CN112926088B - Federal learning privacy policy selection method based on game theory - Google Patents
Federal learning privacy policy selection method based on game theory Download PDFInfo
- Publication number
- CN112926088B CN112926088B CN202110292473.3A CN202110292473A CN112926088B CN 112926088 B CN112926088 B CN 112926088B CN 202110292473 A CN202110292473 A CN 202110292473A CN 112926088 B CN112926088 B CN 112926088B
- Authority
- CN
- China
- Prior art keywords
- participants
- participant
- server
- service
- model
- 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
- 238000010187 selection method Methods 0.000 title claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 18
- 230000007774 longterm Effects 0.000 claims abstract description 15
- 230000008901 benefit Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000006399 behavior Effects 0.000 abstract description 2
- 208000027089 Parkinsonian disease Diseases 0.000 description 5
- 206010034010 Parkinsonism Diseases 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000002776 aggregation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005303 weighing Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012946 outsourcing Methods 0.000 description 1
- 238000004321 preservation 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
- G06N5/042—Backward inferencing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Bioethics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Security & Cryptography (AREA)
- Databases & Information Systems (AREA)
- Computer Hardware Design (AREA)
- Computational Linguistics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a federal learning privacy policy selection method based on game theory, which comprises the following steps: the server provides thresholds with different service costs for the participants, the participants select the optimal threshold according to whether the thresholds meet the service quality, the privacy disclosure cost and the like, and the server updates the service cost in the next iterative training; the server obtains optimal model parameters through multiple iterations, so that a long-term stable service state of the model is maintained, and the model is provided for the participants. The method effectively avoids malicious behaviors such as taking a car and the like of the participants, so that the server can obtain service cost to the maximum extent, and the participants can obtain long-term high-quality service.
Description
Technical Field
The invention relates to the field of federal learning and privacy protection, in particular to a federal learning privacy policy selection method based on game theory.
Background
Federal learning is a machine learning process that can cooperatively train a model without collecting all data of each participant, and when the participants do not adopt privacy protection settings, the model trained by the server has the best service quality, but the problem of revealing the privacy of the participants exists, and the interests of the participants are seriously damaged; when the participants adopt very strong privacy protection settings, the personal privacy of the participants can be guaranteed, but the service quality of the model can be influenced. Therefore, there is a need at this time to have a threshold to adjust the privacy preserving strength of the participants and the server trained model quality of service.
However, in model training, since most participants are selfish, they choose the maximum threshold for privacy protection when they have the right to secure personal privacy information, and they usually consider the interests in front of the eyes rather than long term when choosing the threshold.
Although many incentive mechanisms have been proposed to address the selfish problem in model quality of service and privacy preserving trade-offs, most focus on disposable gaming models, such as disposable gaming in outsourcing services, to improve k-anonymous privacy preservation by designing a coalition policy and then sharing revenue results among cooperating users. Meanwhile, most of the game processes are based on the assumption of complete information, namely, each game party can know the strategies and rewards of other game parties. The complete information assumption is difficult to implement in reality.
Therefore, the existing strategy weighing method does not consider long-term service benefits, namely a high-quality model which can serve the participants for a long time and can guarantee privacy of the participants cannot be trained; most of the existing strategy weighing methods only consider one game, and in fact, multiple games are possible for both parties, and the influence of the current game situation on the following game process is required to be considered.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a federal learning privacy policy selection method based on a game theory, which provides thresholds with different service costs for participants through a server, wherein the participants select an optimal threshold according to whether the thresholds meet the service quality, the privacy revealing cost and the like, and the server updates the service cost in the next iterative training; the server obtains optimal model parameters through multiple iterations, so that a long-term stable service state of the model is maintained, and the model is provided for the participants.
The aim of the invention is achieved by the following technical scheme:
a federal learning privacy policy selection method based on game theory, the method comprising: the server provides the participants with model quality trade-off parameters lambda with different service costs MQ Trade-off parameter lambda with privacy strength PI Ratio lambda of (2) MQ /λ PI The participators select the optimal threshold according to whether the service quality and the privacy revealing cost are met or not, train the model by using own data sets, and send the selected threshold and model parameters updated by training to the server; the server collects the model parameters of each participant to further train the model, obtain optimized model parameters, and update the service threshold lambda MQ /λ PI And the service cost is sent to the participators, so that the server keeps the long-term stable service state of the model and provides the service state for the participators.
Further, in each iteration, the server calculates the utility of each participant i And utility average for all participantsutility,utility=∑utility i N; utility availability when a party is i Greater than or equal to utility average of overall participantsutilityWhen the method is used for the next iteration, the strategy of the method is not changed; otherwise, the participant reselects the threshold lambda at the next iteration MQ /λ PI 。
Further, when the participants adopt the low-quality data training model, no rewards are obtained, namely, rewards are 0, and when the participants adopt the low-quality data training model, rewards are obtained, wherein the rewards are as follows:
where x is the number of iterations of the evolution game, b is the revenue awarded to the participants by the server, qos (ρ, r) =1- (1- ρ) r Is a modelLocal QoS contribution function, ρ is global QoS contribution, r x To select the number of participants for the high quality threshold in the xth iteration, r fix Qos for the value of outdated data in each round of new training expect P is the expected value of global service quality, pdr (sigma, r) is the contribution function of the privacy of the participants, sigma is the contribution of the privacy of the participants, Z is the probability of the participants selecting a high quality threshold, and N is the total number of the participants.
Further, the server finally stabilizes the whole evolution game model at a probability Z by adjusting the budget b of rewards; the probability Z refers to the probability of searching for a high quality threshold selected by a participant in the process of training and updating the model, so as to ensure the long-term service quality of the model; through multiple iterations, finally, the participants trained by the whole service select a high-quality data training model with the probability of Z, and select a low-quality data training model with the probability of 1-Z, wherein the calculation method of Z is as follows: z=utility/utility i 。
Further, utility of the participant i The data quality is distinguished according to the following specific calculation formula:
where j represents the level of high quality data, reorder (x) represents the rewards per participant per iteration round, and cost (x) is the service cost paid out per participant per iteration round.
Further, the method comprises the following steps:
step S1: by using different thresholds lambda MQ /λ PI Training a model and for each threshold lambda MQ /λ PI Setting different service fees; wherein lambda is MQ Lambda is a trade-off parameter for model quality PI A trade-off parameter for privacy intensity;
step S2: the server sends the trained models, different thresholds, corresponding service fees and the like to each participant for selection;
step S3: each participant selects a proper threshold lambda according to the requirements of the participants on the service quality, the privacy protection strength and the like MQ /λ PI And pays the service fee corresponding to the server;
step S4: after the participants finish the selection, using a local data set training model, and sending the threshold value, the corresponding cost and the updated parameters to a server;
step S5: the server gathers and adjusts the model parameters of each participant, trains the model again, and resets the service charge according to the selection condition of the threshold value;
step S6: the server selects a threshold lambda for each participant MQ /λ PI Compared with the threshold range set by the server, when the threshold lambda is selected by the participant MQ /λ PI If the service quality is not within the threshold range set by the server, the participant is considered to select low service quality, and no rewards are obtained, and the benefit is 0; when the participant selects the threshold lambda MQ /λ PI When within the threshold set by the server, the participant is considered to be selecting a high quality of service,
step S7: the server uses the utility of each participant i Calculating utility average value of overall participants of the iterationutility,utility=∑utility i N, and send to each participant, calculate the utility of the participant i The following are provided:
step S8: repeating the steps S2 to S7, and keeping the threshold selection of the current iteration unchanged when the utility of the participant is greater than or equal to the average utility value of the population; otherwise, the participant needs to reselect the threshold.
The beneficial effects of the invention are as follows:
(1) Good participant privacy protection effect: the method effectively avoids the problem of privacy leakage of data of the model participants, takes the federal learning training aggregation update model as a scene, uses the evolutionary game in the game theory, ensures the privacy of the participants through the selection of the participants, and obtains long-term high-quality service.
(2) Good server long-term service effect: according to the invention, malicious behaviors such as 'taking a car for convenience' of a participant are effectively avoided, the federal learning training aggregation update model is taken as a scene, the evolution game in the game theory is used, and high-quality data is obtained through the selection of the participant, so that the aggregation update model is trained, and long-term high-quality service is provided for the participant.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flow chart of a federal learning privacy policy selection method based on game theory of the present invention.
Fig. 2 is a schematic diagram of the interaction of a server and a participant in the method of the invention.
Detailed Description
Other advantages and effects of the present invention will become readily apparent to those skilled in the art from the following disclosure, when considered in light of the accompanying drawings, by describing embodiments of the present invention with specific embodiments thereof. The invention may be practiced or carried out in other embodiments and details within the scope and range of equivalents of the various features and advantages of the invention.
As shown in fig. 1-2, a flow chart of a federal learning privacy policy selection method based on game theory according to an embodiment of the present invention includes the following steps:
step S1: by using different thresholds lambda MQ /λ PI Training a model and for each threshold lambda MQ /λ PI Setting different service fees; which is a kind ofIn lambda, lambda MQ Lambda is a trade-off parameter for model quality PI A trade-off parameter for privacy intensity;
step S2: the server sends the trained models, different thresholds, corresponding service fees and the like to each participant for selection;
step S3: each participant selects a proper threshold lambda according to the requirements of the participants on the service quality, the privacy protection strength and the like MQ /λ PI And pays the service fee corresponding to the server;
step S4: after the participants finish the selection, using a local data set training model, and sending the threshold value, the corresponding cost and the updated parameters to a server;
step S5: the server gathers and adjusts the model parameters of each participant, trains the model again, and resets the service charge according to the selection condition of the threshold value;
step S6: the server selects a threshold lambda for each participant MQ /λ PI Compared with the threshold range set by the server, when the threshold lambda is selected by the participant MQ /λ PI If the service quality is not within the threshold range set by the server, the participant is considered to select low service quality, and no rewards are obtained, and the benefit is 0; when the participant selects the threshold lambda MQ /λ PI When within the threshold set by the server, the participant is considered to be selecting a high quality of service,
step S7: the server uses the utility of each participant i Calculating utility average value of overall participants of the iterationutility,utility=∑utility i N, and send to each participant;
where j represents the level of high quality data, x is the number of iterations of the evolutionary game, reward (x) represents the awards of the iterative participants of the x-th round, cost (x) is the expense of the iterative participants of the x-th round, and b is the benefit awarded to the participants by the server,qos(ρ,r)=1-(1-ρ) r As a function of the contribution degree of the global service quality of the model, ρ is the contribution degree of the global service quality, r x To select the number of participants for the high quality threshold in the xth iteration, r fix Qos for the value of outdated data in each round of new training expect P is the expected value of global service quality, pdr (sigma, r) is the contribution function of the privacy of the participants when privacy leakage occurs, sigma is the contribution degree of the privacy of the participants, Z is the probability of the participants selecting a high quality threshold value, and N is the total number of the participants;
step S8: repeating the steps S2 to S7, and keeping the threshold selection of the current iteration unchanged when the utility of the participant is greater than or equal to the average utility value of the population; otherwise, the participant needs to reselect the threshold. The server eventually stabilizes the entire evolutionary game model at a probability Z by adjusting the budget b of the rewards.
Due to p, ρ, σ, qos of the respective participants expect 、r x-1 、r fix Parameters such as (because the amount of time-shifted data of the obsolete data is statistically stable), the self utility can be easily calculated if they get the number of participants to choose a high quality threshold. However, due to incomplete information and assumptions of limited rational participants, they cannot get the choice of other participants until the iteration is completed.
One embodiment of the application of the present invention in a smart medical scenario is as follows.
In the smart medical scenario, the diagnosis data of the hospital server and each participant trains a system suitable for parkinsonism monitoring through federal learning, so as to realize the diagnosis of on-line parkinsonism. The specific implementation process is as follows:
step S1: the hospital server trains the model by using different thresholds (the trade-off weight ratio of the parkinsonism monitoring system service quality and the participant privacy classifier), and at the same time, each threshold is set by the hospital server to different service fees;
step S2: the hospital server sends the trained model and different thresholds and corresponding service fees to each participant for selection and training;
step S3: each participant selects a threshold according to whether the service effect of parkinsonism monitoring, the privacy protection intensity of the participant and the like are met, and pays the service fee corresponding to the hospital server;
step S4: each participant uses own data set training model, and sends the selected threshold value and the model parameters after training update to the server;
step S5: the hospital server retrains and updates the parkinsonism system according to the threshold value selected by each participant and the submitted parameters, and resets the service fee of each threshold value;
step S6: the hospital server compares the threshold selected by the participant with the threshold range set by the server, and when the threshold selected by the participant is not in the threshold range set by the server, the hospital server considers that the participant selects low service quality, and rewards are not obtained, and the benefit is 0; when the threshold selected by the participant is within the threshold range set by the server, the participant is considered to be selected by high service quality;
step S7: the hospital server calculates the utility of each participant i And ensemble averaging utility for this iterationutilityAnd send to each participant;
step S8: repeating S2-S7, and obtaining the utility of the party i Greater than or equal to utility averageutilityWhen the current iteration threshold is selected as the previous iteration, the utility of the participator is kept the same i Less than the utility averageutilityAt this time, the threshold is reselected for training.
In a real scenario, since the federally learning-based application is a long-term service, the model for concept drift should be updated continuously, but long-term service cannot be provided without a certain number of participants providing high quality data. Therefore, the modeling of the invention is an evolution game mode, and the evolution stability strategy (which means the strategy adopted by most participants is different from other strategies) is found through repeated iteration of the participants and the server, so that the stable service state of the model is ensured, the service cost is obtained to the maximum extent, and long-term benefits are brought to the participants.
It will be appreciated by persons skilled in the art that the foregoing description is a preferred embodiment of the invention, and is not intended to limit the invention, but rather to limit the invention to the specific embodiments described, and that modifications may be made to the technical solutions described in the foregoing embodiments, or equivalents may be substituted for elements thereof, for the purposes of those skilled in the art. Modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (3)
1. A federal learning privacy policy selection method based on game theory is characterized by comprising the following steps: the server provides the participants with model quality trade-off parameters lambda with different service costs MQ Trade-off parameter lambda with privacy strength PI Ratio lambda of (2) MQ /λ PI The participators select the optimal threshold according to whether the service quality and the privacy revealing cost are met or not, train the model by using own data sets, and send the selected threshold and model parameters updated by training to the server; the server collects the model parameters of each participant to further train the model, obtain optimized model parameters, and update the service threshold lambda MQ /λ PI And the service cost is sent to the participators, so that the server keeps the long-term stable service state of the model and provides the service state for the participators;
in each iteration, the server calculates the utility of each participant i And utility average for all participantsutility,utility=∑utility i N; utility availability when a party is i Greater than or equal to utility average of overall participantsutilityWhen the method is used for the next iteration, the strategy of the method is not changed; otherwise, the participant reselects the threshold lambda at the next iteration MQ /λ PI ;
When the participants adopt the low-quality data training model, rewards are not obtained, namely, rewards are 0, and when the participants adopt the low-quality data training model, rewards are obtained, wherein the rewards are as follows:
where x is the number of iterations of the evolution game, b is the revenue awarded to the participants by the server, qos (ρ, r) =1- (1- ρ) r As a function of the contribution degree of the global service quality of the model, ρ is the contribution degree of the global service quality, r x To select the number of participants for the high quality threshold in the xth iteration, r fix Qos for the value of outdated data in each round of new training expect P is the expected value of global service quality, pdr (sigma, r) is the contribution function of the privacy of the participants when privacy leakage occurs, sigma is the contribution degree of the privacy of the participants, Z is the probability of the participants selecting a high quality threshold value, and N is the total number of the participants;
the server adjusts the budget b of rewards to enable the whole evolution game model to be finally stabilized at a probability Z; the probability Z refers to the probability of searching for a high quality threshold selected by a participant in the process of training and updating the model, so as to ensure the long-term service quality of the model; through multiple iterations, finally, the participants trained by the whole service select a high-quality data training model with the probability of Z, and select a low-quality data training model with the probability of 1-Z, wherein the calculation method of Z is as follows: z=utility/utility i 。
2. The federal learning privacy policy selection method based on game theory of claim 1, wherein: utility of the participant i The data quality is distinguished according to the following specific calculation formula:
where j represents the level of high quality data, reorder (x) represents the rewards per participant per iteration round, and cost (x) is the service cost paid out per participant per iteration round.
3. The federal learning privacy policy selection method based on game theory according to claim 1, wherein the method comprises the steps of:
step S1: by using different thresholds lambda MQ /λ PI Training a model and for each threshold lambda MQ /λ PI Setting different service fees; wherein lambda is MQ Lambda is a trade-off parameter for model quality PI A trade-off parameter for privacy intensity;
step S2: the server sends the trained models, different thresholds, corresponding service fees and the like to each participant for selection;
step S3: each participant selects a proper threshold lambda according to the requirements of the participants on the service quality, the privacy protection strength and the like MQ /λ PI And pays the service fee corresponding to the server;
step S4: after the participants finish the selection, using a local data set training model, and sending the threshold value, the corresponding cost and the updated parameters to a server;
step S5: the server gathers and adjusts the model parameters of each participant, trains the model again, and resets the service charge according to the selection condition of the threshold value;
step S6: the server selects a threshold lambda for each participant MQ /λ PI Compared with the threshold range set by the server, when the threshold lambda is selected by the participant MQ /λ PI If the service quality is not within the threshold range set by the server, the participant is considered to select low service quality, and no rewards are obtained, and the benefit is 0; when the participant selects the threshold lambda MQ /λ PI When the server sets the threshold value range, consider that the participant selects high service quality;
step S7: the server uses the utility of each participant i Calculating utility average value of overall participants of the iterationutility,utility=∑utility i N, and send to each participant;
calculating utility of the party i The following are provided:
step S8: repeating the steps S2 to S7, and keeping the threshold selection of the current iteration unchanged when the utility of the participant is greater than or equal to the average utility value of the population; otherwise, the participant needs to reselect the threshold.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110292473.3A CN112926088B (en) | 2021-03-18 | 2021-03-18 | Federal learning privacy policy selection method based on game theory |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110292473.3A CN112926088B (en) | 2021-03-18 | 2021-03-18 | Federal learning privacy policy selection method based on game theory |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112926088A CN112926088A (en) | 2021-06-08 |
CN112926088B true CN112926088B (en) | 2024-03-19 |
Family
ID=76175100
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110292473.3A Active CN112926088B (en) | 2021-03-18 | 2021-03-18 | Federal learning privacy policy selection method based on game theory |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112926088B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019056572A1 (en) * | 2017-09-25 | 2019-03-28 | 深圳大学 | Model-based collaborative filtering method for collaborative web quality-of-service prediction for privacy protection |
CN111611610A (en) * | 2020-04-12 | 2020-09-01 | 西安电子科技大学 | Federal learning information processing method, system, storage medium, program, and terminal |
CN111754000A (en) * | 2020-06-24 | 2020-10-09 | 清华大学 | Quality-aware edge intelligent federal learning method and system |
CN112257063A (en) * | 2020-10-19 | 2021-01-22 | 上海交通大学 | Cooperative game theory-based detection method for backdoor attacks in federal learning |
-
2021
- 2021-03-18 CN CN202110292473.3A patent/CN112926088B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019056572A1 (en) * | 2017-09-25 | 2019-03-28 | 深圳大学 | Model-based collaborative filtering method for collaborative web quality-of-service prediction for privacy protection |
CN111611610A (en) * | 2020-04-12 | 2020-09-01 | 西安电子科技大学 | Federal learning information processing method, system, storage medium, program, and terminal |
CN111754000A (en) * | 2020-06-24 | 2020-10-09 | 清华大学 | Quality-aware edge intelligent federal learning method and system |
CN112257063A (en) * | 2020-10-19 | 2021-01-22 | 上海交通大学 | Cooperative game theory-based detection method for backdoor attacks in federal learning |
Non-Patent Citations (1)
Title |
---|
博弈论隐私保护方法研究综述;周丹丹;李威伟;孙宇清;;小型微型计算机系统(第12期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN112926088A (en) | 2021-06-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107194723B (en) | Bidirectional matching recommendation method for borrowing items and lenders in network micropayment | |
CN107426621B (en) | A kind of method and system showing any active ues image in mobile terminal direct broadcasting room | |
JP2016048535A (en) | Method and apparatus for prediction based on multi-source heterogeneous data | |
CN111432361B (en) | User incentive strategy determination method and system based on crowd sensing network | |
CN111292001B (en) | Combined decision method and device based on reinforcement learning | |
CN115345317B (en) | Fair reward distribution method facing federal learning based on fairness theory | |
CN113724096B (en) | Group knowledge sharing method based on public evolution game model | |
Panagopoulos et al. | Modeling and evaluating a robust feedback-based reputation system for e-commerce platforms | |
CN111565188B (en) | VANET trust model working method based on combination of message type and trust value confidence | |
CN111861661A (en) | Electric vehicle charging transaction method and device | |
Huang et al. | Crowdsourcing with heterogeneous workers in social networks | |
Cai et al. | 2cp: Decentralized protocols to transparently evaluate contributivity in blockchain federated learning environments | |
CN112926088B (en) | Federal learning privacy policy selection method based on game theory | |
CN113298668B (en) | Mobile crowd-sourcing aware user large-scale rapid recruitment method considering social network | |
Tian et al. | Stick or carrot for traffic demand management? Evidence from experimental economics | |
CN112101528B (en) | Terminal contribution measurement method based on back propagation | |
CN114116705A (en) | Method and device for determining contribution value of participants in joint learning | |
CN116451806A (en) | Federal learning incentive distribution method and device based on block chain | |
Nayak et al. | Dynamic advertising in VANETs using repeated auctions | |
Zhang et al. | The evolution of cooperation in public goods games on the scale-free community network under multiple strategy-updating rules | |
CN111030764A (en) | Crowdsourcing user information age management algorithm based on random game online learning | |
Wu et al. | Multi-agent Bayesian Learning with Best Response Dynamics: Convergence and Stability | |
CN117114126B (en) | Web3.0 federal learning cloud architecture and excitation method | |
Garcia Alvarado | Network-based policies versus tax evasion | |
Lv et al. | MODELING and simulation of live streaming e-commerce information dissemination considering opinion leader in social networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |