CN116597498B - Fair face attribute classification method based on blockchain and federal learning - Google Patents

Fair face attribute classification method based on blockchain and federal learning Download PDF

Info

Publication number
CN116597498B
CN116597498B CN202310826138.6A CN202310826138A CN116597498B CN 116597498 B CN116597498 B CN 116597498B CN 202310826138 A CN202310826138 A CN 202310826138A CN 116597498 B CN116597498 B CN 116597498B
Authority
CN
China
Prior art keywords
model
fairness
local model
client
local
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
Application number
CN202310826138.6A
Other languages
Chinese (zh)
Other versions
CN116597498A (en
Inventor
古天龙
王梦圆
李龙
郝峰锐
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.)
Jinan University
Original Assignee
Jinan University
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 Jinan University filed Critical Jinan University
Priority to CN202310826138.6A priority Critical patent/CN116597498B/en
Publication of CN116597498A publication Critical patent/CN116597498A/en
Application granted granted Critical
Publication of CN116597498B publication Critical patent/CN116597498B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • 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
    • G06F21/602—Providing cryptographic facilities or services
    • 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
    • G06F21/64—Protecting data integrity, e.g. using checksums, certificates or signatures
    • 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
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • 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
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computer Security & Cryptography (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Bioethics (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Databases & Information Systems (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Human Computer Interaction (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The application discloses a fair face attribute classification method based on block chain and federal learning, which belongs to the technical field of computer vision and comprises the following steps: the model demander issues a federal learning task, and the blockchain transaction contract transmits the federal learning task to the client; after receiving the federation learning task, the client trains the local model and encrypts and sends the local model parameters to the blockchain node; the block chain link point verifies fairness of the local model parameters, generates client reputation, encrypts the local model parameters which are successfully verified, and packages the client reputation to generate a new block; broadcasting the new block to other block chain nodes for verification by using a block chain transaction contract, and collecting and aggregating local model parameters successfully verified by a central server to obtain a global model; model demander carries out face attribute recognition classification based on the global model. The application realizes the fair enhancement of the attributes of the global model and maintains the ideal accuracy of the classification of the face attributes.

Description

Fair face attribute classification method based on blockchain and federal learning
Technical Field
The application belongs to the technical field of computer vision, and particularly relates to a fair face attribute classification method based on block chain and federal learning.
Background
The human face is an important biological feature, has complex, rich and changeable structure, can provide specific and diversified individual or group information, and is therefore a typical basis for expressing human identity features. The classification of the face attribute can predict the attribute information of a given face picture, such as gender, is important for the fields of computer vision such as face recognition, face verification and the like, is widely studied at present and is applied to a plurality of actual scenes.
Related events and scientific research occurring in recent years have shown that face attribute classification systems may prejudice populations or individuals possessing certain attributes. Buolamwini et al state that the conventional face recognition API presents serious bias problems in Pilot Parliaments Benchmark (PPB) test datasets, e.g., up to 20.6% difference in gender (male/female) determinations.
To solve the above classification variability problem, related scholars have proposed solutions for centralized machine learning, such as balancing data sets, meta-balancing networks, feature distillation, etc. Although these schemes achieve a certain effect, they cannot be directly applied to a distributed system represented by an urban monitoring system, for two main reasons: first, urban monitoring systems train on multiple types of large-scale face data with a large number of monitoring devices, which are often decentralized and difficult to consolidate. Secondly, the unified collection and centralized management of face images rich in user sensitive privacy information violates the increasingly enhanced privacy protection wish of users and even violates the requirements of GDPR and other regulations on the compliance of data services. Therefore, scholars propose a fairness enhancing scheme for federal learning (Federated Learning, FL) to efficiently complete model training and improve attribute fairness on the premise of realizing privacy protection.
It is important to note that FL applied to urban monitoring systems is vulnerable to attacks from malicious clients, and locking and responsibility chasing of malicious clients is difficult to achieve because the training process of FL cannot be traced. BCFL (Blockchain-based Federated Learning) combines Blockchain to advantage in addressing the above problems, however BCFL also has the isomerism that is common in FL and the system architecture is more complex, so there are prejudice and discrimination problems as well. In addition, BCFL is subject to a new type of attack for model bias amplification, in addition to attacks aimed at reducing global model accuracy. However, there is a gap in research on the fairness of BCFL, especially the fairness of face attributes. Therefore, there is a need to propose a fair face attribute classification method based on blockchain and federal learning.
Disclosure of Invention
The application aims to provide a fair face attribute classification method based on blockchain and federal learning so as to solve the problems in the prior art.
In order to achieve the above object, the present application provides a fair face attribute classification method based on blockchain and federal learning, comprising the following steps:
when determining that a federation learning task triggers a blockchain transaction contract, transmitting the federation learning task to all online clients by the blockchain transaction contract, wherein the federation learning task at least comprises an initial model;
training the initial model through the client based on a fairness constraint function to obtain a local model, and sending local model parameters to a blockchain node;
verifying fairness of the local model parameters through the blockchain node, generating a client reputation, encrypting the local model parameters successfully verified, packaging the client reputation to generate a new block, and verifying;
collecting and aggregating the local model parameters successfully verified by a central server, and adjusting the weight of the local model parameters successfully verified based on fairness so as to obtain a global model;
and carrying out face attribute identification and classification based on the global model.
Optionally, the fairness constraint function is as follows:
wherein ,for a real label->Is to->Is used to determine the predicted value of (c),Qin order to train the total amount of samples,representing the authenticity of the overall dataTwo-class cross entropy loss between value and predicted value, [ Y ]]Value set for real tag +_>For sensitive property, ++>Representing the probability that the real tag is 1, +.>Mean value representing sample sensitivity properties, +.>Indicating when the predicted value is +.>Time of dayIs used for the prediction probability of (1).
Optionally, verifying fairness of the local model parameters based on a blockchain node includes: after receiving the local model parameters, the blockchain node acquires the matching degree between the local attribute fairness and the average attribute fairness of the local model; and presetting a fairness threshold range of the local model, and if the matching degree is in the threshold range, successfully verifying fairness of the local model parameters.
Optionally, the process of generating the client reputation comprises: the blockchain node generates a fairness verification score for the client, then forms a client reputation with the blockchain node's signature validity score, and assigns a weight to the client reputation.
Optionally, the process of verifying the new block by other block link points includes: other blockchain nodes check the signature of the new block, the number of the local model parameters and the rationality of the verification condition of the local model parameters based on a preset rule, if the verification condition is reasonable, voting to the blockchain node generating the new block, obtaining the accounting right by the blockchain node with the largest number of votes, and storing the corresponding new block in the distributed account book of the rest blockchain nodes.
Optionally, the process of obtaining the global model includes: acquiring accuracy, average accuracy, attribute fairness and average attribute fairness of each local model based on a central server, and obtaining a first comparison result of the accuracy and the average accuracy through calculation, wherein a second comparison result of the attribute fairness and the average attribute fairness is obtained; determining a weight of each local model based on the first comparison result and the second comparison result; aggregating the local models based on the weight of each local model to obtain the current local model aggregation; and controlling the current local model aggregation and the proportion of the last global model based on the super parameters, and updating the global model until training is finished, so as to obtain the latest global model.
Optionally, the process after obtaining the latest global model includes: and obtaining the compression ratio of the latest global model based on the difference between the client reputation of the current training round and the maximum client reputation, and distributing model rewards to the clients based on the latest global model and the corresponding compression ratio.
Optionally, the process after obtaining the latest global model further includes: the latest global model is returned to the model demander by the blockchain trading contract, and monetary rewards are sent to the client based on the historical value of the client reputation.
The application has the technical effects that:
according to the method, a fairness constraint function is added in the local training process of the client, evaluation of attribute fairness is added in the local model verification link, attention to attribute fairness is added in a plurality of links such as fairness-based weight adjustment in the global model aggregation process, the enhancement of the attribute fairness of the global model is realized, and ideal accuracy is maintained.
The application calculates the client reputation in the model verification link by using the blockchain node, takes the local model parameters with poor fairness as the local model parameters which are included in the aggregation link according to refusal, and can allocate a global model with corresponding compression ratio for the client according to the reputation by the central server, thereby effectively preventing the fairness attack of the malicious client and enhancing the reliability of the system.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:
FIG. 1 is a schematic diagram of the overall structure of a fair face attribute classification method based on blockchain and federal learning in an embodiment of the present application;
fig. 2 is an overall workflow diagram of a fair face attribute classification method based on blockchain and federal learning in an embodiment of the present application.
Detailed Description
It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other. The application will be described in detail below with reference to the drawings in connection with embodiments.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
Example 1
As shown in fig. 1, the embodiment provides a fair face attribute classification method based on blockchain and federal learning, and as shown in the figure, the system includes model demanders and blockchain nodes in addition to the entities commonly found in the two federal learning of a central server and a client. Which will be described separately below:
1. model requirers are publishers of federal learning tasks requesting global models through collaborative training. Specifically, the model demander needs to specify the functional requirements of the federal learning model (FL model), training termination conditions, and the like, while providing monetary rewards.
2. The client is a trainer of the local model, firstly collects enough training data, then trains the local model based on the global model provided by the central server, and finally encrypts the obtained local model parameters and sends the encrypted local model parameters to the blockchain network.
3. The blockchain node is a verifier of the local model, and after receiving the local model parameters, the blockchain node verifies fairness of the local model parameters to generate client reputation, and then packages the model parameters and the reputation to generate a new block. New blocks after inspection and consensus by other blockchain nodes will be added to the blockchain.
4. The central server is responsible for collecting the local model parameters which pass verification, executing an aggregation mechanism to generate a global model, and distributing model rewards according to the credit of each client.
Fig. 2 is an overall workflow diagram of a fair face attribute classification method based on blockchain and federal learning according to the present embodiment. The flow of this embodiment is as follows:
s1: model requirers present FL tasks that provide content such as initial models, learning rates, optimizers, and end conditions to trigger blockchain trade contracts. Federal learning tasks include initial models, but do not include local models. The initial model refers to the initial model which is not trained at the beginning and is the basis of the local model.
S2: if the FL task is successfully started, the trade contract will automatically pass the task to all online clients and block link points.
S3: after receiving the FL task, the client trains its initial model using the local data and the latest global model, then signs the local model parameters with the private key, and sends them to the blockchain node.
S3.1: a fairness constraint function is added during the client local training process, as shown in formula (1). The function consists of two parts, the first part is a target classification loss for optimizing the accuracy of the local model and the second part is a fair regularization loss for removing the bias between the property groups.
(1)
wherein ,for a real label->Is to->Is used to determine the predicted value of (c),Qin order to train the total amount of samples,representing a two-class cross entropy loss between the true and predicted values of the overall data [ Y ]]Value set for real tag +_>For sensitive property, ++>Representing the probability that the real tag is 1, +.>Mean value representing sample sensitivity properties, +.>Indicating when the predicted value is +.>Time of dayIs used for the prediction probability of (1).
S4: the block link propagates the received local model parameters to other nodes and decrypts and validates them to check their fairness and calculate the reputation of each client. The block chain link points place local model parameters with reasonable fairness into the transaction pool.
S4.1: after receiving the local model parameters of the client, the blockchain node calculates the matching degree between the local attribute fairness and the average attribute fairness so as to judge whether the fairness of the local model is within a threshold range. The degree of matching can be calculated from equation (2).
(2)
wherein ,is a local model->Property of->Is fair to the average property of all local models. Further, the blockchain node checks whether equation (3) holds.
(3)
wherein Expressed in the absence of clientiIn the case of (a), the average of the matching degree of other clients. />Representing the matching degree and +.>Average difference between them.
S4.2: if the verification is successful, the block chain link point puts the local model parameters into a transaction pool, generates fairness verification scores for the clients, and then forms each round of reputation of the clients together with the signature effectiveness scores. If verification fails, the block link will discard the local model parameters, will not verify the validity of the signature, and will set the verification score for the client to 0. The client reputation per round can be derived from equation (4).
(4)
wherein ,is a block chain link pointjTo the clientiVerification score of model fairness, ++>Is a block chain link pointjTo the clientiVerification scoring of signature validity. The reputation of the client varies along with the increase of training rounds, and a closer reputation can more indicate whether the client is trustworthy than an old reputation, so that the system selects weighting according to the freshness of the reputation, and adopts a Newton cooling formula to give a larger weight to the nearest reputation and a smaller weight to the earliest reputation.
The block link point calculation client history reputation can be derived from equation (5).
(5)
(6)
wherein t’The current turn of the day is indicated,the weight of the reputation is calculated by a formula (10). />Representing the initial value, i.e. the maximum value of the weights. />>0 is referred to as an exponential decay constant.
S5: when all the local model parameters are verified, the blockchain node signs the local model parameters which are successfully verified by using a private key, packages the local model parameters with the credit to generate a new block, and broadcasts the new block to other blockchain nodes.
S6: other blockchain nodes check whether the signature of the new block, the number of local model parameters and the verification condition of the local model parameters are reasonable, if so, vote to the generator of the block, and terminate the verification of the same batch of blocks. The blockchain node with the most tickets obtains the accounting rights, while the other blockchain nodes store their blocks in their own distributed ledgers.
S7: the central server collects the local model parameters successfully verified, and adopts an aggregation mechanism to conduct model aggregation so as to obtain a global model. Further, the central server distributes the corresponding sparse global model to the client according to the reputation of the client.
S7.1: the central server first calculates the weights of the local model parameters, and from the aspect of accuracy, the weight calculation is shown in formula (7).
(7)
wherein Is a local model->Accuracy of (A)>The average accuracy for all local models. The proposed aggregation mechanism compares the accuracy of each local model with the average accuracy to determine the weights. Since all local model parameters for aggregation are verified by the blockchain node, unreasonable local model parameters do not exist, and therefore the average accuracy can well represent the real condition of the accuracy of all local models. In addition, in order to prevent the weight from affecting the global model too much or too little, the weight is restricted to be within a certain interval.
Similarly, from the aspect of fairness, the weight calculation is as shown in formula (8)
(8)
S7.2: aggregation according to equation (9)LAnd obtaining the latest local model aggregation by verifying the local model parameters.
(9)
S7.3: the global model is updated as shown in equation (10).
(10)
Wherein the super parameterControlling the specific gravity of the current local model aggregate and the last global model,/->∈(0,1)。
S7.4: after the aggregation work is completed, the central server calculates model rewards which the central server needs to acquire according to the reputation of the current turn of the client, and a specific formula is shown as (11).
(11)
(12)
wherein Representing global model compression ratio, calculated by the difference between the reputation of the current round of the client and the maximum reputation, and limiting the value to the interval [0,1 ]]Is calculated by the formula (12).
The system uses the GRACE framework to sparsify the global model and selects depth gradient compression (Deep Gradient Compression, DGC) methods therein to extract important gradient information. Specifically, the DGC determines the importance of a gradient element based on whether the size of the element exceeds a threshold, which is determined by the smallest absolute value in the selected compression gradient, while small gradient elements that do not exceed the threshold are accumulated by iterating the momentum correction multiple times in order to prevent losing a large amount of information.
S8: and after the federal learning task is finished, the transaction contract is returned to the latest global model of the model demander, and corresponding monetary rewards are distributed according to the historical reputation of the client.
S8.1: after the federal learning task is completed, the trade contract calculates the monetary rewards it should acquire based on the client's historical reputation, as shown in equation (13).
(13)
Where CY is the sum of money submitted by the model requirers.
S8.2: the model demander obtains the latest global model and classifies the face attribute by using the global model. Specifically, the model demander takes face data as input of the global model, and the global model correspondingly obtains attribute values, such as gender and the like, of the face data, and then classifies the face data into appropriate categories according to the attribute values.
The embodiment discloses a fair face attribute classification method based on block chain and federal learning, which comprises the following steps: the model demander puts forward a face attribute classification task; the blockchain transaction contract is triggered and delivers tasks; the client trains a local model locally, encrypts model parameters and sends the model parameters to the blockchain node; the block chain link points perform fairness verification on the local model parameters and generate client reputation according to verification results; the block chain link point generates a new block and broadcasts the new block to other nodes; other nodes verify the blocks and achieve consensus; the central server aggregates the local model parameters and distributes model rewards for the client; repeating the steps until the training ending condition is met; the trade contracts return the model to the model requirers and distribute monetary rewards to the clients. The embodiment realizes traceability and non-falsification of federal learning by using the distributed account book, and introduces a reputation mechanism to excite the enthusiasm of the client to participate in the training process. Meanwhile, the fairness of the local model is verified by the blockchain nodes, the central server determines the aggregation weight according to the accuracy and fairness of the local model, and the attack of the malicious client side on the model bias is effectively prevented while the global model quality is ensured.
The present application is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present application are intended to be included in the scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims (6)

1. A fair face attribute classification method based on block chain and federal learning is characterized by comprising the following steps:
when determining that a federation learning task triggers a blockchain transaction contract, transmitting the federation learning task to all online clients by the blockchain transaction contract, wherein the federation learning task at least comprises an initial model;
training the initial model through the client based on a fairness constraint function to obtain a local model, and sending local model parameters to a blockchain node;
verifying fairness of the local model parameters through the blockchain node, generating a client reputation, encrypting the local model parameters successfully verified, packaging the client reputation to generate a new block, and verifying;
collecting and aggregating the local model parameters successfully verified by a central server, and adjusting the weight of the local model parameters successfully verified based on fairness so as to obtain a global model;
performing face attribute identification and classification based on the global model;
the process of verifying fairness of the local model parameters based on blockchain nodes includes: after receiving the local model parameters, the blockchain node acquires the matching degree between the attribute fairness and the average attribute fairness of the local model; presetting a fairness threshold range of a local model, and if the matching degree is in the threshold range, successfully verifying fairness of the local model parameters;
the process of obtaining the global model includes: acquiring accuracy, average accuracy, attribute fairness and average attribute fairness of each local model based on a central server, and obtaining a first comparison result of the accuracy and the average accuracy through calculation, wherein a second comparison result of the attribute fairness and the average attribute fairness is obtained; determining a weight of each local model based on the first comparison result and the second comparison result; aggregating the local models based on the weight of each local model to obtain the current local model aggregation; and controlling the current local model aggregation and the proportion of the last global model based on the super parameters, and updating the global model until training is finished, so as to obtain the latest global model.
2. The fair face attribute classification method according to claim 1, wherein,
the fairness constraint function is shown as follows:
wherein ,y q for a real label->Is tox q Is used to determine the predicted value of (c),Qin order to train the total amount of samples,representing a two-class cross entropy loss between the true and predicted values of the overall data [ Y ]]A set of values is taken for the real tags,a q for sensitive property, ++>Representing the probability that the real tag is 1, +.>Mean value representing sample sensitivity properties, +.>Indicating when the predicted value is +.>Time->Is used for the prediction probability of (1).
3. The fair face attribute classification method according to claim 1, wherein,
the process of generating a client reputation includes: the blockchain node generates a fairness verification score for the client, then forms a client reputation with the blockchain node's signature validity score, and assigns a weight to the client reputation.
4. The fair face attribute classification method according to claim 1, wherein,
the process of verifying the new block includes: other blockchain nodes check the signature of the new block, the number of the local model parameters and the rationality of the verification condition of the local model parameters based on a preset rule, if the verification condition is reasonable, voting to the blockchain node generating the new block, obtaining the accounting right by the blockchain node with the largest number of votes, and storing the corresponding new block in the distributed account book of the rest blockchain nodes.
5. The fair face attribute classification method according to claim 1, wherein,
the process after obtaining the latest global model includes: and obtaining the compression ratio of the latest global model based on the difference between the client reputation of the current training round and the maximum client reputation, and distributing model rewards to the clients based on the latest global model and the corresponding compression ratio.
6. The fair face attribute classification method according to claim 5, wherein,
the process after obtaining the latest global model further comprises: the latest global model is returned to the model demander by the blockchain trading contract, and monetary rewards are sent to the client based on the historical value of the client reputation.
CN202310826138.6A 2023-07-07 2023-07-07 Fair face attribute classification method based on blockchain and federal learning Active CN116597498B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310826138.6A CN116597498B (en) 2023-07-07 2023-07-07 Fair face attribute classification method based on blockchain and federal learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310826138.6A CN116597498B (en) 2023-07-07 2023-07-07 Fair face attribute classification method based on blockchain and federal learning

Publications (2)

Publication Number Publication Date
CN116597498A CN116597498A (en) 2023-08-15
CN116597498B true CN116597498B (en) 2023-10-24

Family

ID=87604708

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310826138.6A Active CN116597498B (en) 2023-07-07 2023-07-07 Fair face attribute classification method based on blockchain and federal learning

Country Status (1)

Country Link
CN (1) CN116597498B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113570065A (en) * 2021-07-08 2021-10-29 国网河北省电力有限公司信息通信分公司 Data management method, device and equipment based on alliance chain and federal learning
CN115640305A (en) * 2022-12-22 2023-01-24 暨南大学 Fair and credible federal learning method based on block chain
CN116362328A (en) * 2023-04-19 2023-06-30 西北工业大学 Federal learning heterogeneous model aggregation method based on fairness characteristic representation

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113570065A (en) * 2021-07-08 2021-10-29 国网河北省电力有限公司信息通信分公司 Data management method, device and equipment based on alliance chain and federal learning
CN115640305A (en) * 2022-12-22 2023-01-24 暨南大学 Fair and credible federal learning method based on block chain
CN116362328A (en) * 2023-04-19 2023-06-30 西北工业大学 Federal learning heterogeneous model aggregation method based on fairness characteristic representation

Also Published As

Publication number Publication date
CN116597498A (en) 2023-08-15

Similar Documents

Publication Publication Date Title
CN113794675B (en) Distributed Internet of things intrusion detection method and system based on block chain and federal learning
CN112949837B (en) Target recognition federal deep learning method based on trusted network
CN107580022B (en) Data sharing system and method
CN112784994B (en) Block chain-based federated learning data participant contribution value calculation and excitation method
CN111988137B (en) DPoS (dual port service) consensus method and system based on threshold signature and fair reward
CN109993233B (en) Method and system for predicting data auditing objective based on machine learning
CN115510494B (en) Multiparty safety data sharing method based on block chain and federal learning
Gao et al. FGFL: A blockchain-based fair incentive governor for Federated Learning
CN112418520A (en) Credit card transaction risk prediction method based on federal learning
CN111860865B (en) Model construction and analysis method, device, electronic equipment and medium
CN105718526A (en) Data search method, device and system based on lawyer information
CN114418109A (en) Node selection and aggregation optimization system and method for federal learning under micro-service architecture
CN114817946A (en) Credible execution environment-based federated learning gradient boosting decision tree training method
CN111105240A (en) Resource-sensitive combined financial fraud detection model training method and detection method
WO2023045691A1 (en) Object recognition method and apparatus, and electronic device and storage medium
CN112116103A (en) Method, device and system for evaluating personal qualification based on federal learning and storage medium
Xu et al. Bitcoin miners: Exploring a covert community in the Bitcoin ecosystem
CN110060157B (en) Reputation evaluation method and system
CN116597498B (en) Fair face attribute classification method based on blockchain and federal learning
CN107070954B (en) Anonymous-based trust evaluation method
CN116186629B (en) Financial customer classification and prediction method and device based on personalized federal learning
CN115640305B (en) Fair and reliable federal learning method based on blockchain
Hsueh et al. EPoW: Solving blockchain problems economically
JP6762057B1 (en) Video / image content posting management system in SNS
Aggarwal et al. A Structural Analysis of Bitcoin Cash's Emergency Difficulty Adjustment Algorithm

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