Apparatus and method for implementing AI / ML model management and exposure services

By implementing the registration, storage and management services of AI/ML models in network devices, the problems of inflexible and secure management of AI/ML models in the prior art are solved, and the functions of third-party applications in non-real-time RAN environments are supported. Improves the functional scalability and management efficiency of network equipment.

CN118945619APending Publication Date: 2024-11-12INTEL CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410578013.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-05-11
Filing Date
2024-05-10
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In the existing wireless communication network, the registration, storage and management mechanism of artificial intelligence/machine learning models is not flexible and secure enough, making it difficult to effectively manage and expose AI/ML models, resulting in the use of functions of third-party applications in non-real-time RAN environments. limit.

Method used

Provides a network device that executes computer program code through processor circuits and memory to realize the registration, storage and management services of AI/ML models, including decoding registration requests, verifying application permissions, registering and storage models, and provides API interfaces to support The registration, update, cancellation, discovery and subscription of the model ensures the safe and effective management of the model.

Benefits of technology

It realizes the secure registration, storage and management of AI/ML models, supports the flexible utilization of third-party applications in non-real-time RAN environments, and improves the functional scalability and management efficiency of network devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118945619A_ABST
    Figure CN118945619A_ABST
Patent Text Reader

Abstract

The invention relates to an apparatus and method for implementing AI / ML model management and exposure services. The AI / ML model management and exposure service may include AI / ML model registration, registration update, and logout; carrying out AI / ML model discovery; registering and querying the AI / ML model; the AI / ML model registers change subscription, subscription updating and subscription canceling; the AI / ML model registers a change notification; storing and acquiring an AI / ML model; and storing and acquiring the weight of the AI / ML model. Cases, service operations, and API designs for these services are described herein.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of the present disclosure relate generally to wireless communications, and more particularly to apparatus and methods for implementing artificial intelligence / machine learning (AI / ML) model management and exposure services in a non-real-time (non-RT) radio access network (RAN) intelligent controller (RIC). Background Art

[0002] Open RAN (O-RAN) is working to embed AI and ML capabilities into wireless communication networks by introducing non-RT RIC or near real-time (Near-RT) RIC. Third-party applications can be deployed into non-RT RIC, and such non-RT RIC applications are called "rApps". rApps leverage the available functions in the non-RT RIC framework to provide value-added services related to RAN operation and optimization. Summary of the invention

[0003] One aspect of the present application provides a network device, comprising a processor circuit and a memory comprising a computer program code, wherein the processor circuit is coupled to the memory and is configured to execute the computer program code so that the network device performs the following operations: decoding a registration model request from a first application, the registration model request comprising an application identifier of the first application and model registration information of an artificial intelligence / machine learning AI / ML model; checking whether the first application is authorized to register the AI / ML model; verifying the registration model request; registering the AI / ML model when it is determined that the first application is authorized to register the AI / ML model and the verification of the registration model request is successful; and encoding a registration model response for transmission to the first application, the registration model response comprising the registration identifier of the AI / ML model.

[0004] Another aspect of the present application provides a network device, comprising a processor circuit and a memory comprising a computer program code, wherein the processor circuit is coupled to the memory and is configured to execute the computer program code so that the network device performs the following operations: decoding a storage model request from a third application, the storage model request comprising an application identifier of the third application, a model identifier of an artificial intelligence / machine learning AI / ML model, and model file information of the AI / ML model; checking whether the third application is authorized to store the AI / ML model; verifying the storage model request; when it is determined that the third application is authorized to store the AI / ML model and the verification of the storage model request is successful, encoding a storage model response for transmission to the third application; and obtaining the model file information of the AI / ML model from the third application to store the AI / ML model in a model repository. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Embodiments of the present disclosure will be illustrated by way of example and not limitation in the accompanying figures in which like reference numerals refer to similar elements.

[0006] Figure 1 is a schematic diagram of an example network device according to various embodiments of the present disclosure.

[0007] Figure 2 is a schematic diagram of an example non-RT RIC architecture according to various embodiments of the present disclosure.

[0008] Figure 3 An exemplary overall end-to-end (E2E) sequence diagram for AI / ML model registration, registration update and deregistration, AI / ML model discovery, AI / ML model registration query, and AI / ML model registration change subscription, subscription update and unsubscription according to some embodiments of the present disclosure is shown.

[0009] Figure 4 An exemplary overall E2E sequence diagram for AI / ML model storage / retrieval and AI / ML model weight storage / retrieval according to some embodiments of the present disclosure is shown.

[0010] Figure 5A An example use case flow diagram for AI / ML model registration according to some embodiments of the present disclosure is shown.

[0011] Figure 5B An example use case flow diagram of an AI / ML model registration update according to some embodiments of the present disclosure is shown.

[0012] Figure 5C An example use case flow diagram of AI / ML model deregistration according to some embodiments of the present disclosure is shown.

[0013] Figure 5D Example service operations for registering AI / ML models according to some embodiments of the present disclosure are shown.

[0014] Figure 5E Example service operations for updating registration information of an AI / ML model according to some embodiments of the present disclosure are shown.

[0015] Fig. 5F Example service operations for partially updating registration information of an AI / ML model according to some embodiments of the present disclosure are shown.

[0016] Figure 5G Example service operations for unregistering an AI / ML model according to some embodiments of the present disclosure are shown.

[0017] Figure 5H An exemplary resource uniform resource identifier (URI) structure defined for an AI / ML model registration application program interface (API) according to some embodiments of the present disclosure is shown.

[0018] Fig.5I Another exemplary resource URI structure defined for an AI / ML model registration API according to some embodiments of the present disclosure is shown.

[0019] Fig. 6A An example use case flow diagram for AI / ML model discovery according to some embodiments of the present disclosure is shown.

[0020] Figure 6B An example use case flow diagram of an AI / ML model registration query according to some embodiments of the present disclosure is shown.

[0021] Figure 6C An example use case flow diagram of an AI / ML model registration change subscription according to some embodiments of the present disclosure is shown.

[0022] Fig.6D An example use case flow diagram of AI / ML model registration change subscription updates according to some embodiments of the present disclosure is shown.

[0023] Fig. 6E An example use case flow diagram of unsubscribing from AI / ML model registration changes according to some embodiments of the present disclosure is shown.

[0024] Fig. 6F An example use case flow diagram of AI / ML model registration change notification according to some embodiments of the present disclosure is shown.

[0025] Figure 6G Example service operations for discovering registered AI / ML models according to some embodiments of the present disclosure are shown.

[0026] Figure 6H Example service operations for querying registered AI / ML models according to some embodiments of the present disclosure are shown.

[0027] Fig.6I Example service operations for subscribing to AI / ML model registry changes according to some embodiments of the present disclosure are shown.

[0028] Figure 6J Example service operations for updating an AI / ML model registry change subscription are shown in accordance with some embodiments of the present disclosure.

[0029] Figure 6K Example service operations for partially updating an AI / ML model registry change subscription according to some embodiments of the present disclosure are shown.

[0030] Figure 6L Example service operations for unsubscribing from AI / ML model registration changes are shown in accordance with some embodiments of the present disclosure.

[0031] Figure 6M Example service operations for notifying AI / ML model registration changes according to some embodiments of the present disclosure are shown.

[0032] Figure 6N An exemplary resource URI structure defined for an AI / ML model discovery API according to some embodiments of the present disclosure is shown.

[0033] Fig. 7A An example use case flow diagram for AI / ML model storage according to some embodiments of the present disclosure is shown.

[0034] Figure 7B An example use case flow diagram for AI / ML model weight storage according to some embodiments of the present disclosure is shown.

[0035] Figure 7C An example use case flow chart for AI / ML model acquisition according to some embodiments of the present disclosure is shown.

[0036] Fig.7D An example use case flowchart for AI / ML model weight acquisition according to some embodiments of the present disclosure is shown.

[0037] Fig. 7E Example service operations for storing AI / ML models according to some embodiments of the present disclosure are shown.

[0038] Figure 7F Example service operations for storing weights of an AI / ML model according to some embodiments of the present disclosure are shown.

[0039] Figure 7G Example service operations for acquiring AI / ML models according to some embodiments of the present disclosure are shown.

[0040] Figure 7H Example service operations for obtaining weights of an AI / ML model according to some embodiments of the present disclosure are shown.

[0041] Fig.7I An exemplary resource URI structure defined for an AI / ML model storage API according to some embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0042] The various aspects of the illustrative embodiments will be described using terms commonly used by those skilled in the art to convey the essence of the present disclosure to other persons skilled in the art. However, it will be readily understood by those skilled in the art that many alternative embodiments may be practiced using portions of the described aspects. For purposes of explanation, specific numbers, materials, and configurations are set forth to provide a thorough understanding of the illustrative embodiments. However, it will be readily understood by those skilled in the art that alternative embodiments may be practiced without these specific details. In other cases, well-known features may be omitted or simplified to avoid blurring the illustrative embodiments.

[0043] Furthermore, various operations will be described as multiple discrete operations in a manner that is most helpful for understanding the illustrative embodiments; however, the order of description should not be construed as implying that these operations are necessarily order dependent. In particular, these operations do not need to be performed in the order presented.

[0044] The phrases "in an embodiment," "in one embodiment," and "in some embodiments" are used repeatedly herein. The phrase generally does not refer to the same embodiment; however, it may. The terms "comprising," "having," and "including" are synonymous unless the context dictates otherwise. The phrases "A or B" and "A / B" mean "(A), (B), or (A and B)."

[0045] O-RAN is working on embedding AI and ML capabilities into wireless communication networks by introducing non-RT RIC or near-RT RIC. Non-RT RIC is an internal function of the Service Management and Coordination (SMO) framework in the overall O-RAN architecture. Generally speaking, non-RTRIC includes a non-RT RIC framework and a non-RT RIC application (rApp). The non-RT RIC framework is an internal function of the SMO framework, which logically terminates the A1 interface with the near-RT RIC and exposes a set of R1 services to rApp. rApp leverages the available functions in the SMO / non-RT RIC framework to provide value-added services related to RAN operation and optimization. The scope of rApp is not limited to radio resource management, data analysis, and provision of rich information.

[0046] Figure 1is a schematic diagram of an example device 100 for implementing the method of the present disclosure. Device 100 can implement any rApp disclosed herein, as well as non-RT RIC, AI / ML management and exposure service consumers or producers. Device 100 can correspond to any type of known computer, server, or data processing device. For example, device 100 may include a processor, a personal computer (PC), a printed circuit board (PCB) including a computing device, a minicomputer, a mainframe computer, a microcomputer, a telephone computing device, a wired / wireless computing device (e.g., a smartphone, a personal digital assistant (PDA)), a laptop, a tablet device, a smart device, or any other device with similar functions.

[0047] In certain embodiments, such as Figure 1 As shown, device 100 may include a set of components, such as a processor 120 , a memory 130 , a storage component 140 , an input component 150 , an output component 160 , and a communication interface 170 .

[0048] The bus 110 may include one or more components that allow communication between a group of components of the device 100. For example, the bus 110 may be a communication bus, a crossbar, a network, etc. Although the bus 110 is Figure 1 Although described as a single line in FIG. 1 , bus 110 may be implemented by multiple (two or more) connections between a set of components of device 100. The present disclosure is not limited in this regard.

[0049] The device 100 may include one or more processors, such as a processor 120. The processor 120 may be implemented by hardware, firmware, and / or a combination of hardware and software. For example, the processor 120 may include a central processing unit (CPU), a graphics processing unit (GPU), an accelerated processing unit (APU), a microprocessor, a microcontroller, a digital signal processor (DSP), a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a general-purpose single-chip or multi-chip processor, or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof intended to implement the functions described herein. A general-purpose processor may be a microprocessor, or any conventional processor, controller, microcontroller, or state machine. The processor 120 may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, a combination of one or more microprocessors and a DSP core, or any other such configuration. In some embodiments, specific processes and methods may be performed by circuits dedicated to a given function.

[0050] Processor 120 may control the overall operation of device 100 and / or a group of components of device 100 (eg, memory 130 , storage component 140 , input component 150 , output component 160 , communication interface 170 ).

[0051] The device 100 may further include a memory 130. In some embodiments, the memory 130 may include a random access memory (RAM), a read-only memory (ROM), an electrically erasable programmable ROM (EEPROM), flash memory, magnetic memory, optical memory, and / or other types of dynamic or static storage devices. The memory 130 may store information and / or instructions for use (e.g., execution) by the processor 120.

[0052] The storage component 140 of the device 100 may store information and / or computer-readable instructions and / or code related to the operation and use of the device 100. For example, the storage component 140 may include a hard disk (such as a magnetic disk, optical disk, magneto-optical disk, and / or solid-state disk), a compact disk (CD), a digital versatile disk (DVD), a universal serial bus (USB) flash drive, a Personal Computer Memory Card International Association (PCMCIA) card, a floppy disk, a cassette tape, and / or other types of non-transitory computer-readable media and corresponding drives.

[0053] Device 100 may further include input component 150. Input component 150 may include one or more components that allow device 100 to receive information, such as receiving information through user input (e.g., a touch screen, keyboard, keypad, mouse, stylus, button, switch, microphone, camera, etc.). In addition, input component 150 may also include a sensor for sensing information (e.g., a global positioning system (GPS) component, an accelerometer, a gyroscope, an actuator, etc.).

[0054] Output components 160 of device 100 may include one or more components that can provide output information from device 100 (e.g., a display, a liquid crystal display (LCD), a light emitting diode (LED), an organic light emitting diode (OLED), a tactile feedback device, a speaker, etc.).

[0055] The device 100 may further include a communication interface 170. The communication interface 170 may include a receiver component, a transmitter component and / or a transceiver component. The communication interface 170 may enable the device 100 to establish a connection and / or transmit communication with other devices (such as a server, other devices). Communication may be achieved by a wired connection, a wireless connection, or a combination of a wired and wireless connection. The communication interface 170 may allow the device 100 to receive information from other devices and / or provide information to other devices. In some embodiments, the communication interface 170 may communicate with another device through a network, such as a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a private network, an ad hoc network, an intranet, the Internet, a fiber-based network, a cellular network (such as a fifth generation (5G) network), a fifth generation (5G) network, a long-term evolution (LTE) network, a third generation (3G) network, a code division multiple access (CDMA) network, etc.), a public land mobile network (PLMN), a telephone network (such as a public switched telephone network (PSTN)) or a similar network, and / or a combination of these networks or other types of networks. Alternatively or additionally, the communication interface 170 may also communicate with other devices via a device-to-device (D2D) communication link, such as FlashLinQ, WiMedia, Bluetooth, ZigBee, Wi-Fi, LTE, 5G, etc. In other embodiments, the communication interface 170 may include an Ethernet interface, an optical interface, a coaxial interface, an infrared interface, a radio frequency (RF) interface, or the like.

[0056] The device 100 may be included in the core network 240 and perform one or more processes described herein. The device 100 may perform operations based on the processor 120 executing computer-readable instructions and / or codes, which may be stored in a non-temporary computer-readable medium, such as a memory 130 and / or a storage component 140. A computer-readable medium may refer to a non-temporary storage device. A storage device may include storage space in a single physical storage device and / or storage space distributed in multiple physical storage devices. The computer-readable instructions and / or code may be read into the memory 130 and / or the storage component 140 from another computer-readable medium or another device via the communication interface 170. The computer-readable instructions and / or code stored in the memory 130 and / or the storage component 140, if or when executed by the processor 120, may cause the device 100 to perform one or more processes described herein.

[0057] As an alternative or addition, hard-wired circuits may also be used instead of software instructions or in combination with software instructions to perform one or more processes described herein. Therefore, the embodiments described herein are not limited to any specific combination of hardware circuits and software.

[0058] Figure 1The number and arrangement of components shown is only an example. In practice, there may be more than Figure 1 More components, fewer components, different components, or components in different arrangements may be shown. Figure 1 Two or more of the components shown in may be implemented in a single component, or Figure 1 The individual components shown in can be implemented as multiple distributed components. In addition, Figure 1 The set (one or more) of components shown may also perform Figure 1 One or more functions performed by another group of components shown.

[0059] Figure 2 FIG. 1 shows a non-RT RIC architecture diagram according to some embodiments of the present disclosure. In O-RAN, the non-RT RIC will provide policy-based guidance and rich information to the near-RT RIC to achieve intelligent RAN optimization and operation. Figure 2 As shown, the non-RT RIC architecture is part of the SMO framework. There are three types of logical functions in the non-RT RIC framework and the SMO framework: functions anchored inside the non-RT RIC framework, functions anchored outside the non-RT RIC framework, and non-anchored functions. For example, non-anchored functions may include AI / ML workflow functions that can provide AI / ML workflow services exposed through the R1 interface, such as AI / ML training services, AI / ML model management and exposure (AIMME) services, and AI / ML model performance monitoring services.

[0060] Specifically, the AI / ML training service can allow AI / ML training service consumers to request training of AI / ML models by specifying training requirements (such as required data, models, verification criteria, etc.). The AI / ML training service can be produced by the SMO / non-RT RIC framework or related rApps. The AI / ML model performance monitoring service allows authorized AI / ML performance monitoring service consumers to request monitoring of the performance of deployed AI / ML models.

[0061] AIMME service can be produced by SMO / non-RT RIC framework and can implement AI / ML model registration / unregistration, AI / ML model discovery, AI / ML model change subscription, AI / ML model storage, AI / ML model training capability registration / unregistration and other functions. AI / ML model registration / unregistration allows rApp to register, update registration and unregister the AI / ML models it provides. rApp needs to be authorized to register, update registration or unregister AI / ML models. AI / ML model discovery function allows rApp to discover registered AI / ML models based on filter conditions. rApp needs to be authorized to discover registered AI / ML models. AI / ML model change subscription allows rApp to subscribe to changes in registered AI / ML models and receive related notifications. AI / ML model storage allows rApp to request storage of AI / ML models and related information of AI / ML models. AI / ML model training capability registration / unregistration can be an optional service that allows AI / ML training service producers to register / update / unregister their ability to train AI / ML models.

[0062] In this disclosure, the AIMME service will be introduced in detail from the aspects of use cases, service operations and API design. According to the embodiments of the present disclosure, the following AIMME service operations are proposed: AI / ML model registration, registration update and deregistration; AI / ML model discovery; AI / ML model registration query; AI / ML model registration change subscription, subscription update and unsubscription; AI / ML model registration change notification; AI / ML model storage and acquisition; and AI / ML model weight storage and acquisition. Figure 3 An exemplary overall E2E sequence diagram for AI / ML model registration, registration update and deregistration, AI / ML model discovery, AI / ML model registration query, and AI / ML model registration change subscription, subscription update and unsubscription according to some embodiments of the present disclosure is shown. Figure 4 An exemplary overall E2E sequence diagram for AI / ML model storage / retrieval and AI / ML model weight storage / retrieval according to some embodiments of the present disclosure is shown.

[0063] like Figure 3 As shown, according to some embodiments of the present disclosure, services associated with AI / ML model registration (including AI / ML model registration, registration update, and deregistration) can be produced by the AI / ML workflow function for the AI / ML model producer rApp (herein referred to as the first application). FIG. 5A to FIG. 5I Describe the specific use cases, service operations, and API design of these services.

[0064] Figure 5A An example use case flow diagram for AI / ML model registration according to some embodiments of the present disclosure is shown. Figure 5A The AI / ML Model Registration use case shown allows the model producer rApp to register an AI / ML model. The context of the AI / ML Model Registration use case is shown in Table 5.1 below.

[0065]

[0066] Table 5.1: AI / ML model registration

[0067] according to Figure 5A As shown in the diagram of Table 5.1, from the perspective of the AI / ML workflow function, the workflow of AI / ML model registration can be described as follows: the AI / ML workflow function decodes the registration model request from the model producer rApp, which includes the application identifier (rApp id) of the model producer rApp and the model registration information of the AI / ML model; checks whether the model producer rApp is authorized to register the AI / ML model; verifies the registration model request; when it is determined that the model producer rApp is authorized to register the AI / ML model and the verification of the registration model request is successful, registers the AI / ML model; and encodes the registration model response for transmission to the model producer rApp, which includes the registration identifier (registration id) of the AI / ML model.

[0068] Figure 5B An example use case flow diagram of an AI / ML model registration update according to some embodiments of the present disclosure is shown. Figure 5B The AI / ML Model Registration Update use case shown allows the model producer rApp to update the registration of an AI / ML model. The context of the AI / ML Model Registration Update use case is shown in Table 5.2 below.

[0069]

[0070] Table 5.2: AI / ML Model Registration Updates

[0071] according to Figure 5BAs shown in the diagram of Table 5.2, from the perspective of the AI / ML workflow function, the workflow of the AI / ML model registration update can be described as follows: the AI / ML workflow function decodes the update model registration request from the model producer rApp, which includes the application identifier of the model producer rApp, the registration identifier of the AI / ML model, and the updated model registration information of the AI / ML model; checks whether the model producer rApp is authorized to update the registration of the AI / ML model; verifies the update model registration request; when it is determined that the model producer rApp is authorized to update the registration of the AI / ML model and the verification of the update model registration request is successful, updates the registration of the AI / ML model; and encodes the update model registration response for transmission to the model producer rApp.

[0072] Figure 5C An example use case flow diagram of AI / ML model deregistration according to some embodiments of the present disclosure is shown. Figure 5C The AI / ML model deregistration use case shown allows the model producer rApp to deregister the AI / ML model. The context of the AI / ML model deregistration use case is shown in Table 5.3 below.

[0073]

[0074] Table 5.3: AI / ML model deregistration

[0075] according to Figure 5C As shown in the diagram of Table 5.3, from the perspective of the AI / ML workflow function, the workflow of AI / ML model deregistration can be described as follows: the AI / ML workflow function decodes the model deregistration request from the model producer rApp, which includes the application identifier of the model producer rApp and the registration identifier of the AI / ML model; checks whether the model producer rApp is authorized to deregister the AI / ML model; verifies the model deregistration request; when it is determined that the model producer rApp is authorized to deregister the AI / ML model and the verification of the model deregistration request is successful, deregisters the AI / ML model; and encodes the model deregistration response for transmission to the model producer rApp.

[0076] In addition to the above-mentioned use cases related to AI / ML model registration, some embodiments of the present disclosure also provide service operations associated with AI / ML model registration at an AI / ML model registration API and definitions of the AI / ML model registration API, and will refer to FIG. 5D to FIG. 5I Describe in detail.

[0077] Figure 5D Example service operations for registering AI / ML models according to some embodiments of the present disclosure are shown. Figure 5DAs shown, the service operation can be used by an API consumer (e.g., a model producer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to register an AI / ML model.

[0078] According to some embodiments of the present disclosure, the service operation may be based on an HTTP POST method. Specifically, an API consumer may send an HTTP POST request including model registration information to an API producer. The API producer may process the message body received in the HTTP POST request and determine whether the registration request from the API consumer can be accepted. If so, the API producer may generate a registration identifier and construct a URI for the created model registration. The API producer may then return an HTTP POST response. If the model registration operation is successful, "201 Created" may be returned. The message body of the HTTP POST response may carry the created model registration information, and the "Location" HTTP header may be presented and may carry the URI for the created model registration identified by the registration identifier. If the model registration operation fails, an appropriate error code may be returned, and the message body of the HTTP POST response may contain additional error information.

[0079] Figure 5E FIG. 4 shows an example service operation for updating registration information of an AI / ML model according to some embodiments of the present disclosure. Figure 5E As shown, the service operation can be used by an API consumer (e.g., a model producer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to update the registration information of the AI / ML model.

[0080] According to some embodiments of the present disclosure, the service operation may be based on the HTTP PUT method. Specifically, the API consumer may send an HTTP PUT request to the API producer, which includes the registration identifier of the AI / ML model in the URI and the updated model registration information in the message body of the HTTP PUT request. The API producer may process the message body received in the HTTP PUT request and determine whether the updated registration request from the API consumer can be accepted. If so, the API producer may update the model registration information. The API producer may then return an HTTP PUT response. If the model registration update operation is successful, "200OK" may be returned. The message body of the HTTP PUT response may carry the updated model registration information. If the model registration update operation fails, an appropriate error code may be returned, and the message body of the HTTP PUT response may contain additional error information.

[0081] Fig. 5F FIG. 4 shows an example service operation for partially updating registration information of an AI / ML model according to some embodiments of the present disclosure. Fig. 5F As shown, the service operation can be used by an API consumer (e.g., a model producer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to partially update the registration information of the AI / ML model.

[0082] According to some embodiments of the present disclosure, the service operation may be based on the HTTP PATCH method. Specifically, the API consumer may send an HTTP PATCH request to the API producer, which includes the registration identifier of the AI / ML model in the URI and the model registration information patch in the message body of the HTTP PATCH request. The API producer may process the message body received in the HTTP PATCH request and determine whether the update registration request from the API consumer can be accepted. If so, the API producer may update the model registration information. The API producer may then return an HTTP PATCH response. If the model registration partial update operation is successful, "200OK" may be returned. The message body of the HTTP PATCH response may carry the updated model registration information. If the model registration update operation fails, an appropriate error code may be returned, and the message body of the HTTP PATCH response may contain additional error information.

[0083] Figure 5G Example service operations for deregistering an AI / ML model according to some embodiments of the present disclosure are shown. Figure 5G As shown, this service operation can be used by an API consumer (e.g., a model producer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to unregister an AI / ML model.

[0084] According to some embodiments of the present disclosure, the service operation may be based on the HTTP DELETE method. Specifically, the API consumer may send an HTTP DELETE request to the API producer, which includes the registration identifier of the AI / ML model in the URI. The API producer may determine whether the logout request from the API consumer can be accepted. If so, the API producer may remove the registration information of the AI / ML model. The API producer may then return an HTTP DELETE response. If the model logout operation is successful, "204 No Content" may be returned. If the model logout operation fails, an appropriate error code may be returned, and the message body of the HTTP DELETE response may contain additional error information.

[0085] Figure 5H An example resource URI structure defined for an AI / ML model registration API according to some embodiments of the present disclosure is shown. Fig.5I Another example resource URI structure defined for an AI / ML model registration API according to some embodiments of the present disclosure is shown. In the following description, Figure 5H The URI structure shown in is used as an example for illustration. Figure 5H Description of the URI structure in , using Fig.5I The adaptation of the shown alternative URI structure including the rAppld is straightforward.

[0086] Table 5.4 lists the resources defined for this API, the applicable HTTP methods, and the associated service operations.

[0087]

[0088]

[0089] Table 5.4

[0090] like Figure 5H As shown in Table 5.4, the resource URI for AI / ML model registration can be defined as {apiRoot} / modelregistration / <apiversion> / modelregistrations. Applicable HTTP methods associated with AI / ML model registrations may include the HTTP POST method.

[0091] The data structure supported by the message body of the HTTP POST request on this resource is shown in Table 5.5.

[0092]

[0093] Table 5.5

[0094] The data structure supported by the message body of the HTTP POST response on this resource is shown in Table 5.6.

[0095]

[0096] Table 5.6

[0097] The headers supported by the 201 response code on this resource are shown in Table 5.7.

[0098]

[0099] Table 5.7

[0100] In addition, if Figure 5H As shown in Table 5.4, the resource URI for individual AI / ML model registration is: {apiRoot} / modelregistration / <apiversion> / modelregistrations / {registrationId}. Applicable HTTP methods for individual AI / ML model registration may include HTTP PUT method, HTTP PATCH method, and HTTP DELETE method.

[0101] The data structure supported by the message body of the HTTP PUT request on this resource is shown in Table 5.8.

[0102] Data Types P Cardinality describe ModelRegistrationInfo M 1 Update AI / ML model registration information

[0103] Table 5.8

[0104] The data structure supported by the message body of the HTTP PUT response on this resource is shown in Table 5.9.

[0105]

[0106]

[0107] Table 5.9

[0108] The data structure supported by the message body of the HTTP PATCH request on this resource is shown in Table 5.10.

[0109] Data Types P Cardinality describe ModelRegistrationPatch M 1 Modify AI / ML model registration information

[0110] Table 5.10

[0111] The data structure supported by the message body of the HTTP PATCH response on this resource is shown in Table 5.11.

[0112]

[0113] Table 5.11

[0114] The data structure supported by the message body of the HTTP DELETE request on this resource is shown in Table 5.12.

[0115] Data Types P Cardinality describe not applicable

[0116] Table 5.12

[0117] The data structure supported by the message body of the HTTP DELETE response on this resource is shown in Table 5.13.

[0118]

[0119] Table 5.13

[0120] Further references Figure 3 According to some embodiments of the present disclosure, services related to AI / ML model discovery (including AI / ML model discovery, AI / ML model registration query, and AI / ML model registration change subscription, subscription update, and unsubscription) can be produced by the AI / ML workflow function for the AI / ML model consumer rApp (herein referred to as the second application). FIG. 6A to FIG. 6N Describe the specific use cases, service operations, and API design of these services.

[0121] Fig. 6A An example use case flow diagram for AI / ML model discovery according to some embodiments of the present disclosure is shown. Fig. 6A The AI / ML model discovery use case shown allows the model consumer rApp to discover registered AI / ML models. The context of the AI / ML model discovery use case is shown in Table 6.1 below.

[0122]

[0123] Table 6.1: AI / ML Model Discovery

[0124] according to Fig. 6A As shown in Table 6.1, from the perspective of the AI / ML workflow function, the workflow of AI / ML model discovery can be described as follows: the AI / ML workflow function decodes a discovery model request from a model consumer rApp, the discovery model request including an application identifier of the model consumer rApp and filtering conditions for discovering one or more registered AI / ML models; checks whether the model consumer rApp is authorized to discover registered AI / ML models; validates the discovery model request; and when it is determined that the model consumer rApp is authorized to discover registered AI / ML models and the validation of the discovery model request is successful, encodes a discovery model response for transmission to the model consumer rApp, the discovery model response including a list of model identifiers of registered AI / ML models that meet the filtering conditions and corresponding model metadata.

[0125] Figure 6B An example use case flow diagram of an AI / ML model registration query according to some embodiments of the present disclosure is shown. Figure 6B The AI / ML model registration query use case shown allows the model consumer rApp to query the registered AI / ML models. The context of the AI / ML model registration query use case is shown in Table 6.2 below.

[0126]

[0127] Table 6.2: AI / ML model registration query

[0128] according to Figure 6B As shown in Table 6.2, from the perspective of the AI / ML workflow function, the workflow of the AI / ML model registration query can be described as follows: the AI / ML workflow function decodes the query model registration request from the model consumer rApp, which query model registration request includes the application identifier of the model consumer rApp and the model identifier of the queried AI / ML model; checks whether the model consumer rApp is authorized to obtain the metadata of the queried AI / ML model; verifies the query model registration request; and when it is determined that the model consumer rApp is authorized to obtain the metadata of the queried AI / ML model and the verification of the query model registration request is successful, encodes the query model registration response for transmission to the model consumer rApp, and the query model registration response includes the metadata of the queried AI / ML model.

[0129] Figure 6C An example use case flow diagram of an AI / ML model registration change subscription according to some embodiments of the present disclosure is shown. Figure 6C The AI / ML Model Registration Change Subscription use case shown allows a model consumer rApp to subscribe to the registration changes of one or more registered AI / ML models. The context of the AI / ML Model Registration Change Subscription use case is shown in Table 6.3 below.

[0130]

[0131] Table 6.3: AI / ML Model Registration Change Subscription

[0132] according to Figure 6C As shown in Table 6.3, from the perspective of the AI / ML workflow function, the workflow of the AI / ML model registration change subscription can be described as follows: the AI / ML workflow function decodes the subscription model registration change request from the model consumer rApp, which subscription model registration change request includes the application identifier of the model consumer rApp and a list of model registration change events; checks whether the model consumer rApp is authorized to subscribe to the model registration changes of one or more identified AI / ML models; verifies the subscription model registration change request; when it is determined that the model consumer rApp is authorized to subscribe to the model registration changes of one or more identified AI / ML models and the verification of the subscription model registration change request is successful, creates a subscription to the model registration changes of one or more identified AI / ML models for the model consumer rApp; and encodes the subscription model registration change response for transmission to the model consumer rApp, which subscription model registration change response includes the subscription identifier of the model consumer rApp.

[0133] Fig.6D An example use case flow diagram of AI / ML model registration change subscription updates according to some embodiments of the present disclosure is shown. Fig.6D The AI / ML Model Registration Change Subscription Update use case shown allows a model consumer rApp to update its subscription to registration changes of one or more registered AI / ML models. The context of the AI / ML Model Registration Change Subscription Update use case is shown in Table 6.4 below.

[0134]

[0135] Table 6.4

[0136] according to Fig.6D As shown in the diagram of Table 6.4, from the perspective of the AI / ML workflow function, the workflow of the AI / ML model registration change subscription update can be described as follows: the AI / ML workflow function decodes the update model registration change subscription request from the model consumer rApp, which includes the application identifier, subscription identifier and updated model registration change event list of the model consumer rApp; checks whether the model consumer rApp is authorized to update the model registration change subscription; verifies the update model registration change subscription request; when it is determined that the model consumer rApp is authorized to update the model registration change subscription and the verification of the update model registration change subscription request is successful, updates the subscription of the model consumer rApp; and encodes the update model registration change subscription response for transmission to the model consumer rApp, which includes the updated model registration change event list.

[0137] Fig. 6E An example use case flow diagram of unsubscribing from AI / ML model registration changes according to some embodiments of the present disclosure is shown. Fig. 6E The AI / ML Model Registration Change Unsubscription use case shown allows a model consumer rApp to unsubscribe from the registration changes of one or more registered AI / ML models. The context of the AI / ML Model Registration Change Unsubscription use case is shown in Table 6.5 below.

[0138]

[0139] Table 6.5

[0140] according to Fig. 6E As shown in the diagram in Table 6.5, from the perspective of the AI / ML workflow function, the workflow of AI / ML model registration change unsubscription can be described as follows: the AI / ML workflow function decodes the unsubscribe model registration change request from the model consumer rApp, which unsubscribe model registration change request includes the application identifier and subscription identifier of the model consumer rApp; checks whether the model consumer rApp is authorized to unsubscribe from the model registration changes; verifies the unsubscribe model registration change request; when it is determined that the model consumer rApp is authorized to unsubscribe from the model registration changes and the verification of the unsubscribe model registration change request is successful, removes the subscription of the model consumer rApp; and encodes the unsubscribe model registration change response for transmission to the model consumer rApp.

[0141] Fig. 6F An example use case flow diagram of AI / ML model registration change notification according to some embodiments of the present disclosure is shown. Fig. 6F The AI / ML Model Registration Change Notification use case shown allows a model consumer rApp to receive notifications about registration changes to one or more registered AI / ML models. The context of the AI / ML Model Registration Change Notification use case is shown in Table 6.6 below.

[0142]

[0143] Table 6.6

[0144] according to Fig. 6F As shown in the diagram in Table 6.6, from the perspective of the AI / ML workflow function, the workflow of the AI / ML model registration change notification can be described as follows: The AI / ML workflow function encodes a notification about the model registration change for transmission to the model consumer rApp, and the notification includes a subscription identifier and details of the model registration change to which the model consumer rApp is subscribed.

[0145] In addition to the use cases related to AI / ML model discovery described above, some embodiments of the present disclosure also provide service operations associated with AI / ML model discovery at an AI / ML model discovery API and a definition of the AI / ML model discovery API, and will refer to Figures 6G to 6N Describe in detail.

[0146] Figure 6G Example service operations for discovering registered AI / ML models according to some embodiments of the present disclosure are shown. Figure 6G As shown, the service operation can be used by an API consumer (e.g., a model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to discover one or more registered AI / ML models.

[0147] According to some embodiments of the present disclosure, the service operation may be based on the HTTP GET method. Specifically, the API consumer may send an HTTP GET request to the API producer, and may include filtering conditions as query parameters for discovering registered AI / ML models. The message body of the HTTP GET request may be empty. The API producer may determine whether the discovery request from the API consumer can be accepted. The API producer may then return an HTTP GET response. If the AI / ML model discovery operation is successful, "200OK" may be returned, and the message body of the HTTP GET response may carry an array of model metadata including a model identifier. If the AI / ML model discovery operation fails, an appropriate error code may be returned, and the message body of the HTTP GET response may include additional error information.

[0148] Figure 6H FIG. 4 shows an example service operation for querying registered AI / ML models according to some embodiments of the present disclosure. Figure 6H As shown, the service operation can be used by an API consumer (e.g., a model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to query registered AI / ML models.

[0149] According to some embodiments of the present disclosure, the service operation may be based on the HTTP GET method. Specifically, the API consumer may send an HTTP GET request to the API producer, and may include the model identifier of the registered AI / ML model in the URI. The message body of the HTTP GET request may be empty. The API producer may determine whether the query request from the API consumer can be accepted. The API producer may then return an HTTP GET response. If the AI / ML model registration query operation is successful, "200OK" may be returned, and the message body of the HTTP GET response may carry metadata of the AI / ML model identified by the model identifier. If the AI / ML model registration query operation fails, an appropriate error code may be returned, and the message body of the HTTP GET response may contain additional error information.

[0150] Fig.6I An example service operation for subscribing to AI / ML model registration changes according to some embodiments of the present disclosure is shown. Fig.6I As shown, this service operation can be used by an API consumer (e.g., a model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to subscribe to AI / ML model registration changes.

[0151] According to some embodiments of the present disclosure, the service operation may be based on the HTTP POST method. Specifically, the API consumer may send an HTTP POST request to the API producer, and may include ModelRegistrationEvents, which represents the model registration change events that the API consumer is interested in receiving notifications of. The API producer may determine whether the subscription request from the API consumer can be accepted. The API producer may then return an HTTP POST response. If the AI / ML model registration change subscription operation is successful, "201Created" may be returned. The message body of the HTTP POST response may carry information about the created subscription for the model registration change event, and the "Location" HTTP header may be presented and may carry the URI for the created subscription identified by the subscription identifier. If the AI / ML model registration change subscription operation fails, an appropriate error code may be returned, and the message body of the HTTP POST response may contain additional error information.

[0152] Figure 6J Example service operations for updating AI / ML model registration change subscriptions according to some embodiments of the present disclosure are shown. Figure 6J As shown, this service operation can be used by an API consumer (e.g., a model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to update an AI / ML model registration change subscription.

[0153] According to some embodiments of the present disclosure, the service operation may be based on the HTTP PUT method. Specifically, the API consumer may send an HTTP PUT request to the API producer, and may include a subscription identifier in the URI and an updated subscription to the model registration change event of interest in the message body of the HTTP PUT request. The API producer may determine whether the update request from the API consumer is acceptable. The API producer may then return an HTTP PUT response. If the AI / ML model registration change subscription update operation is successful, "200OK" may be returned. The message body of the HTTP PUT response may carry the updated subscription. If the AI / ML model registration change subscription update operation fails, an appropriate error code may be returned, and the message body of the HTTP PUT response may include additional error information.

[0154] Figure 6K FIG. 4 shows an example service operation for partially updating an AI / ML model registration change subscription according to some embodiments of the present disclosure. Figure 6K As shown, this service operation can be used by an API consumer (e.g., a model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to partially update an AI / ML model registration change subscription.

[0155] According to some embodiments of the present disclosure, the service operation may be based on the HTTP PATCH method. Specifically, the API consumer may send an HTTP PATCH request to the API producer, and may include a subscription identifier in the URI and a subscription patch about the model registration change event of interest in the message body of the HTTP PATCH request. The API producer may determine whether a partial update request from the API consumer is acceptable. The API producer may then return an HTTP PATCH response. If the AI / ML model registration change subscription partial update operation is successful, "200OK" may be returned. The message body of the HTTP PATCH response may carry the updated subscription. If the AI / ML model registration change subscription partial update operation fails, an appropriate error code may be returned, and the message body of the HTTP PATCH response may include additional error information.

[0156] Figure 6L Example service operations for unsubscribing from AI / ML model registration changes according to some embodiments of the present disclosure are shown. Figure 6L As shown, this service operation can be used by an API consumer (e.g., a model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to unsubscribe from AI / ML model registration changes.

[0157] According to some embodiments of the present disclosure, the service operation may be based on the HTTP DELETE method. Specifically, the API consumer may send an HTTP DELETE request to the API producer, and may include a registration identifier in the URI. The message body of the HTTP DELETE request may be empty. The API producer may determine whether the unsubscribe request from the API consumer is acceptable. If so, the API producer may return an HTTP DELETE response. If the AI / ML model registration change unsubscribe operation is successful, "204 No Content" may be returned. The message body of the HTTP DELETE response may be empty. If the AI / ML model registration change unsubscribe operation fails, an appropriate error code may be returned, and the message body of the HTTP DELETE response may contain additional error information.

[0158] Figure 6M Example service operations for notifying AI / ML model registration changes according to some embodiments of the present disclosure are shown. Figure 6M As shown, the service operation can be used by an API producer (e.g., an AI / ML workflow function) to interact with an API consumer (e.g., a model consumer rApp) to notify the API consumer of changes to a subscribed AI / ML model registration.

[0159] According to some embodiments of the present disclosure, the service operation may be based on the HTTP POST method. Specifically, the API producer may send an HTTP POST request to the API consumer. The target URI may identify the destination of the notification. The message body of the HTTP POST request may carry a notification containing details about the model registration change. The API producer may then return an HTTP POST response. If the AI / ML model registration change unsubscription operation is successful, "204 No Content" may be returned. The message body of the HTTP POST response may be empty. If the AI / ML model registration change notification operation fails, an appropriate error code may be returned, and the message body of the HTTP POST response may contain additional error information.

[0160] Figure 6N An example resource URI structure defined for an AI / ML model discovery API according to some embodiments of the present disclosure is shown. Table 6.7 lists the resources defined for the API, applicable HTTP methods, and associated service operations.

[0161]

[0162]

[0163] Table 6.7

[0164] like Figure 6N As shown in Table 6.7, the resource URI for a registered AI / ML model can be defined as {apiRoot} / modeldiscovery / <apiversion> / models. Applicable HTTP methods associated with registered AI / ML models may include HTTP GET methods.

[0165] The list of URI query parameters supported by the HTTP GET method on this resource is shown in Table 6.8.

[0166]

[0167] Table 6.8

[0168] The data structure supported by the message body of the HTTP GET request on this resource is shown in Table 6.9.

[0169] Data Types P Cardinality describe not applicable

[0170] Table 6.9

[0171] The data structure supported by the message body of the HTTP GET response on this resource is shown in Table 6.10.

[0172]

[0173] Table 6.10

[0174] In addition, if Figure 6N As shown in Table 6.7, the resource URI for an individual registered AI / ML model can be defined as: {apiRoot} / modeldiscovery / <apiversion> / models / {modelId}. Applicable HTTP methods associated with an individual registered AI / ML model may include an HTTP GET method.

[0175] The list of URI query parameters supported by the HTTP GET method on this resource is shown in Table 6.11.

[0176] name Data Types P Cardinality describe not applicable

[0177] Table 6.11

[0178] The data structure supported by the message body of the HTTP GET request on this resource is shown in Table 6.12.

[0179] Data Types P Cardinality describe not applicable

[0180] Table 6.12

[0181] The data structure supported by the message body of the HTTP GET response on this resource is shown in Table 6.13.

[0182]

[0183] Table 6.13

[0184] In addition, if Figure 6N As shown in Table 6.7, the resource URI for AI / ML model registration change subscription can be defined as: {apiRoot} / modeldiscovery / <apiversion> / subscriptions. Applicable HTTP methods associated with model registration change subscriptions may include HTTP POST methods.

[0185] The data structure supported by the message body of the HTTP POST request on this resource is shown in Table 6.14.

[0186]

[0187] Table 6.14

[0188] The data structure supported by the message body of the HTTP POST response on this resource is shown in Table 6.15.

[0189]

[0190] Table 6.15

[0191] The headers supported by the 201 response code on this resource are shown in Table 6.16.

[0192]

[0193] Table 6.16

[0194] In addition, if Figure 6N As shown in Table 6.7, the resource URI for registering change subscriptions for individual AI / ML models can be defined as: {apiRoot} / modeldiscovery / <apiversion> / subscriptions / {subscriptionId}. Applicable HTTP methods associated with an individual AI / ML model registration change subscription may include an HTTP PUT method, an HTTP PATCH method, and an HTTP DELETE method.

[0195] The data structure supported by the message body of the HTTP PUT request on this resource is shown in Table 6.17.

[0196] Data Types P Cardinality describe ModelRegistrationEvents M 1 Subscription to updated AI / ML model registration change events

[0197] Table 6.17

[0198] The data structure supported by the message body of the HTTP PUT response on this resource is shown in Table 6.18.

[0199]

[0200] Table 6.18

[0201] The data structure supported by the message body of the HTTP PATCH request on this resource is shown in Table 6.19.

[0202] Data Types P Cardinality describe ModelRegistrationEventsPatch M 1 Modification of AI / ML model registration information

[0203] Table 6.19

[0204] The data structure supported by the message body of the HTTP PATCH response on this resource is shown in Table 6.20.

[0205]

[0206]

[0207] Table 6.20

[0208] The data structure supported by the message body of the HTTP DELETE request on this resource is shown in Table 6.21.

[0209] Data Types P Cardinality describe not applicable

[0210] Table 6.21

[0211] The data structure supported by the message body of the HTTP DELETE response on this resource is shown in Table 6.22.

[0212]

[0213] Table 6.22

[0214] In addition, Table 6.23 shows an overview of the notifications defined for the AI / ML model discovery API, including the notification name, Callback URI, applicable HTTP method, and associated service operations.

[0215]

[0216] Table 6.23

[0217] As shown in Table 6.23, the Callback URI for AI / ML model registration change notification may be defined as: {modelRegistrationNotification}. Applicable HTTP methods associated with model registration change notification may include an HTTP POST method.

[0218] The data structure supported by the message body of the HTTP POST request on this resource is shown in Table 6.24.

[0219]

[0220] Table 6.24

[0221] The data structure supported by the message body of the HTTP POST response on this resource is shown in Table 6.25.

[0222]

[0223]

[0224] Table 6.25

[0225] Back to Figure 4 According to some embodiments of the present disclosure, services associated with AI / ML model storage (including AI / ML model storage, AI / ML model weight storage, AI / ML model acquisition, and AI / ML model weight acquisition) can be produced by the AI / ML workflow function for the training service consumer rApp (herein referred to as the third application) or the training service producer rApp (herein referred to as the fourth application). 7A to 7I Describe the specific use cases, service operations, and API design of these services.

[0226] Fig. 7A An example use case flow diagram for AI / ML model storage according to some embodiments of the present disclosure is shown. Fig. 7A The AI / ML model storage use case shown allows rApps (e.g. Figure 4 The training service consumer rApp in the storage of AI / ML models. The context of the AI / ML model storage use case is shown in Table 7.1 below.

[0227]

[0228] Table 7.1

[0229] according to Fig. 7A As shown in the diagram of Table 7.1, from the perspective of the AI / ML workflow function, the workflow of AI / ML model storage can be described as follows: the AI / ML workflow function decodes the storage model request from the rApp, which includes the application identifier of the rApp, the model identifier of the AI / ML model, and the model file information of the AI / ML model; checks whether the rApp is authorized to store the AI / ML model; verifies the storage model request; when it is determined that the rApp is authorized to store the AI / ML model and the verification of the storage model request is successful, encodes the storage model response for transmission to the rApp; and obtains the model file information of the AI / ML model from the rApp to store the AI / ML model in the model repository.

[0230] Figure 7B An example use case flow diagram for AI / ML model weight storage according to some embodiments of the present disclosure is shown. Figure 7B The AI / ML model weight storage use case shown allows rApps (e.g. Figure 4 The training service producer rApp in the training service producer rApp stores the weights of the AI / ML model. The context of the AI / ML model weight storage use case is shown in Table 7.2 below.

[0231]

[0232] Table 7.2

[0233] according to Figure 7B As shown in the diagram of Table 7.2, from the perspective of the AI / ML workflow function, the workflow of AI / ML model weight storage can be described as follows: the AI / ML workflow function decodes the storage model weight request from the rApp, which includes the application identifier of the rApp, the model identifier of the stored AI / ML model, and the model weight file information of the stored AI / ML model; checks whether the rApp is authorized to store the weights of the stored AI / ML model; verifies the storage model weight request; when it is determined that the rApp is authorized to store the AI / ML model and the verification of the storage model weight request is successful, encodes a storage model weight response for transmission to the rApp, which includes the weight version identifier of the stored weights of the stored AI / ML model; and obtains the model weight file information of the stored AI / ML model from the rApp to store the weights of the stored AI / ML model in the model repository.

[0234] Figure 7C An example use case flow chart for AI / ML model acquisition according to some embodiments of the present disclosure is shown. Figure 7C The AI / ML model acquisition use case shown allows rApps (e.g. Figure 4 The training service producer rApp in the AI / ML model acquisition use case is shown in Table 7.3.

[0235]

[0236] Table 7.3

[0237] according to Figure 7C As shown in Table 7.3, from the perspective of the AI / ML workflow function, the workflow of AI / ML model weight storage can be described as follows: the AI / ML workflow function decodes the model acquisition request from the fourth application, which includes the application identifier of the rApp and the model identifier of the AI / ML model to be acquired; checks whether the rApp is authorized to acquire the AI / ML model to be acquired; verifies the model acquisition request; when it is determined that the rApp is authorized to acquire the AI / ML model to be acquired and the verification of the model acquisition request is successful, encodes the model acquisition response for transmission to the rApp, the model acquisition response includes the model file information of the AI / ML model to be acquired; and acquires the AI / ML model to be acquired from the model repository to provide it to the rApp.

[0238] Fig.7D An example use case flowchart for AI / ML model weight acquisition according to some embodiments of the present disclosure is shown. Fig.7D The AI / ML model weight acquisition use case shown allows rApps (e.g. Figure 4 The training service consumer rApp in the training service consumer rApp obtains a set of weights of the AI / ML model. The context of the AI / ML model weight acquisition use case is shown in Table 7.4 below.

[0239]

[0240] Table 7.4

[0241] according to Fig.7D As shown in the diagram of Table 7.4, from the perspective of the AI / ML workflow function, the workflow for obtaining AI / ML model weights can be described as follows: the AI / ML workflow function decodes a request for obtaining model weights from rApp, which includes the application identifier of rApp, the model identifier of the stored AI / ML model, and the weight version identifier of the weights of the stored AI / ML model; checks whether rApp is authorized to obtain the weights of the identified version of the stored AI / ML model; verifies the request for obtaining model weights; when it is determined that rApp is authorized to obtain the weights of the identified version of the stored AI / ML model and the verification of the request for obtaining model weights is successful, encodes a response for obtaining model weights for transmission to rApp, which includes model weight file information of the weights of the identified version of the stored AI / ML model; and obtains the weights of the identified version of the stored AI / ML model from the model repository to provide to rApp.

[0242] In addition to the above-mentioned use cases related to AI / ML model storage, some embodiments of the present disclosure also provide service operations associated with AI / ML model storage at an AI / ML model storage API and the definition of the AI / ML model storage API, and will refer to FIG. 7E to FIG. 7I Describe in detail.

[0243] Fig. 7E Example service operations for storing AI / ML models according to some embodiments of the present disclosure are shown. Fig. 7E As shown, the service operation can be used by an API consumer (e.g., a training model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to store AI / ML models.

[0244] According to some embodiments of the present disclosure, the service operation may be based on the HTTP PUT method. Specifically, the API consumer may send an HTTP PUT request to the API producer. The message body of the HTTP PUT request may contain ModelFileInfo, representing the model file information. The API producer may determine whether the storage request from the API consumer can be accepted. If so, the API producer may return an HTTP PUT response. If the AI / ML model storage operation is successful, "201Created" may be returned, and the message body of the HTTP PUT response may carry the model file information. If the AI / ML model storage operation fails, an appropriate error code may be returned, and the message body of the HTTP PUT response may contain additional error information.

[0245] Figure 7F An example service operation for storing weights of an AI / ML model according to some embodiments of the present disclosure is shown. Figure 7F As shown, the service operation can be used by an API consumer (e.g., a training model producer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to store a set of weights for an AI / ML model.

[0246] According to some embodiments of the present disclosure, the service operation may be based on the HTTP POST method. Specifically, the API consumer may send an HTTP POST request to the API producer, and may include ModelWeightsFileInfo, representing the model weight file information. The API producer may determine whether the storage request from the API consumer can be accepted. If so, the API producer may return an HTTP POST response. If the AI / ML model weight storage operation is successful, "201Created" may be returned. The message body of the HTTP POST response may carry the model weight file information, and the "Location" HTTP header may be presented and may carry the URI for the created weight file resource identified by the weight version identifier (also referred to here as weightsId). If the AI / ML model weight storage operation fails, an appropriate error code may be returned, and the message body of the HTTP POST response may contain additional error information.

[0247] Figure 7G FIG. 4 shows an example service operation for acquiring an AI / ML model according to some embodiments of the present disclosure. Figure 7G As shown, the service operation can be used by an API consumer (e.g., a training model producer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to obtain an AI / ML model.

[0248] According to some embodiments of the present disclosure, the service operation may be based on the HTTP GET method. Specifically, the API consumer may send an HTTP GET request to the API producer, and may include a model identifier in the URI. The message body of the HTTP GET may be empty. The API producer may determine whether the acquisition request from the API consumer can be accepted. If so, the API producer may return an HTTP GET response. If the AI / ML model acquisition operation is successful, "200OK" may be returned, and the message body of the HTTP GET response may carry the model file information of the model identified by the model identifier. If the AI / ML model acquisition operation fails, an appropriate error code may be returned, and the message body of the HTTP GET response may include additional error information.

[0249] Figure 7H An example service operation for obtaining weights of an AI / ML model according to some embodiments of the present disclosure is shown. Figure 7H As shown, the service operation can be used by an API consumer (e.g., a training model consumer rApp) to interact with an API producer (e.g., an AI / ML workflow function) to obtain the weights of an AI / ML model.

[0250] According to some embodiments of the present disclosure, the service operation may be based on the HTTP GET method. Specifically, the API consumer may send an HTTP GET request to the API producer, and may include a model identifier and a weight version identifier in the URI. The message body of the HTTP GET may be empty. The API producer may determine whether the acquisition request from the API consumer can be accepted. If so, the API producer may return an HTTP GET response. If the AI / ML model weight acquisition operation is successful, "200OK" may be returned, and the message body of the HTTP GET response may carry the model weight file information identified by the model identifier and the weight version identifier. If the AI / ML model weight acquisition operation fails, an appropriate error code may be returned, and the message body of the HTTP GET response may contain additional error information.

[0251] Fig.7I An example resource URI structure defined for an AI / ML model storage API according to some embodiments of the present disclosure is shown. Table 7.5 lists the resources defined for the API, applicable HTTP methods, and associated service operations.

[0252]

[0253] Table 7.5

[0254] like Fig.7I As shown in Table 7.5, the resource URI for individual AI / ML model file information can be defined as: {apiRoot} / modelrepository / <apiversion> / models / {modelId}. Applicable HTTP methods associated with individual AI / ML model file information may include HTTP PUT method and HTTP GET method.

[0255] The data structure supported by the message body of the HTTP PUT request on this resource is shown in Table 7.6.

[0256] Data Types P Cardinality describe ModelFileInfo M 1 AI / ML model file information

[0257] Table 7.6

[0258] The data structure supported by the message body of the HTTP PUT response on this resource is shown in Table 7.7.

[0259]

[0260] Table 7.7

[0261] The URI query parameters supported by the HTTP GET method on this resource are shown in Table 7.8.

[0262] name Data Types P Cardinality describe not applicable

[0263] Table 7.8

[0264] The data structure supported by the message body of the HTTP GET request on this resource is shown in Table 7.9.

[0265] Data Types P Cardinality describe not applicable

[0266] Table 7.9

[0267] The data structure supported by the message body of the HTTP GET response on this resource is shown in Table 7.10.

[0268]

[0269] Table 7.10

[0270] In addition, if Fig.7I As shown in Table 7.5, the resource URI for the model weight file can be defined as {apiRoot} / modelrepository / <apiversion> / models / {modelid} / weights. Applicable HTTP methods associated with the model weights file may include an HTTP POST method.

[0271] The data structure supported by the message body of the HTTP POST request on this resource is shown in Table 7.11.

[0272]

[0273] Table 7.11

[0274] The data structure supported by the message body of the HTTP POST response on this resource is shown in Table 7.12.

[0275]

[0276] Table 7.12

[0277] The headers supported by the 201 response code on this resource are shown in Table 7.13.

[0278]

[0279] Table 7.13

[0280] In addition, if Fig.7I As shown in Table 7.5, the resource URI for individual AI / ML model weight file information can be defined as: {apiRoot} / modelrepository / <apiversion> / models / {modelId} / weights / {weightsId}. Applicable HTTP methods associated with individual AI / ML model weights file information may include an HTTP GET method.

[0281] The URI query parameters supported by the HTTP GET method on this resource are shown in Table 7.14.

[0282] name Data Types P Cardinality describe not applicable

[0283] Table 7.14

[0284] The data structure supported by the message body of the HTTP GET request on this resource is shown in Table 7.15.

[0285] Data Types P Cardinality describe not applicable

[0286] Table 7.15

[0287] The data structure supported by the message body of the HTTP GET response on this resource is shown in Table 7.16.

[0288]

[0289]

[0290] Table 7.16

[0291] The following paragraphs describe examples of various embodiments.

[0292] Example 1 includes a network device comprising a processor circuit and a memory comprising a computer program code, wherein the processor circuit is coupled to the memory and is configured to execute the computer program code so that the network device performs the following operations: decoding a registration model request from a first application, the registration model request comprising an application identifier of the first application and model registration information of an artificial intelligence / machine learning AI / ML model; checking whether the first application is authorized to register the AI / ML model; verifying the registration model request; registering the AI / ML model when it is determined that the first application is authorized to register the AI / ML model and the verification of the registration model request is successful; and encoding a registration model response for transmission to the first application, the registration model response comprising a registration identifier of the AI / ML model.

[0293] Example 2 includes a network device according to Example 1, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding an update model registration request from the first application, the update model registration request including an application identifier of the first application, a registration identifier of the AI / ML model, and updated model registration information of the AI / ML model; checking whether the first application is authorized to update the registration of the AI / ML model; verifying the update model registration request; when it is determined that the first application is authorized to update the registration of the AI / ML model and the verification of the update model registration request is successful, updating the registration of the AI / ML model; and encoding an update model registration response for transmission to the first application.

[0294] Example 3 includes a network device according to Example 1 or 2, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a deregistration model request from the first application, the deregistration model request including an application identifier of the first application and a registration identifier of the AI / ML model; checking whether the first application is authorized to deregister the AI / ML model; verifying the deregistration model request; when it is determined that the first application is authorized to deregister the AI / ML model and the verification of the deregistration model request is successful, deregistering the AI / ML model; and encoding a deregistration model response for transmission to the first application.

[0295] Example 4 includes a network device according to any one of Examples 1 to 3, wherein the processor circuit is configured to execute the computer program code to cause the network device to register the AI / ML model based on an HTTP POST method.

[0296] Example 5 includes a network device according to any one of Examples 1 to 4, wherein the processor circuit is configured to execute the computer program code so that the network device updates the registration of the AI / ML model based on an HTTP PUT method or partially updates the registration of the AI / ML model based on an HTTP PATCH method.

[0297] Example 6 includes a network device according to any one of Examples 1 to 5, wherein the processor circuit is configured to execute the computer program code to cause the network device to deregister the AI / ML model based on an HTTP DELETE method.

[0298] Example 7 includes a network device according to any one of Examples 1 to 6, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a model discovery request from a second application, the model discovery request including an application identifier of the second application and filtering conditions for discovering one or more registered AI / ML models; checking whether the second application is authorized to discover the registered AI / ML models; verifying the model discovery request; and when it is determined that the second application is authorized to discover the registered AI / ML models and the verification of the model discovery request is successful, encoding a model discovery response for transmission to the second application, the model discovery response including a list of model identifiers of the registered AI / ML models that meet the filtering conditions and corresponding model metadata.

[0299] Example 8 includes a network device according to any one of Examples 1 to 7, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a query model registration request from a second application, the query model registration request including an application identifier of the second application and a model identifier of the queried AI / ML model; checking whether the second application is authorized to obtain the metadata of the queried AI / ML model; verifying the query model registration request; and when it is determined that the second application is authorized to obtain the metadata of the queried AI / ML model and the verification of the query model registration request is successful, encoding a query model registration response for transmission to the second application, the query model registration response including the metadata of the queried AI / ML model.

[0300] Example 9 includes a network device according to any one of Examples 1 to 8, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a subscription model registration change request from a second application, the subscription model registration change request including an application identifier of the second application and a list of model registration change events; checking whether the second application is authorized to subscribe to model registration changes of one or more identified AI / ML models; verifying the subscription model registration change request; when it is determined that the second application is authorized to subscribe to model registration changes of the one or more identified AI / ML models and the verification of the subscription model registration change request is successful, creating a subscription to model registration changes of the one or more identified AI / ML models for the second application; and encoding a subscription model registration change response for transmission to the second application, the subscription model registration change response including the subscription identifier of the second application.

[0301] Example 10 includes a network device according to any one of Examples 1 to 9, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding an update model registration change subscription request from the second application, the update model registration change subscription request including an application identifier of the second application, the subscription identifier and a list of updated model registration change events; checking whether the second application is authorized to update the model registration change subscription; verifying the update model registration change subscription request; when it is determined that the second application is authorized to update the model registration change subscription and the verification of the update model registration change subscription request is successful, updating the subscription of the second application; and encoding an update model registration change subscription response for transmission to the second application, the update model registration change subscription response including the list of updated model registration change events.

[0302] Example 11 includes a network device according to any one of Examples 1 to 10, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a request to unsubscribe from model registration changes from the second application, the request to unsubscribe from model registration changes including an application identifier and the subscription identifier of the second application; checking whether the second application is authorized to unsubscribe from the subscription to model registration changes; verifying the request to unsubscribe from model registration changes; when it is determined that the second application is authorized to unsubscribe from the subscription to model registration changes and the verification of the request to unsubscribe from model registration changes is successful, removing the subscription of the second application; and encoding a response to unsubscribe from model registration changes for transmission to the second application.

[0303] Example 12 includes a network device according to any one of Examples 1 to 11, wherein the processor circuit is configured to execute the computer program code so that the network device further performs the following operations: encoding a notification about a model registration change for transmission to the second application, the notification including the subscription identifier and details of the model registration change to which the second application subscribes.

[0304] Example 13 includes a network device according to any one of Examples 1 to 12, wherein the processor circuit is configured to execute the computer program code so that the network device discovers the registered AI / ML model based on an HTTP GET method.

[0305] Example 14 includes a network device according to any one of Examples 1 to 13, wherein the processor circuit is configured to execute the computer program code so that the network device obtains the metadata of the queried AI / ML model based on an HTTP GET method.

[0306] Example 15 includes the network device of any of Examples 1 to 14, wherein the processor circuit is configured to execute the computer program code to cause the network device to create the subscription to model registration changes based on an HTTP POST method.

[0307] Example 16 includes a network device according to any one of Examples 1 to 15, wherein the processor circuit is configured to execute the computer program code so that the network device updates the subscription to model registration changes based on an HTTP PUT method or partially updates the subscription to model registration changes based on an HTTP PATCH method.

[0308] Example 17 includes the network device of any of Examples 1 to 16, wherein the processor circuit is configured to execute the computer program code to cause the network device to remove the subscription to model registration changes based on an HTTP DELETE method.

[0309] Example 18 includes the network device of any of Examples 1 to 17, wherein the processor circuit is configured to execute the computer program code to cause the network device to notify the second application about the subscribed model registration change based on an HTTP POST method.

[0310] Example 19 includes a network device according to any one of Examples 1 to 18, wherein the first application includes an AI / ML model producer application deployed in a non-RT RIC.

[0311] Example 20 includes a network device according to any one of Examples 1 to 19, wherein the second application includes an AI / ML model consumer application deployed in a non-RT RIC.

[0312] Example 21 includes a network device comprising a processor circuit and a memory comprising a computer program code, wherein the processor circuit is coupled to the memory and is configured to execute the computer program code so that the network device performs the following operations: decoding a storage model request from a third application, the storage model request comprising an application identifier of the third application, a model identifier of an artificial intelligence / machine learning AI / ML model, and model file information of the AI / ML model; checking whether the third application is authorized to store the AI / ML model; verifying the storage model request; when it is determined that the third application is authorized to store the AI / ML model and the verification of the storage model request is successful, encoding a storage model response for transmission to the third application; and obtaining the model file information of the AI / ML model from the third application to store the AI / ML model in a model repository.

[0313] Example 22 includes a network device according to Example 21, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a storage model weight request from a fourth application, the storage model weight request including an application identifier of the fourth application, a model identifier of a stored AI / ML model, and model weight file information of the stored AI / ML model; checking whether the fourth application is authorized to store the weights of the stored AI / ML model; verifying the storage model weight request; when it is determined that the fourth application is authorized to store the AI / ML model and the verification of the storage model weight request is successful, encoding a storage model weight response for transmission to the fourth application, the storage model weight response including a weight version identifier of the stored weights of the stored AI / ML model; and obtaining the model weight file information of the stored AI / ML model from the fourth application to store the weights of the stored AI / ML model in a model repository.

[0314] Example 23 includes a network device according to Example 21 or 22, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a model acquisition request from a fourth application, the model acquisition request including an application identifier of the fourth application and a model identifier of the AI / ML model to be acquired; checking whether the fourth application is authorized to acquire the AI / ML model to be acquired; verifying the model acquisition request; when it is determined that the fourth application is authorized to acquire the AI / ML model to be acquired and the verification of the model acquisition request is successful, encoding a model acquisition response for transmission to the fourth application, the model acquisition response including model file information of the AI / ML model to be acquired; and acquiring the AI / ML model to be acquired from the model repository to provide it to the fourth application.

[0315] Example 24 includes a network device according to any one of Examples 21 to 23, wherein the processor circuit is configured to execute the computer program code so that the network device also performs the following operations: decoding a request for obtaining model weights from the third application, the request for obtaining model weights including an application identifier of the third application, a model identifier of a stored AI / ML model, and a weight version identifier of the weights of the stored AI / ML model; checking whether the third application is authorized to obtain the weights of the identified version of the stored AI / ML model; verifying the request for obtaining model weights; when it is determined that the third application is authorized to obtain the weights of the identified version of the stored AI / ML model and the verification of the request for obtaining model weights is successful, encoding a response for obtaining model weights for transmission to the third application, the response for obtaining model weights including model weight file information of the weights of the identified version of the stored AI / ML model; and obtaining the weights of the identified version of the stored AI / ML model from the model repository to provide to the third application.

[0316] Example 25 includes a network device according to any one of Examples 21 to 24, wherein the processor circuit is configured to execute the computer program code to cause the network device to store the AI / ML model based on an HTTP PUT method.

[0317] Example 26 includes a network device according to any one of Examples 21 to 25, wherein the processor circuit is configured to execute the computer program code so that the network device stores the weights of the stored AI / ML model based on an HTTP POST method.

[0318] Example 27 includes a network device according to any one of Examples 21 to 26, wherein the processor circuit is configured to execute the computer program code so that the network device obtains the AI / ML model to be obtained based on an HTTP GET method.

[0319] Example 28 includes a network device according to any one of Examples 21 to 27, wherein the processor circuit is configured to execute the computer program code so that the network device obtains the weight of the identified version of the stored AI / ML model based on an HTTP GET method.

[0320] Example 29 includes a method, comprising: receiving a registration model request from a first application, the registration model request including an application identifier of the first application and model registration information of an artificial intelligence / machine learning AI / ML model; checking whether the first application is authorized to register the AI / ML model; verifying the registration model request; when it is determined that the first application is authorized to register the AI / ML model and the verification of the registration model request is successful, registering the AI / ML model; and sending a registration model response to the first application, the registration model response including the registration identifier of the AI / ML model.

[0321] Example 30 includes the method according to Example 29, further including: receiving an update model registration request from the first application, the update model registration request including an application identifier of the first application, a registration identifier of the AI / ML model, and updated model registration information of the AI / ML model; checking whether the first application is authorized to update the registration of the AI / ML model; verifying the update model registration request; when it is determined that the first application is authorized to update the registration of the AI / ML model and the verification of the update model registration request is successful, updating the registration of the AI / ML model; and sending an update model registration response to the first application.

[0322] Example 31 includes the method according to Example 29 or 30, further comprising: receiving a model deregistration request from the first application, the model deregistration request including the application identifier of the first application and the registration identifier of the AI / ML model; checking whether the first application is authorized to deregister the AI / ML model; verifying the model deregistration request; when it is determined that the first application is authorized to deregister the AI / ML model and the verification of the model deregistration request is successful, deregistering the AI / ML model; and sending a model deregistration response to the first application.

[0323] Example 32 includes a method according to any one of Examples 29 to 31, wherein registering the AI / ML model is based on an HTTP POST method.

[0324] Example 33 includes a method according to any one of Examples 29 to 32, wherein updating the registration of the AI / ML model includes updating the registration of the AI / ML model based on an HTTP PUT method or partially updating the registration of the AI / ML model based on an HTTP PATCH method.

[0325] Example 34 includes a method according to any one of Examples 29 to 33, wherein deregistering the AI / ML model is based on an HTTP DELETE method.

[0326] Example 35 includes the method according to any one of Examples 29 to 34, further including: receiving a model discovery request from a second application, the model discovery request including an application identifier of the second application and filtering conditions for discovering one or more registered AI / ML models; checking whether the second application is authorized to discover the registered AI / ML models; verifying the model discovery request; and when it is determined that the second application is authorized to discover the registered AI / ML models and the verification of the model discovery request is successful, sending a model discovery response to the second application, the model discovery response including a list of model identifiers of the registered AI / ML models that meet the filtering conditions and corresponding model metadata.

[0327] Example 36 includes the method according to any one of Examples 29 to 35, further including: receiving a query model registration request from a second application, the query model registration request including an application identifier of the second application and a model identifier of the queried AI / ML model; checking whether the second application is authorized to obtain the metadata of the queried AI / ML model; verifying the query model registration request; and when it is determined that the second application is authorized to obtain the metadata of the queried AI / ML model and the verification of the query model registration request is successful, sending a query model registration response to the second application, the query model registration response including the metadata of the queried AI / ML model.

[0328] Example 37 includes the method according to any one of Examples 29 to 36, further including: receiving a subscription model registration change request from a second application, the subscription model registration change request including an application identifier of the second application and a list of model registration change events; checking whether the second application is authorized to subscribe to model registration changes of one or more identified AI / ML models; verifying the subscription model registration change request; when it is determined that the second application is authorized to subscribe to model registration changes of the one or more identified AI / ML models and the verification of the subscription model registration change request is successful, creating a subscription to model registration changes of the one or more identified AI / ML models for the second application; and sending a subscription model registration change response to the second application, the subscription model registration change response including the subscription identifier of the second application.

[0329] Example 38 includes the method according to any one of Examples 29 to 37, further including: receiving an update model registration change subscription request from the second application, the update model registration change subscription request including the application identifier of the second application, the subscription identifier and a list of updated model registration change events; checking whether the second application is authorized to update the model registration change subscription; verifying the update model registration change subscription request; when it is determined that the second application is authorized to update the model registration change subscription and the verification of the update model registration change subscription request is successful, updating the subscription of the second application; and sending an update model registration change subscription response to the second application, the update model registration change subscription response including the list of updated model registration change events.

[0330] Example 39 includes the method according to any one of Examples 29 to 38, further including: receiving a request to unsubscribe from model registration changes from the second application, the request to unsubscribe from model registration changes including the application identifier and the subscription identifier of the second application; checking whether the second application is authorized to unsubscribe from the subscription to model registration changes; verifying the request to unsubscribe from model registration changes; when it is determined that the second application is authorized to unsubscribe from the subscription to model registration changes and the verification of the request to unsubscribe from model registration changes is successful, removing the subscription of the second application; and sending a response to unsubscribe from model registration changes to the second application.

[0331] Example 40 includes the method according to any one of Examples 29 to 39, further comprising: sending a notification about a model registration change to the second application, the notification including the subscription identifier and details of the model registration change to which the second application subscribes.

[0332] Example 41 includes a method according to any one of Examples 29 to 40, wherein the finding of the registered AI / ML model is based on an HTTP GET method.

[0333] Example 42 includes a method according to any one of Examples 29 to 41, wherein querying the registered AI / ML models is based on an HTTP GET method.

[0334] Example 43 includes the method according to any one of Examples 29 to 42, wherein creating a subscription to model registration changes is based on an HTTP POST method.

[0335] Example 44 includes the method according to any one of Examples 29 to 43, wherein updating the subscription to the model registration changes includes updating the subscription to the model registration changes based on an HTTP PUT method or partially updating the subscription to the model registration changes based on an HTTP PATCH method.

[0336] Example 45 includes the method of any one of Examples 29 to 44, wherein removing a subscription to model registration changes is based on an HTTP DELETE method.

[0337] Example 46 includes the method according to any one of Examples 29 to 45, wherein the notification about the subscribed model registration change is based on an HTTP POST method.

[0338] Example 47 includes a method according to any one of Examples 29 to 46, wherein the first application includes an AI / ML model producer application deployed in a non-RT RIC.

[0339] Example 48 includes a method according to any one of Examples 29 to 47, wherein the second application includes an AI / ML model consumer application deployed in a non-RT RIC.

[0340] Example 49 includes a method, comprising: receiving a storage model request from a third application, the storage model request including an application identifier of the third application, a model identifier of an artificial intelligence / machine learning AI / ML model, and model file information of the AI / ML model; checking whether the third application is authorized to store the AI / ML model; verifying the storage model request; when it is determined that the third application is authorized to store the AI / ML model and the verification of the storage model request is successful, sending a storage model response to the third application; and obtaining the model file information of the AI / ML model from the third application to store the AI / ML model in a model repository.

[0341] Example 50 includes the method according to Example 49, further including: receiving a storage model weight request from a fourth application, the storage model weight request including an application identifier of the fourth application, a model identifier of a stored AI / ML model, and model weight file information of the stored AI / ML model; checking whether the fourth application is authorized to store the weights of the stored AI / ML model; verifying the storage model weight request; when it is determined that the fourth application is authorized to store the AI / ML model and the verification of the storage model weight request is successful, sending a storage model weight response to the fourth application, the storage model weight response including a weight version identifier of the stored weights of the stored AI / ML model; and obtaining the model weight file information of the stored AI / ML model from the fourth application to store the weights of the stored AI / ML model in a model repository.

[0342] Example 51 includes the method according to Example 49 or 50, further including: receiving a model acquisition request from a fourth application, the model acquisition request including an application identifier of the fourth application and a model identifier of the AI / ML model to be acquired; checking whether the fourth application is authorized to acquire the AI / ML model to be acquired; verifying the model acquisition request; when it is determined that the fourth application is authorized to acquire the AI / ML model to be acquired and the verification of the model acquisition request is successful, sending a model acquisition response to the fourth application, the model acquisition response including model file information of the AI / ML model to be acquired; and acquiring the AI / ML model to be acquired from the model repository to provide it to the fourth application.

[0343] Example 52 includes the method according to any one of Examples 49 to 51, further including: receiving a request for obtaining model weights from the third application, the request for obtaining model weights including an application identifier of the third application, a model identifier of a stored AI / ML model, and a weight version identifier of the weights of the stored AI / ML model; checking whether the third application is authorized to obtain the weights of the identified version of the stored AI / ML model; verifying the request for obtaining model weights; when it is determined that the third application is authorized to obtain the weights of the identified version of the stored AI / ML model and the verification of the request for obtaining model weights is successful, sending a response for obtaining model weights to the third application, the response for obtaining model weights including model weight file information of the weights of the identified version of the stored AI / ML model; and obtaining the weights of the identified version of the stored AI / ML model from the model repository to provide to the third application.

[0344] Example 53 includes a method according to any one of Examples 49 to 52, wherein storing the AI / ML model is based on an HTTP PUT method.

[0345] Example 54 includes a method according to any one of Examples 49 to 53, wherein storing the weights of the stored AI / ML model is based on an HTTP POST method.

[0346] Example 55 includes a method according to any one of Examples 49 to 54, wherein obtaining the AI / ML model to be obtained is based on an HTTP GET method.

[0347] Example 56 includes a method according to any one of Examples 49 to 55, wherein obtaining the weight of the identified version of the stored AI / ML model is based on an HTTP GET method.

[0348] Example 57 includes a computer-readable medium storing a computer program, wherein the computer program, when executed by a processor, implements the method according to any one of Examples 29 to 56.

[0349] Example 58 includes an apparatus comprising means for implementing the method according to any one of Examples 29 to 56.

[0350] Although certain embodiments have been illustrated and described herein for purposes of description, the embodiments illustrated and described may be replaced with a variety of alternative and / or equivalent embodiments or implementations that achieve the same purpose without departing from the scope of the present disclosure. This application is intended to cover any adaptation or variation of the embodiments discussed herein. Therefore, the embodiments described herein are expressly limited only by the appended claims and their equivalents.< / apiversion> < / apiversion> < / apiversion> < / apiversion> < / apiversion> < / apiversion> < / apiversion> < / apiversion> < / apiversion>

Claims

1. A network device comprising a processor circuit and a memory including computer program code, wherein: The processor circuit is coupled to the memory and configured to execute the computer program code to cause the network device to perform the following operations: Decoding a registration model request from a first application, the registration model request including an application identifier of the first application and model registration information of an artificial intelligence / machine learning AI / ML model; Checking whether the first application is authorized to register the AI / ML model; Verifying the registration model request; When it is determined that the first application is authorized to register the AI / ML model and the verification of the registration model request is successful, registering the AI / ML model; and A register model response is encoded for transmission to the first application, the register model response including a registration identifier of the AI / ML model.

2. The network device according to claim 1, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding an update model registration request from the first application, the update model registration request including an application identifier of the first application, a registration identifier of the AI / ML model, and updated model registration information of the AI / ML model; checking whether the first application is authorized to update the registration of the AI / ML model; Verifying the update model registration request; When it is determined that the first application is authorized to update the registration of the AI / ML model and the verification of the update model registration request is successful, updating the registration of the AI / ML model; and An updated model registration response is encoded for transmission to the first application.

3. The network device according to claim 1, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding a model deregistration request from the first application, the model deregistration request including an application identifier of the first application and a registration identifier of the AI / ML model; Checking whether the first application is authorized to deregister the AI / ML model; Verifying the logout model request; When it is determined that the first application is authorized to deregister the AI / ML model and verification of the deregister model request is successful, deregistering the AI / ML model; and A logout model response is encoded for transmission to the first application.

4. The network device according to claim 1, wherein: The processor circuit is configured to execute the computer program code to cause the network device to register the AI / ML model based on an HTTP POST method.

5. The network device according to claim 2, wherein: The processor circuit is configured to execute the computer program code to cause the network device to update the registration of the AI / ML model based on the HTTP PUT method or to partially update the registration of the AI / ML model based on the HTTP PATCH method.

6. The network device according to claim 3, wherein: The processor circuit is configured to execute the computer program code to cause the network device to deregister the AI / ML model based on an HTTP DELETE method.

7. The network device according to claim 1, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding a discover model request from a second application, the discover model request comprising an application identifier of the second application and a filter for discovering one or more registered AI / ML models; Checking whether the second application is authorized to discover the registered AI / ML model; Validating the discovery model request; and When it is determined that the second application is authorized to discover the registered AI / ML models and the verification of the discovery model request is successful, a discovery model response is encoded for transmission to the second application, the discovery model response including a list of model identifiers of the registered AI / ML models that meet the filtering conditions and corresponding model metadata.

8. The network device according to claim 1, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding a query model registration request from a second application, the query model registration request including an application identifier of the second application and a model identifier of the queried AI / ML model; Checking whether the second application is authorized to obtain metadata of the queried AI / ML model; Verifying the query model registration request; and When it is determined that the second application is authorized to obtain the metadata of the queried AI / ML model and the verification of the query model registration request is successful, a query model registration response is encoded for transmission to the second application, and the query model registration response includes the metadata of the queried AI / ML model.

9. The network device according to claim 1, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding a subscription model registration change request from a second application, the subscription model registration change request including an application identifier of the second application and a list of model registration change events; checking whether the second application is authorized to subscribe to model registration changes of one or more identified AI / ML models; Verifying the subscription model registration change request; When it is determined that the second application is authorized to subscribe to model registration changes of the one or more identified AI / ML models and verification of the subscription model registration changes request is successful, creating a subscription to model registration changes of the one or more identified AI / ML models for the second application; and A subscription model registration change response is encoded for transmission to the second application, the subscription model registration change response including a subscription identifier of the second application.

10. The network device according to claim 9, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding an update model registration change subscription request from the second application, the update model registration change subscription request including an application identifier of the second application, the subscription identifier, and a list of updated model registration change events; checking whether the second application is authorized to update the model registration change subscription; Verifying the update model registration change subscription request; When it is determined that the second application is authorized to update the model registration change subscription and verification of the update model registration change subscription request is successful, updating the subscription of the second application; and An updated model registration change subscription response is encoded for transmission to the second application, wherein the updated model registration change subscription response includes the updated model registration change event list.

11. The network device according to claim 9, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding a cancel subscription model registration change request from the second application, the cancel subscription model registration change request including an application identifier of the second application and the subscription identifier; checking whether the second application is authorized to cancel the subscription to model registration changes; Verifying the unsubscribe model registration change request; When it is determined that the second application is authorized to unsubscribe from the subscription to model registration changes and verification of the unsubscribe request to model registration changes succeeds, removing the subscription of the second application; and The unsubscribe model registration change response is encoded for transmission to the second application.

12. The network device according to claim 9, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: A notification about a model registry change is encoded for transmission to the second application, the notification including the subscription identifier and details of the model registry change to which the second application subscribes.

13. The network device according to claim 7, wherein: The processor circuit is configured to execute the computer program code to cause the network device to discover the registered AI / ML model based on an HTTP GET method.

14. The network device according to claim 8, wherein: The processor circuit is configured to execute the computer program code to cause the network device to obtain metadata of the queried AI / ML model based on an HTTP GET method.

15. The network device according to claim 9, wherein: The processor circuit is configured to execute the computer program code to cause the network device to create the subscription to model registry changes based on an HTTP POST method.

16. The network device according to claim 10, wherein: The processor circuit is configured to execute the computer program code to cause the network device to update the subscription to model registry changes based on an HTTP PUT method or to partially update the subscription to model registry changes based on an HTTP PATCH method.

17. The network device according to claim 11, wherein: The processor circuit is configured to execute the computer program code to cause the network device to remove the subscription to model registry changes based on an HTTP DELETE method.

18. The network device according to claim 12, wherein: The processor circuit is configured to execute the computer program code to cause the network device to notify the second application about the subscribed model registration change based on an HTTP POST method.

19. A network device comprising a processor circuit and a memory including computer program code, wherein: The processor circuit is coupled to the memory and configured to execute the computer program code to cause the network device to perform the following operations: Decoding a storage model request from a third application, the storage model request including an application identifier of the third application, a model identifier of an artificial intelligence / machine learning AI / ML model, and model file information of the AI / ML model; Checking whether the third application is authorized to store the AI / ML model; Verifying the storage model request; When it is determined that the third application is authorized to store the AI / ML model and the verification of the store model request is successful, encoding a store model response for transmission to the third application; and The model file information of the AI / ML model is obtained from the third application to store the AI / ML model in a model repository.

20. The network device according to claim 19, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: Decoding a storage model weight request from a fourth application, the storage model weight request including an application identifier of the fourth application, a model identifier of a stored AI / ML model, and model weight file information of the stored AI / ML model; checking whether the fourth application is authorized to store the weights of the stored AI / ML model; Verifying the storage model weight request; When it is determined that the fourth application is authorized to store the AI / ML model and the validation of the store model weight request is successful, encoding a store model weight response for transmission to the fourth application, the store model weight response including a weight version identifier of the stored weights of the stored AI / ML model; and The model weight file information of the stored AI / ML model is obtained from the fourth application to store the weight of the stored AI / ML model in a model repository.

21. The network device according to claim 19, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding a request for obtaining a model from a fourth application, the request for obtaining a model including an application identifier of the fourth application and a model identifier of an AI / ML model to be obtained; Checking whether the fourth application is authorized to obtain the AI / ML model to be obtained; Verifying the request to obtain the model; When it is determined that the fourth application is authorized to obtain the AI / ML model to be obtained and the verification of the obtain model request is successful, encoding an obtain model response for transmission to the fourth application, the obtain model response including model file information of the AI / ML model to be obtained; and The AI / ML model to be obtained is obtained from the model repository to provide to the fourth application.

22. The network device according to claim 19, wherein: The processor circuit is configured to execute the computer program code so that the network device further performs the following operations: decoding a request for obtaining model weights from the third application, the request for obtaining model weights comprising an application identifier of the third application, a model identifier of a stored AI / ML model, and a weight version identifier of the weights of the stored AI / ML model; checking whether the third application is authorized to obtain the weight of the identified version of the stored AI / ML model; Verifying the request to obtain the model weights; When it is determined that the third application is authorized to obtain the weights of the identified version of the stored AI / ML model and the verification of the obtain model weights request is successful, encoding an obtain model weights response for transmission to the third application, the obtain model weights response including model weight file information of the weights of the identified version of the stored AI / ML model; and The weights of the identified version of the stored AI / ML model are obtained from the model repository to be provided to the third application.

23. The network device according to claim 19, wherein: The processor circuit is configured to execute the computer program code to cause the network device to store the AI / ML model based on an HTTP PUT method.

24. The network device according to claim 20, wherein: The processor circuit is configured to execute the computer program code to cause the network device to store the weights of the stored AI / ML model based on an HTTP POST method.

25. The network device according to claim 21, wherein: The processor circuit is configured to execute the computer program code to enable the network device to obtain the AI / ML model to be obtained based on an HTTP GET method.

26. The network device according to claim 22, wherein: The processor circuit is configured to execute the computer program code to cause the network device to obtain the weights of the identified version of the stored AI / ML model based on an HTTP GET method.