CN111984960A - Privacy protection equipment identification model design and use method based on homomorphic encryption - Google Patents
Privacy protection equipment identification model design and use method based on homomorphic encryption Download PDFInfo
- Publication number
- CN111984960A CN111984960A CN202010666718.XA CN202010666718A CN111984960A CN 111984960 A CN111984960 A CN 111984960A CN 202010666718 A CN202010666718 A CN 202010666718A CN 111984960 A CN111984960 A CN 111984960A
- Authority
- CN
- China
- Prior art keywords
- homomorphic encryption
- identification model
- layer
- data
- privacy protection
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000013461 design Methods 0.000 title claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 25
- 230000008569 process Effects 0.000 claims abstract description 14
- 230000000977 initiatory effect Effects 0.000 claims abstract description 12
- 230000003993 interaction Effects 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 45
- 230000005284 excitation Effects 0.000 claims description 31
- 238000011176 pooling Methods 0.000 claims description 21
- 238000004422 calculation algorithm Methods 0.000 claims description 14
- 230000005540 biological transmission Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 238000004891 communication Methods 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/44—Program or device authentication
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- 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
- G06N3/045—Combinations of networks
-
- 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
-
- 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
- G06N3/08—Learning methods
Abstract
A privacy protection device identification model design and use method based on homomorphic encryption collects device fingerprint characteristics required by training a privacy protection device identification model based on homomorphic encryption, a Convolutional Neural Network (CNN) is used in a training process of using the homomorphic encryption device identification model, credible interaction of an initiating inquirer and a cloud device is guaranteed, and invisibility of original data in an interaction process from encryption to decryption of data by a cloud end is guaranteed.
Description
Technical Field
The invention relates to the technical field of data security and industrial control, in particular to a privacy protection equipment identification model design and use method based on homomorphic encryption.
Background
Due to the high complexity of decryption, homomorphic encryption can effectively protect sensitive data from being decrypted and stolen, and in deep learning, homomorphic encryption provides a model with the ability of predicting encrypted data, so that the homomorphic encryption is mainly used for protecting prediction input and results.
Phong et al propose a privacy-Preserving Deep Learning system using asynchronous stochastic gradient descent applied to neural network connection Deep Learning and cryptography, in combination with Homomorphic cryptography (Le Trieu Phong, Yoshinorri Aono, Takuya Hayashi, Lihua Wang, Shiho Moraii.privacy-Preserving Deep Learning via Additive Homorphillic encryption. IEEE Transactions on Information Forenses&Security, 2018). Orlandi et al propose hiding the scalar product result and adding some dummy neurons to the structure of the Neural Network before delegating the evaluation of the activation function to the client, which may play a role in that the model is not stolen (Orlandi C, Piva a, Barni m. Aslett et al propose a method to perform statistical machine learning on the encrypted data and implement an extremely random forest and naive bayes classifier (l.j.m.aslett, p.m) on more than 20 data sets.C.C.Holmes, Encrypted static machine learning: new privacy-preserving methods. CoRR, 2015). Zhang et al propose to use a BGV encryption scheme to effectively support secure computation of a high-order back propagation algorithm for deep computation model training on the cloud. Liu et al propose MiniONN, a neural network that supports privacy protection and ensures that servers do not know the input and clients do not know the modelThe clients additionally share input and output values for each layer of the Neural Network (Jian Liu, Mika Juuti, Yao Lu, n.hookan.obrevious Neural Network prediction via minion transformations.acm signal Conference, 2017). Hesamifar et al developed CryptoDL for running DNN on Encrypted Data, which designed an approximation function in CNN using a low order polynomial, then trained CNN using the approximation polynomial instead of the original activation function, and finally implemented CNN on Encrypted Data (Hesamifar, Ehsan, Takabi, Hassan, Ghasemi, Mehdi. CryptoDL: Deep Neural Networks over Encrypted Data).
The existing homomorphic encryption can be divided into partial homomorphic encryption, similar homomorphic encryption and full homomorphic encryption, which respectively support addition or multiplication homomorphism only, limited-time ciphertext domain addition and multiplication homomorphism and arbitrary-time ciphertext addition and multiplication homomorphic operation. Using homomorphic encryption only enables a limited very limited set of add and multiply operations to be performed on encrypted data, which enables a device identification model to be applied to the encrypted data and return the encrypted results without affecting security and privacy. However, due to the many constraints associated with these cryptographic schemes, homomorphic encryption cannot be directly used in the deep learning model, and therefore, it is necessary to design a practical and effective solution to run the device identification model on the encrypted data and apply the device identification model in the scenario where device privacy needs to be protected. The machine learning algorithm based on the convolutional neural network obtains a lot of achievements, is widely applied to different fields, and is also taken as a feasible scheme particularly in the fine-grained identification process of the Internet of things equipment. The invention describes a construction method of a homomorphic encryption equipment identification model by taking a convolutional neural network algorithm training equipment identification model as an example. The convolutional neural network algorithm carries out very complex calculation on privacy sensitive data, and the limited calculation capacity used by homomorphic encryption is difficult to support network training, so that the deep neural network model cannot directly apply the existing homomorphic encryption method. Therefore, the invention provides an encrypted equipment identification model trained under homomorphic encryption constraint to support the predictive identification of encrypted data, and simultaneously, the precision of the original equipment identification model is kept unchanged.
Disclosure of Invention
The invention aims to provide a privacy protection equipment identification model design and a using method based on homomorphic encryption.
In order to achieve the purpose, the invention adopts the following technical scheme:
a privacy protection equipment identification model design method based on homomorphic encryption is characterized in that equipment fingerprint characteristics required by training a privacy protection equipment identification model based on homomorphic encryption are collected, and a Convolutional Neural Network (CNN) is used for training the equipment identification model using homomorphic encryption;
in training, the Convolutional Neural Network (CNN) comprises: the device comprises an input layer, a convolution layer, an excitation layer, a pooling layer, a full-connection layer and an output layer;
the input layer is used for inputting and collecting equipment fingerprint characteristics required by the training model, and the equipment fingerprint characteristics comprise coarse-grained characteristics and fine-grained characteristics;
the convolution layer is used for learning features from data, taking a subset output by the input layer for calculation, the step only comprises addition and multiplication, and the encrypted data is calculated through a homomorphic encryption mechanism;
the excitation layer obtains polynomial approximation of the excitation function by using a method for simulating a derivative of the excitation function, so that a homomorphic encryption scheme is used in the calculation process of the excitation layer;
the pooling layer is used for sub-sampling from the data and reducing the length of the processed data, and specifically, an amplified version of average pooling is adopted to calculate the sum of all values, but the sum is not divided by the number of the values, and the average pooling can be realized by adding without influencing the depth of the algorithm;
and the full connection layer is used for connecting each neuron in the full connection layer to all neurons in the pooling layer, each connection is represented by a value called weight, and each node outputs a weighted sum on the whole pooling layer so as to meet the addition operation requirement of the homomorphic encryption scheme.
Optionally, in the input layer, the coarse-grained features refer to a device type, a device brand, a device model, a firmware version number, a port number used by the device, and the fine-grained features refer to a protocol header length field Len in a header field of a TCP protocol used for device communication, a first-time-out-time field RTT, a header option TCP Segment, and a longest-latency-to-message RTO.
Optionally, the homomorphic encryption mechanism of the convolutional layer is a DGHV fully homomorphic encryption algorithm.
Optionally, in the excitation layer, an excitation function ReLU (rectified Linear unit) function is simulated, and the amount of computation can be reduced to some extent by using the ReLU, so as to avoid the problem of increasing the number of layers, and the formula is as follows:
the derivative of the ReLU function is similar to the Step function and is not differentiable at point 0, with the post-derivative formula as follows:
if the function is continuous and infinitely differentiated, it can be approximated more accurately than a discontinuous or non-infinite differential function, and then the resulting approximation polynomial is integrated to use it as the excitation function.
Optionally, the pooling layer can sub-sample from the data and reduce the length of the processed data.
The invention also discloses a use method of the privacy protection equipment identification model based on homomorphic encryption, which is designed according to the privacy protection equipment identification model based on homomorphic encryption, and the use method comprises the following steps:
step (1), deploying a privacy protection equipment identification model based on homomorphic encryption obtained through the training at a cloud end;
step (2), when equipment identification is needed, encrypting the equipment characteristic data by using a homomorphic encryption algorithm and then sending the encrypted equipment characteristic data to a cloud end;
step (3), after the cloud receives the encrypted characteristic data, directly inputting the encrypted data into a homomorphic encryption equipment identification model of the cloud;
step (4), the cloud transmits the encryption prediction result to the inquiry initiating terminal;
and (5) the initiating query end receives the prediction result returned by the cloud end.
Optionally, in step (1), since the device identification model is homomorphically encrypted, neither the specific way of model training nor parameter information is revealed.
Optionally, in step (2), all data in the transmission process is encrypted homomorphically, and even if the data is intercepted in the transmission process, the data cannot be easily cracked.
Optionally, in the step (3), the model can predict and identify the encrypted device characteristic data, and the encrypted device characteristic data does not need to be decrypted and then input into the device identification model, so that the cloud end is ensured not to obtain a transmitted data plaintext, and the device identification model at the cloud end is also ensured not to cause model parameter leakage due to a prediction process;
in the step (4), an available result cannot be obtained even if the result is intercepted in the transmission process, and the privacy of the interaction of the prediction result is further ensured.
Optionally, in step (5), the plaintext prediction result may be obtained only after being decrypted by a private key corresponding to the querying party.
According to the method, the compactness of the ciphertext length and the simplicity of the construction scheme of the fully homomorphic encryption scheme are utilized, so that the cloud model is ensured to have the capacity of predicting the ciphertext, and the implementation cost of the scheme is reduced.
The cloud end of the method cannot obtain any information related to the prediction result and any parameter information related to the equipment identification model, and data in all interaction processes are transmitted in an encrypted mode.
Drawings
FIG. 1 is a schematic diagram of a convolutional neural network used in a privacy preserving device identification model design method based on homomorphic encryption according to a specific embodiment of the present invention;
fig. 2 is a method for using a privacy preserving apparatus identification model based on homomorphic encryption according to a specific embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.
The invention is characterized in that:
(1) a Convolutional Neural Network (CNN) is used in the training and prediction process of a homomorphic encryption equipment identification model, and a low-order polynomial is adopted to approximate an excitation function so as to solve the problem that homomorphic encryption cannot be directly applied to the excitation function. Therefore, the method utilizes the compactness of the ciphertext length and the simplicity of the construction scheme of the fully homomorphic encryption scheme, ensures that the cloud model has the capacity of predicting the ciphertext, and reduces the implementation cost of the scheme.
(2) The homomorphic encryption is used in the data interaction process, the initiating inquiry end sends the encrypted data packet to the cloud end, the cloud end conducts inference prediction by using the encrypted equipment identification model, then the edge server obtains an encrypted prediction result, and then a returned result is decrypted by using a private key, so that a prediction result true value is obtained. Therefore, the cloud end of the method cannot obtain any information related to the prediction result and any parameter information related to the equipment identification model, and data in all interaction processes are transmitted in an encrypted mode.
The present invention uses a Convolutional Neural Network (CNN) in the training and prediction process of device identification models using homomorphic encryption. The convolutional neural network is a multi-neuron combination arranged in an ordered layer, each neuron obtains input, operates on the input and then outputs the result of the function, the functions can be Sigmoid or ReLU functions and are called excitation functions (or transfer functions), but the functions cannot directly use homomorphic encryption, because homomorphic encryption can only use two operation methods of addition and multiplication, in order to use homomorphic encryption, the excitation functions must be replaced by functions (such as polynomials) only comprising the addition and the multiplication, and therefore, for the encrypted data operation deep neural network algorithm, the approximation of the excitation functions by low-order polynomials is an important solution.
In particular, referring to fig. 1, there is shown a convolutional neural network used in the method for designing a privacy preserving device identification model based on homomorphic encryption of the present invention,
a privacy protection equipment identification model design method based on homomorphic encryption is characterized in that equipment fingerprint characteristics required by training a privacy protection equipment identification model based on homomorphic encryption are collected, and a Convolutional Neural Network (CNN) is used for training the equipment identification model using homomorphic encryption;
in training, the Convolutional Neural Network (CNN) comprises: the device comprises an input layer, a convolution layer, an excitation layer, a pooling layer, a full-connection layer and an output layer;
the input layer is used for inputting and collecting equipment fingerprint features required by the training model, and the equipment fingerprint features comprise coarse-grained features and fine-grained features.
Specifically, the coarse-grained characteristics refer to the type of the device, the brand of the device, the model of the device, the version number of the firmware, and the port number used by the device; the fine-grained characteristics refer to a protocol header length field Len, a first waiting timeout field RTT, a header optional TCP Segment and a message longest waiting time RTO in a header field of a TCP protocol used for device communication.
The convolution layer is used for learning features from data, a subset of input layer output is taken for calculation, the step only comprises addition and multiplication, and encrypted data is calculated through a homomorphic encryption mechanism.
Further preferably, the homomorphic encryption mechanism is a DGHV fully homomorphic encryption algorithm.
The excitation layer obtains the polynomial approximation of the excitation function by using a method of simulating the derivative of the excitation function, so that the homomorphic encryption scheme is used in the calculation process of the excitation layer.
Compared with the traditional method, such as Taylor series or Chebyshev inequality directly approximating the polynomial expression of the excitation function, the method does not directly adopt the simulation excitation function, but adopts the derivative of the simulation excitation function. For example: the ReLU (rectified Linear Unit) function, i.e. the Linear rectification function, is a nonlinear excitation function commonly used in the neural network structure, and the ReLU can reduce the operation amount to a certain extent and avoid the problem of layer number increase, and the formula is as follows:
the derivative of the ReLU function is similar to the Step function and is not differentiable at point 0, with the post-derivative formula as follows:
if the function is continuous and infinitely differentiated, it can be approximated more accurately than a discontinuous or non-infinite differential function, and then the resulting approximation polynomial is integrated to use it as the excitation function.
The pooling layer is used for sub-sampling from the data and reducing the length of the processed data, and specifically, the sum of all values is calculated by adopting an amplified version of average pooling, but the sum is not divided by the number of values, and the step can realize average pooling by adding without influencing the depth of the algorithm.
Further preferably, the pooling layer is capable of sub-sampling from the data and reducing the length of the processed data.
And the full connection layer is used for connecting each neuron in the full connection layer to all neurons in the pooling layer, each connection is represented by a value called weight, and each node outputs a weighted sum on the whole pooling layer so as to meet the addition operation requirement of the homomorphic encryption scheme.
Further, referring to fig. 2, a method for using a privacy protecting device identification model based on homomorphic encryption, which is designed by using the above privacy protecting device identification model based on homomorphic encryption design method, is shown.
And (1) deploying the privacy protection equipment recognition model based on homomorphic encryption obtained through the training at the cloud.
In this step, since the device identification model is homomorphically encrypted, the specific way and parameter information of model training are not revealed.
And (2) when equipment identification is required, encrypting the equipment characteristic data by using a homomorphic encryption algorithm and then sending the encrypted equipment characteristic data to the cloud.
In the step, all data in the transmission process are homomorphic encrypted, and even if the data are intercepted in the transmission process, the data cannot be easily cracked.
And (3) directly inputting the encrypted data into a homomorphic encryption equipment identification model of the cloud after the cloud receives the encrypted characteristic data.
In this step, the model can predict and recognize the encrypted device characteristic data, the encrypted device characteristic data does not need to be decrypted and then input into the device identification model, the cloud end is guaranteed not to obtain the transmitted data plaintext, and the device identification model at the cloud end is also guaranteed not to cause model parameter leakage due to the prediction process. Since the generated prediction result is also encrypted, the privacy of the prediction result can be ensured.
And (4) the cloud transmits the encryption prediction result to the inquiry initiating terminal.
In the step, an available result cannot be obtained even if the prediction result is intercepted in the transmission process, and the privacy of the interaction of the prediction result is further ensured.
And (5) the initiating query end receives the prediction result returned by the cloud end.
In this step, the prediction result of the plaintext can only be obtained after the private key corresponding to the initiating inquirer is decrypted.
In the drawings, the query initiating end is an edge server, but this is merely an example, and the present invention is not limited thereto, as long as the device capable of initiating the query can be implemented.
While the invention has been described in further detail with reference to specific preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A privacy protection equipment identification model design method based on homomorphic encryption is characterized in that,
collecting device fingerprint characteristics required for training a privacy protection device identification model based on homomorphic encryption, and using a Convolutional Neural Network (CNN) for training the device identification model using homomorphic encryption;
in training, the Convolutional Neural Network (CNN) comprises: the device comprises an input layer, a convolution layer, an excitation layer, a pooling layer, a full-connection layer and an output layer;
the input layer is used for inputting and collecting equipment fingerprint characteristics required by the training model, and the equipment fingerprint characteristics comprise coarse-grained characteristics and fine-grained characteristics;
the convolution layer is used for learning features from data, taking a subset output by the input layer for calculation, the step only comprises addition and multiplication, and the encrypted data is calculated through a homomorphic encryption mechanism;
the excitation layer obtains polynomial approximation of the excitation function by using a method for simulating a derivative of the excitation function, so that a homomorphic encryption scheme is used in the calculation process of the excitation layer;
the pooling layer is used for sub-sampling from the data and reducing the length of the processed data, and specifically, an amplified version of average pooling is adopted to calculate the sum of all values, but the sum is not divided by the number of the values, and the average pooling can be realized by adding without influencing the depth of the algorithm;
and the full connection layer is used for connecting each neuron in the full connection layer to all neurons in the pooling layer, each connection is represented by a value called weight, and each node outputs a weighted sum on the whole pooling layer so as to meet the addition operation requirement of the homomorphic encryption scheme.
2. The privacy protection device identification model design method based on homomorphic encryption as claimed in claim 1,
in the input layer, the coarse-grained characteristics refer to device type, device brand, device model and firmware version number, port number used by the device, and the fine-grained characteristics refer to protocol header length field Len in header field of TCP protocol used for device communication, first-time-out-waiting field RTT, header option TCP Segment and message longest-waiting time RTO.
3. The privacy protection device identification model design method based on homomorphic encryption as claimed in claim 1,
the homomorphic encryption mechanism used by the convolutional layer is a DGHV fully homomorphic encryption algorithm.
4. The privacy protection device identification model design method based on homomorphic encryption as claimed in claim 1,
in the excitation layer, the simulation excitation function is a ReLU (rectified Linear unit) function, the use of the ReLU can reduce the operation amount to a certain extent, and avoid the problem of layer number increase, and the formula is as follows:
the derivative of the ReLU function is similar to the Step function and is not differentiable at point 0, with the post-derivative formula as follows:
if the function is continuous and infinitely differentiated, it can be approximated more accurately than a discontinuous or non-infinite differential function, and then the resulting approximation polynomial is integrated to use it as the excitation function.
5. The privacy protection device identification model design method based on homomorphic encryption as claimed in claim 1,
the pooling layer enables sub-sampling from the data and reduces the length of the processed data.
6. A method for using the privacy protection device identification model based on homomorphic encryption, which is designed according to the privacy protection device identification model based on homomorphic encryption design method based on any one of claims 1-5, and is characterized by comprising the following steps:
step (1), deploying a privacy protection equipment identification model based on homomorphic encryption obtained through the training at a cloud end;
step (2), when equipment identification is needed, encrypting the equipment characteristic data by using a homomorphic encryption algorithm and then sending the encrypted equipment characteristic data to a cloud end;
step (3), after the cloud receives the encrypted characteristic data, directly inputting the encrypted data into a homomorphic encryption equipment identification model of the cloud;
step (4), the cloud transmits the encryption prediction result to the inquiry initiating terminal;
and (5) the initiating query end receives the prediction result returned by the cloud end.
7. The method for using the privacy protection device identification model based on homomorphic encryption according to claim 6,
in the step (1), since the device identification model is homomorphically encrypted, the specific mode and parameter information of model training are not disclosed.
8. The method for using the privacy protection device identification model based on homomorphic encryption according to claim 6,
in the step (2), all data in the transmission process are homomorphic encrypted, and even if the data are intercepted in the transmission process, the data cannot be easily cracked.
9. The method for using the privacy protection device identification model based on homomorphic encryption according to claim 6,
in the step (3), the model can predict and identify the encrypted device characteristic data, and the encrypted device characteristic data does not need to be decrypted and then input into the device identification model, so that the cloud end can not obtain the transmitted data plaintext, and the device identification model at the cloud end can not cause model parameter leakage due to the prediction process;
in the step (4), an available result cannot be obtained even if the result is intercepted in the transmission process, and the privacy of the interaction of the prediction result is further ensured.
10. The method for using the privacy protection device identification model based on homomorphic encryption according to claim 6,
in step (5), the prediction result of the plaintext can be obtained only after being decrypted by the private key corresponding to the initiating inquirer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010666718.XA CN111984960B (en) | 2020-07-13 | Privacy protection equipment identification model design and use method based on homomorphic encryption |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010666718.XA CN111984960B (en) | 2020-07-13 | Privacy protection equipment identification model design and use method based on homomorphic encryption |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111984960A true CN111984960A (en) | 2020-11-24 |
CN111984960B CN111984960B (en) | 2024-05-17 |
Family
ID=
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112734004A (en) * | 2020-12-31 | 2021-04-30 | 国家电网有限公司大数据中心 | Neural network construction method under homomorphic encryption, image processing method and system |
CN112818360A (en) * | 2021-01-18 | 2021-05-18 | 西安交通大学 | Deep neural network encryption reasoning method based on homomorphic encryption technology |
CN112910624A (en) * | 2021-01-14 | 2021-06-04 | 东北大学 | Ciphertext prediction method based on homomorphic encryption |
CN113111726A (en) * | 2021-03-18 | 2021-07-13 | 浙江大学 | Vibration motor equipment fingerprint extraction and identification method based on homologous signals |
CN113128568A (en) * | 2021-03-26 | 2021-07-16 | 广州大学 | Excavator activity identification method, system, device and storage medium |
CN113705825A (en) * | 2021-07-16 | 2021-11-26 | 杭州医康慧联科技股份有限公司 | Data model sharing method suitable for multi-party use |
CN116800906A (en) * | 2023-08-22 | 2023-09-22 | 北京电子科技学院 | Ciphertext convolutional neural network image classification method based on mode component homomorphism |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110543901A (en) * | 2019-08-22 | 2019-12-06 | 阿里巴巴集团控股有限公司 | image recognition method, device and equipment |
CN110598438A (en) * | 2019-07-19 | 2019-12-20 | 福州大学 | In-cloud protection outsourcing data privacy protection system based on deep convolutional neural network |
CN111245610A (en) * | 2020-01-19 | 2020-06-05 | 浙江工商大学 | Data privacy protection deep learning method based on NTRU homomorphic encryption |
US20200204341A1 (en) * | 2017-01-20 | 2020-06-25 | Enveil, Inc. | Secure Machine Learning Analytics Using Homomorphic Encryption |
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200204341A1 (en) * | 2017-01-20 | 2020-06-25 | Enveil, Inc. | Secure Machine Learning Analytics Using Homomorphic Encryption |
CN110598438A (en) * | 2019-07-19 | 2019-12-20 | 福州大学 | In-cloud protection outsourcing data privacy protection system based on deep convolutional neural network |
CN110543901A (en) * | 2019-08-22 | 2019-12-06 | 阿里巴巴集团控股有限公司 | image recognition method, device and equipment |
CN111245610A (en) * | 2020-01-19 | 2020-06-05 | 浙江工商大学 | Data privacy protection deep learning method based on NTRU homomorphic encryption |
Non-Patent Citations (1)
Title |
---|
谢四江;许世聪;章乐;: "基于同态加密的卷积神经网络前向传播方法", 计算机应用与软件, no. 02, 12 February 2020 (2020-02-12) * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112734004A (en) * | 2020-12-31 | 2021-04-30 | 国家电网有限公司大数据中心 | Neural network construction method under homomorphic encryption, image processing method and system |
CN112910624A (en) * | 2021-01-14 | 2021-06-04 | 东北大学 | Ciphertext prediction method based on homomorphic encryption |
CN112910624B (en) * | 2021-01-14 | 2022-05-10 | 东北大学 | Ciphertext prediction method based on homomorphic encryption |
CN112818360A (en) * | 2021-01-18 | 2021-05-18 | 西安交通大学 | Deep neural network encryption reasoning method based on homomorphic encryption technology |
CN112818360B (en) * | 2021-01-18 | 2023-03-28 | 西安交通大学 | Deep neural network encryption reasoning method based on homomorphic encryption technology |
CN113111726A (en) * | 2021-03-18 | 2021-07-13 | 浙江大学 | Vibration motor equipment fingerprint extraction and identification method based on homologous signals |
CN113128568A (en) * | 2021-03-26 | 2021-07-16 | 广州大学 | Excavator activity identification method, system, device and storage medium |
CN113128568B (en) * | 2021-03-26 | 2023-08-11 | 广州大学 | Excavator activity recognition method, system, device and storage medium |
CN113705825A (en) * | 2021-07-16 | 2021-11-26 | 杭州医康慧联科技股份有限公司 | Data model sharing method suitable for multi-party use |
CN116800906A (en) * | 2023-08-22 | 2023-09-22 | 北京电子科技学院 | Ciphertext convolutional neural network image classification method based on mode component homomorphism |
CN116800906B (en) * | 2023-08-22 | 2023-11-07 | 北京电子科技学院 | Ciphertext convolutional neural network image classification method based on mode component homomorphism |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Additively homomorphical encryption based deep neural network for asymmetrically collaborative machine learning | |
Li et al. | Privacy-preserving machine learning with multiple data providers | |
Ma et al. | PDLM: Privacy-preserving deep learning model on cloud with multiple keys | |
CN108259158A (en) | Efficient and secret protection individual layer perceptron learning method under a kind of cloud computing environment | |
US10635824B1 (en) | Methods and apparatus for private set membership using aggregation for reduced communications | |
US20150381349A1 (en) | Privacy-preserving ridge regression using masks | |
WO2021092977A1 (en) | Vertical federated learning optimization method, appartus, device and storage medium | |
CN110059501B (en) | Safe outsourcing machine learning method based on differential privacy | |
CN114936650A (en) | Method and device for jointly training business model based on privacy protection | |
US20220374544A1 (en) | Secure aggregation of information using federated learning | |
Nassar et al. | Securing aggregate queries for DNA databases | |
CN116032667B (en) | Online trace query method, system and related equipment supporting efficient update | |
Abdul Hussien et al. | A secure environment using a new lightweight AES encryption algorithm for e-commerce websites | |
Chen et al. | Securenets: Secure inference of deep neural networks on an untrusted cloud | |
Wei et al. | Privacy-preserving two-parties logistic regression on vertically partitioned data using asynchronous gradient sharing | |
Selvarajan et al. | A quantum trust and consultative transaction-based blockchain cybersecurity model for healthcare systems | |
Wang et al. | Privacy-preserving analytics on decentralized social graphs: The case of eigendecomposition | |
Malik et al. | A homomorphic approach for security and privacy preservation of Smart Airports | |
Wang et al. | Protecting data privacy in federated learning combining differential privacy and weak encryption | |
Xiong et al. | Decentralized privacy-preserving truth discovery for crowd sensing | |
Baig et al. | Privacy-preserving continuous authentication using behavioral biometrics | |
Takeshita et al. | SLAP: simple lattice-based private stream aggregation protocol | |
Belej et al. | The need to use a hash function to build a crypto algorithm for blockchain | |
CN111984960A (en) | Privacy protection equipment identification model design and use method based on homomorphic encryption | |
CN111984960B (en) | Privacy protection equipment identification model design and use method based on homomorphic encryption |
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 |