CN110008696A - A kind of user data Rebuilding Attack method towards the study of depth federation - Google Patents

A kind of user data Rebuilding Attack method towards the study of depth federation Download PDF

Info

Publication number
CN110008696A
CN110008696A CN201910249056.3A CN201910249056A CN110008696A CN 110008696 A CN110008696 A CN 110008696A CN 201910249056 A CN201910249056 A CN 201910249056A CN 110008696 A CN110008696 A CN 110008696A
Authority
CN
China
Prior art keywords
user
model
data
study
federation
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
Application number
CN201910249056.3A
Other languages
Chinese (zh)
Inventor
王志波
宋梦凯
郑思言
王骞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN201910249056.3A priority Critical patent/CN110008696A/en
Publication of CN110008696A publication Critical patent/CN110008696A/en
Pending legal-status Critical Current

Links

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
    • G06F21/50—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems
    • G06F21/55—Detecting local intrusion or implementing counter-measures
    • G06F21/552—Detecting local intrusion or implementing counter-measures involving long-term monitoring or reporting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
    • G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
    • G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computer Security & Cryptography (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Bioethics (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a kind of user data Rebuilding Attack methods towards the study of depth federation, it compares previous attack method and can only rebuild classification and represent data, this method can rebuild the private data of specific user, and considers to implement to attack by malicious service end, be negatively affected so as to avoid being introduced to former Share Model.In addition, this method, which describes a kind of multitask, generates the distribution that confrontation model carrys out analog subscriber data, which is trained by the differentiation of authenticity, classification and owning user identity to input sample, promotes the sample quality of generation.In order to preferably distinguish different user, this method introduces a kind of user data based on optimization and represents calculation method to portray the user characteristics for participating in federal study, to prefect at the training of confrontation model.For the federal study framework of existing concern secret protection, the data reconstruction attack proposed by the present invention for generating confrontation model based on multitask can cause it to reveal privacy.

Description

A kind of user data Rebuilding Attack method towards the study of depth federation
Technical field
The present invention relates to a kind of user data Rebuilding Attack methods towards the study of depth federation, belong to artificial intelligence safety Field.
Background technique
In recent years, depth learning technology is increasingly used in network field, such as knows that the study of perception is appointed in conjunction with group Business.Crowdsourcing data are being locally stored using traditional centralization training method needs, this would generally bring large-capacity data transmission, The problems such as high computational requirements and privacy compromise.Therefore, the mobile edge calculations frame as a deep learning, synergetics It practises frame to receive significant attention and study, multiparty data source co-benefit can be made to be total in what is obtained by the training of all data Model is enjoyed, and data need not be uploaded in central store.
Federal learning framework is one of Cooperative Study frame of current mainstream: each data source (i.e. user) just learns first Target and model framework are reached an agreement, and server-side initializes a shared learning model, and users are locally with respective hidden The private data set training learning model, server-side are collected model/parameter after training and are updated to update globally shared model.Then World model is issued to each user again, this local training process of iteration terminates until training.The advantage of federation's study It is, avoids explicit access of the center service end to training data, improve the secret protection degree of deep learning.
Recent studies have shown that Cooperative Study is subject to by inference attack and leak data privacy, as Rebuilding Attack and at Member's inference attack.This is because the update of Share Model is derived from these private data collection, data pattern is " encoded " as ginseng Number, which updates, to be uploaded, as soon as therefore construct corresponding " decoder " as attacker, private data can be resumed.Existing attacker Method usually considers that malicious user implements attack to a target, thus it is speculated that its data is typically only capable to deduce classification generation with privacy Table data, such as a face identification mission, existing attack method can speculate the general sample of some corresponding ID, but nothing Method reconstructs the sample for the ID that specific user is possessed, and the general sample of even now can characterize such another characteristic, but It actually can not really cause privacy compromise.This is because updated model can only be accessed in malicious user from server-side, This is to polymerize to obtain by the last round of update of total user, can not implement attack for specific objective.In addition, existing attacker Method requires the structure of modification Share Model to implement to attack, and malice influence is introduced into conventional Cooperative Study process, this The hypothesis of sample has exceeded the attacking ability of a malicious user, and will cause the reduced performance of Share Model, is easy by system Discover, that is, the concealment attacked is poor.
It is considered herein that the existing data reconstruction attack method towards federation's study is established in too strong attacker's ability It on assuming that, can not implement in actual scene, and orientation, valuable privacy information cannot be inferred, therefore expedite the emergence of out A kind of data reconstruction attack for user class private data based on malicious service end.
Summary of the invention
The purpose of the present invention is overcome the deficiencies of the prior art and provide a kind of user that confrontation network is generated based on multitask Data reconstruction attack method.
User data Rebuilding Attack method towards the study of depth federation comprises the following steps:
1) malicious service end participates in conventional federal learning process, and user first reaches the target of Cooperative Study, model Unanimously, then iteration executes: server-side issues Share Model, and user locally trains this model and model parameter is uploaded to clothes Business end, server-side polymerize the update of these parameters again.Until model is restrained；
2) malicious service end is locally constructing a multitask generation confrontation network model, which includes a generation mould A type G and discrimination model D, the differentiation that wherein D is performed simultaneously the authenticity to input sample, classification and owning user identity are appointed Business, G is for synthesizing user data；
3) parameter that malicious service end is submitted according to user, which is more newly arrived, calculates the data representative of different user, user data generation Table is used to supervise the sample owning user identity task training of D model；
4) malicious service end is more newly arrived using the parameter of target user and other users in every wheel iterative process and substitutes D The training of the category task of model；
5) malicious service end utilizes local additional real data set, the false data collection of G synthesis, the sample of Lai Xunlian D model This authenticity differentiates task.By the dual training of G and D, G is enabled to generate the private data of specific user.
A kind of user data Rebuilding Attack method towards the study of depth federation, malicious service end is considered as " honest and curious ", this shows that attacker can go to execute conventional learning tasks according to the rule of federation's study.Specifically, It is assumed that N number of user respectively possesses a private data collection, these data are not usually independent same distribution in actual scene.User To the target of federation's study, after model structure reaches an agreement, distributed training is realized via malicious service end.Specifically, The update of Share Model each round can be expressed as
MtIndicate that t takes turns updated Share Model,It indicates that parameter of the t wheel from user k updates, is existed by user k Locally by private data to MtIt is calculated.
A kind of user data Rebuilding Attack method towards the study of depth federation, attacker establish one in server-side A multitask generates confrontation network model, and model includes an a generation model G and discrimination model D.D is used to complete three Task: 1) the true and false differentiation task of sample；2) classification task of sample；3) the owning user identity of sample differentiates task.It compares Confrontation network model is generated in standard, the present invention extraly differentiates identity as condition entry to G model, it is therefore an objective to generate special Determine the privacy sample of user.
A kind of user data Rebuilding Attack method towards the study of depth federation, the structure of D model can indicate For
Dreal=Sigmoid (FCreal(Lshare))
Dcat=Sigmoid (FCcat(Lshare))
Did=Sigmoid (FCid(Lshare))
Dreal, Dcat, DidRespectively indicate three above-mentioned tasks, FCreal, FCcatAnd FCidIt is corresponding to respectively indicate three tasks The full articulamentum of neural network, LshareIndicate the layer structure in addition to the last layer of Share Model.Sigmoid is activation primitive.
A kind of user data Rebuilding Attack method towards the study of depth federation, G model receive classification and user Identity label and random noise synthesize the sample of specific user.The training that multitask generates confrontation network can be expressed as
X~pvicttmWith x~potherThe sample sampled with victim user and other users is respectively indicated, CE indicates cross entropy Loss function.Attacker can not directly acquire the private data of user, each use of approximate substitution of more newly arriving uploaded by user The data at family represent.
A kind of user data Rebuilding Attack method towards the study of depth federation, attacker uploads according to user to join The data of number building different user represent, data represent can line flag be: it is total as private data training when using data to represent When enjoying model, parameter similar with true private data can be generated and updated, be expressed as
Indicate Share Model MtParameter, γ be scale factor be used to balanceMagnitude, XkIndicate different user Data represent.
A kind of user data Rebuilding Attack method towards the study of depth federation, in order to avoid the calculation based on optimization Method introduces excessive noise in calculating user data representative, and local difference summation is added into as regular terms, is expressed as
xijIndicate pixel value of the image x at (i, j).The distance of the adjacent pixels of image is calculated, realizes and generates figure Picture it is smooth
A kind of user data Rebuilding Attack method towards the study of depth federation, attacker is in each round to G, D The training of model is expressed as
η1And η2Indicate learning rate, θDAnd θGRespectively indicate the parameter of model D and G.
Compared to the prior art the present invention, has the beneficial effect that and 1) represents with existing towards reconstruction categorical data Aggressive mechanism it is different, the present invention can implement attack for specific user in federal study, and rebuild its private data, cause Privacy compromise.Since previous mechanism can only reconstruct the characterization sample of some classification, without similar with any authentic specimen, because And substantially without causing privacy compromise.2) present invention introduces a kind of multitasks to generate confrontation network model, and wherein arbiter is same Shi Zhihang is to the differentiation task of the authenticity of sample, classification and owning user identity, and generator is for synthesizing user data.It improves The authenticity of synthesis sample.3) present invention introduces a kind of class of subscriber method for computing data, so as to depict different use The feature of user data collection, supervision confrontation network model can synthesize the data of specific user.4) compared to previous attack pattern It is required that attacker can modify the structure of Share Model and learn to introduce negative effect to federation, the present invention is not influence often Implement attack under the premise of advising study course, improves the concealment of attack.
Detailed description of the invention
Fig. 1 is that the data reconstruction based on malicious service end towards federation's study attacks frame diagram.
Fig. 2 is the model structure that multitask generates antagonism network.
Fig. 3 is the reconstruction effect picture of the invention towards MNIST data set.
Fig. 4 is the reconstruction effect picture of the invention towards AT&T data set.
Fig. 5 is present invention figure compared with the reconstruction effect of existing attack method.
Specific embodiment
User data Rebuilding Attack method towards the study of depth federation comprises the following steps:
1) malicious service end participates in conventional federal learning process, and user first reaches the target of Cooperative Study, model Unanimously, then iteration executes: server-side issues Share Model, and user locally trains this model and model parameter is uploaded to clothes Business end, server-side polymerize the update of these parameters again.Until model is restrained；
2) malicious service end is locally constructing a multitask generation confrontation network model, which includes a generation mould A type G and discrimination model D, the differentiation that wherein D is performed simultaneously the authenticity to input sample, classification and owning user identity are appointed Business, G is for synthesizing user data；
3) parameter that malicious service end is submitted according to user, which is more newly arrived, calculates the data representative of different user, user data generation Table is used to supervise the sample owning user identity task training of D model；
4) malicious service end is more newly arrived using the parameter of target user and other users in every wheel iterative process and substitutes D The training of the category task of model；
5) malicious service end utilizes local additional real data set, the false data collection of G synthesis, the sample of Lai Xunlian D model This authenticity differentiates task.By the dual training of G and D, G is enabled to generate the private data of specific user.
A kind of user data Rebuilding Attack method towards the study of depth federation, malicious service end is considered as " honest and curious ", this shows that attacker can go to execute conventional learning tasks according to the rule of federation's study.Specifically, It is assumed that N number of user respectively possesses a private data collection, these data are not usually independent same distribution in actual scene.User To the target of federation's study, after model structure reaches an agreement, distributed training is realized via malicious service end.Specifically, The update of Share Model each round can be expressed as
MtIndicate that t takes turns updated Share Model,It indicates that parameter of the t wheel from user k updates, is existed by user k Locally by private data to MtIt is calculated.
A kind of user data Rebuilding Attack method towards the study of depth federation, attacker establish one in server-side A multitask generates confrontation network model, and model includes an a generation model G and discrimination model D.D is used to complete three Task: 1) the true and false differentiation task of sample；2) classification task of sample；3) the owning user identity of sample differentiates task.It compares Confrontation network model is generated in standard, the present invention extraly differentiates identity as condition entry to G model, it is therefore an objective to generate special Determine the privacy sample of user.
A kind of user data Rebuilding Attack method towards the study of depth federation, the structure of D model can indicate For
Dreal=Sigmoid (FCreal(Lshare))
Dcat=Sigmoid (FCcat(Lshare))
Did=Sigmoid (FCid(Lshare))
Dreal, Dcat, DidRespectively indicate three above-mentioned tasks, FCreal, FCcatAnd FCidIt is corresponding to respectively indicate three tasks The full articulamentum of neural network, LshareIndicate the layer structure in addition to the last layer of Share Model.Sigmoid is activation primitive.
A kind of user data Rebuilding Attack method towards the study of depth federation, G model receive classification and user Identity label and random noise synthesize the sample of specific user.The training that multitask generates confrontation network can be expressed as
X~pvictimWith x~potherThe sample sampled with victim user and other users is respectively indicated, CE indicates cross entropy Loss function.Attacker can not directly acquire the private data of user, each use of approximate substitution of more newly arriving uploaded by user The data at family represent.
A kind of user data Rebuilding Attack method towards the study of depth federation, attacker uploads according to user to join The data of number building different user represent, data represent can line flag be: it is total as private data training when using data to represent When enjoying model, parameter similar with straight reality private data can be generated and updated, be expressed as
Indicate Share Model MtParameter, γ be scale factor be used to balanceMagnitude, XkIndicate different user Data represent.
A kind of user data Rebuilding Attack method towards the study of depth federation, in order to avoid the calculation based on optimization Method introduces excessive noise in calculating user data representative, and local difference summation is added into as regular terms, is expressed as
xijIndicate pixel value of the image x at (i, j).The distance of the adjacent pixels of image is calculated, realizes and generates figure Picture it is smooth
A kind of user data Rebuilding Attack method towards the study of depth federation, attacker is in each round to G, D The training of model is expressed as
η1And η2Indicate learning rate, θDAnd θGRespectively indicate the parameter of model D and G.
Embodiment 1
1) malicious service end participates in conventional federal learning process, and user first reaches the target of Cooperative Study, model Unanimously, then iteration executes: server-side issues Share Model, and user locally trains this model and model parameter is uploaded to clothes Business end, server-side polymerize the update of these parameters again, until model is restrained.The update of each round can be expressed as
MtIndicate that t takes turns updated Share Model,It indicates that parameter of the t wheel from user k updates, is existed by user k Locally by private data to MtIt is calculated.
2) malicious service end is locally constructing a multitask generation confrontation network model, which includes a generation mould A type G and discrimination model D, the differentiation that wherein D is performed simultaneously the authenticity to input sample, classification and owning user identity are appointed Business, the structure of model can be expressed as
Dreal=Sigmoid (FCreal(Lshare))
Dcat=Sigmoid (FCcat(Lshare))
Did=Sigmoid (FCid(Lshare))
Dreal, Dcat, DidRespectively indicate three above-mentioned tasks, FCreal, FCcatAnd FCidIt is corresponding to respectively indicate three tasks The full articulamentum of neural network, LshareIndicate the layer structure in addition to the last layer of Share Model.Sigmoid is activation primitive.
3) parameter that malicious service end is submitted according to user, which is more newly arrived, calculates the data representative of different user, user data generation Table is used to supervise the sample owning user identity task training of D model, is expressed as
Indicate Share Model MtParameter, γ be scale factor be used to balanceMagnitude, XkIndicate different user Data represent.
4) malicious service end is more newly arrived using the parameter of target user and other users in every wheel iterative process and substitutes D The training of the category task of model, is expressed as
Wherein MtIndicate that t takes turns updated Share Model,Indicate that parameter of the t wheel from user k updates.
5) malicious service end utilizes local additional real data set, the false data collection of G synthesis, the sample of Lai Xunlian D model This authenticity differentiates task.By the dual training of G and D, G is enabled to generate the private data of specific user, each round is to G, D The training of model is expressed as
η1And η2Indicate learning rate, θDAnd θGRespectively indicate the parameter of model D and G.
Specific embodiment described herein is only an example for the spirit of the invention.The neck of technology belonging to the present invention The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.

Claims (8)

1. a kind of user data Rebuilding Attack method towards the study of depth federation, which is characterized in that comprise the following steps:
Step 1, malicious service end participate in conventional federal learning process, and user first reaches the target of Cooperative Study, model Unanimously, then iteration executes: server-side issues Share Model, and user locally trains this model and model parameter is uploaded to clothes Business end, server-side polymerize the update of these parameters again；Until model is restrained；
Step 2, malicious service end are locally constructing a multitask generation confrontation network model, which includes a generation mould A type G and discrimination model D, the differentiation that wherein D is performed simultaneously the authenticity to input sample, classification and owning user identity are appointed Business, G is for synthesizing user data；
The parameter that step 3, malicious service end are submitted according to user, which is more newly arrived, calculates the data representative of different user, user data generation Table is used to supervise the sample owning user identity task training of D model；
Step 4, malicious service end are more newly arrived using the parameter of target user and other users in every wheel iterative process and substitute D The training of the category task of model；
Step 5, malicious service end utilize local additional real data set, the false data collection of G synthesis, the sample of Lai Xunlian D model This authenticity differentiates task；By the dual training of G and D, G is enabled to generate the private data of specific user.
2. a kind of user data Rebuilding Attack method towards the study of depth federation as described in claim 1, which is characterized in that Malicious service end is considered as " honest and curious ", this shows that attacker can go to execute routine according to the rule of federation's study Learning tasks；Specifically, assuming that N number of user respectively possesses a private data collection, these data are usual in actual scene It is not independent same distribution；After users reach an agreement to target, the model structure of federation's study, realized via malicious service end Distribution training；Specifically, the update of Share Model each round can be expressed as
MtIndicate that t takes turns updated Share Model,Indicate that parameter of the t wheel from user k updates, by user k in local By private data to MtIt is calculated.
3. a kind of user data Rebuilding Attack method towards the study of depth federation as described in claim 1, which is characterized in that Attacker establishes a multitask in server-side and generates confrontation network model, and model includes a generation model G and a differentiation Model D；D is used to complete three tasks: 1) the true and false differentiation task of sample；2) classification task of sample；3) the affiliated use of sample Family identity differentiates task；Confrontation network model is generated compared to standard, the present invention, which extraly differentiates identity, is used as condition entry To G model, it is therefore an objective to generate the privacy sample of specific user.
4. a kind of user data Rebuilding Attack method towards the study of depth federation as described in claim 1, which is characterized in that The structure of D model can be expressed as
Dreal=Sigmoid (FCreal(Lshare))
Dcat=Sigmoid (FCcat(Lshare))
Did=Sigmoid (FCid(Lshare))
Dreal, Dcat, DidRespectively indicate three above-mentioned tasks, FCreal, FCcatAnd FCidRespectively indicate the corresponding mind of three tasks Through the full articulamentum of network, LshareIndicate the layer structure in addition to the last layer of Share Model；Sigmoid is activation primitive.
5. a kind of user data Rebuilding Attack method towards the study of depth federation as described in claim 1, which is characterized in that G model receives classification and user identity label and random noise, synthesizes the sample of specific user；Multitask generates confrontation net The training of network can be expressed as
X~pvictimWith x~potherThe sample sampled with victim user and other users is respectively indicated, CE indicates to intersect entropy loss Function；Attacker can not directly acquire the private data of user, approximate substitution of more the newly arriving each user uploaded by user Data represent.
6. a kind of user data Rebuilding Attack method towards the study of depth federation as described in claim 1, which is characterized in that Attacker represents according to the data that user's upload parameter constructs different user, data represent can line flag be: when with data generation When table is as private data training Share Model, parameter similar with true private data can be generated and updated, be expressed as
Indicate Share Model MtParameter, γ be scale factor be used to balanceMagnitude, XkIndicate the data of different user It represents.
7. a kind of user data Rebuilding Attack method towards the study of depth federation as described in claim 1, which is characterized in that In order to avoid the algorithm based on optimization introduces excessive noise in calculating user data representative, local difference summation is as regular terms It is added into, is expressed as
xijIndicate pixel value of the image x at (i, j)；The distance of the adjacent pixels of image is calculated, realizes and generates the flat of image It is sliding.
8. a kind of user data Rebuilding Attack method towards the study of depth federation as described in claim 1, attacker is every To G, the training of D model is expressed as one wheel
η1And η2Indicate learning rate, θDAnd θGRespectively indicate the parameter of model D and G.
CN201910249056.3A 2019-03-29 2019-03-29 A kind of user data Rebuilding Attack method towards the study of depth federation Pending CN110008696A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910249056.3A CN110008696A (en) 2019-03-29 2019-03-29 A kind of user data Rebuilding Attack method towards the study of depth federation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910249056.3A CN110008696A (en) 2019-03-29 2019-03-29 A kind of user data Rebuilding Attack method towards the study of depth federation

Publications (1)

Publication Number Publication Date
CN110008696A true CN110008696A (en) 2019-07-12

Family

ID=67168881

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910249056.3A Pending CN110008696A (en) 2019-03-29 2019-03-29 A kind of user data Rebuilding Attack method towards the study of depth federation

Country Status (1)

Country Link
CN (1) CN110008696A (en)

Cited By (62)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110378749A (en) * 2019-07-25 2019-10-25 深圳前海微众银行股份有限公司 Appraisal procedure, device, terminal device and the storage medium of user data similitude
CN110443378A (en) * 2019-08-02 2019-11-12 深圳前海微众银行股份有限公司 Feature correlation analysis method, device and readable storage medium storing program for executing in federation's study
CN110443375A (en) * 2019-08-16 2019-11-12 深圳前海微众银行股份有限公司 A kind of federation's learning method and device
CN110472745A (en) * 2019-08-06 2019-11-19 深圳前海微众银行股份有限公司 Information transferring method and device in a kind of federal study
CN110569663A (en) * 2019-08-15 2019-12-13 深圳市莱法照明通信科技有限公司 Method, device, system and storage medium for educational data sharing
CN110569911A (en) * 2019-09-11 2019-12-13 深圳绿米联创科技有限公司 Image recognition method, device, system, electronic equipment and storage medium
CN110647765A (en) * 2019-09-19 2020-01-03 济南大学 Privacy protection method and system based on knowledge migration under collaborative learning framework
CN110751291A (en) * 2019-10-29 2020-02-04 支付宝(杭州)信息技术有限公司 Method and device for realizing multi-party combined training neural network of security defense
CN110942154A (en) * 2019-11-22 2020-03-31 深圳前海微众银行股份有限公司 Data processing method, device, equipment and storage medium based on federal learning
CN111079946A (en) * 2019-12-20 2020-04-28 支付宝(杭州)信息技术有限公司 Model training method, member detection device training method and system
CN111079977A (en) * 2019-11-18 2020-04-28 中国矿业大学 Heterogeneous federated learning mine electromagnetic radiation trend tracking method based on SVD algorithm
CN111091199A (en) * 2019-12-20 2020-05-01 哈尔滨工业大学（深圳） Federal learning method and device based on differential privacy and storage medium
CN111245903A (en) * 2019-12-31 2020-06-05 烽火通信科技股份有限公司 Joint learning method and system based on edge calculation
CN111260061A (en) * 2020-03-09 2020-06-09 厦门大学 Differential noise adding method and system in federated learning gradient exchange
CN111340614A (en) * 2020-02-28 2020-06-26 深圳前海微众银行股份有限公司 Sample sampling method and device based on federal learning and readable storage medium
CN111447083A (en) * 2020-03-10 2020-07-24 中国人民解放军国防科技大学 Federal learning framework under dynamic bandwidth and unreliable network and compression algorithm thereof
CN111445031A (en) * 2020-03-31 2020-07-24 深圳前海微众银行股份有限公司 Attack coping method and federal learning device
CN111460443A (en) * 2020-05-28 2020-07-28 南京大学 Security defense method for data manipulation attack in federated learning
CN111464568A (en) * 2020-06-17 2020-07-28 广东电网有限责任公司佛山供电局 Method and system for enhancing network attack prevention capability of multiple network ports
CN111581648A (en) * 2020-04-06 2020-08-25 电子科技大学 Method of federal learning to preserve privacy in irregular users
CN111598143A (en) * 2020-04-27 2020-08-28 浙江工业大学 Credit evaluation-based defense method for federal learning poisoning attack
CN111985562A (en) * 2020-08-20 2020-11-24 复旦大学 End cloud collaborative training system for protecting end-side privacy
CN112039702A (en) * 2020-08-31 2020-12-04 中诚信征信有限公司 Model parameter training method and device based on federal learning and mutual learning
CN112100659A (en) * 2020-09-14 2020-12-18 电子科技大学 Block chain federal learning system and Byzantine attack detection method
CN112101403A (en) * 2020-07-24 2020-12-18 西安电子科技大学 Method and system for classification based on federate sample network model and electronic equipment
CN112118099A (en) * 2020-09-16 2020-12-22 西安电子科技大学 Distributed multi-task learning privacy protection method and system for resisting inference attack
CN112162959A (en) * 2020-10-15 2021-01-01 深圳技术大学 Medical data sharing method and device
CN112203282A (en) * 2020-08-28 2021-01-08 中国科学院信息工程研究所 5G Internet of things intrusion detection method and system based on federal transfer learning
CN112214342A (en) * 2020-09-14 2021-01-12 德清阿尔法创新研究院 Efficient error data detection method in federated learning scene
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
CN112329009A (en) * 2020-10-12 2021-02-05 南京理工大学 Defense method for noise attack in joint learning
CN112434758A (en) * 2020-12-17 2021-03-02 浙江工业大学 Cluster-based federal learning casual vehicle attack defense method
WO2021036014A1 (en) * 2019-08-28 2021-03-04 深圳前海微众银行股份有限公司 Federated learning credit management method, apparatus and device, and readable storage medium
CN112784990A (en) * 2021-01-22 2021-05-11 支付宝(杭州)信息技术有限公司 Training method of member inference model
WO2021090142A1 (en) * 2019-11-05 2021-05-14 International Business Machines Corporation Intelligent agent to simulate customer data
CN112819180A (en) * 2021-01-26 2021-05-18 华中科技大学 Multi-service data generation method and device based on federal generation model
CN112949670A (en) * 2019-12-10 2021-06-11 京东数字科技控股有限公司 Data set switching method and device for federal learning model
CN113051608A (en) * 2021-03-11 2021-06-29 佳讯飞鸿（北京）智能科技研究院有限公司 Method for transmitting virtualized sharing model for federated learning
WO2021142627A1 (en) * 2020-01-14 2021-07-22 Oppo广东移动通信有限公司 Resource scheduling method and apparatus, and readable storage medium
CN113159332A (en) * 2020-01-23 2021-07-23 华为技术有限公司 Method and device for realizing model updating
CN113239351A (en) * 2020-12-08 2021-08-10 武汉大学 Novel data pollution attack defense method for Internet of things system
CN113297573A (en) * 2021-06-11 2021-08-24 浙江工业大学 Vertical federal learning defense method and device based on GAN simulation data generation
TWI745958B (en) * 2019-11-19 2021-11-11 大陸商支付寶（杭州）信息技術有限公司 Training method and device of neural network model for protecting privacy and safety
CN113934578A (en) * 2021-10-28 2022-01-14 电子科技大学 Method for data recovery attack in federated learning scene
WO2022033579A1 (en) * 2020-08-13 2022-02-17 华为技术有限公司 Federated learning method, device and system
US11461728B2 (en) 2019-11-05 2022-10-04 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for consortium sharing
US11461793B2 (en) 2019-11-05 2022-10-04 International Business Machines Corporation Identification of behavioral pattern of simulated transaction data
US11475467B2 (en) 2019-11-05 2022-10-18 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for realistic modeling
US11475468B2 (en) 2019-11-05 2022-10-18 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for detection model sharing across entities
US11488185B2 (en) 2019-11-05 2022-11-01 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for consortium sharing
US11488172B2 (en) 2019-11-05 2022-11-01 International Business Machines Corporation Intelligent agent to simulate financial transactions
US11494835B2 (en) 2019-11-05 2022-11-08 International Business Machines Corporation Intelligent agent to simulate financial transactions
CN115438753A (en) * 2022-11-03 2022-12-06 电子科技大学 Method for measuring security of federal learning protocol data based on generation
CN115600250A (en) * 2022-12-12 2023-01-13 阿里巴巴（中国）有限公司（Cn） Data processing method, storage medium and electronic device
US11556734B2 (en) 2019-11-05 2023-01-17 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for realistic modeling
CN115719085A (en) * 2023-01-10 2023-02-28 武汉大学 Deep neural network model inversion attack defense method and equipment
US11599884B2 (en) 2019-11-05 2023-03-07 International Business Machines Corporation Identification of behavioral pattern of simulated transaction data
TWI800303B (en) * 2022-03-16 2023-04-21 英業達股份有限公司 Fedrated learning method using synonym
WO2023097602A1 (en) * 2021-12-02 2023-06-08 东莞理工学院 Inference method and apparatus for cooperative training data attribute, and device and storage medium
US11676218B2 (en) 2019-11-05 2023-06-13 International Business Machines Corporation Intelligent agent to simulate customer data
US11842357B2 (en) 2019-11-05 2023-12-12 International Business Machines Corporation Intelligent agent to simulate customer data
CN112214342B (en) * 2020-09-14 2024-05-24 德清阿尔法创新研究院 Efficient error data detection method in federal learning scene

Cited By (92)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110378749B (en) * 2019-07-25 2023-09-26 深圳前海微众银行股份有限公司 Client similarity evaluation method and device, terminal equipment and storage medium
CN110378749A (en) * 2019-07-25 2019-10-25 深圳前海微众银行股份有限公司 Appraisal procedure, device, terminal device and the storage medium of user data similitude
CN110443378A (en) * 2019-08-02 2019-11-12 深圳前海微众银行股份有限公司 Feature correlation analysis method, device and readable storage medium storing program for executing in federation's study
CN110443378B (en) * 2019-08-02 2023-11-03 深圳前海微众银行股份有限公司 Feature correlation analysis method and device in federal learning and readable storage medium
CN110472745A (en) * 2019-08-06 2019-11-19 深圳前海微众银行股份有限公司 Information transferring method and device in a kind of federal study
CN110569663A (en) * 2019-08-15 2019-12-13 深圳市莱法照明通信科技有限公司 Method, device, system and storage medium for educational data sharing
CN110443375A (en) * 2019-08-16 2019-11-12 深圳前海微众银行股份有限公司 A kind of federation's learning method and device
WO2021036014A1 (en) * 2019-08-28 2021-03-04 深圳前海微众银行股份有限公司 Federated learning credit management method, apparatus and device, and readable storage medium
CN110569911A (en) * 2019-09-11 2019-12-13 深圳绿米联创科技有限公司 Image recognition method, device, system, electronic equipment and storage medium
CN110647765A (en) * 2019-09-19 2020-01-03 济南大学 Privacy protection method and system based on knowledge migration under collaborative learning framework
WO2021082633A1 (en) * 2019-10-29 2021-05-06 支付宝(杭州)信息技术有限公司 Multi-party joint neural network training method and apparatus for achieving security defense
CN110751291A (en) * 2019-10-29 2020-02-04 支付宝(杭州)信息技术有限公司 Method and device for realizing multi-party combined training neural network of security defense
US11461728B2 (en) 2019-11-05 2022-10-04 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for consortium sharing
US11488172B2 (en) 2019-11-05 2022-11-01 International Business Machines Corporation Intelligent agent to simulate financial transactions
US11676218B2 (en) 2019-11-05 2023-06-13 International Business Machines Corporation Intelligent agent to simulate customer data
US11599884B2 (en) 2019-11-05 2023-03-07 International Business Machines Corporation Identification of behavioral pattern of simulated transaction data
US11461793B2 (en) 2019-11-05 2022-10-04 International Business Machines Corporation Identification of behavioral pattern of simulated transaction data
US11475467B2 (en) 2019-11-05 2022-10-18 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for realistic modeling
US11475468B2 (en) 2019-11-05 2022-10-18 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for detection model sharing across entities
US11556734B2 (en) 2019-11-05 2023-01-17 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for realistic modeling
US11488185B2 (en) 2019-11-05 2022-11-01 International Business Machines Corporation System and method for unsupervised abstraction of sensitive data for consortium sharing
GB2605054A (en) * 2019-11-05 2022-09-21 Ibm Intelligent agent to simulate customer data
WO2021090142A1 (en) * 2019-11-05 2021-05-14 International Business Machines Corporation Intelligent agent to simulate customer data
US11842357B2 (en) 2019-11-05 2023-12-12 International Business Machines Corporation Intelligent agent to simulate customer data
US11494835B2 (en) 2019-11-05 2022-11-08 International Business Machines Corporation Intelligent agent to simulate financial transactions
CN111079977A (en) * 2019-11-18 2020-04-28 中国矿业大学 Heterogeneous federated learning mine electromagnetic radiation trend tracking method based on SVD algorithm
TWI745958B (en) * 2019-11-19 2021-11-11 大陸商支付寶（杭州）信息技術有限公司 Training method and device of neural network model for protecting privacy and safety
CN110942154B (en) * 2019-11-22 2021-07-06 深圳前海微众银行股份有限公司 Data processing method, device, equipment and storage medium based on federal learning
CN110942154A (en) * 2019-11-22 2020-03-31 深圳前海微众银行股份有限公司 Data processing method, device, equipment and storage medium based on federal learning
CN112949670A (en) * 2019-12-10 2021-06-11 京东数字科技控股有限公司 Data set switching method and device for federal learning model
CN111079946A (en) * 2019-12-20 2020-04-28 支付宝(杭州)信息技术有限公司 Model training method, member detection device training method and system
WO2021120854A1 (en) * 2019-12-20 2021-06-24 支付宝(杭州)信息技术有限公司 Model training method, and method and system for training member detection device
CN111091199A (en) * 2019-12-20 2020-05-01 哈尔滨工业大学（深圳） Federal learning method and device based on differential privacy and storage medium
CN111245903B (en) * 2019-12-31 2022-07-01 烽火通信科技股份有限公司 Joint learning method and system based on edge calculation
CN111245903A (en) * 2019-12-31 2020-06-05 烽火通信科技股份有限公司 Joint learning method and system based on edge calculation
WO2021142627A1 (en) * 2020-01-14 2021-07-22 Oppo广东移动通信有限公司 Resource scheduling method and apparatus, and readable storage medium
WO2021147373A1 (en) * 2020-01-23 2021-07-29 华为技术有限公司 Method and device for implementing model update
CN113159332A (en) * 2020-01-23 2021-07-23 华为技术有限公司 Method and device for realizing model updating
CN113159332B (en) * 2020-01-23 2024-01-30 华为技术有限公司 Method and equipment for realizing model update
CN111340614A (en) * 2020-02-28 2020-06-26 深圳前海微众银行股份有限公司 Sample sampling method and device based on federal learning and readable storage medium
CN111340614B (en) * 2020-02-28 2021-05-18 深圳前海微众银行股份有限公司 Sample sampling method and device based on federal learning and readable storage medium
CN111260061A (en) * 2020-03-09 2020-06-09 厦门大学 Differential noise adding method and system in federated learning gradient exchange
CN111260061B (en) * 2020-03-09 2022-07-19 厦门大学 Differential noise adding method and system in federated learning gradient exchange
CN111447083B (en) * 2020-03-10 2022-10-21 中国人民解放军国防科技大学 Federal learning framework under dynamic bandwidth and unreliable network and compression algorithm thereof
CN111447083A (en) * 2020-03-10 2020-07-24 中国人民解放军国防科技大学 Federal learning framework under dynamic bandwidth and unreliable network and compression algorithm thereof
WO2021196701A1 (en) * 2020-03-31 2021-10-07 深圳前海微众银行股份有限公司 Attack coping method and federated learning device
CN111445031B (en) * 2020-03-31 2021-07-27 深圳前海微众银行股份有限公司 Attack coping method and federal learning device
CN111445031A (en) * 2020-03-31 2020-07-24 深圳前海微众银行股份有限公司 Attack coping method and federal learning device
CN111581648B (en) * 2020-04-06 2022-06-03 电子科技大学 Method of federal learning to preserve privacy in irregular users
CN111581648A (en) * 2020-04-06 2020-08-25 电子科技大学 Method of federal learning to preserve privacy in irregular users
CN111598143A (en) * 2020-04-27 2020-08-28 浙江工业大学 Credit evaluation-based defense method for federal learning poisoning attack
CN111598143B (en) * 2020-04-27 2023-04-07 浙江工业大学 Credit evaluation-based defense method for federal learning poisoning attack
CN111460443B (en) * 2020-05-28 2022-09-23 南京大学 Security defense method for data manipulation attack in federated learning
CN111460443A (en) * 2020-05-28 2020-07-28 南京大学 Security defense method for data manipulation attack in federated learning
CN111464568A (en) * 2020-06-17 2020-07-28 广东电网有限责任公司佛山供电局 Method and system for enhancing network attack prevention capability of multiple network ports
CN112101403B (en) * 2020-07-24 2023-12-15 西安电子科技大学 Classification method and system based on federal few-sample network model and electronic equipment
CN112101403A (en) * 2020-07-24 2020-12-18 西安电子科技大学 Method and system for classification based on federate sample network model and electronic equipment
WO2022033579A1 (en) * 2020-08-13 2022-02-17 华为技术有限公司 Federated learning method, device and system
CN111985562A (en) * 2020-08-20 2020-11-24 复旦大学 End cloud collaborative training system for protecting end-side privacy
CN111985562B (en) * 2020-08-20 2022-07-26 复旦大学 End cloud collaborative training system for protecting end-side privacy
CN112203282B (en) * 2020-08-28 2022-02-18 中国科学院信息工程研究所 5G Internet of things intrusion detection method and system based on federal transfer learning
CN112203282A (en) * 2020-08-28 2021-01-08 中国科学院信息工程研究所 5G Internet of things intrusion detection method and system based on federal transfer learning
CN112039702A (en) * 2020-08-31 2020-12-04 中诚信征信有限公司 Model parameter training method and device based on federal learning and mutual learning
CN112039702B (en) * 2020-08-31 2022-04-12 中诚信征信有限公司 Model parameter training method and device based on federal learning and mutual learning
CN112100659A (en) * 2020-09-14 2020-12-18 电子科技大学 Block chain federal learning system and Byzantine attack detection method
CN112214342A (en) * 2020-09-14 2021-01-12 德清阿尔法创新研究院 Efficient error data detection method in federated learning scene
CN112214342B (en) * 2020-09-14 2024-05-24 德清阿尔法创新研究院 Efficient error data detection method in federal learning scene
CN112100659B (en) * 2020-09-14 2023-04-07 电子科技大学 Block chain federal learning system and Byzantine attack detection method
CN112118099B (en) * 2020-09-16 2021-10-08 西安电子科技大学 Distributed multi-task learning privacy protection method and system for resisting inference attack
CN112118099A (en) * 2020-09-16 2020-12-22 西安电子科技大学 Distributed multi-task learning privacy protection method and system for resisting inference attack
CN112329009B (en) * 2020-10-12 2022-12-06 南京理工大学 Defense method for noise attack in joint learning
CN112329009A (en) * 2020-10-12 2021-02-05 南京理工大学 Defense method for noise attack in joint learning
CN112162959A (en) * 2020-10-15 2021-01-01 深圳技术大学 Medical data sharing method and device
CN112162959B (en) * 2020-10-15 2023-10-10 深圳技术大学 Medical data sharing method and device
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
CN112257063B (en) * 2020-10-19 2022-09-02 上海交通大学 Cooperative game theory-based detection method for backdoor attacks in federal learning
CN113239351A (en) * 2020-12-08 2021-08-10 武汉大学 Novel data pollution attack defense method for Internet of things system
CN113239351B (en) * 2020-12-08 2022-05-13 武汉大学 Novel data pollution attack defense method for Internet of things system
CN112434758B (en) * 2020-12-17 2024-02-13 浙江工业大学 Clustering-based federal learning pick-up car attack defense method
CN112434758A (en) * 2020-12-17 2021-03-02 浙江工业大学 Cluster-based federal learning casual vehicle attack defense method
CN112784990A (en) * 2021-01-22 2021-05-11 支付宝(杭州)信息技术有限公司 Training method of member inference model
CN112819180A (en) * 2021-01-26 2021-05-18 华中科技大学 Multi-service data generation method and device based on federal generation model
CN113051608A (en) * 2021-03-11 2021-06-29 佳讯飞鸿（北京）智能科技研究院有限公司 Method for transmitting virtualized sharing model for federated learning
CN113297573A (en) * 2021-06-11 2021-08-24 浙江工业大学 Vertical federal learning defense method and device based on GAN simulation data generation
CN113934578A (en) * 2021-10-28 2022-01-14 电子科技大学 Method for data recovery attack in federated learning scene
WO2023097602A1 (en) * 2021-12-02 2023-06-08 东莞理工学院 Inference method and apparatus for cooperative training data attribute, and device and storage medium
TWI800303B (en) * 2022-03-16 2023-04-21 英業達股份有限公司 Fedrated learning method using synonym
CN115438753B (en) * 2022-11-03 2023-01-06 电子科技大学 Method for measuring security of federal learning protocol data based on generation
CN115438753A (en) * 2022-11-03 2022-12-06 电子科技大学 Method for measuring security of federal learning protocol data based on generation
CN115600250B (en) * 2022-12-12 2023-03-21 阿里巴巴（中国）有限公司 Data processing method, storage medium and electronic device
CN115600250A (en) * 2022-12-12 2023-01-13 阿里巴巴（中国）有限公司（Cn） Data processing method, storage medium and electronic device
CN115719085A (en) * 2023-01-10 2023-02-28 武汉大学 Deep neural network model inversion attack defense method and equipment

Similar Documents

Publication Publication Date Title
CN110008696A (en) A kind of user data Rebuilding Attack method towards the study of depth federation
CN109815893B (en) Color face image illumination domain normalization method based on cyclic generation countermeasure network
Ren et al. Low-light image enhancement via a deep hybrid network
CN111145116B (en) Sea surface rainy day image sample augmentation method based on generation of countermeasure network
CN109859288B (en) Image coloring method and device based on generation countermeasure network
CN110460600A (en) The combined depth learning method generated to network attacks can be resisted
CN110458765B (en) Image quality enhancement method based on perception preserving convolution network
CN110097178A (en) It is a kind of paid attention to based on entropy neural network model compression and accelerated method
CN111625820A (en) Federal defense method based on AIoT-oriented security
CN110660020B (en) Image super-resolution method of antagonism generation network based on fusion mutual information
CN109064422A (en) A kind of underwater image restoration method based on fusion confrontation network
Hsu et al. A high-capacity QRD-based blind color image watermarking algorithm incorporated with AI technologies
CN114997420B (en) Federal learning system and method based on segmentation learning and differential privacy fusion
CN108460720A (en) A method of changing image style based on confrontation network model is generated
CN115481431A (en) Dual-disturbance-based privacy protection method for federated learning counterreasoning attack
CN114362948B (en) Federated derived feature logistic regression modeling method
CN113724149A (en) Weak supervision visible light remote sensing image thin cloud removing method
Geng et al. Improved gradient inversion attacks and defenses in federated learning
CN108492275B (en) No-reference stereo image quality evaluation method based on deep neural network
Wang et al. Data hiding during image processing using capsule networks
CN114639174A (en) Privacy type deep forgery detection method under federal cooperation
CN115329388B (en) Privacy enhancement method for federally generated countermeasure network
CN115510472B (en) Multi-difference privacy protection method and system for cloud edge aggregation system
CN116187469A (en) Client member reasoning attack method based on federal distillation learning framework
CN116011597A (en) Personalized federal learning method and device based on graph data

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