CN117390664A - Federal learning oriented game-driven privacy self-adaptive pricing method and device - Google Patents
Federal learning oriented game-driven privacy self-adaptive pricing method and device Download PDFInfo
- Publication number
- CN117390664A CN117390664A CN202311091588.1A CN202311091588A CN117390664A CN 117390664 A CN117390664 A CN 117390664A CN 202311091588 A CN202311091588 A CN 202311091588A CN 117390664 A CN117390664 A CN 117390664A
- Authority
- CN
- China
- Prior art keywords
- privacy
- client
- server
- federal learning
- utility function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 76
- 230000006870 function Effects 0.000 claims abstract description 75
- 230000008569 process Effects 0.000 claims abstract description 21
- 230000003044 adaptive effect Effects 0.000 claims description 17
- 238000010276 construction Methods 0.000 claims description 9
- 230000008901 benefit Effects 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 5
- 238000004886 process control Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 abstract description 5
- 230000007246 mechanism Effects 0.000 description 25
- 238000012549 training Methods 0.000 description 15
- 238000010586 diagram Methods 0.000 description 10
- 238000004891 communication Methods 0.000 description 9
- 230000002776 aggregation Effects 0.000 description 6
- 238000004220 aggregation Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 230000035945 sensitivity Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 230000008450 motivation Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G06F21/60—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
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0283—Price estimation or determination
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Finance (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Medical Informatics (AREA)
- Accounting & Taxation (AREA)
- Bioethics (AREA)
- Computer Security & Cryptography (AREA)
- Entrepreneurship & Innovation (AREA)
- Computer Hardware Design (AREA)
- Game Theory and Decision Science (AREA)
- Databases & Information Systems (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The invention provides a game-driven privacy self-adaptive pricing method and device oriented to federal learning, wherein the method comprises the following steps: constructing a federal learning framework based on differential privacy; obtaining the privacy payment upper limit and income of a federal learning framework server, and constructing a utility function of the server; acquiring privacy budget and privacy loss of the federal learning framework client, and constructing a utility function of the client; inputting the utility function of the server and the utility function of the client into a two-stage Stark game model to obtain a privacy self-adaptive pricing strategy; and controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy. The method can maximize the utility of the server and the client and balance the privacy and model performance of the client. The invention designs utility functions of the client and the server aiming at the problem of mutual balance of the client privacy and the global server performance in federal learning, and realizes the balance of the client privacy and the server performance by optimizing the utility functions.
Description
Technical Field
The invention relates to the technical field of network security, in particular to a game-driven privacy self-adaptive pricing method and device for federal learning.
Background
Since in federal learning, clients performing local training inevitably take on the risks of computation and communication overhead and local data privacy exposure. In the federal learning process, in order to motivate customers to actively participate in federal learning, many researchers design corresponding motivation mechanisms to increase the enthusiasm of customers to participate in model training.
However, existing incentive mechanisms only consider the training costs and communication costs of customers in joint learning, where the risk of personal privacy disclosure is not compensated.
Disclosure of Invention
The invention provides a game-driven privacy self-adaptive pricing method and device oriented to federal learning, which are used for solving the defect that in the prior art, only training cost and communication cost generated by a client in joint learning are considered by a federal learning incentive mechanism, so that personal privacy leakage risk of the client in joint learning is not compensated, and the self utility is maximized and the client privacy and model performance are balanced by selecting total privacy payment of a server and personal privacy budget of the client respectively.
The invention provides a game-driven privacy self-adaptive pricing method oriented to federal learning, which comprises the following steps:
constructing a federal learning framework based on differential privacy;
obtaining the privacy payment upper limit and income of the federal learning framework server to construct a utility function of the server;
obtaining privacy budget and privacy loss of the federal learning framework client to construct a utility function of the client;
inputting the utility function of the server and the utility function of the client into a two-stage Stark game model to obtain a privacy self-adaptive pricing strategy;
and controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
According to the game-driven privacy self-adaptive pricing method facing federal learning, the differential privacy is Gaussian differential privacy.
According to the federal learning-oriented game-driven privacy adaptive pricing method provided by the invention, the utility function of the server and the utility function of the client are input into a two-stage stark-berg game model to obtain a privacy adaptive pricing strategy, and the method comprises the following steps:
s31, solving the maximum value of the utility function of the server and the value of the upper payment limit under the condition, and subtracting the customer privacy loss from the value of the upper payment limit under the condition to obtain the customer privacy budget;
s32, based on the upper payment limit and the privacy budget of the client, which are obtained in the S31, when the upper payment limit is set, solving the privacy budget of the client when the maximum value of the client utility function is solved.
According to the game-driven privacy self-adaptive pricing method facing federal learning provided by the invention, the method for obtaining the privacy payment upper limit and the income of the federal learning framework server to construct the utility function of the server comprises the following steps:
s41, if the upper payment limit of the server is B, the server sends the client c i The payment is expressed as:
wherein, E is i Is client c i The expected privacy budget submitted to the server, representing client c i A level of privacy protection desired to be achieved;
s42, in the federal learning framework based on differential privacy, obtaining benefits E of the server S Revenue f (E) expressed as global model g ) The method comprises the following steps:
wherein,average privacy budget for global model, alpha, beta>0 is the coefficient of the model profit function, N is the number of clients participating in federal learning, E i Is client c i The expected privacy budget submitted to the server, representing client c i A level of privacy protection desired to be achieved;
s43, defining the total payment of the server to the client privacy as the cost of the server as follows:
wherein,to client c for server i And the payment, B, is the upper limit of payment for the server.
S44, utility function U of server S Expressed as:
according to the game-driven privacy self-adaptive pricing method for federal learning provided by the invention, the privacy budget and the privacy loss of the federal learning framework client are obtained to construct a utility function of the client, and the method comprises the following steps:
s51, customer c i Revenue E of (E) C,i Rewards P for server reporting privacy budget payments thereto i The method comprises the following steps:
s52, customer c i The cost of (1) is the loss of privacy for the client to participate in federal learning, expressed as:
C C,i ＝ρ i ∈ i
wherein ρ is i >0 is a customer cost parameterA number describing a privacy loss of the privacy budget of the client unit;
s53, customer c i Utility function U of (2) C,i The definition is as follows:
the invention also provides a game-driven privacy self-adaptive pricing device facing federal learning, which comprises:
the federal learning framework construction module is used for constructing a federal learning framework based on differential privacy;
the server utility function construction module is used for acquiring the privacy payment upper limit and the income of the federal learning framework server so as to construct a utility function of the server;
the client utility function construction module is used for acquiring the privacy budget and the privacy loss of the federal learning framework client so as to construct a utility function of the client;
the privacy self-adaptive pricing strategy output module is used for inputting the utility function of the server and the utility function of the client into a two-stage Stark game model so as to obtain a privacy self-adaptive pricing strategy;
and the privacy pricing process control module is used for controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes any federal learning oriented game driving privacy self-adaptive pricing method when executing the program.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements any of the federal learning oriented game driven privacy adaptive pricing methods described above.
According to the federal learning-oriented game-driven privacy self-adaptive pricing method and device, the federal learning mechanism under the game-driven privacy self-adaptive pricing is constructed, so that the privacy safety of clients is guaranteed, the attacker is prevented from stealing gradient information in the gradient aggregation and transmission process, and the privacy data of the clients are restored. Aiming at the problem that the customer privacy and the global server performance are balanced mutually in federal learning containing privacy, the invention designs a game-driven privacy self-adaptive pricing method facing federal learning, and the privacy leakage risk born by the participation of customers in federal learning is compensated by paying a reward, so that the customers are stimulated to actively participate in joint learning. Meanwhile, the utility function of the client and the server is designed, so that the client and the server realize the balance of the client privacy and the server performance while optimizing the respective utility.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a federally learning oriented game-driven privacy adaptive pricing method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a federal learning system according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a privacy pricing mechanism based on two-stage Stark primary game according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a privacy pricing process based on two-stage Stark primary game according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a federally learning oriented game-driven privacy adaptive pricing device according to an embodiment of the present invention;
fig. 6 is a schematic entity structure diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The existing research works mostly adopt methods such as auction, contract theory, game and the like to design the motivation mechanism of federal learning. An incentive mechanism for federal learning based auctions makes customer selections in the form of auction games. The practical problem of motivating customers to participate in federal learning is solved by the game between customer bids and server selection payment strategies. Aiming at the problem of mismatching of incentives between a client and a model owner and between model owners, the hierarchical incentives mechanism framework utilizes a self-revealing mechanism in contractual agreements and a alliance game theory method to rewards the model owners according to marginal contributions of the model owners. Zhan Yufeng et al propose a game-based incentive mechanism that combines distributed deep learning with crowd sensing for big data analysis on mobile clients. The platform issues tasks and rewards, and the enthusiasm of the client training data is improved by maximizing the self utility of the mobile client. The structural characteristics of a server and a plurality of clients in the federal learning are consistent with a leader and a plurality of follow-up structures in the stark-berg game model, and some researchers use the stark-berg game model to design a federal learning excitation mechanism according to the characteristics of federal learning. As an incentive mechanism for the two-stage stark-berg game of the federal learning system, the TSG can ensure that the server achieves the best utility while incentive staff trains the joint model with the best effort. Yunus Sarikaya et al devised a method of motivation based on stark-berg gaming to motivate employees to allocate more computing resources for local training. For joint learning of edge networks, latif U.Khan et al model incentive-based interactions between a global server and participating devices through Stark gaming to facilitate the participation of the devices in the joint learning process. Feng Shaohan et al devised a service pricing mechanism based on joint learning of stark-berg. For training tasks of model owners, the mobile device non-cooperatively decides prices for unit training data to maximize individual utility and cooperatively transmit model updates. However, the incentive mechanism only considers the training cost and the communication cost of the client in joint learning, and the personal privacy leakage risk of the client in joint learning is not compensated.
In view of the above problems, the embodiment of the invention provides a game-driven privacy self-adaptive pricing method for federal learning, which provides a game-driven privacy self-adaptive pricing algorithm for federal learning, builds a federal learning mechanism under game-driven privacy self-adaptive pricing according to personalized privacy requirements of clients in federal learning, and maximizes self utility and balances client privacy and server performance by selecting total privacy payment and personal privacy budget respectively for a server and clients.
The game-driven privacy-adaptive pricing method for federal learning provided by the embodiment of the invention can be applied to the scene of federal learning for a plurality of clients in certain specific fields, wherein the specific fields can comprise financial fields, internet fields, scientific research fields and the like.
The federally learning oriented game-driven privacy adaptive pricing method and apparatus of the present invention are described below in conjunction with fig. 1-6.
Fig. 1 is a schematic flow chart of a game-driven privacy adaptive pricing method for federal learning, which is provided by an embodiment of the present invention, and as shown in fig. 1, the method includes:
step 101: constructing a federal learning framework based on differential privacy;
step 102: obtaining the privacy payment upper limit and income of the federal learning framework server to construct a utility function of the server;
step 103: obtaining privacy budget and privacy loss of the federal learning framework client to construct a utility function of the client;
step 104: inputting the utility function of the server and the utility function of the client into a two-stage Stark game model to obtain a privacy self-adaptive pricing strategy;
step 105: and controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
The above steps are described in detail below in connection with specific embodiments.
Step 101: constructing a federal learning framework based on differential privacy;
in this step, the federal learning framework includes four parts, namely, local training, global aggregation, parameter broadcasting and model updating;
the local training is specifically as follows:
all clients perform local training according to the local data, and model parameters obtained through training are uploaded to a server.
The global aggregation specifically comprises the following steps:
the server performs secure aggregation of parameters uploaded by the N clients without knowing the local information.
The parameter broadcasting specifically comprises the following steps:
the server broadcasts the aggregated parameters to the clients.
The model updating is specifically as follows:
all clients update their respective models using the aggregate parameters and test the performance of the updated models.
By way of example, a federal learning architecture consisting of one server and N clients may be used. Fig. 2 is a schematic structural diagram of a federal learning system according to an embodiment of the present invention, as shown in fig. 2, assuming that a local data set of each client is D i Where i e {1,2,..N }. The client finds the appropriate model parameter vector w by minimizing the loss function based on the local data i . And the server obtains a global model according to the parameter aggregation uploaded by the N clients. The weights received by the server from the N clients are aggregated as follows:
wherein w is i Is the parameter vector trained on the ith client and N is the number of clients.Wherein the method comprises the steps ofIs the total number of all data samples. Therefore, the specific calculation formula of the optimization problem of the global model parameter vector is as follows:
wherein F is i (. Cndot.) is the local loss function of the ith customer.
In an embodiment of the present invention, the differential privacy (Differential Privacy) is a privacy protection technique, which aims to protect personal privacy when sharing or publishing data. It can effectively prevent data abuse and leakage while allowing meaningful analysis of the data.
Specifically, the differential privacy technique adds a certain amount of noise during the data distribution process to confuse the information of individuals, so that the data of individual individuals cannot be accurately restored or identified. By adding appropriate noise, differential privacy can ensure that the shared data remains accurate and useful as a whole, but without exposing the actual information of any particular individual.
For data of differential privacy effect, there are the following definitions:
adjoining data sets: if for both data sets, D, D ', there is D-D' | 1 And +.1, data sets D and D' are referred to as contiguous data sets. There is at most one record difference between adjoining databases. The embodiment of the invention selects privacy at the user level, so that adjacent databases in the text have differences in one data sample at most.
Gaussian differential privacy:
gaussian differential privacy provides relaxed (epsilon, delta) -DP, where epsilon represents the privacy budget and delta represents the relaxation term. The gaussian mechanism calculates the magnitude of the random noise added based on the sensitivity of the query data. Sensitivity of gaussian differential privacy is defined as:
for the query function f: N |X| →R k The sensitivity of (c) is defined as:
where D, D' is the contiguous dataset. Sensitivity refers to the degree of sensitivity of the query results to changes in individual data. Higher sensitivity means that the data is more vulnerable to privacy attacks and therefore more noise needs to be added at the time of distribution.
Based on the above analysis, when the privacy budget ε (0, 1), for arbitrary function f: N |X| →R k The Gaussian mechanism for adding Gaussian differential privacy is defined as follows:
M G (x,f(·),ε)＝f(x)+n
wherein N corresponds to a Gaussian distribution N-N (0, sigma) 2 ) Standard deviation of noiseThe constant b satisfies the algorithmWhen M is G (x, f (·), ε) =f (x) +n satisfies (ε, δ) -differential privacy.
Specifically, the (ε, δ) -differential privacy is defined as:
X-R for any random mechanism, if for any measurable setAnd for any two contiguous data sets D, D' ∈x, satisfying the following formula:
Pr[M(D)∈S]≤Pr[M(D′)∈S]+δ
it will be appreciated that with the gaussian mechanism described above, we can ensure that the pre-aggregate parameters meet differential privacy by adding appropriate noise.
Step 102: obtaining the privacy payment upper limit and income of the federal learning framework server to construct a utility function of the server;
in this step, an initial upper payment limit of the server is predefined, and a customer privacy payment policy is defined on the basis of this. Illustratively, a server initial payment upper limit may be defined as B, with the total payment for all customers being affected by changing the server payment upper limit B.
Specifically, for any client c i E C, server sends to client C i The payment is expressed as:
wherein, E is i Is client c i The expected privacy budget submitted to the server, representing client c i A level of privacy protection that is desired to be achieved. The consideration that the client gets is proportional to the privacy budget that he reports. Both the client and the server may influence the final payment, but the server determines the upper bound of the payment.
In this step, for the federal learning model with added (∈, δ) -differential privacy, its model yield is defined as:
f(∈)＝βln(1+α∈)
where α, β >0 is a coefficient of the model yield function. From equation (4), the model benefit is proportional to the privacy budget e of the model, and as the privacy budget e increases, the model increase revenue will gradually decrease.
Alternatively, the benefits of the server may be represented by the benefits of the global model:
wherein,is the average privacy budget of the global model.
The cost of the server can then be expressed as the total payment for the customer's privacy:
utility function U of server S Expressed as:
step 103: obtaining privacy budget and privacy loss of the federal learning framework client to construct a utility function of the client;
in this step, on the one hand, the consideration paid by the server to the client is taken as the initial reporting privacy budget of the client, namely:
on the other hand, define client c i The cost of (1) is the loss of privacy for the client to participate in federal learning, expressed as:
C C,i ＝ρ i ∈ i
wherein ρ is i >0 is a customer cost parameter describing the privacy loss of a customer unit privacy budget. It is assumed herein that the cost of privacy loss for a client is proportional to the privacy budget, with a correlation coefficient ρ i 。
Based on the above calculation result, client c i Utility U of (C) C,i The definition is as follows:
further, after adding differential privacy to the parameters after local training, aiming at the privacy pricing problem of the client in federal learning containing differential privacy, the embodiment provides a privacy pricing mechanism based on two-stage stark-berg game, which specifically comprises the following steps:
step 104: inputting the utility function of the server and the utility function of the client into a two-stage Stark game model to obtain a privacy self-adaptive pricing strategy;
step 105: and controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
Specifically, fig. 3 is a schematic structural diagram of a privacy pricing mechanism based on two-stage stark-berg game according to an embodiment of the present invention, and as shown in fig. 3, the present invention provides a customer privacy payment policy, and then the utility of a customer and a server are defined respectively, and the utility optimization problem is modeled as a two-stage stark-berg game model with the server as a leader and all the customers as followers.
In the step, the differential privacy request of the client and the pricing of the server are modeled as a two-stage Stark game model, and the optimal strategy for the game between the server and the client is obtained through the Stark game model based on the utility function.
Specifically, fig. 4 is a schematic diagram of a privacy pricing process based on two-stage stark-berg game according to an embodiment of the present invention, and as shown in fig. 4, the pricing process of the privacy pricing mechanism is as follows:
firstly, in order to protect local data privacy, clients submit differential privacy requests to a server based on privacy loss cost and personalized privacy requirements;
secondly, according to privacy budget submitted by the client, the server calculates corresponding rewards of the client to be paid, and further calculates the utility of the server;
again, the server and client learn each other's policies and respective parameter information over the past several rounds of transactions. Then, the server calculates an optimal policy to maximize utility according to the known information, and sends an optimal upper payment limit to the client;
finally, the client determines the optimal strategy for applying the privacy budget according to the optimal upper payment limit of the server by considering the respective cost.
In particular, since individuals in the privacy pricing mechanism are both rational and selfish, each party wants to maximize their own utility, and thus the privacy pricing process described above can be expressed as a utility optimization problem.
In one aspect, for a server, the utility optimization problem can be expressed as:
on the other hand, for a customer, the utility optimization problem can be expressed as:
it can be appreciated that the privacy pricing process described above, the participating parties are aware of each other's own rewards and decisions, consistent with the characteristics of stark-berg gaming. The utility optimization problem is modeled as a two-stage Stark game, a server is used as a leader in a Stark game model of a privacy pricing mechanism, privacy payment upper limits of all clients are determined in a first stage of the game, all clients are used as followers, and privacy budgets capable of maximizing respective utilities are selected according to the payment upper limits and local privacy loss costs determined by the server; meanwhile, in the second stage of game, non-cooperative game exists among clients, under the condition of a certain upper limit of payment, the clients possibly obtain higher payment by improving the self privacy budget proportion, and in the competition of the clients for privacy payment, the respective privacy budget strategies are mutually influenced.
Further, it willAs an optimal equilibrium solution, we can obtain:
wherein,the process of the stark game can be converted into a process of solving the above-described optimal equilibrium solution.
According to the game-driven privacy self-adaptive pricing method for federal learning, differential privacy is added in the parameters after local training, total privacy payment of a server and personal privacy budget of a client are considered, self-utility is maximized, and client privacy and model performance are balanced. Specifically, a federal learning mechanism under game-driven privacy self-adaptive pricing is constructed, privacy federal learning is constructed, privacy safety of clients is guaranteed, attackers are prevented from stealing gradient information in the gradient aggregation and transmission process, and client privacy data are restored. Aiming at the problem that the customer privacy and the global model performance are balanced mutually in the privacy federal learning, a game-driven privacy self-adaptive pricing algorithm for federal learning is designed, and the privacy risk born by the participation of the customers in federal learning is compensated in a payment mode, so that the customers are stimulated to actively participate in joint learning. Meanwhile, utility functions of the client and the server are designed, so that the client and the server realize the balance of the privacy of the client and the performance of the model while optimizing respective utilities.
The game-driven privacy self-adaptive pricing device for federal learning provided by the invention is described below, and the game-driven privacy self-adaptive pricing device for federal learning described below and the game-driven privacy self-adaptive pricing method for federal learning described above can be referred to correspondingly.
Fig. 5 is a schematic structural diagram of a federally learning oriented game driving privacy adaptive pricing device according to an embodiment of the present invention, where, as shown in fig. 5, the device includes:
the federal learning framework construction module 51 is configured to construct a federal learning framework based on differential privacy;
a server utility function construction module 52, configured to obtain a privacy payment upper limit and a benefit of the federal learning framework server, so as to construct a utility function of the server;
a client utility function construction module 53, configured to obtain a privacy budget and a privacy loss of the federal learning framework client, so as to construct a utility function of the client;
the privacy adaptive pricing strategy output module 54 is configured to input the utility function of the server and the utility function of the client into a two-stage stark-berg game model to obtain a privacy adaptive pricing strategy;
and the privacy pricing process control module 55 is used for controlling the federal learning-oriented privacy pricing process according to the privacy adaptive pricing strategy.
The apparatus of the present embodiment may be used to execute the method of any one of the foregoing electronic device side method embodiments, and specific implementation processes and technical effects of the apparatus are similar to those of the electronic device side method embodiments, and specific details of the electronic device side method embodiments may be referred to in the detailed description of the electronic device side method embodiments, which are not repeated herein.
Fig. 6 is a schematic physical structure diagram of an electronic device according to an embodiment of the present invention, where, as shown in fig. 6, the electronic device may include: processor 610, communication interface (Communications Interface) 620, memory 630, and communication bus 640, wherein processor 610, communication interface 620, and memory 630 communicate with each other via communication bus 640. The processor 610 can invoke logic instructions in the memory 630 to perform a federally learning oriented game-driven privacy-adaptive pricing method that includes: constructing a federal learning framework based on differential privacy; obtaining the privacy payment upper limit and income of the federal learning framework server to construct a utility function of the server; obtaining privacy budget and privacy loss of the federal learning framework client to construct a utility function of the client; inputting the utility function of the server and the utility function of the client into a two-stage Stark game model to obtain a privacy self-adaptive pricing strategy; and controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
Further, the logic instructions in the memory 630 may be implemented in the form of software functional units and stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention further provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the federal learning oriented game drive privacy adaptive pricing method provided by the above methods, the method comprising: constructing a federal learning framework based on differential privacy; obtaining the privacy payment upper limit and income of the federal learning framework server to construct a utility function of the server; obtaining privacy budget and privacy loss of the federal learning framework client to construct a utility function of the client; inputting the utility function of the server and the utility function of the client into a two-stage Stark game model to obtain a privacy self-adaptive pricing strategy; and controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. A game-driven privacy self-adaptive pricing method facing federal learning is characterized by comprising the following steps:
constructing a federal learning framework based on differential privacy;
obtaining the privacy payment upper limit and income of the federal learning framework server to construct a utility function of the server;
obtaining privacy budget and privacy loss of the federal learning framework client to construct a utility function of the client;
inputting the utility function of the server and the utility function of the client into a two-stage Stark game model to obtain a privacy self-adaptive pricing strategy;
and controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
2. The federally learning oriented game-driven privacy adaptive pricing method of claim 1, wherein the differential privacy is gaussian differential privacy.
3. The federally learning oriented game-driven privacy-adaptive pricing method of claim 1, wherein the entering the server utility function and the client utility function into a two-stage stark-berg game model to obtain the privacy-adaptive pricing strategy comprises:
s31, solving the maximum value of the utility function of the server and the value of the upper payment limit under the condition, and subtracting the customer privacy loss from the value of the upper payment limit under the condition to obtain the customer privacy budget;
s32, based on the upper payment limit and the privacy budget of the client, which are obtained in the S31, when the upper payment limit is set, solving the privacy budget of the client when the maximum value of the client utility function is solved.
4. The federally learning oriented game-driven privacy adaptive pricing method of claim 1, wherein the obtaining privacy payment upper limit and revenue for the federally learning framework server to construct a utility function for the server comprises:
s41, if the upper payment limit of the server is B, the server sends the client c i The payment is expressed as:
wherein, E is i Is client c i Expected hidden submitted to serverPrivate budget, representing customer c i A level of privacy protection desired to be achieved;
s42, in the federal learning framework based on differential privacy, obtaining benefits E of the server S Revenue f (E) expressed as global model g ) The method comprises the following steps:
wherein,average privacy budget for global model, alpha, beta>0 is the coefficient of the model profit function, N is the number of clients participating in federal learning, E i Is client c i The expected privacy budget submitted to the server, representing client c i A level of privacy protection desired to be achieved;
s43, defining the total payment of the server to the client privacy as the cost of the server as follows:
wherein,to client c for server i And the payment, B, is the upper limit of payment for the server.
S44, utility function U of server S Expressed as:
5. the federally learning oriented game-driven privacy adaptive pricing method of claim 1, wherein the obtaining the privacy budget and privacy loss of the federally learning framework client to construct the client's utility function comprises:
s51, customer c i Revenue E of (E) C,i Rewards P for server reporting privacy budget payments thereto i The method comprises the following steps:
s52, customer c i The cost of (1) is the loss of privacy for the client to participate in federal learning, expressed as:
C C,i ＝ρ i ∈ i
wherein ρ is i >0 is a customer cost parameter describing the privacy loss of a customer unit privacy budget;
s53, customer c i Utility function U of (2) C,i The definition is as follows:
6. game-driven privacy self-adaptive pricing device oriented to federal learning, and is characterized by comprising:
the federal learning framework construction module is used for constructing a federal learning framework based on differential privacy;
the server utility function construction module is used for acquiring the privacy payment upper limit and the income of the federal learning framework server so as to construct a utility function of the server;
the client utility function construction module is used for acquiring the privacy budget and the privacy loss of the federal learning framework client so as to construct a utility function of the client;
the privacy self-adaptive pricing strategy output module is used for inputting the utility function of the server and the utility function of the client into a two-stage Stark game model so as to obtain a privacy self-adaptive pricing strategy;
and the privacy pricing process control module is used for controlling the federal learning-oriented privacy pricing process according to the privacy self-adaptive pricing strategy.
7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the federally learning oriented game-driven privacy adaptive pricing method of any of claims 1-5 when the program is executed.
8. A non-transitory computer readable storage medium having stored thereon a computer program which when executed by a processor implements the federally learning oriented game driven privacy adaptive pricing method of any of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311091588.1A CN117390664A (en) | 2023-08-28 | 2023-08-28 | Federal learning oriented game-driven privacy self-adaptive pricing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311091588.1A CN117390664A (en) | 2023-08-28 | 2023-08-28 | Federal learning oriented game-driven privacy self-adaptive pricing method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117390664A true CN117390664A (en) | 2024-01-12 |
Family
ID=89437956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311091588.1A Pending CN117390664A (en) | 2023-08-28 | 2023-08-28 | Federal learning oriented game-driven privacy self-adaptive pricing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117390664A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117808123A (en) * | 2024-02-28 | 2024-04-02 | 东北大学 | Edge server allocation method based on multi-center hierarchical federal learning |
-
2023
- 2023-08-28 CN CN202311091588.1A patent/CN117390664A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117808123A (en) * | 2024-02-28 | 2024-04-02 | 东北大学 | Edge server allocation method based on multi-center hierarchical federal learning |
| CN117808123B (en) * | 2024-02-28 | 2024-07-05 | 东北大学 | Edge server allocation method based on multi-center hierarchical federal learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lyu et al. | Collaborative fairness in federated learning | |
| Goodspeed | A re-examination of the use of ability to pay taxes by local governments | |
| EP3923207A2 (en) | Clustering techniques for machine learning models | |
| CN114297722A (en) | Privacy protection asynchronous federal sharing method and system based on block chain | |
| CA3081079A1 (en) | Information processing device, information processing method, and computer program | |
| CN117390664A (en) | Federal learning oriented game-driven privacy self-adaptive pricing method and device | |
| Hoelzlein | Two-sided sorting and spatial inequality in cities | |
| Fenichel et al. | Heterogeneity and the fragility of the first best: putting the “micro” in bioeconomic models of recreational resources | |
| Chen et al. | A Mechanism Design Approach for Multi-party Machine Learning | |
| Ma et al. | RTRC: a reputation-based incentive game model for trustworthy crowdsourcing service | |
| Chen et al. | DIM-DS: Dynamic incentive model for data sharing in federated learning based on smart contracts and evolutionary game theory | |
| Tang et al. | Credit and quality intelligent learning based multi-armed bandit scheme for unknown worker selection in multimedia MCS | |
| Cheng et al. | Dynamic games for social model training service market via federated learning approach | |
| Zheng et al. | FL-market: Trading private models in federated learning | |
| Yin et al. | A game-theoretic approach for federated learning: A trade-off among privacy, accuracy and energy | |
| Imakura et al. | Another use of SMOTE for interpretable data collaboration analysis | |
| Peng et al. | Truth based three-tier Combinatorial Multi-Armed Bandit ecosystems for mobile crowdsensing | |
| Yang et al. | Dynamic distributed iterative computational model for payment information management in shared logistics using blockchain-assisted Internet of Things approach | |
| Suter et al. | Common pool resource management at the extensive and intensive margins: experimental evidence | |
| Xu et al. | Optimal control of an online reputation dynamic feedback incentive model | |
| US20140344020A1 (en) | Competitor pricing strategy determination | |
| US20140344022A1 (en) | Competitor response model based pricing tool | |
| US20140344021A1 (en) | Reactive competitor price determination using a competitor response model | |
| d’Andria | Tax policy and entrepreneurial entry with information asymmetry and learning | |
| Witt et al. | Decentralized and Incentivized Federated Learning: A Blockchain-Enabled Framework Utilising Compressed Soft-Labels and Peer Consistency |
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 |