Federated Learning-Based Model Management System
Patent Information
- Application Number
- TW114106077
- Authority / Receiving Office
- TW · TW
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-02-19
- Publication Date
- 2026-09-01
- Estimated Expiration
- 2045-02-18
AI Technical Summary
Current federated learning technologies face challenges in effectively controlling users' access rights to artificial intelligence models, leading to potential security and legal risks due to misuse or abuse.
A model management system is introduced that includes a server device storing user accounts, passwords, and trained AI models with usage permissions, verifying user credentials, and authorizing access based on these permissions to manage AI model usage.
The system ensures secure and controlled access to AI models, enhancing the practicality, security, and adaptability of federated learning in commercial scenarios by managing usage rights effectively.
Smart Images
Figure 00000000_0000_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a management system, and more particularly to a model management system applied to federated learning. Prior Technology
[0002] Federated learning is a decentralized machine learning approach that allows multiple devices or organizations to collaboratively train an artificial intelligence model without sharing data. This approach not only protects data privacy but also fully utilizes the computing power of distributed resources, promoting cross-domain data collaboration.
[0003] However, current federated learning technologies still face challenges in access control after the artificial intelligence model has been trained. Existing mechanisms cannot effectively control users' access rights to the artificial intelligence model, which may lead to its misuse or abuse, resulting in potential security and legal risks.
[0004] Therefore, there is an urgent need to develop a comprehensive management system to solve the problems related to the management of artificial intelligence models, ensure that the usage rights of artificial intelligence models are effectively controlled, and thereby improve the practicality, security and adaptability of federated learning to commercial scenarios. Summary of the Invention
[0005] Therefore, the object of the present invention is to provide a management system for managing models generated by federated learning.
[0006] Therefore, the model management system based on federated learning of the present invention includes at least one model management unit.
[0007] Each model management unit includes a server device and multiple user devices that are connected to the server device via signals.
[0008] The server device stores multiple user accounts, multiple user passwords corresponding to each user account, and multiple trained artificial intelligence models. Each artificial intelligence model has a model number, and each user account includes multiple usage permissions corresponding to each model number.
[0009] Each of these terminal devices corresponds to a corresponding terminal account.
[0010] In this system, a target server device within a target model management unit, upon receiving a login request from a target user device containing a user account to be verified and a user password to be verified, determines whether the user password corresponding to the user account to be verified is the same as the user password to be verified. The target model management unit is one of the at least one model management unit, the target server device is the server device within the target model management unit, the target user device is one of the user devices within the target model management unit, and the user account to be verified is stored by the target server device. When the target server device determines that the password corresponding to the user account to be verified is the same as the password of the user account to be verified, for each model number stored in the target server device, the target server device determines whether the usage permission corresponding to the model number indicates that the artificial intelligence model corresponding to the model number can be used. When the target server device determines that the usage permission corresponding to the model number indicates that the artificial intelligence model corresponding to the model number can be used, the target server device authorizes the target user device to use the artificial intelligence model corresponding to the model number.
[0011] The advantage of this invention is that after the target server device verifies the user account and password to be verified to confirm login, the target server device can determine the artificial intelligence model that can be authorized to be used by the target user device according to the usage permissions corresponding to each stored model number, so as to effectively manage the use of such artificial intelligence models. Simple Explanation of the Diagram
[0012] Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: Figure 1 is a block diagram illustrating one embodiment of the model management system based on federated learning according to the present invention; Figure 2 is a flowchart illustrating a user login procedure for executing a model management method in this embodiment of the federated learning-based model management system of the present invention. Figure 3 is a flowchart illustrating a model usage record procedure that performs a model management method in this embodiment; Figure 4 is a flowchart illustrating a management terminal login procedure for executing a model management method in this embodiment; Figure 5 is a flowchart illustrating a model permission adjustment procedure that executes a model management method in this embodiment; and Figure 6 is a flowchart illustrating a model training procedure for performing a model management method in this embodiment. Implementation
[0013] Before the invention is described in detail, it should be noted that similar elements are represented by the same numbers in the following description.
[0014] Referring to Figure 1, the federated learning-based model management system 100 of the present invention includes at least one model management unit 1 and a third-party computing unit 2 connected to the at least one model management unit 1 via a communication network 900. Each model management unit 1 is interconnected via the communication network 900. Notably, both the server devices 11 and the third-party computing unit 2 use encryption algorithms to encrypt the information to be transmitted before transmission.
[0015] Each model management unit 1 includes a server device 11, a management device 12 connected to the server device 11 by a signal, and multiple user devices 13 connected to the server device 11 by a signal.
[0016] Each server device 11 also stores a training dataset for training artificial intelligence models, multiple user accounts corresponding to the corresponding user devices 13, multiple user passwords corresponding to the corresponding user accounts, a management account corresponding to the corresponding management device 12, a management password corresponding to the corresponding management account, and multiple trained artificial intelligence models. Each artificial intelligence model corresponds to a model number, and each user account includes multiple usage permissions corresponding to the corresponding model number. It is worth noting that each usage permission indicates whether the corresponding user account can use the artificial intelligence model associated with that usage permission. It is particularly important to note that the training dataset, user accounts, management accounts, and artificial intelligence models stored in each server device 11 are all different from each other.
[0017] Each server device 11 may be implemented as a server or a human computer, but is not limited thereto.
[0018] Each management terminal device 12 may be implemented as, for example, a personal computer, a laptop, a tablet computer, or a smartphone, but is not limited thereto.
[0019] Each user device 13 may be implemented as, for example, a personal computer, a laptop, a tablet computer, or a smartphone, but is not limited thereto.
[0020] Each third-party computing unit 2 may be implemented as a server or a human computer, but is not limited thereto.
[0021] The following will use a model management method to illustrate the operation details of the federated learning-based model management system 100. The model management method includes a user login program, a model usage record program, an administrator login program, a model permission adjustment program, and a model training program.
[0022] Referring to Figure 2, the client login procedure includes steps S201 to S209.
[0023] In step S201, a target user device 13A in a target model management unit 1A generates and transmits a login request containing a user account to be verified and a user password to be verified to a target server device 11A in the target model management unit 1A. The target model management unit 1A is one of the at least one model management unit 1, and the target server device 11A and the target user device 13A are both included in the target model management unit 1A.
[0024] In step S202, after receiving the login request from the target user device 13A, the target server device 11A determines whether the password corresponding to the user account to be verified is the same as the password of the user account to be verified. The user account to be verified is one of the user accounts stored by the target server device 11A. When the target server device 11A determines that the password corresponding to the user account to be verified is different from the password of the user account to be verified, it proceeds to step S203; when the target server device 11A determines that the password corresponding to the user account to be verified is the same as the password of the user account to be verified, it proceeds to step S205.
[0025] In step S203, the target server device 11A generates and transmits a login failure message indicating that the target user device 13A has failed to log in to the target user device 13A.
[0026] In step S204, after receiving the login failure message, the target user device 13A displays the login failure message.
[0027] In step S205, the target server device 11A generates and transmits a login success message indicating that the target user device 13A has successfully logged in to the target user device 13A.
[0028] In step S206, after receiving the login success message, the target user device 13A displays the login success message.
[0029] In step S207, for each AI model stored in the target server device 11A, the target server device 11A determines whether the usage permissions corresponding to the model number included in the user account to be verified indicate that the AI model corresponding to the model number can be used. When the target server device 11A determines that the usage permissions corresponding to the model number included in the user account to be verified indicate that the AI model corresponding to the model number cannot be used, proceed to step S208; when the target server device 11A determines that the usage permissions corresponding to the model number included in the user account to be verified indicate that the AI model corresponding to the model number can be used, proceed to step S209.
[0030] In step S208, the target server device 11A does not authorize the target user device 13A to use the artificial intelligence model corresponding to the model number.
[0031] In step S209, the target server device 11A authorizes the target user device 13A to use the artificial intelligence model corresponding to the model number.
[0032] Referring to Figure 3, the model uses a recording procedure that includes steps S301 to S302.
[0033] In step S301, the target user device 13A generates and transmits a model usage request containing data to be analyzed and a model number to be used to the target server device 11A.
[0034] In step S302, after receiving the model usage request from the target user device 13A, the target server device 11A generates and stores a log file showing that the target user device 13A uses the artificial intelligence model corresponding to the model number to be used to analyze the data to be analyzed. The artificial intelligence model corresponding to the model number to be used includes one of the artificial intelligence models corresponding to all model numbers authorized by the target server device 11A to be used by the target user device 13A.
[0035] Referring to Figure 4, the management terminal login procedure includes steps S401 to S406.
[0036] In step S401, a target management terminal device 12A in the target model management unit 1A generates and transmits another login request containing a management terminal account and a management terminal password to be verified to the target server device 11A. The target management terminal device 12A is included in the target model management unit 1A.
[0037] In step S402, after receiving the other login request from the target management device 12A, the target server device 11A determines whether the management password is the same as the management password to be verified based on the other login request. When the target server device 11A determines that the management password is not the same as the management password to be verified, it proceeds to step S403; when the target server device 11A determines that the management password is the same as the management password to be verified, it proceeds to step S405.
[0038] In step S403, the target server device 11A generates and transmits a management terminal login failure message indicating that the target management terminal device 12A has failed to log in to the target management terminal device 12A.
[0039] In step S404, after receiving the login failure message from the management terminal, the target user device 13A displays the login failure message.
[0040] In step S405, the target server device 11A generates and transmits a management terminal login success message indicating that the target management terminal device 12A has successfully logged in to the target management terminal device 12A.
[0041] In step S406, after receiving the login success message from the management terminal, the target user device 13A displays the login success message.
[0042] Referring to Figure 5, the model permission adjustment procedure includes steps S501 to S502.
[0043] In step S501, the target management terminal device 12A generates and transmits a permission adjustment request containing a user account to be adjusted and a model number to be adjusted to the target server terminal device 11A.
[0044] In step S502, after receiving the permission adjustment request from the target management device 12A, the target server device 11A adjusts the usage permissions of the model number to be adjusted in the user account to be adjusted according to the permission adjustment request. In this way, the target server device 11A can authorize or deauthorize the user device 13 corresponding to the user account to be adjusted to use the artificial intelligence model corresponding to the model number to be adjusted, according to the permission adjustment request from the target management device 12A. Here, the user account to be adjusted is one of the user accounts, and the model number to be adjusted is one of the model numbers corresponding to the artificial intelligence models stored in the target server device 11A.
[0045] Referring to Figure 6, the model training procedure includes steps S601 to S606.
[0046] In step S601, the target user device 13A generates and transmits a model training request containing an algorithm and at least one specified user account to the target server device 11A. The at least one specified user account is at least one of the user accounts stored by the server devices 11.
[0047] In step S602, after receiving the model training request from the target user device 13A, the target server device 11A generates and transmits a model training requirement containing the algorithm and at least one specified user account to the server device 11 other than the target server device 11A and the third-party computing unit 2, according to the model training request. Based on its stored training dataset, the target server device 11A uses the algorithm to generate and transmit multiple model adjustment parameters to the third-party computing unit 2. These model adjustment parameters include multiple gradient parameters, but are not limited to this. It is worth noting that the target server device 11A only transmits these model adjustment parameters to the third-party computing unit 2, and does not transmit its own stored training dataset to the third-party computing unit 2, thus effectively protecting the privacy and security of the training data.
[0048] In step S603, for each server device 11 other than the target server device 11A, the server device 11 generates and transmits multiple additional model adjustment parameters to the third-party computing unit 2 based on its stored training data set using the algorithm. It is worth noting that each server device 11 other than the target server device 11A only transmits these additional model adjustment parameters to the third-party computing unit 2, and does not transmit its own stored training data set to the third-party computing unit 2, thus effectively protecting the privacy and security of the training data.
[0049] In step S604, after receiving the model adjustment parameters transmitted by all server devices 11, the third-party computing unit 2 trains a target artificial intelligence model corresponding to the algorithm based on the model adjustment parameters and the algorithm, and then transmits the target artificial intelligence model to the server devices 11. It is worth noting that in other embodiments, when the model training request does not include the at least one specified user account, it indicates that only the target user device 13A has the right to use the target artificial intelligence model. Therefore, the third-party computing unit 2 does not need to transmit the target artificial intelligence model to server devices 11 other than the target server device 13A, and step S606 is not required subsequently.
[0050] In step S605, after receiving the target artificial intelligence model, the target server device 11A sets the usage permissions of the target artificial intelligence model in the user account corresponding to the target user device, so as to authorize the target user device to use the target artificial intelligence model.
[0051] In step S606, for each designated user account, the server device 11 storing the designated user account sets the usage permissions of the designated user account for the target artificial intelligence model after receiving the target artificial intelligence model, so as to authorize the user device 13 corresponding to the designated user account to use the target artificial intelligence model.
[0052] In summary, the federated learning-based model management system 100 of this invention manages the usage permissions of model numbers in each user account through the target server device 11A, and determines whether each user device 13 has permission to use any of the artificial intelligence models. Furthermore, the target management device 12A can adjust the corresponding usage permissions by executing the model permission adjustment procedure, and the target user device 13A can use any authorized artificial intelligence model, or initiate the training of the artificial intelligence model by executing the model training procedure. These methods not only ensure effective control over the use of artificial intelligence models, but also enhance the practicality, security, and adaptability of federated learning to commercial scenarios, thus effectively achieving the objectives of this invention.
[0053] However, the above description is merely an embodiment of the present invention and should not be construed as limiting the scope of the present invention. Any simple equivalent changes and modifications made in accordance with the scope of the patent application and the contents of the patent specification of the present invention shall still fall within the scope of the patent of the present invention.
[0054] 900: Communication Network 100: Federated Learning-Based Model Management System 1: Model Management Unit 11: Servo device 12: Management terminal device 13: End device 1A: Target Model Management Unit 11A: Target server device 12A: Target Management Terminal Device 13A: Target user device 2: Third-party computing unit S201~S209: Steps S301~S302: Steps S401~S406: Steps S501~S502: Steps S601~S606: Steps
Claims
1. A model management system based on federated learning, comprising: a third-party computing unit; and at least one model management unit, each model management unit being signal-connected to each other, each model management unit including: A server-side device is connected to the third-party computing unit. Each server-side device stores a training dataset, multiple user accounts, multiple user passwords corresponding to the user accounts, and multiple trained artificial intelligence models. Each artificial intelligence model has a model number, and each user account includes multiple usage permissions corresponding to the model number. There are also multiple user-side devices connected to the server-side device and corresponding to the user accounts.In this system, a target server device within a target model management unit, upon receiving a login request from a target user device containing a user account and a user password to be verified, determines whether the user password corresponding to the user account to be verified is the same as the user password to be verified. The target model management unit is one of the at least one model management unit, the target server device is the server device within the target model management unit, the target user device is one of the user devices within the target model management unit, and the user account to be verified is one of the user accounts stored by the target server device. When the target server device determines that the user password corresponding to the user account to be verified is the same as the user password to be verified, for each AI model stored by the target server device, the target server device determines whether the usage permissions corresponding to the model number included in the user account to be verified indicate that the AI model corresponding to the model number can be used. When the target server determines that the usage permission corresponding to the model number included in the user account to be verified indicates that the user can use the artificial intelligence model corresponding to the model number, the target server authorizes the target user to use the artificial intelligence model corresponding to the model number. After receiving a model training request containing an algorithm from the target user, the target server generates and transmits a model training requirement containing the algorithm to other server devices and the third-party computing unit according to the model training request. Based on its stored training dataset, the target server uses the algorithm to generate and transmit multiple model adjustment parameters to the third-party computing unit. For each server device other than the target server, the server generates and transmits multiple other model adjustment parameters to the third-party computing unit based on its stored training dataset and the algorithm. After receiving the model adjustment parameters transmitted by all server devices, the third-party computing unit trains a target artificial intelligence model corresponding to the algorithm based on the model adjustment parameters and the algorithm. After the third-party computing unit completes training and obtains the target AI model, it transmits the target AI model to the target server device. Upon receiving the target AI model, the target server device sets the usage permissions for the target AI model in the user account corresponding to the target user device, thereby authorizing the target user device to use the target AI model.
2. The federated learning-based model management system as described in claim 1, wherein, Each model management unit also includes a management terminal device that is signal-connected to the server device. The server device also stores a management terminal account corresponding to the management terminal device and a management terminal password corresponding to the management terminal account. After receiving another login request from a target management terminal device that includes a management terminal account to be verified and a management terminal password to be verified, the target server device determines whether the management terminal password is the same as the management terminal password to be verified. The target management terminal device is the management terminal device in the target model management unit. When the target server device determines that the management terminal password is the same as the management terminal password to be verified, it generates and sends a management terminal login success message indicating that the target management terminal device has successfully logged in to the target management terminal device.
3. The federated learning-based model management system as described in claim 2, wherein, After receiving a permission adjustment request from the target management device, which includes a user account to be adjusted and a model number to be adjusted, the target server device adjusts the usage permission of the model number to be adjusted in the user account to be adjusted according to the permission adjustment request. The user account to be adjusted is one of the user accounts, and the model number to be adjusted is one of the model numbers corresponding to the artificial intelligence models stored in the target server device.
4. The federated learning-based model management system as described in claim 1, wherein, After receiving a model usage request from the target user device that includes data to be analyzed and a model number to be used, the target server device generates and stores a log file about the target user device using the artificial intelligence model corresponding to the model number to be used. The artificial intelligence model corresponding to the model number to be used includes one of the artificial intelligence models corresponding to all model numbers that the target server device has authorized the target user device to use.
5. A federated learning-based model management system as described in claim 1, wherein, The model training request also includes at least one specified user account. The model training requirement also includes the at least one specified user account. The at least one specified user account is at least one of the user accounts stored in the server devices. After the third-party computing unit completes the training and obtains the target artificial intelligence model, it transmits the target artificial intelligence model to the server devices. For each specified user account, after receiving the target artificial intelligence model, the server device storing the specified user account sets the usage permissions of the specified user account for the target artificial intelligence model, so as to authorize the user device corresponding to the specified user account to use the target artificial intelligence model.