CN117892843A - Machine learning data forgetting method based on game theory and cryptography - Google Patents
Machine learning data forgetting method based on game theory and cryptography Download PDFInfo
- Publication number
- CN117892843A CN117892843A CN202410303506.3A CN202410303506A CN117892843A CN 117892843 A CN117892843 A CN 117892843A CN 202410303506 A CN202410303506 A CN 202410303506A CN 117892843 A CN117892843 A CN 117892843A
- Authority
- CN
- China
- Prior art keywords
- data
- user
- model
- game
- machine learning
- 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
- 238000010801 machine learning Methods 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000006870 function Effects 0.000 claims abstract description 91
- 238000012549 training Methods 0.000 claims abstract description 45
- 239000011159 matrix material Substances 0.000 claims abstract description 25
- 230000008901 benefit Effects 0.000 claims description 15
- 238000012417 linear regression Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000003064 k means clustering Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 4
- 230000001419 dependent effect Effects 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 abstract description 5
- 230000005540 biological transmission Effects 0.000 abstract description 2
- 230000000875 corresponding effect Effects 0.000 description 21
- 230000000694 effects Effects 0.000 description 4
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of machine learning, in particular to a machine learning data forgetting method based on game theory and cryptography, which comprises the following steps: s1, acquiring a training set and constructing a learning system; s2, initializing iteration of a learning system; s3, constructing a necessary function, and constructing and judging a game matrix according to the necessary function; s4, inputting all data, and performing model training by using a SISA frame; s5, users game with the machine, and self game space is allocated; s6, encrypting the game space by using PKI to finish the forgetting protection of the information category; s7, extracting machine forgetting learning results. The game space of the user is encrypted through the machine learning technology of game theory and cryptography, so that confidentiality and integrity of data in the storage and transmission processes are ensured. The encryption measures can effectively prevent unauthorized access and data leakage, thereby protecting the privacy of users and improving the data security, model performance and expandability and flexibility of the method.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a machine learning data forgetting method based on game theory and cryptography.
Background
With the rapid development of artificial intelligence technology, machine learning is increasingly widely applied in various fields, however, a great amount of data is generally required to construct accurate prediction and decision models by traditional machine learning algorithms, which brings about an important problem that in some cases, user data may contain sensitive information, such as personal privacy or business confidentiality along with release of privacy protection regulations and improvement of people's attention to privacy, users have higher requirements on control rights and forgetting rights of own data, users want to be able to autonomously decide that own data can be completely deleted or forgotten when related platforms or software is no longer used, but only deleting data in original training data sets is insufficient, because an attacker can still acquire user information from a trained model, and retraining a model from the head is a perfect information deletion method, but is often a process with high calculation cost. In view of the above, the present invention proposes a machine learning data forgetting method based on game theory and cryptography to solve the above-mentioned problems.
Disclosure of Invention
The invention mainly aims to provide a machine learning data forgetting method based on game theory and cryptography so as to solve the problems in the related art.
To achieve the above object, according to one aspect of the present invention, there is provided a machine learning data forgetting method based on game theory and cryptography, comprising the steps of:
s1, acquiring a training set and constructing a learning system;
S2, initializing iteration of a learning system;
S3, constructing a necessary function, and constructing and judging a game matrix according to the necessary function;
s4, inputting all data, and performing model training by using a SISA frame;
S5, users game with the machine, and self game space is allocated;
s6, encrypting the game space by using PKI to finish the forgetting protection of the information category;
S7, extracting machine forgetting learning results.
Further, in the step S1, small-sized typical data is obtained from complete data provided by a plurality of users And constructing a data classifier including K-means clustering (K-means clustering) and a pre-trained small machine learning model G, wherein/> ,/> Representing the nth data provided from the user.
Further, the step of learning the system initialization iteration in S2 is as follows:
Initializing t times, wherein the machine learning model obtained by the t times is as follows According to/> defining a model cost function/>, by linear regression wherein y is the data obtained by training the machine learning model,/> Representing y corresponds to a data set .
Further, three functions are defined according to the machine learning iteration result in S3: regarding the cost function of the model, regarding the weight function of the data and regarding the benefit function of the user, fitting the machine learning result and the corresponding user by using linear regression to construct the cost function, and obtaining the weight function of the data through the cost function.
Further, the calculation formula of the cost function is as follows:
。
Wherein the method comprises the steps of For/> Index of/> The cost function is set by the result as regression coefficient data set/>, as dependent variable And obtaining a basic linear regression equation for the independent variable fitting.
The cost function has n subfunctions, and the calculation formula is as follows:
；
Wherein the method comprises the steps of Namely, data category/> weight function for the present cost function.
Further, the specific steps of constructing and judging the game matrix according to the necessary function in S3 are as follows:
S3.1, constructing a judgment game matrix ;
S3.2, reusing the cost function and the benefit function The individual data sets are filled into the matrix/> ;
s3.3. definition matrix the element in (a) is/> Calculating/>, based on the cost function and the benefit function of the construction Is the value of (1): /(I) ;
And S3.4, finding the optimal solution of the forgetting model according to the Nash equilibrium odd rule.
Further, the specific steps of inputting all the data in S4 and performing model training by using the SISA framework are as follows:
S4.1: dividing an original data set containing all user information into a plurality of disjoint sliced data sets Di and putting the disjoint sliced data sets Di into a database;
S4.2: the SISA model frame algorithm is utilized to put the information which is to be forgotten finally into any one In (a) and (b);
s4.3: model for each piece of data from when there was initially no forgotten data Beginning training, first use/> updating the model as a training set is/> second use/> Update/>, as training set When the training set is updated, the model updated by the original training set is saved, and the model set/>, which is obtained by updating the model through n times of iteration, is updated continuously and incrementally wherein/> We are the final trained model.
Further, the specific operation steps in S5 are as follows:
S5.1: when the user does not desire to save the data to the model, the access system provides an interface for inquiring the user about the option of deleting the data, and the user selects the option of not saving the data;
S5.2: the system provides the deleted data category set corresponding to the optimal solution for the user, and the user decides whether to select the optimal solution;
S5.3: if the user selects to change the expected unreserved data into the data category set corresponding to the optimal solution, distributing a game space of the optimal solution corresponding to the user and the system; if the user does not select the optimal solution, the system searches all parts containing the information which the user does not expect to keep in the game matrix, finds and provides the data category set corresponding to the optimal solution belonging to the two parties from the parts, and the user decides whether to select the optimal solution; distributing self game space of the user and the system according to the user decision result;
S5.4: if the user does not select the optimal solution, the system searches for all parts containing information that the user does not want to keep, e.g. the user wants to delete the data Both data, then the model will be provided to the include/> Obtaining the profit function and the cost function of the corresponding set, finding and providing the optimal solution belonging to both the user and the model from the information, and deciding whether to select the optimal solution by the user.
Further, in the step S6, the game space of the two parties is calculated through the optimal solution, and the game space of the user is calculated And model game space/> Distributing authentication certificates by PKI (public key infrastructure) to gaming parties and generating the respective keys/>, of both parties and encrypting the game space belonging to the machine model after the training of the machine model is finished.
Further, before the certificate in S7 expires, the original model is used, and after the certificate expires, the information belonging to the game space part of the user is deleted in the database.
Compared with the prior art, the invention has the following beneficial effects:
The game space of the user is encrypted through the machine learning technology of game theory and cryptography, so that confidentiality and integrity of data in the storage and transmission processes are ensured. The encryption measures can effectively prevent unauthorized access and data leakage, thereby protecting the privacy of users and improving the data security, model performance and expandability and flexibility of the method.
Drawings
FIG. 1 is a flow chart of a method of a preferred embodiment;
Detailed Description
In order to further describe the technical means and effects adopted by the present invention for achieving the intended purpose, the following detailed description will refer to the specific implementation, structure, characteristics and effects according to the present invention with reference to the accompanying drawings and preferred embodiments.
Referring to fig. 1, the invention provides a machine learning data forgetting method based on game theory and cryptography, comprising the following steps:
s1, acquiring a training set and constructing a learning system;
S2, initializing iteration of a learning system;
S3, constructing a necessary function, and constructing and judging a game matrix according to the necessary function;
s4, inputting all data, and performing model training by using a SISA frame;
S5, users game with the machine, and self game space is allocated;
s6, encrypting the game space by using PKI to finish the forgetting protection of the information category;
S7, extracting machine forgetting learning results.
S1, acquiring small-sized typical data from complete data provided by a plurality of users And constructing a data classifier including K-means clustering (K-means clustering) and a pre-trained small machine learning model G, wherein/> ,/> Representing the nth data provided from the user.
The step of learning the system initialization iteration in S2 is as follows:
Initializing t times, wherein the machine learning model obtained by the t times is as follows According to/> defining a model cost function/>, by linear regression wherein y is the data obtained by training the machine learning model,/> Representing y corresponds to a data set .
In S3, three functions are defined according to the machine learning iteration result: regarding the cost function of the model, regarding the weight function of the data and regarding the benefit function of the user, fitting the machine learning result and the corresponding user by using linear regression to construct the cost function, and obtaining the weight function of the data through the cost function.
The calculation formula of the cost function is as follows:
。
Wherein the method comprises the steps of For/> Index of/> The cost function is set by the result as regression coefficient data set/>, as dependent variable And obtaining a basic linear regression equation for the independent variable fitting.
The cost function has n subfunctions, and the calculation formula is as follows:
；
Wherein the method comprises the steps of Namely, data category/> weight function for the present cost function.
S3, constructing and judging a game matrix according to the necessary functions, wherein the specific steps are as follows:
S3.1, constructing a judgment game matrix ;
S3.2, reusing the cost function and the benefit function The individual data sets are filled into the matrix/> ;
s3.3. definition matrix the element in (a) is/> Calculating/>, based on the cost function and the benefit function of the construction Is the value of (1): /(I) ;
And S3.4, finding the optimal solution of the forgetting model according to the Nash equilibrium odd rule.
S4, inputting all data, and performing model training by using the SISA frame, wherein the specific steps are as follows:
S4.1: dividing an original data set containing all user information into a plurality of disjoint sliced data sets Di and putting the disjoint sliced data sets Di into a database;
S4.2: the SISA model frame algorithm is utilized to put the information which is to be forgotten finally into any one In (a) and (b);
s4.3: model for each piece of data from when there was initially no forgotten data Beginning training, first use/> updating the model as a training set is/> second use/> Update/>, as training set when the training set is updated, the model updated by the original training set is saved, and the model set is obtained by updating the model through n times of iteration continuously and incrementally wherein/> We are the final trained model.
The specific operation steps in S5 are as follows:
S5.1: when the user does not desire to save the data to the model, the access system provides an interface for inquiring the user about the option of deleting the data, and the user selects the option of not saving the data;
S5.2: the system provides the deleted data category set corresponding to the optimal solution for the user, and the user decides whether to select the optimal solution;
S5.3: if the user selects to change the expected unreserved data into the data category set corresponding to the optimal solution, distributing a game space of the optimal solution corresponding to the user and the system; if the user does not select the optimal solution, the system searches all parts containing the information which the user does not expect to keep in the game matrix, finds and provides the data category set corresponding to the optimal solution belonging to the two parties from the parts, and the user decides whether to select the optimal solution; distributing self game space of the user and the system according to the user decision result;
S5.4: if the user does not select the optimal solution, the system searches for all parts containing information that the user does not want to keep, e.g. the user wants to delete the data Both data, then the model will be provided to the include/> Obtaining the profit function and the cost function of the corresponding set, finding and providing the optimal solution belonging to both the user and the model from the information, and deciding whether to select the optimal solution by the user.
S6, calculating game spaces of the two parties through the optimal solution, and calculating the game space of the user And model game space/> Distributing authentication certificates by PKI (public key infrastructure) to gaming parties and generating the respective keys/>, of both parties and encrypting the game space belonging to the machine model after the training of the machine model is finished.
And S7, before the certificate expires, the original model is used, and after the certificate expires, the information belonging to the game space part of the user is deleted in the database.
In this embodiment, a machine learning data forgetting method based on game theory and cryptography comprises the following specific steps: acquiring training set and constructing and learning system thereof, by acquiring small-sized typical data from complete data provided by a plurality of users And constructing a data classifier including K-means clustering (K-means clustering) and a pre-trained small machine learning model G, wherein/> ,/> Representing the nth data provided from the user. The data classifier is the data input by the user/> Feature extraction, if/> the method is characterized in that the method is determined to be N different data, the data is set and divided into N clusters, namely K=N, the clustering center is initialized and then is subjected to gradual iterative optimization, and a data set obtained by classifying according to categories is recorded as/>, wherein the data set is obtained by iterative cutoff And its data class collection/> Wherein/> For the ith data we note/> Is a 0-1 variable, i.e. when/> Training data of this type This type of data is not trained, data set/> Put into circulation and train, data set is from circulation to/> Obtaining result array/> And (3) arranging the data in descending order according to a sorting algorithm to obtain and print descending order results and corresponding data sets. At the moment, the corresponding machine performance conditions in different data sets can be obtained, and the maximum group of machine performance benefits is found. Evaluation criteria for machine performance: when the algorithm is designed, the two game parties are game main bodies, whether the finally trained result is satisfactory or not, whether the finally trained result reaches a preset target or not, the user feedback can intuitively reflect the training effect of the user, and the data is collected through user feedback in user investigation, questionnaire investigation or practical application, and is analyzed and tidied; by setting up an experimental scene, the application situation of machine forgetting learning is simulated, and the performance and behavior of the model after forgetting user data are evaluated. Real user data or artificially generated data can be used for setting according to actual conditions; the comparison analysis can be used for verifying the good degree of our results, the existing machine forgetting methods such as random forest, GBDT forgetting, disturbance technology, federal learning removal and the like are compared, and the reference condition can be a model obtained by training by using all user data or the effects of other forgetting methods; through comparative analysis, whether the performance of machine forgetting learning meets the requirement consistent with the requirement that no user information is obtained or not is evaluated; the effectiveness and efficiency are combined. The learning system initiates the iteration as follows: initializing t times, wherein a machine learning model obtained by the t times is/> According to/> defining a model cost function/>, by linear regression wherein y is the data obtained by training the machine learning model,/> representing y corresponds to data set/> . Constructing a necessary function, and constructing and judging a game matrix according to the necessary function; three functions are defined according to the machine learning iteration result: regarding the cost function of the model, regarding the weight function of the data and regarding the benefit function of the user, fitting the machine learning result and the corresponding user by using linear regression to construct the cost function, and obtaining the weight function of the data through the cost function. The user benefit is positively correlated with the information value of the data. Thereby defining a user's benefit function as/> . The calculation formula of the cost function is as follows:
。
Wherein the method comprises the steps of For/> Index of/> The cost function is set by the result as regression coefficient data set/>, as dependent variable And obtaining a basic linear regression equation for the independent variable fitting.
The cost function has n subfunctions, and the calculation formula is as follows:
；
Wherein the method comprises the steps of Namely, data category/> weight function for the present cost function.
The specific steps for constructing and judging the game matrix according to the necessary functions are as follows: s3.1, constructing a judgment game matrix ; s3.2/>, by reusing the cost function and the benefit function The individual data sets are filled into the matrix/> ; s3.3 definition matrix/> the element in (a) is/> Calculating/>, based on the cost function and the benefit function of the construction Is the value of (1): ; and S3.4, finding the optimal solution of the forgetting model according to the Nash equilibrium odd rule. Inputting all data, and performing model training by using a SISA frame, wherein the steps are as follows: s4.1: dividing an original data set containing all user information into a plurality of disjoint sliced data sets Di and putting the disjoint sliced data sets Di into a database; the sliced data set Di is one of sets obtained by dividing all original data in our users, each subset contains part of data in the original data set, each data exists in one of all sets, and the union of the data in all sets is the original data set of our users. S4.2: placing the information which is finally forgotten into any one/>, by using SISA model frame algorithm in (a) and (b); s4.3: from the model/>, when there is no forgetting data initially, each piece of data Beginning training, first use/> updating the model as a training set is/> second use/> Update/>, as training set When the training set is updated, the model updated by the original training set is saved, and the model set/>, which is obtained by updating the model through n times of iteration, is updated continuously and incrementally wherein/> we are the final trained model. The user plays games with the machine, and self game space is allocated, and S5.1: when the user does not desire to save the data to the model, the access system provides an interface for inquiring the user about the option of deleting the data, and the user selects the option of not saving the data; s5.2: the system provides the deleted data category set corresponding to the optimal solution for the user, and the user decides whether to select the optimal solution; s5.3: if the user selects to change the expected unreserved data into the data category set corresponding to the optimal solution, distributing a game space of the optimal solution corresponding to the user and the system; if the user does not select the optimal solution, the system searches all parts containing the information which the user does not expect to keep in the game matrix, finds and provides the data category set corresponding to the optimal solution belonging to the two parties from the parts, and the user decides whether to select the optimal solution; distributing self game space of the user and the system according to the user decision result; s5.4: if the user does not select the optimal solution, the system searches for all parts containing information that the user does not want to keep, e.g. the user wants to delete the data Both data, then the model will be provided to the include/> Obtaining the profit function and the cost function of the corresponding set, finding and providing the optimal solution belonging to both the user and the model from the information, and deciding whether to select the optimal solution by the user. Encrypting the game space by using PKI to finish the forgetting protection of the information category; s6, calculating game spaces of the two parties through the optimal solution, and calculating user game space/> And model game space/> Distributing authentication certificates by PKI (public key infrastructure) to gaming parties and generating the respective keys/>, of both parties And encrypting the game space belonging to the machine model after the training of the machine model is finished. After expiration of the certificate, the information belonging to the user game space part is deleted by the database. Public key infrastructure PKI is introduced, a digital certificate is issued by a certificate authentication mechanism (Certification Authority, CA), identity information of PKI users and a public key are bound, so that two game parties are securely encrypted, the game spaces of the two game parties are encrypted by using the public key, and the information which needs to be forgotten is deleted by using a database regular timing deletion technology. Extracting machine forgetting learning results, using an original model before a certificate expires, and deleting part of information belonging to a user game space in a database after the certificate expires. Taking out a model corresponding to the forgotten data and judging the forgetting performance of the machine; judging according to the machine forgetting performance evaluation standard, judging whether the trained new model is consistent with the model performance obtained after complete retraining, and obtaining a new model by applying the result consistent; the result is inconsistent and the model is retrained here.
The present invention is not limited to the above embodiments, but is capable of modification and variation in detail, and other modifications and variations can be made by those skilled in the art without departing from the scope of the present invention.
Claims (10)
1. The machine learning data forgetting method based on game theory and cryptography is characterized by comprising the following steps of:
s1, acquiring a training set and constructing a learning system;
S2, initializing iteration of a learning system;
S3, constructing a necessary function, and constructing and judging a game matrix according to the necessary function;
s4, inputting all data, and performing model training by using a SISA frame;
S5, users game with the machine, and self game space is allocated;
s6, encrypting the game space by using PKI to finish the forgetting protection of the information category;
S7, extracting machine forgetting learning results.
2. The machine learning data forgetting method based on game theory and cryptography according to claim 1, wherein the small-sized typical data is obtained from the complete data provided by several users in S1 And constructing a data classifier including K-means clustering (K-means clustering) and a pre-trained small machine learning model G, wherein/> ,/> Representing the nth data provided from the user.
3. The machine learning data forgetting method based on game theory and cryptography according to claim 2, wherein the step of learning the system initialization iteration in S2 is as follows:
Initializing t times, wherein the machine learning model obtained by the t times is as follows According to/> defining a model cost function/>, by linear regression wherein y is the data obtained by training the machine learning model,/> Representing y corresponds to a data set .
4. The machine learning data forgetting method based on game theory and cryptography according to claim 3, wherein three functions are defined according to the machine learning iteration result in S3: regarding the cost function of the model, regarding the weight function of the data and regarding the benefit function of the user, fitting the machine learning result and the corresponding user by using linear regression to construct the cost function, and obtaining the weight function of the data through the cost function.
5. The machine learning data forgetting method based on game theory and cryptography according to claim 4, wherein the calculation formula of the cost function is:
；
Wherein the method comprises the steps of For/> Index of/> As regression coefficients, the cost function is obtained by using the result array/> data set/>, as dependent variable Fitting the independent variables to obtain a basic linear regression equation;
The cost function has n subfunctions, and the calculation formula is as follows:
；
Wherein the method comprises the steps of Namely, data category/> weight function for the present cost function.
6. The machine learning data forgetting method based on game theory and cryptography according to claim 5, wherein the specific steps of determining a game matrix according to the necessary function configuration in S3 are as follows:
S3.1, constructing a judgment game matrix ;
S3.2, reusing the cost function and the benefit function The individual data sets are filled into the matrix/> ;
s3.3. definition matrix the element in (a) is/> Calculating/>, based on the cost function and the benefit function of the construction Is the value of (1): ;
And S3.4, finding the optimal solution of the forgetting model according to the Nash equilibrium odd rule.
7. The machine learning data forgetting method based on game theory and cryptography according to claim 1, wherein the specific steps of inputting all data in S4 and performing model training by using the SISA framework are as follows:
S4.1: dividing an original data set containing all user information into a plurality of disjoint sliced data sets Di and putting the disjoint sliced data sets Di into a database;
S4.2: the SISA model frame algorithm is utilized to put the information which is to be forgotten finally into any one In (a) and (b);
s4.3: model for each piece of data from when there was initially no forgotten data Beginning training, first use/> updating the model as a training set is/> second use/> Update/>, as training set when the training set is updated, the model updated by the original training set is saved, and the model set is obtained by updating the model through n times of iteration continuously and incrementally wherein/> We are the final trained model.
8. The machine learning data forgetting method based on game theory and cryptography according to claim 6, wherein the specific operation steps in S5 are as follows:
S5.1: when the user does not desire to save the data to the model, the access system provides an interface for inquiring the user about the option of deleting the data, and the user selects the option of not saving the data;
S5.2: the system provides the deleted data category set corresponding to the optimal solution for the user, and the user decides whether to select the optimal solution;
S5.3: if the user selects to change the expected unreserved data into the data category set corresponding to the optimal solution, distributing a game space of the optimal solution corresponding to the user and the system; if the user does not select the optimal solution, the system searches all parts containing the information which the user does not expect to keep in the game matrix, finds and provides the data category set corresponding to the optimal solution belonging to the two parties from the parts, and the user decides whether to select the optimal solution; distributing self game space of the user and the system according to the user decision result;
S5.4: if the user does not select the optimal solution, the system searches for all parts containing information that the user does not want to keep, e.g. the user wants to delete the data Both data, then the model will be provided to the include/> Obtaining the profit function and the cost function of the corresponding set, finding and providing the optimal solution belonging to both the user and the model from the information, and deciding whether to select the optimal solution by the user.
9. The machine learning data forgetting method based on game theory and cryptography according to claim 8, wherein in S6, the game space of both parties is calculated through the optimal solution, and the game space of the user is calculated And model game space/> Distributing authentication certificates by PKI (public key infrastructure) to gaming parties and generating the respective keys/>, of both parties and encrypting the game space belonging to the machine model after the training of the machine model is finished.
10. The method for forgetting machine learning data based on game theory and cryptography according to claim 9, wherein the original model is used before the certificate expires in S7, and the part of information belonging to the game space of the user is deleted in the database after the certificate expires.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410303506.3A CN117892843B (en) | 2024-03-18 | Machine learning data forgetting method based on game theory and cryptography |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410303506.3A CN117892843B (en) | 2024-03-18 | Machine learning data forgetting method based on game theory and cryptography |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117892843A true CN117892843A (en) | 2024-04-16 |
| CN117892843B CN117892843B (en) | 2024-06-04 |
Family
ID=
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113779870A (en) * | 2021-08-24 | 2021-12-10 | 清华大学 | Parallelization imperfect information game strategy generation method and device, electronic equipment and storage medium |
| CN114048834A (en) * | 2021-11-05 | 2022-02-15 | 哈尔滨工业大学（深圳） | Continuous reinforcement learning non-complete information game method and device based on after-the-fact review and progressive expansion |
| CN116485430A (en) * | 2023-04-18 | 2023-07-25 | 大连理工大学 | Federal learning forgetting mechanism and method for data circulation |
| CN117056957A (en) * | 2023-05-06 | 2023-11-14 | 浙江大学 | Verifiable data forgetting privacy protection method and device for minimum and maximum learning model |
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113779870A (en) * | 2021-08-24 | 2021-12-10 | 清华大学 | Parallelization imperfect information game strategy generation method and device, electronic equipment and storage medium |
| CN114048834A (en) * | 2021-11-05 | 2022-02-15 | 哈尔滨工业大学（深圳） | Continuous reinforcement learning non-complete information game method and device based on after-the-fact review and progressive expansion |
| CN116485430A (en) * | 2023-04-18 | 2023-07-25 | 大连理工大学 | Federal learning forgetting mechanism and method for data circulation |
| CN117056957A (en) * | 2023-05-06 | 2023-11-14 | 浙江大学 | Verifiable data forgetting privacy protection method and device for minimum and maximum learning model |
Non-Patent Citations (2)
| Title |
|---|
| JOSEP DOMINGO-FERRER等: "Co-utility for digital content protection and digital forgetting", 2016 MEDITERRANEAN AD HOC NETWORKING WORKSHOP (MED-HOC-NET), 4 August 2016 (2016-08-04) * |
| 张玉清;董颖;柳彩云;雷柯楠;孙鸿宇;: "深度学习应用于网络空间安全的现状、趋势与展望", 计算机研究与发展, no. 06, 12 January 2018 (2018-01-12) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11574077B2 (en) | Systems and methods for removing identifiable information | |
| CN111914256B (en) | Defense method for machine learning training data under toxic attack | |
| Wang et al. | A survey on ChatGPT: AI-generated contents, challenges, and solutions | |
| CN109002861A (en) | Federal modeling method, equipment and storage medium | |
| US8301622B2 (en) | Identity analysis and correlation | |
| Lee et al. | Graddiv: Adversarial robustness of randomized neural networks via gradient diversity regularization | |
| Sheng et al. | A biometric key generation method based on semisupervised data clustering | |
| WO2009020482A2 (en) | Hidden markov model ('hmm')-based user authentication using keystroke dynamics | |
| US8346774B1 (en) | Protecting network entity data while preserving network properties | |
| CN111242230A (en) | Image processing method and image classification model training method based on artificial intelligence | |
| CN110348238B (en) | Privacy protection grading method and device for application | |
| CN113761261A (en) | Image retrieval method, image retrieval device, computer-readable medium and electronic equipment | |
| CN109829333B (en) | OpenID-based key information protection method and system | |
| CN114417427B (en) | Deep learning-oriented data sensitivity attribute desensitization system and method | |
| Chen et al. | Challenges and remedies to privacy and security in aigc: Exploring the potential of privacy computing, blockchain, and beyond | |
| CN108804931B (en) | Neural network model encryption protection system and method related to domain transformation data encryption | |
| CN117892843B (en) | Machine learning data forgetting method based on game theory and cryptography | |
| CN113342418A (en) | Distributed machine learning task unloading method based on block chain | |
| Abdrashitov et al. | Sensor data anonymization based on genetic algorithm clustering with L-Diversity | |
| CN117892843A (en) | Machine learning data forgetting method based on game theory and cryptography | |
| CN116861485A (en) | Student information privacy protection method based on deep learning fusion | |
| CN116861152A (en) | Tax data security graph neural network training method based on matrix decomposition | |
| CN112084509A (en) | Block chain key generation method and system based on biological identification technology | |
| CN115189966B (en) | Block chain private data encryption and decryption service system | |
| CN113626854A (en) | Image data privacy protection method based on localized differential privacy |
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 |