CN109347633B - Fuzzy key communication system based on deep learning and countermeasure network system - Google Patents
Fuzzy key communication system based on deep learning and countermeasure network system Download PDFInfo
- Publication number
- CN109347633B CN109347633B CN201811264598.XA CN201811264598A CN109347633B CN 109347633 B CN109347633 B CN 109347633B CN 201811264598 A CN201811264598 A CN 201811264598A CN 109347633 B CN109347633 B CN 109347633B
- Authority
- CN
- China
- Prior art keywords
- neural network
- model
- network model
- fully
- 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
- 230000006854 communication Effects 0.000 title claims abstract description 121
- 238000004891 communication Methods 0.000 title claims abstract description 119
- 238000013135 deep learning Methods 0.000 title claims abstract description 43
- 238000003062 neural network model Methods 0.000 claims abstract description 91
- 230000006870 function Effects 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 11
- 238000005336 cracking Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 10
- 238000000034 method Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000008485 antagonism Effects 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/08—Key distribution or management, e.g. generation, sharing or updating, of cryptographic keys or passwords
- H04L9/0861—Generation of secret information including derivation or calculation of cryptographic keys or passwords
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/002—Countermeasures against attacks on cryptographic mechanisms
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Machine Translation (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a fuzzy key communication system and an confrontation network system based on deep learning, which comprises a communication party Alice and a communication party Bob; the correspondent Alice includes an encryption model; the encryption model is obtained by deep learning of a first neural network model; the communication party Bob comprises a decryption model, the decryption model is obtained by deep learning of the second neural network model and is used for inputting a secret key and a ciphertext and decrypting the input ciphertext according to the fuzzy secret key to obtain plaintext information; the second neural network model comprises a second full connection layer, a third full connection layer and a multilayer convolution layer from input to output; the key entered by the decryption model of the correspondent Bob is a fuzzy key. The countermeasure network system is obtained by adding the communication system into the Eve model and then deeply learning. The system can enhance the communication performance of the network in the fuzzy key communication environment and realize accurate communication in the fuzzy key environment.
Description
Technical Field
The invention belongs to the technical field of computer technology and information security communication, and particularly relates to a fuzzy key communication system and an antagonistic network system based on deep learning.
Background
With the development of deep learning techniques, attempts have also been made to use in various fields. The general encryption algorithm is designed by people, but the encryption algorithm can be made by utilizing a generation countermeasure network in deep learning. The Google Brain countermeasure network is countered by two neural networks, wherein one neural network Eve is responsible for breaking a ciphertext of communication, and the other neural network consists of two parts, namely Alice and Bob; encrypted communication is carried out between Alice and Bob, and Eve is responsible for cracking communication contents; the two neural networks develop network countermeasure between the decryption ciphertext and the protection ciphertext; the neural network enhances respective abilities through continuous antagonism, and finally, networks of Alice and Bob in the generated antagonism network are extracted to serve as an encryption neural network algorithm model. Encrypted communication can be performed by using the trained model. In the prior art, Alice and Bob share a secret key K, Bob decrypts a ciphertext C through the secret key K, and Eve decrypts the ciphertext C by himself. Alice and Bob are the same party, neither learn any encryption knowledge, but simply depend on the consensus between the two parties (a shared secret key K) to cut and design an encryption method, for example, a researcher gives a 16-byte original text P, and Alice encrypts P according to the method designed by himself and outputs a section of ciphertext C. And then, the Bob designs a decryption method by himself by matching the secret key K with the ciphertext C sent by Alice, and decrypts an answer. And Eve plays the role of a listener of an enemy, can eavesdrop on the ciphertext C, and also designs a decryption method by oneself on the premise of no secret key K to guess what the original text is. In the prior art, Alice and Bob generally share a secret key K, and when the secret key K is leaked, a ciphertext of communication between Alice and Bob is easy to break; but if the key given to Bob is different from the key K, i.e. Alice and Bob have similar or different keys, Bob will have difficulty in accurately decrypting the plaintext from the ciphertext. In addition, typically, when Alice and Bob want to communicate securely, Eve wants to eavesdrop on their communication. Thus, the desired security attribute is privacy, and the adversary is a "passive attacker". While Eve can intercept communications, it cannot initiate a session, inject a message, or modify a message. However, this mode is not absolutely secure, and Eve can decrypt the encryption key K of Alice and Bob.
At present, an information protection encryption algorithm can be learned through a countermeasure network of Google Brain, the model can also realize encrypted communication when a small amount of loss or difference occurs in a secret key acquired by Bob, but when a certain amount of information loss exists in the communication process, Bob cannot guarantee that a plaintext can be restored.
Disclosure of Invention
The first purpose of the present invention is to overcome the disadvantages and shortcomings of the prior art, and to provide a fuzzy key communication system based on deep learning, which can enhance the communication performance of the network in the fuzzy key communication environment, and achieve accurate communication in the fuzzy key environment.
A second object of the present invention is to provide a countermeasure network system.
The first purpose of the invention is realized by the following technical scheme: a fuzzy secret key communication system based on deep learning comprises a communication party Alice and a communication party Bob;
the correspondent Alice includes an encryption model; the encryption model is obtained by deep learning of the first neural network model and is used for inputting a plaintext and a secret key and forming a ciphertext by adopting the inputted plaintext and the secret key; the first neural network model comprises a first fully-connected layer and a multi-layer convolutional layer from input to output;
the correspondent Bob includes a decryption model; the decryption model is obtained by deep learning of the second neural network model and is used for inputting a secret key and a ciphertext and decrypting the input ciphertext according to the input secret key to obtain plaintext information; the second neural network model comprises a second fully connected layer and a plurality of convolutional layers from input to output;
and the key input by the decryption model of the correspondent Bob is a fuzzy key, and the second neural network model of the correspondent Bob further comprises a third fully-connected layer adjacent to the second fully-connected layer.
Preferably, the activation function of one of the second fully-connected layer and the third fully-connected layer of the second neural network model, which is close to the input end of the second neural network model, is set as a tanh function.
Preferably, the second fully-connected layer and the third fully-connected layer of the second neural network model and the first fully-connected layer of the first neural network model are all added to the batch normalization process.
Further, the second and third fully-connected layers of the second neural network model and the first fully-connected layer of the first neural network model are batch normalized as follows: normalizing the input to make the average value of the input be 0 and the variance be 1, and then outputting the input to the next layer through the activation function.
Preferably, the first neural network model and the second neural network model are optimized by using AdamaOptimizer.
Preferably, the system is obtained by performing multi-round training on the first neural network model and the second neural network model together; in each round of training, the first neural network model training and the second neural network model are iteratively trained for M1 times, wherein M1 is a constant value.
The second purpose of the invention is realized by the following technical scheme: a countermeasure network system, which is obtained by adding the fuzzy key communication system based on deep learning according to the first object of the claim into the Eve model and then performing deep learning;
the Eve model is obtained by deep learning of a third neural network model and is used for cracking a ciphertext output to a communication party Bob by a communication party Alice; the third neural network model comprises two fully-connected layers and a plurality of convolutional layers from input to output, wherein the two fully-connected layers are adjacent.
Preferably, batch normalization processing is added to both fully connected layers of the third neural network model.
Further, the two fully-connected layers of the third neural network model are respectively subjected to batch normalization as follows: normalizing the input to make the average value of the input be 0 and the variance be 1, and then outputting the input to the next layer through the activation function.
Preferably, the system is obtained by carrying out multi-round training on an encryption model of a communication party Alice, a decryption model of a communication party Bob and an Eve model; in each round of training, the encryption model of the communication party Alice and the decryption model of Bob are trained for M1 times, the Eve model is trained for M2 times in an iterative manner, and M1 and M2 are both constant values.
Compared with the prior art, the invention has the following advantages and effects:
(1) the fuzzy secret key communication system based on deep learning comprises a communication party Alice and a communication party Bob; the encryption model of the communication party Alice is obtained by deep learning of the first neural network model and is used for inputting a plaintext and a secret key and forming a ciphertext by adopting the inputted plaintext and the secret key; the decryption model of the communication party Bob is obtained after the second neural network model is deeply learned and used for inputting the fuzzy key and the ciphertext and decrypting the input ciphertext according to the fuzzy key to obtain plaintext information; adding a full connection layer aiming at a second neural network model of a communication party Bob to enable the full connection layer to comprise two continuous full connection layers; in the invention, the second neural network model with a layer of full connection layer is added to improve the analysis depth of the decryption model to the ciphertext and the fuzzy key, so that the capacity of the decryption model for analyzing the fuzzy key is greatly improved, and when the ciphertext and the key are 16-bit data, the ciphertext can be analyzed by using the fuzzy key with 1 or 2 bit errors basically, therefore, the communication system can enhance the communication performance of the network in the fuzzy key communication environment and realize accurate communication in the fuzzy key environment.
(2) In the fuzzy key communication system based on deep learning, aiming at the second full connection layer and the third full connection layer of the second neural network model, the activation function of one layer close to the input end of the second neural network model is set as a tanh function. The derivative range of the tanh function is 0 to 1, so that the information can be effectively prevented from being greatly reduced, the weight parameter can be quickly adjusted by the second neural network model, the original text can be accurately decrypted, and the error loss value is very close to 0. Therefore, the performance of the decryption model is also improved greatly by the operation; when the ciphertext and the key are 16-bit data, the fuzzy key with 1-4 bit errors can be used for analyzing the ciphertext, namely error-free fuzzy key communication can be realized when the key difference is 25% or less, the decryption accuracy is very high, and therefore the communication performance of the network under the fuzzy key communication environment is further enhanced
(3) In the fuzzy key communication system based on deep learning, the fully-connected layers of the first neural network model and the second neural network model are added in batch normalization processing. The operation greatly improves the stability of the first neural network model and the second neural network model, and reduces the probability that the model cannot find the lowest loss point due to different initialization conditions; in the training process, the learning rate can be improved, the training of the model cannot be influenced, and the batch size can be further reduced, so that the speed of searching the optimal solution for the whole model of the deep learning-based fuzzy key communication system is greatly improved by the processing.
(4) The countermeasure network system is obtained by adding an Eve model into the deep learning-based fuzzy key communication system of the first purpose of the invention and then performing deep learning; the Eve model is obtained by deep learning of the third neural network model and is used for cracking a ciphertext output to the communication party Bob by the communication party Alice; the third neural network model comprises two fully-connected layers and a plurality of convolutional layers from input to output, wherein the two fully-connected layers are adjacent. In the invention, the countermeasure network system can train Alice and Bob to successfully communicate, and Eve can not accurately break the communication information between the Alice and the Bob, so that the Eve recognition system is defeated; the anti-network system ensures that the communication between Alice and Bob can be basically intact, and simultaneously effectively prevents Eve from cracking the communication content, and basically ensures that the error rate of Eve is about half and the communication loss is nearly 0.
(5) In the countermeasure network system, the two fully-connected layers of the Eve model are added with batch normalization processing, wherein the difference results of the models caused by different initialization states can be greatly reduced by adding batch normalization, and the stability of the trained models is further improved.
Drawings
Fig. 1 is a block diagram of a communication system architecture of the present invention.
Fig. 2a is a diagram of a neural network model in a correspondent Bob of a prior art communication system.
Fig. 2b is a diagram of a second neural network model in the correspondent Bob of the communication system in embodiment 1 of the present invention.
Fig. 3a is a test result of decryption using a fuzzy key in a communication system in the prior art.
Fig. 3b shows the result of the decryption test using the fuzzy key in the communication system in embodiment 1 of the present invention.
Fig. 3c shows the decryption test result of the communication system using the fuzzy key in embodiment 2 of the present invention.
Fig. 3d shows the result of the decryption test using the fuzzy key in the communication system in embodiment 3 of the present invention.
Fig. 4 is a diagram of a second neural network model in the correspondent Bob of the communication system in embodiment 2 of the present invention.
Fig. 5 is a graph of the derivative of the Sigmoid function and the tanh function.
Fig. 6 is a block diagram of the countermeasure network system of the present invention.
FIG. 7 is a third neural network model diagram of the Eve model in the countermeasure network system of the present invention.
FIG. 8 shows the test result of the present invention against the Eve model in the network system to break the plaintext.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
Example 1
The embodiment 1 discloses a fuzzy key communication system based on deep learning, which includes a communication party Alice and a communication party Bob as shown in fig. 1.
The correspondent Alice includes an encryption model; the encryption model is obtained by deep learning of the first neural network model and is used for inputting a plaintext P and a key K, and a ciphertext C is formed by adopting the inputted plaintext P and the key K; the first neural network model comprises a first fully-connected layer and a multi-layer convolutional layer from input to output;
the correspondent Bob includes a decryption model; the decryption model is obtained by deep learning of the second neural network model and is used for inputting a key and a ciphertext and decrypting the input ciphertext C according to the input key CK to obtain plaintext information PBob(ii) a As shown in fig. 2a, the second neural network model comprises, from input to output, a first fully-connected layer and a plurality of convolutional layers, wherein the plurality of convolutional layers are convolutional layer 1, convolutional layer 2, convolutional layer 3, and convolutional layer 4, respectively.
In this embodiment, the key input by the decryption model of the correspondent Bob is a fuzzy key, and as shown in fig. 2b, the second neural network model of the correspondent Bob further includes a third fully-connected layer adjacent to the second fully-connected layer, that is, the second neural network model of the correspondent Bob in this embodiment has two fully-connected layers, and compared with the neural network of the correspondent Bob in the prior art, the second neural network model of the correspondent Bob in this embodiment has one more fully-connected layer.
In this embodiment 1, both the first neural network model and the second neural network model are optimized by using an adammoptimizer optimizer.
In the present embodiment, the key K input to the first neural network model, the plaintext P, and the fuzzy key CK input to the second neural network model are all N in length, and K, CK and P are both arrays of-1 and 1 with a length N. Where CK is derived from inverting a random n number of K, e.g., inverting 1 to-1, or inverting-1 to 1. In this embodiment, N may be 16.
In the embodiment, the communication system is obtained by performing multi-round training on a first neural network model and a second neural network model together, and the total number of rounds of training can be selected to be 10; wherein in each round of training, the first neural network model and the second neural network model are iteratively trained for M1 times. In this embodiment, M1 is a constant value, which may be 2000.
When n is selected to be 1 or 2, namely one or two bits of the fuzzy key CK acquired by the decryption model are different from one or two bits of the key K, namely one or two bits of the fuzzy key CK are lost, a decryption error test is carried out on a communication system of which the neural network of the communication party Bob only has one layer of full connection layer in the prior art, and the loss condition of the decrypted ciphertext of the communication system is calculated through the following formula:
loss L of the correspondent Bob compared to the correspondent AliceBobComprises the following steps: d (P, P)Bob). The test result obtained by calculation according to the above formula is shown in fig. 3 a. As can be seen from FIG. 3a, when n is 1 or 2, as in the prior artThe decryption error rate of the communication system is always maintained at about 0.5, and the communication party Bob basically cannot decrypt the plaintext.
When n is selected to be 1, 2, or 3, the communication system obtained in this embodiment is tested by using the above loss calculation formula, and the test result is shown in fig. 3 b. As can be seen from fig. 3b, when n is 1 or 2, the decryption error rate of the communication system of the present embodiment deviates below 0.01 around 0, and the decryption error rate is significantly reduced compared to fig. 3 a. The reason is that the second neural network model with one layer of full connection layer is added, so that the analysis depth of the decryption model on the ciphertext and the fuzzy key can be improved, the capacity of the decryption model for analyzing the fuzzy key is greatly improved, and when the ciphertext and the key are 16-bit data, the ciphertext can be analyzed by using 1 or 2 digital wrong fuzzy keys basically, so that the system can enhance the communication performance of the network in the fuzzy key communication environment, and accurate communication can be realized in the fuzzy key environment.
Example 2
The present embodiment discloses a fuzzy key communication system based on deep learning, and as shown in fig. 4, the present embodiment only differs from embodiment 1 in that the activation function of one of the second fully-connected layer and the third fully-connected layer of the second neural network model, which is close to the input end of the second neural network model, is set as a tanh function.
In embodiment 1, after a full connection layer is added to the second neural network of the communication party Bob, the decryption capability of the entire communication system is improved to some extent, but as shown in fig. 3b, error-free decryption cannot be implemented on the keys with differences of 3 bits or more (n is greater than or equal to 3). After the analysis of the system, it is found that when a layer of fully connected layer close to the input end in the second neural network model uses a Sigmoid function, the weight update of the model is unfavorable, because as shown in fig. 5, the derivative of the Sigmoid function is less than 0.25, when the weight is updated by back propagation, the information amount in the neural network is reduced by 75%, and the influence on the system is increased after passing through the Sigmoid functions of the two layers of fully connected layers.
In this embodiment, the derivative range of the tanh function used by the layer of fully-connected layer of the second neural network model close to the input end is 0 to 1, as shown in fig. 5, it can be effectively ensured that information is not greatly reduced, so that the weight parameter can be quickly adjusted by the second neural network model, the original text can be accurately decrypted, and the error loss value is very close to 0. Therefore, the performance of the decryption model is also improved by the communication system of the embodiment. When n is 1 to 4, the communication system of the present embodiment is tested by the loss calculation formula shown in embodiment 1, and the test result is shown in fig. 3 d. As can be seen from fig. 3c, the communication system of this embodiment can use the fuzzy key with 1 to 4 digital errors to analyze the ciphertext, that is, when the key difference is 25% or less, error-free fuzzy key communication can be realized, and the decryption accuracy is also very high. Therefore, the system of the embodiment further enhances the communication performance of the network in the fuzzy key communication environment.
Example 3
The present embodiment discloses a fuzzy key communication system based on deep learning, which performs the following processing on the basis of embodiment 1 or 2, and adds batch normalization processing to the second fully-connected layer and the third fully-connected layer of the second neural network model and the first fully-connected layer of the first neural network model, where the batch normalization processing is as follows: normalizing the input to make the average value of the input be 0 and the variance be 1, and then outputting the input to the next layer through the activation function.
After batch normalization processing, the stability of the communication system is greatly improved, and the probability that the model cannot find the lowest loss point due to different initialization conditions is reduced. In the training process, the learning rate can be improved, the training of the model can not be influenced, the batch size can be further reduced, and the speed of searching the optimal solution by the whole model is greatly improved.
When n is 3, the communication system obtained on the basis of embodiment 1 in this embodiment is tested by the loss calculation formula shown in embodiment 1, and the test result is shown in fig. 3 d. As can be seen from fig. 3d, although the initialization state at the time of starting is not optimal, the loss value can be rapidly and gradually reduced at the later stage, and finally a better decryption effect is achieved. In addition, in the previous embodiment 1, since the two fully-connected layers both use Sigmoid activation functions, the problem that the difference of 3-bit numerical values of the key cannot be realized well is solved, the performance is improved after batch normalization in this embodiment, and 3-bit fuzzy key communication can be realized.
Example 4
The present embodiment discloses a countermeasure network system, as shown in fig. 6, which is obtained by adding an Eve model to the fuzzy key communication system based on deep learning according to any one of embodiments 1 to 3 and then performing deep learning.
In this embodiment, the Eve model is obtained by deep learning of the third neural network model, and is used for decrypting the ciphertext output to the communication party Bob by the communication party Alice to obtain the plaintext information PEveWhen the plaintext information P input by Alice is 16 bits, the Eve model carries out cracking to obtain the plaintext information PEveAlso 16 bits. In this embodiment, as shown in fig. 7, the third neural network model includes, from input to output, two fully-connected layers and a plurality of convolutional layers, where the plurality of convolutional layers are convolutional layer 1, convolutional layer 2, convolutional layer 3, and convolutional layer 4; wherein two layers of full connection layers are adjacent, and the first layer of full connection layer can expand the length of input data N to 2N.
In this embodiment, the third neural network model of the Eve model is optimized by using an adammoptimizer optimizer, as well as the first neural network model and the second neural network model in the fuzzy key communication system based on deep learning.
In this embodiment, batch normalization processing is added to both fully connected layers of the third neural network model, and the batch normalization processing procedure is as follows: the input is normalized to have a mean value of 0 and a variance of 1, and then output to the next layer via the activation function, where the batch size is 500.
The countermeasure network system in this embodiment is obtained by performing multiple rounds of training on the encryption model of the communication party Alice, the decryption model of the communication party Bob and the Eve model, the number of training rounds may be 60 rounds, and the learning rate is 0.005; in each round of training, the encryption model of the correspondent Alice and the decryption model of Bob are trained for M1 times, the Eve model is trained for M2 times in an iterative manner, and M1 and M2 are both constant values, in this embodiment, M1 may be 10, and M2 may be 2000.
In the countermeasure network system, the communication party Bob and the communication party Alice continuously update and optimize the encryption and decryption model of the communication party Bob and the communication party Alice in the communication process, so that the ciphertext cannot be decrypted by Eve, and finally the effect of preventing Eve is achieved. After the countermeasure network system is trained, Alice and Bob can successfully communicate, and Eve steals and identifies the signal communication between the Alice and the Bob, and defeats the Eve identification system.
In the embodiment, the Eve model is cracked to the plaintext information PEveThe loss can then be calculated by the following equation:
wherein the loss L of the Eve model is capturedEveComprises the following steps: d (P, P)Eve)
Obtaining the loss L of Bob and AliceBobComprises the following steps:
d(P,PBob)+(1-LEve)2；
when n is 3, the loss calculated by the above formula is tested for the communication system acquired in this embodiment, and the test result is shown in fig. 8. As can be seen from fig. 8, the countermeasure network system of the embodiment ensures that communication between Alice and Bob can be basically intact, and at the same time, Eve can be effectively prevented from cracking communication content, so that error rate of Eve can be basically ensured to be about half, and communication loss is approximately 0. Therefore, in the embodiment, while certain security of encryption can be ensured overall, both communication parties can be ensured to accurately receive and send signals.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (10)
1. A fuzzy secret key communication system based on deep learning comprises a communication party Alice and a communication party Bob;
the correspondent Alice includes an encryption model; the encryption model is obtained by deep learning of the first neural network model and is used for inputting a plaintext and a secret key and forming a ciphertext by adopting the inputted plaintext and the secret key; the first neural network model comprises a first fully-connected layer and 4 convolutional layers from input to output;
the correspondent Bob includes a decryption model; the decryption model is obtained by deep learning of the second neural network model and is used for inputting a secret key and a ciphertext and decrypting the input ciphertext according to the input secret key to obtain plaintext information; the second neural network model comprises a second fully-connected layer and 4 convolutional layers from input to output;
the key input by the decryption model of the correspondent Bob is a fuzzy key, and the second neural network model of the correspondent Bob further comprises a third fully-connected layer adjacent to the second fully-connected layer;
the fuzzy key is obtained by turning over the random n number of K; k is the key input to the first neural network model.
2. The fuzzy key communication system based on deep learning of claim 1, wherein the activation function of one of the second fully-connected layer and the third fully-connected layer of the second neural network model near the input end of the second neural network model is set as a tanh function.
3. The fuzzy key communication system based on deep learning of claim 1 or 2, wherein the second fully-connected layer and the third fully-connected layer of the second neural network model and the first fully-connected layer of the first neural network model each incorporate batch normalization processing.
4. The fuzzy key communication system based on deep learning of claim 3, wherein the second and third fully-connected layers of the second neural network model and the first fully-connected layer of the first neural network model are batch normalized as follows: normalizing the input to make the average value of the input be 0 and the variance be 1, and then outputting the input to the next layer through the activation function.
5. The deep learning based fuzzy key communication system of claim 1, wherein said first and second neural network models are optimized using an adammoptimizer.
6. The fuzzy key communication system based on deep learning of claim 1, wherein the system is obtained by performing multiple rounds of training on a first neural network model and a second neural network model; in each round of training, the first neural network model and the second neural network model are iteratively trained for M1 times, wherein M1 is a constant value.
7. A countermeasure network system, characterized in that the system is obtained by adding the fuzzy key communication system based on deep learning of any claim 1 to 6 into an Eve model and then performing deep learning;
the Eve model is obtained by deep learning of a third neural network model and is used for cracking a ciphertext output to a communication party Bob by a communication party Alice; the third neural network model comprises two fully-connected layers and 4 convolutional layers from input to output, wherein the two fully-connected layers are adjacent.
8. The countermeasure network system of claim 7, wherein batch normalization processing is added to both fully connected layers of the third neural network model.
9. The countermeasure network system of claim 8, wherein the two fully-connected layers of the third neural network model are batch normalized as follows: normalizing the input to make the average value of the input be 0 and the variance be 1, and then outputting the input to the next layer through the activation function.
10. The countermeasure network system of claim 7, wherein the system is obtained by performing multiple rounds of training together with the encryption model of the correspondent Alice, the decryption model of the correspondent Bob, and the Eve model; in each round of training, the encryption model of the communication party Alice and the decryption model of Bob are trained for M1 times, the Eve model is trained for M2 times in an iterative manner, and M1 and M2 are both constant values.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811264598.XA CN109347633B (en) | 2018-10-29 | 2018-10-29 | Fuzzy key communication system based on deep learning and countermeasure network system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811264598.XA CN109347633B (en) | 2018-10-29 | 2018-10-29 | Fuzzy key communication system based on deep learning and countermeasure network system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109347633A CN109347633A (en) | 2019-02-15 |
| CN109347633B true CN109347633B (en) | 2020-07-03 |
Family
ID=65310776
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811264598.XA Active CN109347633B (en) | 2018-10-29 | 2018-10-29 | Fuzzy key communication system based on deep learning and countermeasure network system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109347633B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109995520A (en) * | 2019-03-06 | 2019-07-09 | 西北大学 | Cipher key transmission methods, image processing platform based on depth convolutional neural networks |
| CN110035063A (en) * | 2019-03-08 | 2019-07-19 | 佛山市云米电器科技有限公司 | A kind of intelligent appliance data ciphering method neural network based |
| CN110674941B (en) * | 2019-09-25 | 2023-04-18 | 南开大学 | Data encryption transmission method and system based on neural network |
| CN111355587B (en) * | 2020-02-18 | 2021-02-19 | 华南农业大学 | Authenticated encryption secure communication system and method based on countermeasure network |
| CN111382455B (en) * | 2020-03-18 | 2023-05-26 | 北京丁牛科技有限公司 | File protection method and device |
| CN111539012B (en) * | 2020-03-19 | 2021-07-20 | 重庆特斯联智慧科技股份有限公司 | Privacy data distribution storage system and method of edge framework |
| CN112751671B (en) * | 2020-12-30 | 2022-07-05 | 华南农业大学 | Novel key exchange method based on tree parity machine |
| CN113014570B (en) * | 2021-02-22 | 2022-08-02 | 西安理工大学 | Communication data encryption and decryption method based on convolutional neural network |
| CN115884172B (en) * | 2022-12-01 | 2023-07-04 | 同济大学 | Information encryption and decryption method based on countermeasure sample technology |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107248914A (en) * | 2017-08-14 | 2017-10-13 | 四川长虹电器股份有限公司 | Novel symmetrical encryption system and encryption method in a kind of iOS device |
| AU2017101168A4 (en) * | 2017-08-25 | 2017-11-02 | Guan, Zijin MR | A method of testing the robustness of cryptographickey |
| CN107659398A (en) * | 2017-09-28 | 2018-02-02 | 四川长虹电器股份有限公司 | Suitable for Android symmetric encryption method |
-
2018
- 2018-10-29 CN CN201811264598.XA patent/CN109347633B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107248914A (en) * | 2017-08-14 | 2017-10-13 | 四川长虹电器股份有限公司 | Novel symmetrical encryption system and encryption method in a kind of iOS device |
| AU2017101168A4 (en) * | 2017-08-25 | 2017-11-02 | Guan, Zijin MR | A method of testing the robustness of cryptographickey |
| CN107659398A (en) * | 2017-09-28 | 2018-02-02 | 四川长虹电器股份有限公司 | Suitable for Android symmetric encryption method |
Non-Patent Citations (4)
| Title |
|---|
| Learning to Protect Communications with Adversarial Neural Cryptography;Martín Abadi;《https://arxiv.org/abs/1610.06918》;20161021;1-15 * |
| 利用生成对抗网络保护通信（learning to Protect Communications with Adversarial Neural Cryptography）附代码;陶将;《https://blog.csdn.net/weixin_42111770/article/details/81702831》;20180815;1-5 * |
| 深度学习原理及应用综述;付文博;《计算机科学》;20180615;第45卷(第6A期);11-15，40 * |
| 谷歌发布2017年度AI报告;Jeff Dean;《机器人产业》;20180325(第02期);12-25 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109347633A (en) | 2019-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109347633B (en) | Fuzzy key communication system based on deep learning and countermeasure network system | |
| Pagnin et al. | Privacy‐Preserving Biometric Authentication: Challenges and Directions | |
| US11569978B2 (en) | Encrypting and decrypting information | |
| CA2883313C (en) | Multi-factor authentication using quantum communication | |
| US8837727B2 (en) | Method for privacy preserving hashing of signals with binary embeddings | |
| CN110138802B (en) | User characteristic information acquisition method, device, block chain node, network and storage medium | |
| Friha et al. | 2DF-IDS: Decentralized and differentially private federated learning-based intrusion detection system for industrial IoT | |
| Khazbak et al. | MLGuard: Mitigating poisoning attacks in privacy preserving distributed collaborative learning | |
| Liang et al. | Efficient and privacy-preserving decision tree classification for health monitoring systems | |
| Liu et al. | An efficient biometric identification in cloud computing with enhanced privacy security | |
| Torres et al. | Effectiveness of fully homomorphic encryption to preserve the privacy of biometric data | |
| CN114065169B (en) | Privacy protection biometric authentication method and device and electronic equipment | |
| Aseeri et al. | A subspace pre-learning approach to fast high-accuracy machine learning of large xor pufs with component-differential challenges | |
| Erkin et al. | Privacy-preserving user clustering in a social network | |
| Nozaki et al. | Secret sharing schemes based secure authentication for physical unclonable function | |
| CN113114456B (en) | Multi-user quantum privacy query method with authentication | |
| Win et al. | Protecting private data using improved honey encryption and honeywords generation algorithm | |
| He et al. | Differentially private set intersection for asymmetrical id alignment | |
| Zhong et al. | MPC-based privacy-preserving serverless federated learning | |
| CN111355587A (en) | Authenticated encryption secure communication system and method based on countermeasure network | |
| Hu et al. | MASKCRYPT: Federated Learning with Selective Homomorphic Encryption | |
| Meraouche et al. | Learning asymmetric encryption using adversarial neural networks | |
| US20240195618A1 (en) | Method and system for a lattice-based homomorphic proxy re-encryption scheme | |
| Jia et al. | Encrypted Packet Inspection Based on Oblivious Transfer | |
| Sarkar et al. | Cryptanalysis of key exchange method in wireless communication |
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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231215 Address after: Room 204, Building 83, Industrial Zone, No. 83-87 Jiangxia East 1st Road, Huangshi Street, Baiyun District, Guangzhou City, Guangdong Province, 510000 (self declared) Patentee after: Guangdong Fuhui Electronic Technology Co.,Ltd. Address before: 510642 No. five, 483 mountain road, Guangzhou, Guangdong, Tianhe District Patentee before: SOUTH CHINA AGRICULTURAL University |
|
| TR01 | Transfer of patent right |