Federal average learning method and system based on ring signature hiding client identity

Abstract

The invention discloses a federal average learning method and system based on ring signature hiding client identity, and the method comprises the steps: a client downloads a global model through a block chain, and carries out the training through local data, and obtains a local gradient; the client performs anonymization processing on model updating through a ring signature algorithm, and uploads the local gradient and signature information to a block chain network; the client downloads the trained gradient, tests the gradient and uploads a test result; and when the smart contract collects a test result of a preset condition, updating the global model. The system comprises a training unit, an anonymization unit, a testing unit and a global updating unit. According to the invention, data privacy protection can be realized. The method can be widely applied to the field of federal learning.

Description

Technical Field

[0001] The present invention relates to the field of federated learning, and in particular to a method and system for federated average learning based on ring signatures to conceal client identities. Background Art

[0002] Machine learning is highly dependent on scenarios and data. High-quality data directly determines the quality of the model, but data acquisition has become difficult. Data security and user privacy issues are receiving more and more attention, data supervision is becoming more stringent, and data interoperability is becoming more difficult.

[0003] Federated learning is a distributed machine learning solution that is expected to protect data privacy and is gradually being widely used. However, existing federated learning systems rely on centralized servers to aggregate and manage model parameters. This architecture not only faces the risk of single point failure, but also poses a threat to data privacy. Summary of the invention

[0004] In view of this, in order to solve the technical problem that the existing federated learning method relies on a centralized server for data processing, thereby causing the risk of data leakage, the present invention proposes a federated average learning method based on ring signature to conceal the identity of the client, and the method includes the following steps:

[0005] System initialization: The client identity information and public information are verified through smart contracts and uploaded to the chain;

[0006] Model distribution training: The client downloads the global model through the blockchain and trains it with local data to obtain local gradients;

[0007] Ring signature generation: The client anonymizes the model update through the ring signature algorithm and uploads the local gradient and signature information to the blockchain network;

[0008] Model parameter authentication: The client downloads the trained gradients and performs tests, and uploads the test results;

[0009] Model aggregation: When the smart contract collects the test results of the preset conditions, it updates the global model.

[0010] In some embodiments, the steps of system initialization specifically include:

[0011] In the blockchain-based federated learning system of the present invention, the system initialization phase ensures that participating clients can participate in the federated learning process anonymously and securely. The real identity of the client will not be directly disclosed on the blockchain, but the identity will be anonymized through the ring signature technology, thereby verifying the legitimacy of the participant while ensuring privacy. Each client that wishes to participate in federated learning needs to register the system identity and the blockchain will generate a public-private key pair for it for subsequent signing operations. First, select groups G1 and G2, and then use them as bilinear mapping data to select two hash functions H(·) and H1(·). H(·) is used to identify the participant according to their identity (ID i ) generates a public key, and H1(·) is used to perform hash calculations on messages, parameters, and public keys in the ring to ensure the security and anonymity of the signature. The blockchain randomly selects parameters x i With r i ,in ( represents a finite field) to calculate the corresponding public key P pub , which is a global public key disclosed by the system and stored on the blockchain for all participants to use to verify global operations or encrypt certain system messages. The public parameters are: params = {G1, G2, e(·, ·), q, P, P pub ,H(·),H1(·)}. The blockchain generates the public and private keys used by the user for the ring signature, where c i It is a value calculated by the hash function H1(·), which is related to the user's identity, system public key and other parameters. It is used to identify the user's identity and help the system ensure the uniqueness of the identity without directly exposing the user's identity. i =(r i -c i x i )modq,s i It is an intermediate value required in the ring signature generation process. Its purpose is to i and a random number r i The combination of makes the signature different each time it is generated, while ensuring security and validity during the verification process. Private key: usk i ={c i ,s i ,x i}, public key: The real identity of the client will not be directly disclosed. During the system initialization phase, after the client initiates communication with the blockchain, the blockchain will record the user's ring information.

[0012] In some embodiments, the step of model distribution training specifically includes:

[0013] After the smart contract completes the initialization of the model, it uploads these initial model parameters to the blockchain. Each client participating in the federated learning downloads the initial global model parameters from the chain and starts the training process on its local dataset.

[0014] Each client adjusts and optimizes the model according to its own data characteristics to achieve better performance. This process is carried out in the local environment of the client, ensuring the privacy and security of the data, because only the model parameters are transmitted between the client and the smart contract, without revealing sensitive local data. The client conducts multiple rounds of training locally, and the optimization objectives are as follows:

[0015]

[0016] That is, find the local model parameter value w that minimizes the loss function L m (w), and this optimization process is initialized by the global model parameters m . After the training is completed, the model parameters are updated and prepared to be uploaded to the blockchain for the next round of global model update. In some embodiments, the ring signature generation step specifically includes:

[0017] When the client completes the training of the local model, it generates the corresponding model parameters. To protect the privacy of the client during the upload process, the system uses the ring signature technology to sign the model parameters, so that in the communication process, the specific upload client information cannot be directly traced, ensuring the anonymity and non-repudiation of the uploaded model parameters in the federated learning system. Non-repudiation means that in this system, the ring signature technology ensures that the uploaded model parameters actually come from a certain member within the ring, rather than an external entity, and at the same time, it is impossible to clearly identify which member sent the message.

[0018] After the client C

[0019] generates the local model parameter w i m m m it will not directly associate it with its own identity, but generates a signature ring containing the public keys of multiple participating clients through the ring signature technology. The client signs the model parameters with its own private key and includes the public keys of other clients at the same time, so that the observer cannot judge the true source of the signature and can only determine that the signature comes from a certain client in the signature ring. The characteristics of the ring signature ensure the anonymity of the uploader and also ensure the validity of the signature.

[0020] During the ring signature generation process, the model parameter is set as the message m. In the designed scheme, every 5 clients form a ring, so the signature steps are as follows: for 0 < i < 6, calculate Where L represents the set of all user identities, randomly selected And calculate U s =r' s P-∑ i≠s {U i +H i pub i}, where U s It is used to identify the public network identity of the client. Calculate H s =H1(m||L||U s ),v=x s ·(r' s +H s ) -1 , randomly selected And calculate T = tP, c = H1 (P pub ,x s pub s ,c s P pub +s s P), these two parameters are used by the blockchain to verify whether the ring signature is legal, and calculate e = H1 (P pub ,L,m,T,c),π=(t-es s )modq, and finally generate the ring signature format as

[0021] The model parameters with signatures are uploaded to the blockchain, and the smart contract verifies the signature. Through the decentralized nature of the blockchain, all clients in the system can verify whether the signature is valid and ensure that the model parameters have not been tampered with during transmission. This process not only ensures the authenticity of the uploaded model parameters, but also protects the identity privacy of the uploader.

[0022] In some embodiments, the model parameter authentication step specifically includes:

[0023] After the smart contract verifies the legitimacy of the signature, it notifies all users to download the model parameters, which are verified and scored by the user's local data set, and then the verification results are fed back to the blockchain / smart contract using the personal wallet. The blockchain detects the score count. If the score is qualified, the user is set as a whitelist user. The next verification must wait for the whitelist user to feedback the score result. If the score is unqualified, the ring where the user is located is disassembled in the system, and all users in the ring are set as blacklist users. When the number of blacklist users exceeds one-third of the number of system users, all users in the blacklist are removed and the ring is randomly re-formed.

[0024] In some embodiments, the model aggregation step specifically includes:

[0025] When the number of model parameters with qualified scores reaches 5% of the total number of system users, the smart contract automatically triggers the model aggregation process. This process first involves the smart contract collecting all verified qualified model parameters that are trained and uploaded by the client based on the local dataset. The smart contract then executes a weighted average algorithm to generate an updated global model.

[0026] The present invention also proposes a federated average learning system based on ring signature to conceal client identity, the system comprising:

[0027] Training unit: The client downloads the global model through the blockchain and trains it with local data to obtain local gradients;

[0028] Anonymization unit: the client anonymizes the model update through a ring signature algorithm and uploads the local gradient and signature information to the blockchain network;

[0029] Testing unit: the client downloads the trained gradients and performs tests, and uploads the test results;

[0030] The global update unit updates the global model after the smart contract collects the test results of the preset conditions.

[0031] Based on the above scheme, the present invention provides a federated average learning method and system based on ring signature to conceal the client identity, which combines blockchain technology and ring signature algorithm to build a decentralized federated learning environment, avoid centralized dependence, protect data privacy and improve the credibility of model updates. At the same time, through the tamper-proof and traceability of blockchain, the security and transparency of the system are ensured. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flowchart of the steps of a federated average learning method for concealing client identity based on ring signature of the present invention;

[0033] Figure 2 It is a data flow diagram of a specific embodiment of the present invention;

[0034] Figure 3 It is a structural block diagram of a federated average learning system based on ring signature to conceal client identity in the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0036] It should be noted that, for the convenience of description, only the parts related to the relevant invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0037] It should be understood that the "system", "device", "unit" and / or "module" used in this application is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.

[0038] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not refer to the singular, but also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. The elements defined by the sentence "includes a..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

[0039] In the description of the embodiments of the present application, "plurality" means two or more than two. The following terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0040] In addition, flow charts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, the various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0041] Reference Figure 1 , which is a flow chart of an optional example of a federated average learning method for concealing client identity based on ring signature proposed in the present invention. The method can be applied to a computer device. The imaging method proposed in this embodiment may include but is not limited to the following steps:

[0042] Step S0: Verify the client identity information and public information through smart contracts and upload them to the chain;

[0043] Step S1: The client downloads the global model through the blockchain and trains it with local data to obtain local gradients;

[0044] Step S2: The client anonymizes the model update through a ring signature algorithm, and uploads the local gradient and signature information to the blockchain network;

[0045] Step S3: The client downloads the trained gradient and performs a test, and uploads the test result;

[0046] Step S4: After the smart contract collects the test results of the preset conditions, it updates the global model.

[0047] The overall data flow of this method refers to Figure 2 .

[0048] In some feasible embodiments, the step S0 specifically includes:

[0049] During the system initialization phase, user C i Private key usk i The system uses a random number generation algorithm in a finite field Pick a random number x i 、r i Calculate c i 、s i Generate the private key usk i It is important secret information used by the client in subsequent signing operations and will not be disclosed. i It is through the hash function H(·) combined with the user's identity ID i Generate. Generator P is selected from group G1 and used to calculate the public key. Finally, the system returns the public-private key pair to the client, where the public key will be broadcasted to the blockchain.

[0050] In some feasible embodiments, the step S1 specifically includes:

[0051] Client C i The local dataset D i The batch size is B, which is used for the calculation of a model update. The number of local training rounds E is pre-specified. In each round of local training, the local dataset D of client i is calculated. i The gradient of each mini-batch b in It is the loss function with respect to the current model parameters The gradient of the model is then updated using the gradient descent method. The gradient descent algorithm gradually approaches the minimum value of the loss function by iteratively calculating the gradient and updating the parameters. In each iteration, the gradient provides information about how to adjust the parameters to reduce the loss as quickly as possible under the current parameter position.

[0052] In some feasible embodiments, the step S2 specifically includes:

[0053] S2.1. Collect the public keys of all participating clients to build the basis of the signature ring;

[0054] Specifically, the system collects the public keys of all clients participating in the signature ring∑PK=pk1,pk2,…,pk n These public keys are used to build the basis of the signature ring. Next, the signing client selects a random number r' s , and calculate U using this random number s , which is an intermediate value used to identify the client's identity. In this process, the public keys of other clients also participate in the U s , so that all clients in the signature ring are included.

[0055] S2.2. Use hash functions to confuse information such as message m and public key to ensure the anonymity of the ring signature;

[0056] Specifically, the hash function H1 is used to perform a hash operation on the message m, the public key and other information to generate H i The purpose of this step is to ensure that the identity of each ring signature member is obfuscated, thereby ensuring anonymity. The signing client uses its own private key x s And the generated random number r' s Calculate the key variable v of the signature, which is an important component of the ring signature and ensures the validity of the signature.

[0057] S2.3. Generate a new random number t and calculate the intermediate result T as the key point for signature verification;

[0058] Specifically, the signing client generates a new random number t and uses it to calculate the intermediate result T, which will be used as a verification point for the signature. Subsequently, the system calculates the hash values ​​c and e, which are used by the smart contract on the blockchain to verify the legitimacy of the signature and ensure that the uploaded model parameters have not been tampered with during the transmission process.

[0059] S2.4. Complete the generation of the ring signature and upload all calculation results to the blockchain;

[0060] Specifically, after the signature is completed, all calculation results σ include: all U i Value, signature variable v, hash value c, intermediate result T and final signature π. These data constitute a complete ring signature, which is uploaded to the blockchain.

[0061] In some feasible embodiments, the step S3 specifically includes:

[0062] S3.1. Extract the public key set and calculate the hash value to verify the signature consistency;

[0063] Specifically, the public key set ∑PK is extracted from the input information. These public keys are used to build the signature ring during the signing process. The system uses the hash function H1, combined with some parameters of the public parameter params, the message m, the intermediate result T and the hash value c in the signature, to calculate a new hash value e, which is used to verify the consistency of the signing process. Next, t' is obtained through the final proof π of the ring signature and the previously calculated e. This value is used to verify the correctness of the random number. T' is calculated through t' and is used to check the randomness of the entire signing process.

[0064] S3.2. Recalculate the hash value to ensure that the signature parameters have not been tampered with during the verification process;

[0065] Specifically, the system recalculates the hash value c'. This step is to ensure that the signer's public key and other signature parameters have not been tampered with during the verification process. Next, the system verifies each client participating in the signature ring, using the hash value H generated by the message m and the corresponding public key. i To ensure that the signatures of all signature ring members remain consistent.

[0066] S3.3. Calculate the intermediate value and verify the signature integrity to ensure the return of verification success or failure;

[0067] Specifically, the system calculates U' s , which is an intermediate value calculated by T'=t'P and other related public keys. If the calculated U' s and U generated during the signing process s If they are inconsistent, it means that the signature has been tampered with or the verification has failed, and the system will return a verification failure result. s and U s If they are consistent, the signature verification is successful and a success message is returned. This process ensures that the signature can be legally verified while maintaining anonymity, thus ensuring the security and credibility of the entire federated learning system.

[0068] In some feasible embodiments, the step S4 specifically includes:

[0069] S4.1. Initialize global model parameters and accumulator M to collect model updates from clients.

[0070] Specifically, we first initialize the global model parameter w0. Then, we initialize an accumulator M to collect the model updates of all selected clients. For each consecutive round of training t, we collect all the participating clients C i 5% of the selected clients C are randomly selected for local model update, in which the ClientUpdate algorithm is called. i Using its local data D iPerform E rounds of local training, where each round of training uses a small batch of data of size B and updates the model parameters with a learning rate η. The updated model parameters Uploaded to the server.

[0071] The aggregate update process is as follows:

[0072]

[0073] in, represents the updated model, M represents the total number of models uploaded by the client, and D m represents the dataset of the mth client, |D m | represents the number of data points in the dataset of the mth client, |D| represents the total number of data points, i.e., the sum of the number of data points of all clients, and w m Represents the mth model.

[0074] S4.2, the server aggregates client updates and calculates the average model, and broadcasts it to all clients;

[0075] Specifically, the server then traverses S t Each client in the selected client accumulates their model updates into M. Once the updates of all selected clients are accumulated, the algorithm calculates the accumulated value M and the total number of selected clients |S t |, that is, to calculate the average model update. This average value is used as the global model parameter update is returned and used to update the global model. This process ensures that all participating clients have an equal say in global model training while maintaining the consistency and effectiveness of model updates. After aggregation is completed, the server sends the new global model parameters w t+1 Broadcast back to all clients so that each client can use the latest global model for the next round of local updates. This process will be repeated.

[0076] Based on the above scheme, the specific beneficial effects of the specific embodiment of the present invention are as follows:

[0077] 1. Completely decentralized federated learning framework: Traditional federated learning systems rely on centralized servers to aggregate and manage model parameters. This architecture not only faces the risk of single point failure, but also poses a threat to data privacy. This invention uses blockchain technology to build a decentralized federated learning framework, eliminating the reliance on central servers. Through smart contracts on the blockchain, the system can automatically manage model updates submitted by clients and execute the model aggregation process, avoiding the potential performance bottlenecks and trust crises of centralized servers.

[0078] 2. Strong anonymity and privacy protection guaranteed by ring signature technology: Ring signature technology provides strong anonymity protection for model updates submitted by clients. Ring signature technology allows clients to submit signatures anonymously in a set of public keys, and the true source of the signature cannot be traced. This feature ensures that the identity of the client will not be exposed during the entire federated learning process, greatly improving the ability to protect data privacy. Compared with traditional signature technology, ring signatures not only protect personal privacy, but also ensure the authenticity and immutability of signatures.

[0079] 3. Tamper-proof and traceable on-chain data management: The introduction of blockchain technology provides extremely high security for the federated learning system of the present invention. All model updates and verification results are recorded on the blockchain, and the distributed storage characteristics of the blockchain ensure the tamper-proof and traceability of these data. Through the smart contract of the blockchain, the system can automatically verify the legitimacy of the model update and upload qualified model parameters to the chain for storage, avoiding the risk of tampering and forgery of data during transmission.

[0080] 4. Efficient model update and aggregation mechanism: Compared with the prior art, the present invention has significant efficiency improvements in model update and aggregation. Through smart contracts on the blockchain, the update and verification process of model parameters can be completed automatically, reducing the need for manual intervention. In addition, the present invention adopts an efficient weighted average aggregation algorithm to fuse the local model updates of multiple clients to generate a global model. After reaching a certain verification result, the system automatically triggers the global model update and ensures the fairness and effectiveness of the model update process, avoiding inefficient centralized aggregation operations.

[0081] 5. Anti-coercion and system legitimacy verification: In the system of the present invention, the anonymity of the ring signature technology effectively resists external coercion. Since the source of the signature cannot be traced, the attacker cannot influence the model update submitted by a certain client by forcing it. In addition, the present invention ensures the legitimacy of each client in the system through the combination of smart contracts and blockchain. The system assigns an encrypted identity to each client during the initialization phase. Only clients that have passed the legal identity authentication can participate in the federated learning process, which fundamentally eliminates the intrusion of illegal users.

[0082] 6. System transparency and verifiability: The present invention optimizes system transparency. All operations, such as model parameter submission, verification, and model aggregation, are performed openly and transparently through the blockchain. Any node participating in the system can verify the authenticity and legality of each operation through the blockchain. In addition, through the combination of ring signature technology and blockchain smart contracts, any client in the system can verify whether the model parameters submitted by others are processed correctly. This fully verifiable mechanism ensures the fairness and credibility of the system.

[0083] like Figure 3 As shown, a federated average learning system based on ring signature to conceal client identity includes:

[0084] Training unit: The client downloads the global model through the blockchain and trains it with local data to obtain local gradients;

[0085] Anonymization unit: the client anonymizes the model update through a ring signature algorithm and uploads the local gradient and signature information to the blockchain network;

[0086] Testing unit: the client downloads the trained gradients and performs tests, and uploads the test results;

[0087] The global update unit updates the global model after the smart contract collects the test results of the preset conditions.

[0088] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0089] A federated average learning device based on ring signature to conceal client identity:

[0090] at least one processor;

[0091] at least one memory for storing at least one program;

[0092] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned federated average learning method for concealing client identity based on ring signature.

[0093] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0094] A storage medium stores processor-executable instructions, which are used to implement the above-mentioned federated average learning method for concealing client identity based on ring signature when executed by the processor.

[0095] The contents of the above method embodiments are all applicable to the present storage medium embodiments. The functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0096] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A federated average learning method based on ring signature to conceal client identity, characterized in that: The following steps are involved: The client downloads the global model through the blockchain and trains it with local data to obtain local gradients; The client anonymizes the model update through the ring signature algorithm and uploads the local gradient and signature information to the blockchain network; The client downloads the trained gradients and performs tests, and uploads the test results; When the smart contract collects the test results of the preset conditions, it updates the global model.

2. According to claim 1, a method for federated average learning based on ring signature to conceal client identity, characterized in that: The client downloads the global model through the blockchain and trains it with local data to obtain the local gradient, and also includes: The client’s identity information and public information are verified through smart contracts and uploaded to the chain.

3. According to claim 1, a method for federated average learning based on ring signature to conceal client identity, characterized in that: The client anonymizes the model update through the ring signature algorithm and uploads the local gradient and signature information to the blockchain network. This step specifically includes: Obtain the public key information of other members; The client combines the private key, public key information and public parameters in the blockchain, calls the signature generation algorithm to generate an anonymous signature for the local gradient, and obtains the signature information; Upload local gradient and signature information to the blockchain network.

4. According to claim 1, a method for federated average learning based on ring signature to conceal client identity, characterized in that: The client downloads the trained gradient and performs a test, and uploads the test result. The step specifically includes: Verify the legitimacy of the signature information through a smart contract to obtain a verification result; Based on the verification results, the client downloads the trained gradients and uses the corresponding local data for testing to obtain the test results; Upload the test results to the blockchain.

5. According to claim 4, a method for federated average learning based on ring signature to conceal client identity, characterized in that: Also includes: The blockchain determines the scores in the test results; If the rating is qualified, the user will be set as a whitelist user, and the next verification will wait for the whitelist user to feedback the rating result; If the score is not up to standard, all users in the signature ring will be blacklisted; When the number of blacklisted users exceeds the preset threshold, all blacklisted users are removed and the ring is randomly created again.

6. According to claim 1, a method for federated average learning based on ring signature to conceal client identity, characterized in that: The update formula of the global model is expressed as follows: in, represents the updated model, M represents the total number of models uploaded by the client, and D m represents the dataset of the mth client, |D m | represents the number of data points in the dataset of the mth client, |D| represents the total number of data points, i.e., the sum of the number of data points of all clients, and w m Represents the mth model.

7. According to claim 1, a method for federated average learning based on ring signature to conceal client identity, characterized in that: Also includes: Broadcast the updated global model parameters to all clients via the blockchain; After the client receives the global model, it performs a new round of local training based on the corresponding local data.

8. A federated average learning system based on ring signature to conceal client identity, characterized in that: include: Training unit: The client downloads the global model through the blockchain and trains it with local data to obtain local gradients; Anonymization unit: the client anonymizes the model update through a ring signature algorithm and uploads the local gradient and signature information to the blockchain network; Testing unit: the client downloads the trained gradients and performs tests, and uploads the test results; The global update unit updates the global model after the smart contract collects the test results of the preset conditions.

9. A federated average learning device based on ring signature to conceal client identity, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a federated average learning method for concealing client identity based on a ring signature as described in any one of claims 1 to 7.

Images