A system and method to support automatic continuous updates of artificial intelligence algorithms

Abstract

The application relates to a system and method for supporting automatic continuous updating of an artificial intelligence algorithm. The system comprises a computing module, a storage module, an updating module, an input module and an output module. The computing module provides computing power for the system. The storage module stores data related to updating of the artificial intelligence algorithm, including one or more artificial intelligence algorithm models, artificial intelligence algorithm training rules and training data. The updating module trains and updates the artificial intelligence algorithm based on the data related to updating of the artificial intelligence algorithm stored in the storage module. The output module visually outputs data in the storage module to a display device. The input module accepts external operations and external input data and sends the external input data to the storage module.

Description

[0001] This disclosure is a divisional application of the patent application filed on September 18, 2021, with application number 2021110978839 and invention title "A System and Method for Supporting Automatic Continuous Updates by Artificial Intelligence Algorithms". Technical Field

[0002] This application relates to the field of artificial intelligence, and more specifically, to a system and method for automatically and continuously updating artificial intelligence algorithms. Background Technology

[0003] As artificial intelligence algorithms become increasingly sophisticated, more and more business products are deploying them to improve product quality. For example, AI algorithms can be used in security inspection systems deployed at customs and rail transit stations to inspect items, improving inspection quality while saving manpower.

[0004] However, as AI products, device-based artificial intelligence (AI) algorithm models lack generalization capabilities when facing different needs (such as different customs districts or different rail transit stations). In other words, the same AI algorithm model cannot be universally applied to all scenarios and requires modification of the AI ​​algorithm model or parameters to adapt to the various specific needs of each scenario.

[0005] To address the diverse inspection content and requirements across various scenarios and business operations, and to develop more scenario-specific and tailored algorithms, existing technical solutions typically define performance metrics for AI algorithm model products based on regional, scenario, and other business conditions. Generally, only AI algorithm model products that have passed on-site verification and meet the performance metrics are deployed to the business environment. However, if an AI algorithm model product experiences a performance decline during its service at a particular business site, failing to meet business needs, the customer communicates with the developers and provides them with a certain amount of on-site data. The developers then use this data to train a machine learning model or adjust its parameters. Once the developer has completed the machine learning model training based on the on-site data and passed verification, the updated AI algorithm model product is sent to product support personnel, who then deploy it to the business site to replace the previous AI algorithm model product. The updated AI algorithm model product requires a trial period to observe its effectiveness in the business environment.

[0006] However, the existing technologies described above have significant room for improvement. For example, some industry clients have high data security management requirements and cannot allow on-site data transfer or copying from company equipment to be taken off-site. In such cases, existing technical solutions are completely unfeasible. Furthermore, with existing solutions, from discovering on-site issues with AI algorithm model products failing to meet performance metrics to finally updating the AI ​​algorithm model product and observing its effectiveness on-site, multiple rounds of communication and dedicated developers are required to schedule machine learning training based on on-site data. In short, implementing existing technical solutions is a very time-consuming and labor-intensive process.

[0007] Furthermore, in existing technologies, end users (i.e., customers) do not have the authority to operate on daily AI learning algorithm updates; this is typically delegated to developers. This leads to practical problems such as increased communication costs and low efficiency. Summary of the Invention

[0008] This application is made in view of the technical problems existing in the prior art, and its purpose is to provide a system that is user-friendly, convenient to use, and allows for the updating of artificial intelligence algorithms at any time.

[0009] In one aspect of this application, a system supporting automatic continuous updates using artificial intelligence algorithms is provided, the system comprising: The computing module provides computing power to the system. A storage module stores data related to the update of artificial intelligence algorithms, including one or more artificial intelligence algorithm models, artificial intelligence algorithm training rules, and training data. An update module, which trains and updates the artificial intelligence algorithm based on data related to the update of the artificial intelligence algorithm stored in the storage module; The system includes an input module and an output module. The output module visually outputs the data in the storage module to a display device. The input module accepts external operations and external input data and sends the external input data to the storage module.

[0010] According to this technical solution, an interface is provided between the system and users (including developers and end users). Through the output module, data from the current system can be output to a display device in a user-friendly manner for developers or end users to view. This allows developers or end users to understand the existing artificial intelligence algorithm models, training rules, or training data within the system, providing a basis for further system operation. Simultaneously, through the input module, the system can accept input from developers or end users, including external operations and external input data. After receiving external input data, the input module sends it to the storage module for storage.

[0011] According to the second scheme, the system also includes an artificial intelligence algorithm model file library. The input module can receive one or more artificial intelligence algorithm model files from the artificial intelligence algorithm model file library and store them in the storage module.

[0012] According to this scheme, setting up an artificial intelligence algorithm model database within the system provides a way for AI algorithms to enter the system. Its advantage lies in the fact that, without requiring external connections or further user input, the desired model can be downloaded from the model database with a simple one-click operation, facilitating the addition of AI algorithm models to the system.

[0013] According to the third approach, the external operations include user-initiated training of the artificial intelligence algorithm model and user-queried operations of the artificial intelligence algorithm model. The external input data includes one or more artificial intelligence algorithm models, artificial intelligence algorithm training rules, and training data.

[0014] Through user interaction, specifically between the user and the system, users can initiate AI algorithm model training at any time, improving user convenience. Furthermore, through query operations, users can readily access and understand the current status of existing AI algorithm models within the system.

[0015] According to the fourth scheme, the system also includes a user management module. The user management module manages user types and their operation permissions. The user types include developers and end users.

[0016] The user management module allows for the management of user types and permissions. Based on this, ordinary users (such as end users) and developers are assigned different permissions. This means that some operations are restricted to developers, while permissions for other operations can be delegated to end users. This provides necessary guarantees for the security of system data. Furthermore, it facilitates system maintenance.

[0017] According to the fifth scheme, the one or more artificial intelligence algorithm models and the training rules of the artificial intelligence algorithms are set or input only by the developers.

[0018] Without relevant professional knowledge, it is generally difficult to understand artificial intelligence algorithm models and training rules. Therefore, the setting or input permissions for artificial intelligence algorithm models and training rules can be made accessible only to developers. This ensures data security within the system while also reducing the difficulty of use for ordinary users such as end users.

[0019] According to the sixth scheme, the update module automatically updates the current optimal artificial intelligence algorithm model to the artificial intelligence algorithm currently running in the system based on the calculation results, or

[0020] The update module outputs the current optimal artificial intelligence algorithm model to the display device through the output module based on the calculation results.

[0021] According to this scheme, the update module can directly update the calculated optimal AI algorithm model to the AI ​​algorithm actually running in the system, allowing the optimal algorithm to be applied as quickly as possible. Alternatively, the update module can output the optimal AI algorithm model to a display device, allowing users (including developers or end users) to decide whether to use the model as the actual running AI algorithm.

[0022] According to the seventh scheme, the external operation includes the user's selection of training data.

[0023] According to this scheme, users can filter the datasets used for training in the system, thereby enabling training on specific data, such as data from a specific time period or data from a specific terminal, thus improving the flexibility of training and the performance of the trained model for specific situations.

[0024] According to the eighth scheme, the storage module includes a mechanical hard disk and a solid-state drive.

[0025] According to this scheme, the system can store data reasonably based on its popularity, achieving a better cost-effectiveness ratio.

[0026] According to the ninth scheme, the storage module includes file system storage. The file system uses ceph as the underlying storage and cephfs to provide a POSIX-compliant file system.

[0027] This solution optimizes the performance of the file system storage, ensuring the smooth training of artificial intelligence algorithm models.

[0028] According to the tenth scheme, the storage module further includes object storage, which is implemented by ceph.

[0029] This solution improves the reliability and convenience of storage in the system, enables server-side encryption, and facilitates easy expansion.

[0030] According to the eleventh plan, the system is deployed at the customer's site in a cloud-edge-device manner.

[0031] This solution, deployed in a cloud-edge-device manner, improves the system's distributed nature, enabling customers to train AI algorithm models on-site, while also avoiding unnecessary data transmission and enhancing system security.

[0032] According to the twelfth proposal, a method is provided to support automatic continuous updating of artificial intelligence algorithms, the method comprising: Accept developers' settings for artificial intelligence algorithm models, training rules, training datasets, etc. Accept input from end users, including selecting a training dataset, starting training, etc. If the end user's input is to enable training, the target artificial intelligence algorithm model is trained based on the training rules and the training dataset. By comparing the evaluation index values ​​of various target artificial intelligence algorithms, the optimal artificial intelligence algorithm model is determined. Update the currently running artificial intelligence algorithm model to the determined optimal artificial intelligence algorithm model, or output the determined optimal artificial intelligence algorithm model to the display device.

[0033] This method can achieve the same or similar technical effects as the above systems. Attached Figure Description

[0034] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the specification, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0035] Figure 1 A schematic diagram of an exemplary system architecture applied to a method or apparatus according to embodiments of this disclosure is shown.

[0036] Figure 2 A schematic diagram of the system hardware and software coordination suitable for implementing the embodiments of this disclosure is shown.

[0037] Figure 3 A schematic diagram of the system file system storage is shown.

[0038] Figure 4 A schematic diagram of object storage is shown.

[0039] Figure 5 A schematic diagram of each module in the system is shown.

[0040] Figure 6 A flowchart of the model update method of the present invention is shown.

[0041] Figure 7 A schematic diagram of a computer-readable storage medium is shown. Detailed Implementation

[0042] Specific embodiments will now be described more fully with reference to the accompanying drawings. However, embodiments can be implemented in many forms and should not be construed as limited to those set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of this application to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0043] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, or in one or more software-hardened modules, or in different network and / or processor devices and / or microcontroller devices.

[0044] Furthermore, the modules, components, dimensions, etc. involved in the specific embodiments are merely exemplary, and those skilled in the art can conceive of other examples based on them.

[0045] Figure 1 A schematic diagram of an exemplary system architecture supporting automatic continuous updates using artificial intelligence algorithms that can be applied to embodiments of this disclosure is shown.

[0046] like Figure 1As shown, system architecture 100 may include terminal devices 101, 102, and 103, network 104, and server 105. Network 104 is used to provide communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0047] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, desktop computers, etc.

[0048] Server 105 can be a server providing various services, such as a backend management server supporting the devices operated by users using terminal devices 101, 102, and 103. The backend management server can analyze and process received requests and other data, and feed the processing results back to the terminal devices. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. This disclosure does not impose any limitations in this regard.

[0049] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Server 105 can be a single physical server or a combination of multiple servers. Depending on actual needs, it can have any number of terminal devices, networks, and servers.

[0050] The method of this application can be built into and run on terminal devices 101, 102, 103, server 105, etc., or it can be distributed to terminal devices 101, 102, 103 via network through server 105 and run thereon.

[0051] Figure 2 A schematic diagram of the system hardware and software cooperation suitable for implementing the embodiments of this disclosure is shown. It should be noted that... Figure 2 The example shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0052] like Figure 2 As shown, the system consists of two main parts: hardware and software.

[0053] The hardware components mainly include GPU servers and storage servers.

[0054] GPU servers are an example of hardware that provides computing power, supporting the training and evaluation of artificial intelligence algorithms, as well as other system operations. Using one or more servers equipped with multiple GPUs, when the system needs to execute artificial intelligence algorithm training and inference tasks, the task scheduling system will rationally schedule the tasks to GPU servers with sufficient resources. GPU servers provide the basic resources required for task execution, such as CPU and memory, while also providing additional GPU accelerator cards, enabling artificial intelligence algorithm training and inference to run at a faster speed.

[0055] Storage servers can be single or multiple servers with large-capacity disks, serving as the physical medium for the storage system and responsible for actual data write-to-disk functions. In terms of storage server configuration, the system can use all low-speed mechanical hard drives. Alternatively, a combination of high-speed and low-speed disks can be used, with some storage servers equipped with low-speed mechanical hard drives and others using solid-state drives (SSDs). This design allows for efficient data storage based on frequency of use, achieving a better cost-performance ratio.

[0056] As shown in the diagram, in the software portion of the system, a container runtime is required to adapt to various operating environments and systems. Additionally, a task orchestration system is needed to schedule tasks among themselves.

[0057] Specifically, the underlying container runtime in the system can use the Docker container runtime. It provides container runtime capabilities above the operating system, and offers isolation between containers, among other things. Docker is the most commonly used container runtime, but it is not the only one; therefore, those skilled in the art can make substitutions without affecting the scope of this application.

[0058] In addition, various artificial intelligence algorithms require different operating environments during operation, such as the installation of different dependency packages and the running on different operating systems. The system collects the algorithm requirements, builds the user runtime in the form of an image, and provides it directly to the algorithm for use.

[0059] Task orchestration primarily addresses the function of scheduling tasks to appropriate servers for execution and managing the lifecycle of various tasks. Machine learning algorithms on the platform run within container images provided by the platform; therefore, task scheduling in the orchestration system is essentially container scheduling. To achieve container task orchestration, the platform uses a Kubernetes-based task orchestration system. However, in practice, Kubernetes encapsulates tasks into Pods outside the containers during task scheduling.

[0060] A file system is provided on the storage server to provide software support for reading data during the training process of artificial intelligence algorithms. Figure 3 A schematic diagram of the file system storage in this embodiment is shown.

[0061] As shown in the diagram, the system can use Ceph as the underlying storage for its file system storage implementation. Ceph manages multiple storage servers, providing basic functions such as data slicing and data redundancy, ensuring data reliability and high read / write performance. To provide shared file system services on top of Ceph, the storage system uses CephFS, a POSIX-compliant file system.

[0062] In addition, by using the CSI-related interfaces of the Kubernetes system to interface CephFS with Kubernetes, tasks running in Kubernetes can seamlessly use the file system storage services provided by CephFS.

[0063] In addition to the file system storage used for algorithm training tasks, the system can also utilize object storage. Object storage differs from file system storage in that it typically does not provide services through a file system interface, but rather offers file upload and download services via APIs. Object storage offers advantages such as high reliability, simple access, server-side encryption, and easy scalability.

[0064] The artificial intelligence algorithm models, intermediate products, and runtime logs in the system can be stored using object storage. For example... Figure 4 As shown, the system's storage module implements object storage through Ceph, providing an interface that conforms to the S3 protocol proposed by AWS to serve clients. The software in the system interacts with the object storage system through the S3 protocol to complete file upload and download.

[0065] With the cooperation of hardware and basic software, the system allows for the automatic updating of platform software and artificial intelligence algorithm models.

[0066] Figure 5 The diagram shows the various modules in the artificial intelligence algorithm model update system. Figure 5 System modules and Figure 2 The software and hardware system architectures described may overlap; however, it should be understood that this is a description of the same thing from different perspectives, and this does not affect the understanding and reproduction of this implementation method by those skilled in the art.

[0067] like Figure 5As shown, the automatic continuous update system for artificial intelligence algorithms in this embodiment may include a calculation module 1, a storage module 2, an update module 3, an input module 4, an output module 5, and a user management module 6. Additionally, the system may also include an artificial intelligence algorithm model file library 7. It should be understood that these modules are merely exemplary in this embodiment, and those skilled in the art can make any extensions and equivalent substitutions that conform to the spirit of this application.

[0068] The computing module 1 provides computing power to the system. It may include a GPU server or other devices with computing capabilities, such as CPUs and other IC chips.

[0069] Storage module 2 may include a storage server, which includes physical disks, etc. Data related to the updates of artificial intelligence algorithms, such as one or more artificial intelligence algorithm models, artificial intelligence algorithm training rules, and training data, can be stored in storage module 2.

[0070] Different types of artificial intelligence algorithms, their specific structures, and the selection of parameters can all lead to different artificial intelligence algorithm models. In other words, algorithms with slightly different types, structures, and parameters are considered different artificial intelligence algorithm models. For example, intelligent algorithms using artificial neural networks and those using support vector machines are different artificial intelligence algorithm models. Similarly, even within artificial neural network models, algorithms using different activation functions are considered different models. Furthermore, even within artificial neural network models, different numbers of layers result in different structures and are considered different models. Finally, even within artificial neural network models with the same structure, different model parameters also indicate different artificial intelligence algorithm models. Therefore, the automatic updating of the artificial intelligence algorithm model in this application can include updates to the model type, structure, parameters, etc., or may only include updates to one or a few of these aspects.

[0071] Artificial intelligence algorithm training rules are the sum of rules used to train artificial intelligence algorithms, and may include, for example, evaluation metrics. Evaluation metrics can be diverse. For example, in the field of security inspection, metrics could be the accuracy or speed of image or video recognition. Those skilled in the art will understand that metrics can be specifically set according to specific application scenarios, and will not be listed here individually.

[0072] The AI ​​algorithm model automatic update system of this invention can be deployed at the customer's site in a cloud-edge-device manner. Data collected on-site can become training data without leaving the station. Furthermore, from the vast amount of training data, training datasets tailored to specific purposes can be selected as needed. For example, training datasets can be selected based on specific time periods. For instance, if users feel that security check efficiency has decreased in the last six months, then training can be performed using data from the last six months. Alternatively, different training weights can be set for different time periods, with data closer to the training time having a higher weight. Or, if users feel that the security check efficiency or accuracy of a certain station or a specific terminal at that station has decreased, then training can be performed using data from that station or specific terminal. By allowing the training dataset to be set as needed, the relevance of training can be further improved, enhancing the effectiveness of the AI ​​algorithm across various terminals and time periods.

[0073] The update module 3 trains and updates the artificial intelligence algorithm based on the data related to the AI ​​algorithm update stored in the storage module 2. This data may include the AI ​​algorithm model, AI algorithm training rules, and training data. After the user initiates the training of the AI ​​algorithm model, the update module trains and evaluates several AI algorithm models that need training based on the algorithm training rules and the training dataset.

[0074] As a specific embodiment, the update module 3 trains the model iteratively based on the selected training rules and dataset, calculates the evaluation index value, compares the evaluation index value, and thus determines the optimal artificial intelligence algorithm model.

[0075] The system also includes input module 4 and output module 5.

[0076] Output module 5 can use a display device such as a monitor to visually output the data in the storage module to the user, enabling the user to grasp the data in the system in a timely manner. The data may include the artificial intelligence algorithm model currently running in the system, other alternative artificial intelligence algorithm models, training rules, training datasets, and other data related to the updating of the artificial intelligence algorithm. In addition, output module 5 can also output other commonly used system data, such as system time, system load, and system storage utilization, for user confirmation.

[0077] Input module 4 accepts external operations and external input data, and sends the received external input data to storage module 2. Input module 4 may include, but is not limited to, a keyboard, mouse, joystick, voice input device, etc. External operations may include, for example, starting training, stopping training, data querying, etc. External input data may include, for example, new artificial intelligence algorithm models, training rules, datasets from external sources, and new data set for system data such as system time. Data input from external sources can be sent by input module 4 to storage module 2, and stored appropriately by storage module 2.

[0078] The system may also include a user management module 6, which manages user registration, login, and operations. The system can manage users hierarchically, with different levels of users having different access permissions. For example, developers have permissions to input AI algorithm models, training rules, and external datasets, while end users cannot input AI algorithm models, training rules, or external datasets, but can choose training datasets, such as training based on data from a specific time period. The advantage of this approach is that the aspects requiring specialized knowledge are implemented by the developers during the system deployment phase or adjusted during use, allowing end users to train models without requiring highly specialized knowledge, thus balancing system professionalism and ease of use.

[0079] The system may also include a file repository 7. File repository 7 serves as a storage repository for artificial intelligence algorithms and training data. The system can import AI algorithm models or training data from file repository 7 using SFTP. The advantage of this setup is that after the initial system deployment is complete, algorithm models or training data can be imported from file repository 7 using tools like SFTP without needing an external interface, thus improving system data security.

[0080] Figure 6 A flowchart of the model update method of the present invention is shown.

[0081] In step S1, the developers set up the artificial intelligence algorithm model, training rules, and training dataset to complete the system deployment.

[0082] In step S2, input from the end user is accepted, including selecting a training dataset, starting training, etc.

[0083] In step S3, if the terminal user's input is to enable training, the target artificial intelligence algorithm model is trained based on the training rules and the training dataset.

[0084] In step S4, the optimal artificial intelligence algorithm model is determined by comparing the evaluation index values ​​of each target artificial intelligence algorithm.

[0085] In step S5, the currently running artificial intelligence algorithm model is updated to the determined optimal artificial intelligence algorithm model, or the determined optimal artificial intelligence algorithm model is output to the display device for the end user to make a decision.

[0086] The system in this disclosure embodiment may have an operating system. The operating system may be any operating system, such as a UNIX-based operating system, a Windows-based operating system, an Android-based operating system, an iOS operating system, or other possible operating systems. This disclosure embodiment does not specifically limit the operating system.

[0087] Figure 7 A computer-readable storage medium 200 is shown storing a program implementing the methods described in the embodiments of this application. This storage medium may be a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device. However, the storage medium of this application is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0088] The storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0089] Program code for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0090] Those skilled in the art will understand that the electronic devices in the system may also include a power supply (such as a battery) that supplies power to the various components. The power supply can be connected to the processor logic through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system.

[0091] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this disclosure is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0092] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not claimed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims. It should be understood that this disclosure is not limited to the detailed structures, drawing arrangements, or implementations shown herein; rather, this disclosure is intended to cover various modifications and equivalent arrangements contained within the spirit and scope of the appended claims.

Claims

1. A system supporting automatic and continuous updates of artificial intelligence algorithms, the system being deployed at security checkpoints to ensure that on-site data remains within the checkpoint and becomes training data, the system comprising: The storage module stores the artificial intelligence algorithm model, artificial intelligence algorithm training rules, and training data set by the developers. The update module accepts input from the end user to select a training dataset and selected artificial intelligence algorithm training rules from the field data as needed. After the end user starts training the artificial intelligence algorithm model, the artificial intelligence algorithm model is trained based on the selected artificial intelligence algorithm training rules, the training dataset selected by the end user, and the training data stored in the storage module. An input module that accepts external input data and sends the external input data to the storage module, wherein the external input data includes datasets from external sources; The user management module grants the developer permission to input datasets from external sources.

2. The system according to claim 1, characterized in that, The system also includes a library of artificial intelligence algorithm model files. The input module can receive one or more artificial intelligence algorithm model files from the artificial intelligence algorithm model file library and store them in the storage module.

3. The system according to claim 1, characterized in that, The external operations include user-initiated training of the artificial intelligence algorithm model and user-queried operations of the artificial intelligence algorithm model. The external input data includes one or more artificial intelligence algorithm models, artificial intelligence algorithm training rules, and training data.

4. The system according to any one of claims 1-3, characterized in that, The user management module manages user types and their operation permissions. The user types include developers and end users.

5. The system according to claim 4, characterized in that, The one or more artificial intelligence algorithm models and the training rules of the artificial intelligence algorithms are set or input only by the developer.

6. The system according to claim 1, characterized in that, The update module automatically updates the current optimal artificial intelligence algorithm model to the artificial intelligence algorithm currently running in the system based on the calculation results, or The update module outputs the current optimal artificial intelligence algorithm model to the display device through the output module based on the calculation results.

7. The system according to claim 2, characterized in that, The external operations include the user's selection of training data.

8. The system according to claim 1, characterized in that, The storage module includes a mechanical hard drive and a solid-state drive.

9. The system according to claim 1, characterized in that, The storage module includes file system storage. The file system uses ceph as the underlying storage and cephfs to provide a POSIX-compliant file system.

10. The system according to claim 9, characterized in that, The storage module also includes object storage, which is implemented by Ceph.

11. The system according to claim 1, characterized in that, The system is deployed at the customer's site in a cloud-edge-device manner.

12. A method for supporting automatic continuous updates using artificial intelligence algorithms, applied to the system as described in any one of claims 1 to 11, the method comprising: The developer is given permission to input datasets from external sources, and these external datasets are stored in the storage module as training data. Accept the developer's settings for the artificial intelligence algorithm model, training rules, and training dataset; Accept input from end users, including selecting a training dataset that meets a certain purpose from the field data as needed, starting training, and selecting training rules; If the end user's input is to enable training, the target artificial intelligence algorithm model is trained based on the selected training rules, the training dataset selected by the end user, and the training data stored in the storage module.

Images