Privacy-protecting data collaborative analysis platform based on federated learning

Abstract

The application discloses a kind of privacy protection data collaborative analysis platform based on federal learning, it is related to privacy computing field, including: central server, initialization global model and distribute to client, the model update of feedback to client is aggregated, obtains aggregation parameter and updates global model, and updated global model is distributed to client;Multiple clients, each client receives global model and utilizes local data to carry out model training, obtains model update, and is fed back to central server;Intelligent routing gateway cluster, several routing groups are randomly divided, and a group of routing groups is allocated to each client to build anonymous communication path, and model update and global model are transmitted through anonymous communication path.The application can realize privacy protection when cross-institution data joint modeling, improve the security of data collaborative analysis.

Description

Technical Field

[0001] This invention relates to the field of privacy computing technology, and more specifically to a privacy-preserving data collaborative analysis platform based on federated learning. Background Technology

[0002] With the rapid development of information technology, data has become an important resource in modern society. However, "data silos" exist between different industries such as finance and healthcare. Due to privacy regulations (such as GDPR), institutions cannot directly share raw data. Therefore, how to achieve data sharing and how to protect personal privacy and data security during the data sharing process has become a global challenge.

[0003] Traditional data sharing methods often involve the risk of sensitive information leakage, limiting the maximization of data value. Even using traditional encryption or anonymization techniques during data sharing still carries the risk of data leakage and cannot support complex machine learning tasks. To address this issue, federated learning, as an emerging machine learning technology, has gained significant attention in recent years due to its ability to achieve data sharing and model training while protecting user privacy.

[0004] Federated learning, as a distributed machine learning framework, enables multiple participants to collaboratively train a global model without sharing training data, thus effectively ensuring client data privacy and security. However, federated learning still faces risks such as model parameter leakage and identity privacy threats during communication.

[0005] Therefore, how to improve the privacy and security of the data sharing process is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a privacy-preserving data collaborative analysis platform based on federated learning, which can realize privacy protection when jointly modeling data across institutions and improve the security of data collaborative analysis.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A privacy-preserving collaborative data analysis platform based on federated learning includes:

[0009] The central server initializes the global model and distributes it to the clients. It aggregates the model updates reported by the clients, obtains the aggregation parameters, updates the global model, and distributes the updated global model to the clients.

[0010] Multiple clients, each client receives the global model and uses local data to train the model, obtains model updates, and feeds them back to the central server;

[0011] The intelligent routing gateway cluster randomly divides into several routing groups and assigns a routing group to each client to build an anonymous communication path. Model updates and the global model are transmitted through the anonymous communication path.

[0012] Preferably, the process of randomly dividing the intelligent routing gateway cluster into several routing groups is as follows:

[0013] Step 11: Obtain the gateway node list and grouping parameters, and initialize the cryptographic security entropy source;

[0014] Grouping parameters include a limit on the number of gateway nodes in the routing group, with both a maximum and a minimum limit; the cryptographically secure entropy source uses a secure random number generator and establishes an entropy pool connection;

[0015] Step 12: Generate random numbers based on the cryptographic security entropy source;

[0016] Step 13: Based on the random number, use the improved Fisher-Yates shuffling algorithm to randomly permutate the list of gateway nodes to obtain a random sequence, thus achieving shuffling;

[0017] Step 14: Randomly select the number of gateway nodes from the grouping parameters as the segmentation threshold, divide the random sequence into several routing groups according to the segmentation threshold, assign a route ID to each routing group, and assign a node ID to each gateway node in each routing group; when the number of gateway nodes in the gateway node list is not divisible by the number of groups, adjust the number of gateway nodes in the last routing group.

[0018] Preferably, the specific process of step 13 is as follows:

[0019] Step 131: Create an initial gateway node sequence based on the gateway node list;

[0020] Step 132: Use the random number as the key to set the random permutation function;

[0021] Step 133: Permutate the gateway nodes in the initial gateway node sequence according to the random permutation function to obtain a random sequence.

[0022] Preferably, the process of assigning a group of routes to each client to build an anonymous communication path is as follows:

[0023] Step 21: Client C i Generate an access request and send it to the smart routing gateway cluster to request access permissions for the routing group;

[0024] Step 22: The intelligent routing gateway cluster uses a weighted round-robin algorithm to dynamically load balance the routing groups, assigning a corresponding routing group to each client;

[0025] Step 23: Construct anonymous communication paths based on the assigned routing groups, and use the anonymous communication paths to transmit model updates and the global model.

[0026] Preferably, the specific process of step 22 is as follows:

[0027] Step 221: The intelligent routing gateway cluster constructs a routing group weight matrix based on the real-time load status of each routing group;

[0028] Step 222: Construct the routing group state table based on the routing group ID, routing group weight matrix, and number of allocations;

[0029] Step 223: For each client C i Perform a round-robin assignment, select the route group with the highest weight and assign it to the client, and update the assignment count and the weight of the current route group in the route group status table until all clients are assigned the corresponding route group.

[0030] Preferably, the specific process of step 23 is as follows:

[0031] Step 231: For each client's assigned routing group, each gateway node generates a corresponding public key PK and private key SK, and transmits the public key PK of all gateway nodes to the corresponding client and central server.

[0032] Step 232: Arrange all gateway nodes in the routing group according to their communication distance with the client, and construct anonymous communication paths;

[0033] Step 233: The client generates an encrypted transmission command based on the communication direction of the anonymous communication path using the public keys of all gateway nodes and the model update, and transmits the encrypted transmission command to the first gateway node; after being decrypted by each gateway node in the anonymous communication path, the model update is transmitted to the central server.

[0034] The central server generates an update command using the public keys of all gateway nodes and the global model based on the reverse communication direction of the anonymous communication path, and transmits the update command to the last gateway node; after being decrypted by each gateway node in the anonymous communication path, the global model is transmitted to the corresponding client.

[0035] Preferably, the encrypted transmission instruction includes several layers of shared keys and model update ciphertext; a shared key is generated based on the public key of a gateway node and the node ID of the next gateway node in the communication direction; the client encrypts the model update using its local public key to obtain the model update ciphertext; the update instruction includes several layers of shared keys and global model ciphertext; the central server encrypts the global model using its server private key to obtain the global model ciphertext; and the gateway node's private key is used to decrypt the shared key of the corresponding layer to determine the ID of the next gateway node for transmission.

[0036] Preferably, initially, the client generates a registration request based on its local public key and sends it to the central server. The central server receives the registration request and saves its local public key, while simultaneously sending its own public key back to the client. The client receives and saves the server's public key. The client then transmits the encrypted model update to the central server via an anonymous communication path. The central server uses its saved local public key to decrypt the updated model. Finally, the central server transmits the encrypted global model to the client via an anonymous communication path. The client uses its saved server public key to decrypt the global model.

[0037] Preferably, the central server also includes a heterogeneous adapter. During client registration, the heterogeneous adapter collects and analyzes the hardware capabilities, network status, and data characteristics of each client to generate a device performance profile. Based on the device performance profile, it assigns a suitable model architecture to the client: a lightweight model to low-computing-power clients, a pruned model to medium-computing-power clients, and a complete global model to high-computing-power clients. It performs structured pruning operations on the complete global model, freezes the selection layer, and obtains the pruned model. It then performs knowledge distillation on the pruned model to obtain a lightweight model.

[0038] Preferably, the client also includes a differential privacy injection unit, which adds noise layer by layer during model training. The specific process is as follows:

[0039] Step 31: Preset privacy noise allocation principles according to the global model structure;

[0040] Step 32: Calculate the sensitivity for each layer of the global model and perform gradient clipping to ensure that the sensitivity is bounded;

[0041] Step 33: Calculate the noise scale of each layer based on the privacy noise allocation principle and sensitivity, and construct the noise matrix corresponding to each layer;

[0042] Step 34: Add corresponding noise matrices to different layer structures during the forward propagation and backward propagation phases.

[0043] Preferably, the layer structure includes an input layer, an output layer, and an intermediate layer; the privacy noise allocation principle includes: the output layer is allocated the highest amount of noise, the intermediate layer is allocated a medium amount of noise, and the input layer is allocated the lowest amount of noise; at the beginning of the forward propagation phase, the noise matrix corresponding to the input layer is added to the input layer; when calculating the gradient in the backpropagation phase, the noise matrix corresponding to the intermediate layer is added to the intermediate layer; and when the gradient calculation is completed in the backpropagation phase, the noise matrix corresponding to the output layer is added to the output layer.

[0044] Preferably, the central server integrates a TEE (Trusted Execution Environment) module, in which global model initialization, model update aggregation, and global model update are performed.

[0045] Preferably, the client also has a zero-knowledge proof verifier to verify the zero-knowledge proofs generated during the client's model training process, verifying whether the training conforms to the protocol (e.g., no illegal data was used). If it does not conform, the current learning session is stopped.

[0046] As can be seen from the above technical solutions, compared with the prior art, this invention discloses a privacy-preserving data collaborative analysis platform based on federated learning. It dynamically constructs random routing groups through an intelligent routing gateway cluster, employs an improved Fisher-Yates shuffling algorithm to ensure unpredictable gateway allocation, and effectively hides the direct association between clients and servers. Based on a weighted round-robin algorithm, it dynamically allocates routing groups, achieving efficient dynamic resource scheduling, real-time response to changes in gateway cluster load, avoiding single-point congestion, and improving communication efficiency. It implements layered encrypted transmission based on anonymous communication paths, ensuring that model parameters cannot be decrypted even if intercepted during transmission, preventing eavesdropping by intermediate nodes. Heterogeneous adapters dynamically allocate models according to device performance profiles (computing power / network status), significantly reducing the computational load on edge devices and solving the efficiency bottleneck caused by device heterogeneity in federated learning. Layered noise addition avoids excessive global perturbation, improving model accuracy under the same privacy budget. Global model initialization, aggregation, and updates are performed within the central server's TEE module, ensuring that sensitive operations run in a hardware-level secure environment and defending against malicious code tampering. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0048] Figure 1 A schematic diagram of the privacy-preserving data collaborative analysis platform based on federated learning provided by this invention;

[0049] Figure 2 This is a schematic diagram of the process of dividing routing groups provided by the present invention;

[0050] Figure 3 This is a schematic diagram illustrating the process of constructing an anonymous communication path provided by the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] This invention discloses a privacy-preserving data collaborative analysis platform based on federated learning, such as... Figure 1 As shown, it includes:

[0053] The central server initializes the global model and distributes it to the clients. It aggregates the model updates reported by the clients, obtains the aggregation parameters, updates the global model, and distributes the updated global model to the clients.

[0054] Multiple clients, each client receives the global model and uses local data to train the model, obtains model updates, and feeds them back to the central server;

[0055] The intelligent routing gateway cluster randomly divides into several routing groups and assigns a routing group to each client to build an anonymous communication path. Model updates and the global model are transmitted through the anonymous communication path.

[0056] Furthermore, such as Figure 2 As shown, the process of randomly dividing the intelligent routing gateway cluster into several routing groups is as follows:

[0057] S11: Obtain the list of gateway nodes and grouping parameters, and initialize the cryptographic security entropy source;

[0058] Grouping parameters include a limit on the number of gateway nodes in the routing group, with both a maximum and a minimum limit; the cryptographically secure entropy source uses a secure random number generator and establishes an entropy pool connection;

[0059] S12: Generate random numbers based on cryptographic security entropy sources;

[0060] Read the cryptographically secure entropy source: read(entropy_fd, buffer, 256);

[0061] Generate cryptographically secure random numbers based on the cryptographically secure entropy source: seed = SHA3-256(buffer);

[0062] S13: Based on the random number, the improved Fisher-Yates shuffle algorithm is used to randomly permutate the list of gateway nodes to obtain a random sequence, thereby achieving shuffling;

[0063] S14: Randomly select the number of gateway nodes from the grouping parameters as the segmentation threshold. Based on the segmentation threshold, divide the shuffled random sequence into several routing groups, assign a route ID to each routing group, and assign a node ID to each gateway node in each routing group. When the number of gateway nodes in the gateway node list is not divisible by the number of groups, adjust the number of gateway nodes in the last routing group.

[0064] Furthermore, the specific process of S13 is as follows:

[0065] S131: Create an initial gateway node sequence based on the gateway node list; seq = [1,2,3,…,m] represents the ordered arrangement of gateway indices;

[0066] S132: Use the random number as the key to set the random permutation function;

[0067] nonce = seed[16:24];

[0068] PRP = def PRP(key, nonce, i);

[0069] S133: Permutate the gateway nodes in the initial gateway node sequence according to the random permutation function to obtain a random sequence.

[0070] Furthermore, such as Figure 3 As shown, the process of assigning a group of routes to each client to build an anonymous communication path is as follows:

[0071] S21: Client C i Generate an access request and send it to the smart routing gateway cluster to request access permissions for the routing group;

[0072] S22: The intelligent routing gateway cluster uses a weighted round-robin algorithm to dynamically load balance the routing groups and assign a corresponding routing group to each client.

[0073] S23: Construct anonymous communication paths based on the assigned routing groups, and use the anonymous communication paths to transmit model updates and the global model.

[0074] Furthermore, the specific process of S22 is as follows:

[0075] S221: The intelligent routing gateway cluster constructs a routing group weight matrix W based on the real-time load status of each routing group.

[0076]

[0077] α+β=1;

[0078] This represents the weight of the i-th routing group; This represents the average latency within the i-th routing group. This represents the available bandwidth of the i-th routing group; The maximum bandwidth represents the sum of the physical / logical bandwidths of all gateway nodes within the routing group, and can be dynamically configured based on SDN (Software Defined Networking); α and β represent the weights, with α taking a value of 0.3 and β taking a value of 0.7; the weights of all routing groups are normalized, and the normalized weights are used to construct the routing group weight matrix;

[0079] S222: Construct a routing group state table based on the routing group ID, the normalized weights in the routing group weight matrix, the weights of the current routing group, and the number of allocations;

[0080] S223: For each client C i Perform a round-robin assignment, select the route group with the highest current weight and assign it to the client, and update the assignment count and the current weight of the route group in the route group state table until all clients are assigned the corresponding route group; subtract the sum of all normalized weights from the current weight of the assigned route group, and increase the current weight of all route groups by the corresponding normalized weight.

[0081] The weighted round-robin algorithm can effectively balance the load, allowing servers or processors with strong processing capabilities to handle more requests or tasks, while also ensuring that even nodes with weaker processing capabilities will not be completely left without tasks, thereby improving resource utilization and service response speed.

[0082] Furthermore, the specific process of S23 is as follows:

[0083] S231: For each client, a routing group G is assigned. k ={g1, g2, ..., g m}, m represents the number of gateway nodes in the routing group. Each gateway node generates a corresponding public key PK and private key SK, and transmits the public key PK of all gateway nodes to the corresponding client and central server.

[0084] S232: Construct anonymous communication paths by arranging all gateway nodes in the routing group according to their communication distance to the client;

[0085] S233: The client generates an encrypted transmission command using the public keys of all gateway nodes and the model update according to the communication direction of the anonymous communication path, and transmits the encrypted transmission command to the first gateway node; after being decrypted by each gateway node in the anonymous communication path, the model update is transmitted to the central server; the central server generates an update command using the public keys of all gateway nodes and the global model according to the reverse communication direction of the anonymous communication path, and transmits the update command to the last gateway node; after being decrypted by each gateway node in the anonymous communication path, the global model is transmitted to the corresponding client.

[0086] Furthermore, the encrypted transmission command includes several layers of shared keys and model update ciphertext; a shared key is generated based on the public key of a gateway node and the node ID of the next gateway node in the communication direction; the client encrypts the model update using its local public key to obtain the model update ciphertext; the update command includes several layers of shared keys and global model ciphertext; the central server encrypts the global model using its server private key to obtain the global model ciphertext; and the gateway node's private key is used to decrypt the shared key of the corresponding layer to determine the ID of the next gateway node for transmission.

[0087] Furthermore, initially, the client generates a registration request based on its local public key and sends it to the central server. The central server receives the registration request and saves its local public key, while simultaneously sending its own public key back to the client. The client receives and saves the server's public key. The client then transmits the encrypted model update to the central server via an anonymous communication path. The central server uses its saved local public key to decrypt the updated model. Finally, the central server transmits the encrypted global model to the client via an anonymous communication path. The client uses its saved server public key to decrypt the global model.

[0088] On the other hand, in one specific embodiment, the central server also includes a heterogeneous adapter. When a client registers, the heterogeneous adapter collects and analyzes the hardware capabilities, network status, and data characteristics of each client to generate a device performance profile. Based on the device performance profile, it assigns a suitable model architecture to the client: a lightweight model to low-computing-power clients, a pruned model to medium-computing-power clients, and a complete global model to high-computing-power clients. It performs structured pruning operations on the complete global model, freezes the selection layer, and obtains the pruned model. It then performs knowledge distillation on the pruned model to obtain a lightweight model.

[0089] On the other hand, in one specific embodiment, the client also includes a differential privacy injection unit, which adds noise (Gaussian noise distribution or Laplacian noise) in layers during model training. The specific process is as follows:

[0090] S1: Pre-set privacy noise allocation principles according to the global model structure;

[0091] S2: Calculate the sensitivity for each layer of the global model and perform gradient clipping to ensure that the sensitivity is bounded;

[0092] S3: Calculate the noise scale of each layer based on the privacy noise allocation principle and sensitivity, and construct the noise matrix corresponding to each layer;

[0093] S4: Add corresponding noise matrices to different layer structures during the forward propagation and backward propagation phases.

[0094] Furthermore, the layer structure includes an input layer, an output layer, and intermediate layers; the privacy noise allocation principle includes: the output layer is allocated the highest amount of noise, the intermediate layer is allocated a medium amount of noise, and the input layer is allocated the lowest amount of noise; at the beginning of the forward propagation phase, the noise matrix corresponding to the input layer is added to the input layer; when calculating the gradient in the backpropagation phase, the noise matrix corresponding to the intermediate layer is added to the intermediate layer; and when the gradient calculation is completed in the backpropagation phase, the noise matrix corresponding to the output layer is added to the output layer.

[0095] Noise is dynamically allocated according to the model's layer structure (input layer / intermediate layer / output layer). The strongest noise is added to the output layer to protect the prediction results, while the noise is minimized in the input layer to preserve feature validity. Combined with gradient pruning to constrain sensitivity, fine-grained control of the privacy budget is achieved, satisfying differential privacy requirements while ensuring model usability. This layered noise injection scheme works in deep collaboration with the federated learning framework. The client completes noise injection during local training, and the intelligent routing gateway provides supplementary anonymity protection during transmission, forming a dual privacy protection system of "local noise addition + transmission anonymity".

[0096] On the other hand, in one specific embodiment, the central server integrates a TEE (Trusted Execution Environment) module, which performs global model initialization, model update aggregation, and global model update. Performing global model initialization, aggregation, and updates within the central server's TEE module ensures that sensitive operations run in a hardware-level secure environment, defending against malicious code tampering.

[0097] On the other hand, in one specific embodiment, the client also includes a zero-knowledge proof verifier to verify the zero-knowledge proofs generated during the client's model training process. This verifies whether the training conforms to the protocol; if not, the current learning session is stopped. By verifying the compliance of the training process (e.g., the absence of illegal data) through zero-knowledge proofs, the client ensures the credibility of the participants' actions and prevents malicious attacks from the outset.

[0098] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0099] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A privacy-preserving data collaborative analysis platform based on federated learning, characterized in that, include: The central server initializes the global model and distributes it to the clients. It aggregates the model updates reported by the clients, obtains the aggregation parameters, updates the global model, and distributes the updated global model to the clients. Multiple clients, each client receives the global model and uses local data to train the model, obtains model updates, and feeds them back to the central server; The intelligent routing gateway cluster randomly divides into several routing groups and assigns a routing group to each client to build an anonymous communication path. Model updates and the global model are transmitted through the anonymous communication path. The process of randomly dividing a smart routing gateway cluster into several routing groups is as follows: Step 11: Obtain the list of gateway nodes and grouping parameters, and initialize the cryptographic security entropy source; the grouping parameters include the limit on the number of gateway nodes in the routing group; Step 12: Generate random numbers based on the cryptographic security entropy source; Step 13: Using a modified Fisher-Yates shuffling algorithm based on random numbers, randomly permutate the gateway node list to obtain a random sequence; the specific process of step 13 is as follows: Step 131: Create an initial gateway node sequence based on the gateway node list; Step 132: Use the random number as the key to set the random permutation function; Step 133: Permutate the gateway nodes in the initial gateway node sequence according to the random permutation function to obtain a random sequence; Step 14: Randomly select the number of gateway nodes from the grouping parameters as the segmentation threshold, divide the random sequence into several routing groups according to the segmentation threshold, assign a route ID to each routing group, and assign a node ID to each gateway node in each routing group; when the number of gateway nodes in the gateway node list is not divisible by the number of groups, adjust the number of gateway nodes in the last routing group. The process of assigning a group of routes to each client to build an anonymous communication path is as follows: Step 21: The client generates an access request and sends it to the smart routing gateway cluster to request access permission for the routing group; Step 22: The intelligent routing gateway cluster uses a weighted round-robin algorithm to dynamically load balance the routing groups, assigning a corresponding routing group to each client; the specific process of step 22 is as follows: Step 221: The intelligent routing gateway cluster constructs a routing group weight matrix based on the real-time load status of each routing group; Step 222: Construct the routing group state table based on the routing group ID, routing group weight matrix, and number of allocations; Step 223: Perform a round-robin assignment for each client, select the route group with the highest weight and assign it to the client, and update the assignment count and the weight of the current route group in the route group status table until all clients are assigned the corresponding route group; Step 23: Construct anonymous communication paths based on the assigned routing groups, and use the anonymous communication paths to transmit model updates and the global model.

2. The privacy-preserving data collaborative analysis platform based on federated learning according to claim 1, characterized in that, The specific process of step 23 is as follows: Step 231: For each client's assigned routing group, each gateway node generates a corresponding public key and private key, and transmits the public keys of all gateway nodes to the corresponding client and central server; Step 232: Arrange all gateway nodes in the routing group in order of their communication distance to the client from farthest to closest, and construct anonymous communication paths; Step 233: The client generates an encrypted transmission command based on the communication direction of the anonymous communication path using the public keys of all gateway nodes and the model update, and transmits the encrypted transmission command to the first gateway node; after being decrypted by each gateway node in the anonymous communication path, the model update is transmitted to the central server. The central server generates an update command using the public keys of all gateway nodes and the global model based on the reverse communication direction of the anonymous communication path, and transmits the update command to the last gateway node; after being decrypted by each gateway node in the anonymous communication path, the global model is transmitted to the corresponding client.

3. The privacy-preserving data collaborative analysis platform based on federated learning according to claim 2, characterized in that, The encrypted transmission command includes several layers of shared keys and model update ciphertext; a shared key is generated based on the public key of a gateway node and the node ID of the next gateway node in the communication direction; the client encrypts the model update using its local public key to obtain the model update ciphertext; the update command includes several layers of shared keys and global model ciphertext; the central server encrypts the global model using its server private key to obtain the global model ciphertext; the gateway node's private key is used to decrypt the shared key of the corresponding layer to determine the ID of the next gateway node for transmission.

4. The privacy-preserving data collaborative analysis platform based on federated learning according to claim 3, characterized in that, Initially, the client generates a registration request based on its local public key and sends it to the central server. The central server receives the registration request and saves the local public key, while simultaneously sending the server's public key back to the client. The client receives and saves the server's public key. The client transmits the encrypted model update to the central server via an anonymous communication path. The central server decrypts the data using its stored local public key to obtain the model update. The central server also transmits the encrypted global model to the client via an anonymous communication path. The client decrypts the data using its stored server public key to obtain the global model.

5. The privacy-preserving data collaborative analysis platform based on federated learning according to claim 4, characterized in that, The central server also includes a heterogeneous adapter. When a client registers, the heterogeneous adapter collects and analyzes the hardware capabilities, network status, and data characteristics of each client to generate a device performance profile; and allocates a suitable model architecture to the client based on the device performance profile.

6. The privacy-preserving data collaborative analysis platform based on federated learning according to claim 1, characterized in that, The client also includes a differential privacy injection unit, which adds noise in layers during model training. The specific process is as follows: Step 31: Preset privacy noise allocation principles according to the global model structure; Step 32: Calculate the sensitivity for each layer of the global model and perform gradient clipping to ensure that the sensitivity is bounded; Step 33: Calculate the noise scale of each layer based on the privacy noise allocation principle and sensitivity, and construct the noise matrix corresponding to each layer; Step 34: Add corresponding noise matrices to different layer structures during the forward propagation and backward propagation phases.

Images