TWI810560B - Artificial intelligence model service method and server based on cloud microservice - Google Patents

Abstract

An artificial intelligence model service server, after a processing unit receives a project request data from a client via a communication unit, the processing unit executes an information conversion application group stored in a storage unit to analyze the project request data And generate an analysis result in a specific format, and execute a security management application group stored in the storage unit to determine whether the analysis result is legal. When the processing unit determines that the analysis result is legal, according to the analysis result, use a Generate a service result corresponding to the target artificial intelligence model container of the analysis result, and when it is determined that the analysis result is illegal, the processing unit executes the security management application program group to generate and send a verification failure message to the use end.

Description

Artificial intelligence model service method and server based on cloud microservice

The present invention relates to an artificial intelligence model service method, in particular to an artificial intelligence model service method and server based on cloud microservices.

In recent years, artificial intelligence (AI) has become one of the hottest technologies, and it can be applied in various fields, such as medical care, transportation, finance and other fields.

The architecture used in the existing self-built AI model implementation projects is mainly to package the entire self-built AI service into an AI model container dedicated to a single project, and deploy the AI model container on a single virtual machine (Virtual Machine, VM).

However, the entire service is packaged into a container and deployed on a single virtual machine. Different landing projects cannot share resources. The AI model containers of all projects carry out information conversion and authorization management separately. Therefore, each time a project is carried out, Additional manpower and time costs are required, and the support of the inherent system is obtained repeatedly to complete the end-to-end (End to End) connection of the project, which does not have high reusability (High Reusability).

Therefore, the object of the present invention is to provide an artificial intelligence model service method with high reusability.

Therefore, the artificial intelligence model service method of the present invention is implemented by a server, which stores an information conversion application program group, a security management application program group, and a plurality of artificial intelligence model containers, and the server passes a The communication network is connected to a user end, and the artificial intelligence model service method includes a step (A), a step (B), a step (C), and a step (D).

In the step (A), after the server receives a project request data from the client, the server executes the information conversion application program group to parse the project request data and generate a parsing result in a specific format.

In the step (B), the server executes the security management application group to determine whether the analysis result is legal.

In the step (C), when the server determines that the analysis result is legal, the server uses one of the artificial intelligence model containers corresponding to the target artificial intelligence model container of the analysis result according to the analysis result to generate 1. Service Results.

In the step (D), when the server determines that the analysis result is invalid, the server executes the security management application group to generate and send a verification failure message to the client via the communication network.

Another object of the present invention is to provide a highly reusable artificial Smart model service server.

Therefore, the artificial intelligence model service server of the present invention includes a communication unit, a storage unit, and a processing unit.

The communication unit is connected to a communication network, and connected to a user end through the communication network.

The storage unit stores an information conversion application program group, a security management application program group, and multiple artificial intelligence model containers.

The processing unit is electrically connected to the communication unit and the storage unit.

Wherein, after the processing unit receives a project request data from the client via the communication unit, the processing unit executes the information conversion application program group to analyze the project request data and generate a parsing result in a specific format, and executes the The security management application group determines whether the analysis result is legal. When the processing unit determines that the analysis result is legal, the processing unit uses one of the artificial intelligence model containers corresponding to the target artificial intelligence of the analysis result according to the analysis result. The smart model container generates a service result. When the processing unit determines that the analysis result is invalid, the processing unit executes the security management application program group to generate and send a verification failure message to the client via the communication unit.

The effect of the present invention is: the processing unit executes the information conversion application program group to analyze the project request data to generate the analysis result in a unified and specific format, so that the artificial intelligence model container does not need to repeatedly process complex information conversion between different systems And by executing the security management application program group to determine whether the analysis result is legal, so that the artificial intelligence model container does not need to manage authorization information by itself, which has high reusability.

11: Artificial intelligence model service server

111: Communication unit

112: storage unit

113: Processing unit

12: Use end

13: Database

100: Communication network

21~25: Steps

211, 212: sub-steps

231~234: sub-steps

Other features and effects of the present invention will be clearly presented in the implementation manner with reference to the drawings, wherein: FIG. 1 is a block diagram illustrating an embodiment of the artificial intelligence model service server of the present invention; FIG. 2 is a process flow Figure 3 illustrates an embodiment of the artificial intelligence model service method of the present invention; FIG. 3 is a flow chart to assist in explaining the sub-steps of step 21 in FIG. 2; and FIG. 4 is a flow chart to assist in explaining the sub-steps in step 23 in FIG. 2.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are denoted by the same numerals.

Referring to Fig. 1, an embodiment of the artificial intelligence model service server 11 based on cloud microservices of the present invention includes a communication unit 111, a storage unit 112, and a processing unit electrically connected to the communication unit 111 and the storage unit 112 113.

The communication unit 111 communicates with a client via a communication network 100 12 and a plurality of databases 13, these databases 13 are, for example, using a database management system (Database Management System, DBMS), a secure file transfer protocol (Secure File Transfer Protocol, SFTP), a network attached storage (Network Attached Storage, NAS) and other technologies established.

The storage unit 112 stores an information conversion application program group, a security management application program group, a resource allocation application program group, a plurality of artificial intelligence model containers, and a plurality of pieces of authorization information respectively related to the databases 13 . The information conversion application group includes a plurality of information conversion applications, the security management application group includes a plurality of security management applications, the resource allocation application group includes a plurality of resource allocation applications, and the resource allocation applications The programs correspond to the authorization information respectively.

It is worth noting that these artificial intelligence model containers are for example related to natural language processing (Natural Language Processing, NLP) for extracting document stems and obtaining combinations of possible meanings of texts, or for recognizing facial images and obtaining human The unique feature value of the face is related to the artificial intelligence model container of Computer Vision (CV), the authorization information is encoded by Base64 when the processing unit 113 executes the security management application group, and the encoded information The authorization information is stored in a single write to the memory of the underlying Kubernetes environment, but is not limited thereto.

It should be noted that each resource allocation application is independent, and the objects of resource allocation are different, such as the resource allocation related to the database management system The application program will only verify the authorization of the database of the database management system, and the resource allocation application related to the secure file transfer protocol will only verify the authorization of the database of the secure file transfer protocol.

Referring to FIGS. 1 and 2 , an embodiment of the cloud microservice-based artificial intelligence model service method of the present invention is realized by the cloud microservice-based artificial intelligence model service server 11 of the present invention shown in FIG. 1 . The steps of this embodiment of the cloud microservice-based artificial intelligence model service method are described in detail below.

In step 21, after the processing unit 113 receives a project request data from the client 12 via the communication unit 111, the processing unit 113 executes the information conversion application program group to analyze the project request data and generate a specific A parsing result in a format, the parsing result includes an authorization code having a key, a valid time interval, and a model code corresponding to one of the artificial intelligence model containers.

Referring to FIG. 3 , step 21 includes sub-steps 211 and 212 , and the sub-steps included in step 21 are described below.

In sub-step 211, after the processing unit 113 receives the project request data via the communication unit 111, the processing unit 113 executes the information conversion application program group to perform regular regular text analysis (Regular Expression Text) on the project request data Parser) to generate a pre-conversion parsing result.

In sub-step 212, the processing unit 113 executes the information conversion application program group to convert the pre-conversion analysis result into the analysis result.

It is worth noting that, in this embodiment, the processing unit 113 executes the information conversion application program group to convert the pre-conversion parsing result into JSON (JavaScript Object Notation) format, and the parsing result includes a model with the A uniform resource identifier (Uniform Resource Identifier, URI) of the code, but not limited thereto.

It should be noted that, in sub-step 211, the information conversion application group has different analysis methods according to the project request data. For example, if the project request data includes a network path, such as http: //aigo.ctbcbank.com/api/hrface/punch, indicating that the entry point is directly RESTful API (that is, api-gateway), where <aigo.ctbcbank.com> represents the RESTFul entry point, <hrface> represents the project name, <punch > Indicates the service performed. After regular regular text analysis, the information transformation application group will transfer the requirements of this project to the specified model to be used by the project, and generate a uniform resource identifier, such as http://feature .hrface.svc.cluster.local: 18890 , the uniform resource identifier is the native description method of Kubernetes, where <feature> is the model code.

表示進入點為企業服務匯流排(Enterprise Service Bus,ESB)通道，其中<AiQnaInq>表示專案名稱，<01>表示所進行的服務，經過常規正則文本解析後，該資訊轉換應用程式群組會將這個專案的需求，轉給專案指定要使用的模型，產生該統一資源標誌符。 If the project request data includes a message with multiple agreement URLs, as follows:

Indicates that the entry point is the enterprise service bus (Enterprise Service Bus, ESB) channel, where <AiQnaInq> indicates the project name, and <01> indicates the service performed. After regular regular text analysis, the information conversion application group will convert The requirements of this project are transferred to the project to specify the model to be used, and the uniform resource identifier is generated.

If the project request data includes an independent space placement path, such as /in-nfs/hrface/employeeData, it means that the entry point is a file or directory change, where hrface is the project name, employeeData is the service name, and the information conversion application The group directly triggers the project's business logic to generate the URI.

In step 22, the processing unit 113 executes the security management application program group to determine whether the analysis result is legal. When it is determined that the analysis result is legal, the process proceeds to step 23; and when it is determined that the analysis result is illegal, the process proceeds to step 25.

It should be noted that, in this embodiment, the authorization code is composed of random character strings whose length is greater than 128 and less than or equal to 256, and the processing unit 113 executes the security management application group to determine whether the key of the authorization code is legitimate and the authorization code is in the within the valid time interval to determine whether the analysis result is legal, when the key of the authorization code is legal and the authorization code is within the valid time interval, the processing unit 113 determines that the analysis result is legal; and when the authorization code is valid If the key is invalid or the authorization code is not within the valid time interval, the processing unit 113 determines that the parsing result is invalid, but not limited thereto.

It should be noted that, in this embodiment, the processing unit 113 uses a JSON Web Token (JWT) algorithm to determine whether the key of the authorization code is legal, but not limited thereto.

In step 23, the processing unit 113 uses one of the artificial intelligence model containers corresponding to the target artificial intelligence model container of the model code of the analysis result to generate a service result according to the analysis result. It is worth noting that different artificial intelligence model containers are used to generate different service results. The service results are, for example, identification results of human faces, seals, or Anti-Money Laundering (AML) documents, but this does not mean that limit.

With reference to FIG. 4 , step 23 includes sub-steps 231 to 234 , and the sub-steps included in step 23 are described below.

In sub-step 231, the processing unit 113 utilizes the target artificial intelligence model container to generate a resource allocation request for a target resource allocation application among the corresponding resource allocation applications according to the parsing result.

In sub-step 232, the processing unit 113 executes the security management application The formula group obtains target authorization information corresponding to the target resource allocation application program and the target artificial intelligence model container from the authorization information.

It is worth noting that, in this embodiment, the security management application group obtains all authorization information from the underlying Kubernetes when it is activated, and when the target resource allocation application requests authorization information, the cloud-native Decoding method to obtain the authorization information of the target, but not limited thereto.

In sub-step 233, the processing unit 113 executes the target resource allocation application program to obtain the authorization of a target database corresponding to the target authorization information from one of the databases 13 according to the target authorization information, and allocates resources according to the target authorization information. A request is made to retrieve an allocation data related to the resource allocation request from the target database.

In sub-step 234, the processing unit 113 uses the target artificial intelligence model container to generate the service result according to the distribution data.

In step 24 , the processing unit 113 executes the security management application group to encode the authorization code and store it in the storage unit 112 .

In step 25 , the processing unit 113 executes the security management application group to generate and send a verification failure message to the client 12 via the communication unit 111 .

To sum up, the cloud microservice-based artificial intelligence model service method and server of the present invention use the processing unit 113 to execute the information conversion application group to analyze the project request data to generate the analysis result in a unified and specific format, so as to artificial intelligence The intelligent model container does not need to repeatedly process complex information conversion between different systems, and judges whether the analysis result is legal by executing the security management application group, so that the artificial intelligence model container does not need to manage authorization information by itself, which has high reusability and By executing the target resource allocation application program to obtain the authorization of the target database according to the target authorization information, the decentralized resource allocation (Decentralized resource allocation) provides high concurrency (High Concurrency) and high throughput (High Capacity) guarantees , In addition, the information conversion application group, the security management application group, and the resource allocation application group exist in the form of microservice clusters, which realize the horizontal and vertical expansion of microservice design, making the artificial intelligence model It is no longer just a single-point service, but has high availability (High Availability), so it can really achieve the purpose of the present invention.

But the above-mentioned ones are only embodiments of the present invention, and should not limit the scope of the present invention. All simple equivalent changes and modifications made according to the patent scope of the present invention and the content of the patent specification are still within the scope of the present invention. Within the scope covered by the patent of the present invention.

11: Artificial intelligence model service server

111: Communication unit

112: storage unit

113: Processing unit

12: Use end

13: Database

100: Communication network

Claims (8)

An artificial intelligence model service method, implemented by a server, the server stores an information conversion application program group, a security management application program group, multiple artificial intelligence model containers, a resource allocation application program group, and A plurality of pieces of authorization information respectively related to the databases, the server is connected to a client and multiple databases through a communication network, the resource allocation application program group includes multiple resource allocation application programs, and the resource allocation application programs The programs respectively correspond to the authorization information. The artificial intelligence model service method includes the following steps: (A) After the server receives a project request data from the client, the server executes the information conversion application program group to analyze the The project requests data and generates an analysis result in a specific format, and the analysis result includes a model code of one of the artificial intelligence model containers; (B) the server executes the security management application group to determine the analysis result whether it is legal; and (C) when the server determines that the analysis result is legal, the server uses one of the artificial intelligence model containers corresponding to the target artificial intelligence model of the model code of the analysis result according to the analysis result The container generates a service result, and the step (C) includes the following sub-steps: (C-1) When the server determines that the analysis result is legal, the server uses the target artificial intelligence model container to generate a service result according to the analysis result. In response to a resource allocation request from a target resource allocation application among the resource allocation applications, (C-2) the server executes the security management application group from Among the authorization information, a pair of target authorization information of the target resource allocation application and the target artificial intelligence model container is obtained, and (C-3) the server executes the target resource allocation application to obtain the target authorization information. an authorization of one of the databases to a target database of the target authorization information, and accessing an allocation data associated with the resource allocation request from the target database in accordance with the resource allocation request, and (C-4) the server Use the target artificial intelligence model container to generate the service result according to the distribution data; and (D) when the server determines that the analysis result is illegal, the server executes the security management application group to generate and pass through the communication network and send an authentication failure message to the client. The artificial intelligence model service method as described in claim 1, wherein step (A) includes the following sub-steps: (A-1) after the server receives the project request data, the server executes the information conversion application program The group performs regular regular text analysis on the item request data to generate a pre-conversion analysis result; and (A-2) the server executes the information transformation application group and converts the pre-conversion analysis result into the analysis result. The artificial intelligence model service method as described in claim item 1, wherein, in step (A), the analysis result includes an authorization code with a key and a valid time interval, and in step (B), the server Execute the security management application group to determine whether the key of the authorization code is legal and whether the authorization code is within the valid time interval, so as to determine whether the analysis result is legal, when If the key of the authorization code is legal and the authorization code is within the valid time interval, the server determines that the parsing result is legal. The artificial intelligence model service method described in claim 3 further includes the following steps after step (C): (E) the server executes the security management application program group to encode the authorization code and store it. An artificial intelligence model service server, comprising: a communication unit, connected to a communication network, and connected to a client and multiple databases through the communication network; a storage unit, storing an information conversion application program group, and a security management An application program group, a plurality of artificial intelligence model containers, a resource allocation application program group, and multiple pieces of authorization information respectively related to the databases, the resource allocation application program group includes multiple resource allocation application programs, the and other resource allocation applications respectively corresponding to the authorization information; a processing unit electrically connected to the communication unit and the storage unit; wherein, after the processing unit receives a project request data from the client through the communication unit, the processing The unit executes the information conversion application program group to parse the project request data and generate an analysis result in a specific format, the analysis result includes a model code corresponding to one of the artificial intelligence model containers, and executes the security management application program The group judges whether the analysis result is legal, and when the processing unit determines that the analysis result is legal, the processing unit uses one of the artificial intelligence model containers corresponding to the target artificial intelligence model code of the analysis result according to the analysis result. The smart model container generates a service result, when the processing unit determines that the solution When the analysis result is valid, the processing unit uses the target artificial intelligence model container to generate a resource allocation request for one of the resource allocation application programs according to the analysis result, and the processing unit executes the security management application The program group obtains the target authorization information of the target resource allocation application and the target artificial intelligence model container from the authorization information, and executes the target resource allocation application to obtain the target authorization information in the database An authorization of a target database corresponding to the target authorization information, and according to the resource allocation request, accessing an allocation data related to the resource allocation request from the target database, and then using the target artificial intelligence model container according to the allocation data , generating the service result, when the processing unit determines that the parsing result is invalid, the processing unit executes the security management application program group to generate and send a verification failure message to the client via the communication unit. The artificial intelligence model service server as described in claim item 5, wherein, after the processing unit receives the project request data through the communication unit, the processing unit executes the information conversion application program group to perform routine processing on the project request data The regular text is parsed to generate a pre-conversion parsing result, and the processing unit executes the information conversion application program group to convert the pre-conversion parsing result into the parsing result. The artificial intelligence model service server as described in claim item 5, wherein the analysis result includes an authorization code with a key and a valid time interval, and the processing unit executes the security management application group to determine the authorization code Whether the key is legal and whether the authorization code is within the valid time interval is used to determine whether the analysis result is legal. When the key of the authorization code is legal and the If the authorization code is within the valid time interval, the processing unit determines that the parsing result is legal. The artificial intelligence model service server according to claim 7, wherein the processing unit executes the security management application program group to encode the authorization code and store it in the storage unit.

Images