21
- 239000013598 vector Substances 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000001514 detection method Methods 0.000 claims abstract description 5
- 238000005259 measurement Methods 0.000 claims description 14
- 239000002096 quantum dot Substances 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 9
- 230000006835 compression Effects 0.000 claims description 5
- 238000007906 compression Methods 0.000 claims description 5
- 230000006378 damage Effects 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 12
- 238000010801 machine learning Methods 0.000 abstract description 6
- 238000012549 training Methods 0.000 abstract description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 239000003381 stabilizer Substances 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 208000011580 syndromic disease Diseases 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- 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
- G06N10/40—Physical realisations or architectures of quantum processors or components for manipulating qubits, e.g. qubit coupling or qubit control
-
- 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
- G06N10/60—Quantum algorithms, e.g. based on quantum optimisation, quantum Fourier or Hadamard transforms
-
- 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
- G06N10/70—Quantum error correction, detection or prevention, e.g. surface codes or magic state distillation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of quantum neural networks, and discloses an image classification method of a quantum neural network based on error mitigation, which comprises the following steps: preprocessing a data set, loading classical feature vectors, relieving errors, judging a classical layer and outputting data. The invention compresses the input high-dimensional image into low-dimensional characteristic representation by using tensor network to match the input of variable component sub-circuits, maximally utilizes the existing quantum resources and opens up a new road for analyzing the image by quantum machine learning in NISQ era; the error bits are filtered by using the quantum automatic encoder based on detection, so that the influence of noise in a quantum circuit on an algorithm result is greatly relieved, the accuracy of algorithm classification is obviously improved, and the accuracy of the algorithm is ensured; the number of training parameters is reduced, fewer parameters are needed compared with a classical equivalent model, and a mixed classical quantum circuit with millions of parameters can be optimized relatively efficiently.
Description
Technical Field
The invention relates to the technical field of quantum neural networks, in particular to an image classification method of a quantum neural network based on error mitigation.
Background
Recently, quantum computing has been used for machine learning, and it is expected that uncertainty in quantum computing can be a great advantage of probability-based modeling in machine learning, driving new research into noise mesoscale quantum (NISQ) devices. In recent years, variable component sub-circuits (VQC) have become one of the most effective methods of deep quantum learning when used with NISQ devices. With the advent of available quantum computing devices and quantum variation algorithms, quantum machine learning research has begun focusing on hybrid classical quantum algorithms that can be performed in short-term noise mesoscale quantum (NISQ) devices. Due to the limited number of qubits, these pave the way for machine learning applications that ultimately use the NISQ device in practice. VQC can be trained using classical optimization methods, which are considered to have certain expression advantages over classical neural network topologies, including quantum state-based Hopfield networks and their classical implementation using quantum acceleration matrix inversion, recursive quantum neural networks, quantum convolutional neural networks, and peak quantum neural networks. However, due to the large amount of noise in the quantum circuit, erroneous qubits are generated, affecting the final result of the algorithm. Therefore, reducing errors is important in quantum neural networks.
Among the many techniques developed for this purpose, the concept of subspaces is ubiquitous. In quantum error correction, some stabilizers define a subspace in which the quantum states are verified: the damaged condition detected by the syndrome measurement outside the subspace is corrected by the recovery operation. Alternatively, quantum computation may be performed in non-decoherence subspaces, which are chosen to be completely decoupled from certain ambient noise, thereby protecting the desired operation. It is expected that quantum algorithms can run completely successfully on approximately 50-100 qubits in the noisy medium-scale quantum (NISQ) era. However, quantum error correction on this scale of algorithms requires more controllable qubits, which presents a technical challenge. On the other hand, the incoherent subspace exists only in the selective error source, and the noise suppression capability is limited. This means that a more versatile approach is needed to reduce different types of errors within a limited number of controllable qubits. In view of this, we propose an image classification method based on the quantum neural network for error mitigation.
Disclosure of Invention
In order to overcome the problems in the prior art, the invention provides an image classification method of a quantum neural network based on error mitigation.
The technical scheme adopted for solving the technical problems is as follows: an image classification method of a quantum neural network based on error mitigation comprises the following steps:
s1, preprocessing a data set: compressing the input high-dimensional image into a low-dimensional feature representation using a tensor network to match the input of the variable component sub-line;
s2, loading classical eigenvectors into a quantum circuit: encoding the feature vector into a variable component sub-circuit using variable component encoding;
s3, error relief: filtering out erroneous bits using a quantum-based auto-encoder;
s4, outputting a classification result: the first four qubits are measured and the recovered data is subjected to classical processing using Softmax to calculate the probability of each potential class.
The specific method for preprocessing the data set in S1 includes the following steps:
s1.1, input image in data setFlattened into an N-dimensional vector x= (x) 1 ,x 2 ,...,x N ) And normalize each component to x i ∈[0,1];
S1.2, mapping the vector subjected to normalization processing to a product state by using feature mapping, wherein a specific conversion formula is as follows:
where Φ (x) > represents a product state obtained by using the feature map.
S1.3, MPS is further processed to generate a compressed representation, MPS is taken as a feature extraction program, data is encoded into a product state, the product state is contracted with the MPS, class labels or outputs are generated, the product state is obtained in step S1.2, and then input data is loaded into the MPS through the following operations:
wherein f (l) (x) Representing the loaded input dataRepresenting feature extractor approximation, ++>Representing the corresponding product state of each pixel point of the input image.
The specific method for loading the classical feature vector into the quantum circuit in the S2 comprises the following steps:
s2.1, a coding part, firstly creating an unbiased initial state by using a Hadamard gate, and then passing through a single quantum bit gate R y (arctan(x i ) Sum of (d)Implementing phase encodingWherein the characteristic of the rotation angle is derived from an input of the tensor network, and subsequently processing the encoded states using a parametric optimized variable component sub-circuit;
s2.2, a variation part, which consists of a CNOT gate for entangling each qubit quantum state and a unitary gate R (alpha, beta, gamma) representing a general single qubit.
The above-mentioned image classification method based on the quantum neural network of error relief, in said S3, the concrete method of error relief includes the following steps:
s3.1, firstly, coding the unitary operator U e Applied to noise quantum state, the compression state is formed byBecomes as followsWherein ε represents the probability of the quantum state being destroyed and ε [0,1 ]]Sigma represents the compressed quantum state, ρ err Representing error items->Representing a decoding unitary operator, U e Representing encoded unitary operator->Representing the noisy quantum state after compression, +.>Represent noisy quantum states, U e Transferring the error-free term to the potential subspace, while most of the errors remain in the garbage subspace;
s3.2, error detection and measurement are carried out, error quantum bits are detected, corresponding quantum data are discarded, and the state is projected to a potential subspace by measurement, so that errors in a garbage subspace are detected and eliminated;
wherein ρ' represents the error state, ε represents the probability of quantum state destruction, ε [0,1 ]],Trace representing projection of error term to supporting subspace, O (ε) 2 ) Representing the square order of the probability that the quantum state is destroyed, ρ represents the ideal quantum state,representing the projection of the error term into the supporting subspace.
The above-mentioned method for classifying images of quantum neural network based on error mitigation, in the step S4, the specific method for outputting the classification result includes the following steps:
s4.1, measuring the first four quantum bits recovered by the quantum self-encoder, transmitting all measurement results to the next layer by using a full-connection layer, and outputting Pauli-Z expected values by a multi-time operation quantum circuit by using the measurement part;
s4.2, mapping all score values into a probability value by using a sigmoid function;
and S4.3, judging the category of the image according to the probability value.
The beneficial effects of the invention are as follows: 1. in the method for classifying the images based on the quantum neural network for error alleviation, the tensor network is utilized to compress the input high-dimensional images into low-dimensional characteristic representation so as to match the input of the variable component sub-circuit, a new road is opened up for analyzing the images by the quantum machine learning process in the NISQ era, a small number of quantum bits can be ensured to be used, the analysis and the processing of the images can be realized, the classical quantum boundary can be regulated according to the availability of the quantum resources by utilizing the corresponding relation between the tensor network and the quantum circuit, and the existing quantum resources can be utilized to the maximum extent;
2. in the method for classifying the images of the quantum neural network based on error alleviation, due to the error alleviation characteristic of the quantum self-encoder, the error bits are filtered by using the quantum self-encoder based on detection, so that the influence of noise in a quantum circuit on algorithm results is greatly relieved, the accuracy of algorithm classification is obviously improved compared with other quantum classification algorithms, and the accuracy of the algorithm is ensured;
3. the image classification method based on the error alleviation quantum neural network has a simpler overall scheme, ensures simple hardware operation through the algorithmic integrity, and is easy to realize in an actual quantum computer.
4. In a noise environment, the image classification method of the quantum neural network based on error alleviation has higher precision than that of the traditional neural network, the traditional neural network is difficult to modify the network weight of the traditional neural network in the noisy environment, quantum bits have inherent probability characteristics, and the original pixel value is effectively used for approximating noise in a modification quantum layer, so that the classification precision is improved;
5. because the parameterized VQC in the modified quantum layer reduces the number of training parameters, the image classification method of the quantum neural network based on error relief needs fewer parameters than a classical equivalent model thereof to be realized, and when hundreds of quantum bits are possessed in the future, the image classification method of the quantum neural network based on error relief can be easily realized to relatively efficiently optimize a mixed classical quantum circuit with millions of parameters.
Drawings
The invention will be further described with reference to the drawings and examples.
FIG. 1 is a flow chart of an overall scheme of an image classification method of a quantum neural network based on error mitigation in the present invention;
FIG. 2 is a schematic diagram of tensor network feature extraction in accordance with the present invention;
FIG. 3 is a schematic diagram of error mitigation in accordance with the present invention.
Detailed Description
The present invention will be described in detail below with reference to the drawings and detailed description to enable those skilled in the art to better understand the technical scheme of the present invention.
As shown in fig. 1, the present embodiment provides an image classification method of a quantum neural network based on error mitigation, and the overall transmission scheme of the present embodiment is mainly divided into four parts of a data set preprocessing stage, a classical feature vector loading stage, an error mitigation stage, and a classical layer judgment stage, and specifically includes the following steps:
s1, preprocessing a data set: compressing the input high-dimensional image into a low-dimensional feature representation using a tensor network to match the input of the variable component sub-line;
s2, loading classical eigenvectors into a quantum circuit: encoding the feature vector into a variable component sub-circuit using variable component encoding;
s3, error relief: filtering out erroneous bits using a quantum-based auto-encoder;
s4, outputting a classification result: the first four qubits are measured and the recovered data is subjected to classical processing using Softmax to calculate the probability of each potential class.
In this embodiment, in S1, a specific method for preprocessing a data set includes the following steps:
s1.1, flattening input images 0,1,2,3 with the size of 28 x 28 in the mnist data set into 784-dimensional vectors x= (x) respectively 1 ,x 2 ,...,x N ) And normalize each component to x i ∈[0,1];
S1.2, mapping the vector subjected to normalization processing to a product state by using feature mapping;
s1.3, MPS is further processed to generate a compressed representation, MPS is the feature extraction procedure, as shown in fig. 2, the data is encoded into a product state, which contracts with MPS and generates class labels or outputs. The product state is obtained according to step S1.2, and then the input data is loaded into the MPS by the following operations.
As a further improvement of the present technical solution, in S1.2, a specific conversion formula is:
where Φ (x) > represents a product state obtained by using the feature map.
The product state is obtained according to step S1.2, and then the input data is loaded into MPS by:
wherein f (l) (x) Representing the MPS after loading of the input data,representing feature extractor approximation, ++>Representing the corresponding product state of each pixel point of the input image.
In this embodiment, in S2, the specific method for loading the classical feature vector into the quantum circuit includes the following steps:
s2.1, a coding part, firstly creating an unbiased initial state by using a Hadamard gate, and then passing through a single quantum bit gate R y (arctan(x i ) Sum of (d)Implementing phase encoding, wherein the characteristic of the rotation angle is derived from the input of the tensor network, followed by processing the encoded states using a parametric optimized variable component sub-circuit;
s2.2, a variation part, which consists of a CNOT gate used for entanglement of each quantum bit quantum state and a unitary gate R (alpha, beta, gamma) representing a general single quantum bit, wherein three parameters alpha, beta and gamma need to be learned, the three parameters represent the rotation angle of the unitary gate, the three parameters are initialized randomly firstly, and then the three parameters are updated by adopting a counter-propagation algorithm based on an Adam optimizer.
In which S2.1 the outputs of the classical parts need to be encoded in order for the quantum circuits to use them. The general N-qubit quantum state can be expressed as:
wherein |ψ>Represents the N-qubit quantum state, |q N >Representing the Nth quantum stateRepresenting the amplitude of each quantum state.
We choose the variant encoding method to encode classical data into quantum states. Initial quantum state First experience->Operation of creating unbiased State->Where H is Hadamard. Considering an n-qubit system, the corresponding unbiased initial state is:
wherein, the liquid crystal display device comprises a liquid crystal display device,represents an unbiased initial state, h|0>Represents the quantum state in which the H-gate acts in the spin direction, |1>Representing a spin-down quantum state, |0>Represents the quantum state in the spin-up direction, |k>Representing unbiased quantum states, n representing n qubits.
In the present embodiment, S3In the specific method of error mitigation, as shown in fig. 3, the large ellipse in fig. 3 represents the complete hilbert space H, the small ellipse inside represents the potential subspace L, and the rest is the garbage subspace J, we use the points to represent the error-free term (1-epsilon) ρ, and the cross represents the error term epsilon ρ err ,U e Transferring the error-free term to L while most of the error remains in J, measuring the projection of the state to the potential subspace, thereby detecting and eliminating the error in J, and finally, applyingAnd the quantum data is recovered, and the error is reduced. The method comprises the following specific steps:
s3.1, firstly, coding the unitary operator U e Applied to the quantum state processed by the variable component sub-circuit, the quantum state is formed byBecome->Wherein ε represents the probability of the quantum state being destroyed and ε [0,1 ]]Sigma represents the compressed quantum state, ρ err Representing error items->Representing a decoding unitary operator, U e Representing encoded unitary operator->Representing the noisy quantum state after compression, +.>Represent noisy quantum states, U e Transferring the error-free term to the potential subspace, while most of the errors remain in the garbage subspace;
s3.2, error detection and measurement are carried out, error quantum bits are detected, corresponding quantum data are discarded, and the measurement projects states to a potential subspace, so that errors in a garbage subspace are detected and eliminated.
Specifically, in S3.3, the error state may be written as:
wherein ρ' represents the error state, ε represents the probability of quantum state destruction, ε [0,1 ]],Trace representing projection of error term to supporting subspace, O (ε) 2 ) Square order representing probability of destruction of quantum state, ρ representing ideal quantum state, +.>Representing the projection of the error term into the supporting subspace.
In this embodiment, in S4, the specific method for outputting the classification result includes the following steps:
s4.1, measuring the first four quantum bits recovered by the quantum self-encoder, transmitting all measurement results to the next layer by using a full-connection layer, and outputting Pauli-Z expected values by a multi-time operation quantum circuit by using the measurement part;
s4.2, mapping all score values into a probability value by using a sigmoid function;
and S4.3, finally judging the category of the image according to the probability value.
Through the steps of the method, the embodiment combines the general quantum error mitigation protocol with the quantum neural network for image classification for the first time, does not need to use multiple bits as large equipment cost of a line gate, is applicable to NISQ equipment, in which the classification accuracy of the image classification method can reach 94.1%, the accuracy refers to the proportion of correctly classified samples to the total number of samples, the mnist data set is input into TN-VQC and RQNN models for image classification, and the classification accuracy of the two models is calculated, wherein the classification accuracy of TN-VQC is 75.5% and the classification accuracy of RQNN is 71.6%, and therefore, the image classification method of the embodiment is obviously higher than that of the RQNN and TN-VQC models in the prior art.
Wherein quantum data is compressed into a potential subspace while leaving errors outside, the latter then being eliminated by measurement and post-selection. In contrast to previously developed methods, our protocol does not require extra qubits on the one hand and on the other hand it has near optimal denoising capabilities, all errors detected outside the potential subspace can be removed, and errors detected in the subspace cannot be removed anyway, under reasonable requirements. Therefore, our method of image classification based on error mitigation quantum neural networks is particularly useful for recent quantum devices where controllable quantum bit limitation and noise reduction is important.
The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention, the scope of which is defined by the claims. Various modifications and equivalent arrangements of this invention will occur to those skilled in the art, and are intended to be within the spirit and scope of the invention.
Claims (5)
1. An image classification method of a quantum neural network based on error mitigation is characterized by comprising the following steps:
s1, preprocessing a data set: compressing the input high-dimensional image into a low-dimensional feature representation using a tensor network to match the input of the variable component sub-line;
s2, loading classical eigenvectors into a quantum circuit: encoding the feature vector into a variable component sub-circuit using variable component encoding;
s3, error relief: filtering out erroneous bits using a quantum-based auto-encoder;
s4, outputting a classification result: the first four qubits are measured and the recovered data is subjected to classical processing using Softmax to calculate the probability of each potential class.
2. The method for classifying images based on the quantum neural network for error mitigation according to claim 1, wherein in S1, the specific method for preprocessing the data set comprises the following steps:
s1.1, flattening the input image in the dataset into an N-dimensional vector x= (x) 1 ,x 2 ,...,x N ) And normalize each component to x i ∈[0,1];
S1.2, mapping the vector subjected to normalization processing to a product state by using feature mapping, wherein a specific conversion formula is as follows:
where Φ (x) > represents a product state obtained by using the feature map.
S1.3, MPS is further processed to generate a compressed representation, MPS is taken as a feature extraction program, data is encoded into a product state, the product state is contracted with the MPS, class labels or outputs are generated, the product state is obtained in step S1.2, and then input data is loaded into the MPS through the following operations:
3. The method for classifying images of a quantum neural network based on error mitigation according to claim 1, wherein in S2, the specific method for loading classical eigenvectors into a quantum circuit comprises the following steps:
s2.1, a coding part, firstly creating an unbiased initial state by using a Hadamard gate, and then passing through a single quantum bit gate R y (arctan(x i ) Sum of (d)Implementing phase encoding, wherein the characteristic of the rotation angle is derived from the input of the tensor network, followed by processing the encoded states using a parametric optimized variable component sub-circuit;
s2.2, a variation part, which consists of a CNOT gate for entangling each qubit quantum state and a unitary gate R (alpha, beta, gamma) representing a general single qubit.
4. The method for classifying images based on the quantum neural network for error mitigation according to claim 1, wherein in S3, the specific method for error mitigation comprises the following steps:
s3.1, firstly, coding the unitary operator U e Applied to noise quantum state, the compression state is formed byBecomes as followsWherein ε represents the probability of the quantum state being destroyed and ε [0,1 ]]Sigma represents the compressed quantum state, ρ err Representing error items->Representing a decoding unitary operator, U e Representing encoded unitary operator->Representing the noisy quantum state after compression, +.>Represent noisy quantum states, U e Transferring the error-free term to the potential subspace, while most of the errors remain in the garbage subspace;
s3.2, error detection and measurement are carried out, error quantum bits are detected, corresponding quantum data are discarded, and the state is projected to a potential subspace by measurement, so that errors in a garbage subspace are detected and eliminated;
wherein ρ' represents the error state, ε represents the probability of quantum state destruction, ε [0,1 ]],Trace representing projection of error term to supporting subspace, O (ε) 2 ) Square order representing probability of destruction of quantum state, ρ representing ideal quantum state, +.>Representing the projection of the error term into the supporting subspace.
5. The method for classifying images based on the quantum neural network for error mitigation according to claim 1, wherein in S4, the specific method for outputting the classification result comprises the following steps:
s4.1, measuring the first four quantum bits recovered by the quantum self-encoder, transmitting all measurement results to the next layer by using a full-connection layer, and outputting Pauli-Z expected values by a multi-time operation quantum circuit by using the measurement part;
s4.2, mapping all score values into a probability value by using a sigmoid function;
and S4.3, judging the category of the image according to the probability value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211410414.2A CN116310473A (en) | 2022-11-11 | 2022-11-11 | Quantum neural network image classification method based on error mitigation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211410414.2A CN116310473A (en) | 2022-11-11 | 2022-11-11 | Quantum neural network image classification method based on error mitigation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116310473A true CN116310473A (en) | 2023-06-23 |
Family
ID=86776755
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211410414.2A Pending CN116310473A (en) | 2022-11-11 | 2022-11-11 | Quantum neural network image classification method based on error mitigation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116310473A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117456561A (en) * | 2023-12-22 | 2024-01-26 | 华侨大学 | Pedestrian re-identification method based on component quantum learning |
CN117649563A (en) * | 2024-01-29 | 2024-03-05 | 量子科技长三角产业创新中心 | Quantum recognition method, system, electronic device and storage medium for image category |
-
2022
- 2022-11-11 CN CN202211410414.2A patent/CN116310473A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117456561A (en) * | 2023-12-22 | 2024-01-26 | 华侨大学 | Pedestrian re-identification method based on component quantum learning |
CN117456561B (en) * | 2023-12-22 | 2024-03-12 | 华侨大学 | Pedestrian re-identification method based on component quantum learning |
CN117649563A (en) * | 2024-01-29 | 2024-03-05 | 量子科技长三角产业创新中心 | Quantum recognition method, system, electronic device and storage medium for image category |
CN117649563B (en) * | 2024-01-29 | 2024-05-10 | 量子科技长三角产业创新中心 | Quantum recognition method, system, electronic device and storage medium for image category |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11461628B2 (en) | Method for optimizing neural networks | |
CN116310473A (en) | Quantum neural network image classification method based on error mitigation | |
US8463722B2 (en) | Implementing a neural associative memory based on non-linear learning of discrete synapses | |
Qian et al. | Feature learning for steganalysis using convolutional neural networks | |
CN111259904B (en) | Semantic image segmentation method and system based on deep learning and clustering | |
JP2023549579A (en) | Temporal Bottleneck Attention Architecture for Video Behavior Recognition | |
US20200143209A1 (en) | Task dependent adaptive metric for classifying pieces of data | |
Flenner et al. | A deep non-negative matrix factorization neural network | |
CN108985442B (en) | Handwriting model training method, handwritten character recognition method, device, equipment and medium | |
Wadekar et al. | Hybrid CAE-VAE for unsupervised anomaly detection in log file systems | |
US11989162B2 (en) | System and method for optimized processing of information on quantum systems | |
CN116992396A (en) | Redundancy self-adaptive multi-mode robust fusion learning method and system | |
Tao et al. | Efficient incremental training for deep convolutional neural networks | |
Lai et al. | Robust variational autoencoding with wasserstein penalty for novelty detection | |
US11727109B2 (en) | Identifying adversarial attacks with advanced subset scanning | |
Farooq et al. | Noise removal from the image using convolutional neural networks-based denoising auto encoder | |
CN116310644A (en) | Abnormality detection method and device based on normalized stream network and encoder | |
Lai et al. | Novelty detection via robust variational autoencoding | |
CN108009586B (en) | Capping concept decomposition method and image clustering method | |
CN112949344B (en) | Characteristic autoregression method for anomaly detection | |
Ali et al. | A deep convolutional neural wavelet network for classification of medical images | |
CN111797732B (en) | Video motion identification anti-attack method insensitive to sampling | |
CN114972871A (en) | Image registration-based few-sample image anomaly detection method and system | |
CN113011163A (en) | Compound text multi-classification method and system based on deep learning model | |
Wang et al. | Replication study of decaf: Generating fair synthetic data using causally-aware generative networks |
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 |