Methods, apparatuses, and computer programs

Abstract

Example embodiments herein relate to methods, apparatuses, and computer programs. A user equipment is provided comprising means for providing, to an artificial intelligence as a service (AIaaS) function, an artificial intelligence (AI) service discovery request message for discovering the AIaaS function; receiving, from the AIaaS function, an AI service discovery response message based on the AI service discovery request; providing, to the AIaaS function, an AIaaS request message for a service for training a machine learning model in an environment in which the training of the machine learning model is private to a provider of the service, the request message comprising: selection criteria for indicating that the machine learning model is to be deployed to one or more of a plurality of user equipments; and receiving, from the AIaaS function, an AIaaS completion message indicating whether the training of the machine learning model has been performed.

Description

Technical Field

[0001] Various embodiments of this disclosure relate to methods, apparatuses, systems, and computer programs, particularly but not limited to artificial intelligence / machine learning (AI / ML) model training environments as a service. Background Technology

[0002] A communication network can be viewed as a facility that enables communication between two or more communication devices, or provides communication devices with access to a data network. Mobile or wireless communication networks are an example of communication networks. Communication devices can be provided with services through application servers.

[0003] Such communication networks operate according to standards, such as those issued by 3GPP (3rd Generation Partnership Project) or ETSI (European Telecommunications Standards Institute). Examples of standards issued by 3GPP include so-called 3GPP standards for various generations of cellular technology, such as the 3GPP standards for 4G technology and the 3GPP standards for 5G technology. Summary of the Invention

[0004] Some examples of this disclosure will be described with reference to certain aspects. These aspects are not intended to identify key or essential features of the various examples of this disclosure, nor are they intended to limit the scope of this disclosure. Other features, aspects, and elements will be apparent to those skilled in the art. For example, it should be understood that additional aspects may be provided by any two or more of the following aspects.

[0005] According to a first aspect, an apparatus is provided, comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: provide a request message to a network function for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and receive a response message from the network function indicating whether the training of the machine learning model has been performed.

[0006] In some examples, the training of the machine learning model in a private environment of the service provider includes at least one of the following: training of the machine learning model on a shared server implemented by the service provider; or training of the machine learning model on a dedicated server implemented by the service provider.

[0007] In some examples, the device is a user equipment, including a user equipment controller.

[0008] In some examples, the request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available at the network function; a request for the type of environment for training the machine learning model available at the network function; information indicating which one or more user devices among a plurality of user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; configuration requirements for training the machine learning model; or information indicating which one or more user devices among the plurality of user devices the machine learning model will be deployed to.

[0009] In some examples, the response message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model has been validated; the availability of one or more inference instances of the machine learning model; the device's ability to deploy the trained machine learning model to one or more other user devices and / or application functions of an application service provider serving the device; or the device's ability to deploy the availability of one or more inference instances to the one or more other user devices and / or the application function.

[0010] In some examples, when executed by at least one processor, the instruction causes the device to: provide a trained machine learning model to one or more other user devices and / or to an application function of an application service provider serving the device; or provide the availability of one or more inference instances to the one or more other user devices and / or to the application function.

[0011] In some examples, when executed by at least one processor, the instruction causes the device to: receive an acknowledgment message from a network function confirming the request message, the acknowledgment message including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0012] In some examples, when executed by at least one processor, the instruction causes the device to: provide a discovery request message to a network function for discovering the network function, and include at least one of the following: a first identifier value indicating an application service provider serving the device; or a second identifier value indicating the type of the device.

[0013] In some examples, when executed by at least one processor, the instruction causes the device to: receive from the network function at least one of the following: one or more artificial intelligence service types available at the network function; or one or more environment types available at the network function for training the machine learning model.

[0014] In some examples, the device includes application functionality from an application service provider.

[0015] In some examples, the request message includes configuration requirements for training the machine learning model, which include information indicating at least one of the following: one or more uses of the machine learning model; the accuracy of the machine learning model; training data for the machine learning model; information indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; training data for training the machine learning model; validation data for validating the trained machine learning model; one or more training strategies for training the machine learning model; or one or more energy consumption strategies indicating the energy consumption for training the machine learning model.

[0016] In some examples, when the training of the machine learning model has been performed, the response message includes: an indication that the requested service has been approved and / or configuration information of the requested service; and when the training of the machine learning model has not yet been performed, the response message includes: an indication that the requested service has been rejected and / or the reason why the requested service has been rejected.

[0017] According to a second aspect, a method is provided, comprising: providing a network function with a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and receiving a response message from the network function indicating whether the training of the machine learning model has been performed.

[0018] In some examples, the training of the machine learning model in a private environment of the service provider includes at least one of the following: training of the machine learning model on a shared server implemented by the service provider; or training of the machine learning model on a dedicated server implemented by the service provider.

[0019] In some examples, the device is a user equipment, including a user equipment controller.

[0020] In some examples, the request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available at the network function; a request for the type of environment for training the machine learning model available at the network function; information indicating which one or more user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; configuration requirements for training the machine learning model; or information indicating which one or more user devices the machine learning model will be deployed to.

[0021] In some examples, the response message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model and its validation; the availability of one or more inference instances of the machine learning model; the device's ability to deploy the trained machine learning model to one or more other user devices and / or application functions of an application service provider serving the device; or the device's ability to deploy the availability of one or more inference instances to the one or more other user devices and / or the application function.

[0022] In some examples, the method may include: providing a trained machine learning model to one or more other user devices and / or to an application function of an application service provider serving the device; or providing the availability of one or more inference instances to the one or more other user devices and / or to the application function.

[0023] In some examples, the method may include: receiving an acknowledgment message from a network function confirming the request message, the acknowledgment message including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0024] In some examples, the method may include: providing a discovery request message to a network function for discovering the network function, and including at least one of the following: a first identification value indicating an application service provider serving the device; or a second identification value indicating the type of the device.

[0025] In some examples, the method may include receiving from the network function at least one of the following: one or more artificial intelligence service types available at the network function; or one or more environment types available at the network function for training the machine learning model.

[0026] In some examples, the device includes application functionality from an application service provider.

[0027] In some examples, the request message includes configuration requirements for training the machine learning model, which include information indicating at least one of the following: one or more uses of the machine learning model; the accuracy of the machine learning model; training data for the machine learning model; information indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; training data for training the machine learning model; validation data for validating the trained machine learning model; one or more training strategies for training the machine learning model; or one or more energy consumption strategies indicating the energy consumption for training the machine learning model.

[0028] In some examples, when the training of the machine learning model has been performed, the response message includes: an indication that the requested service has been approved and / or configuration information of the requested service; and when the training of the machine learning model has not yet been performed, the response message includes: an indication that the requested service has been rejected and / or the reason why the requested service has been rejected.

[0029] According to a third aspect, an apparatus is provided, the apparatus comprising components for: providing a trained machine learning model to one or more other user devices and / or to an application function of an application service provider serving the apparatus; or receiving a response message from a network function indicating whether training of the machine learning model has been performed.

[0030] In some embodiments, the apparatus includes components for performing at least one or more methods of the second aspect or examples thereof.

[0031] According to a fourth aspect, a computer program is provided, the computer program including instructions that, when executed by a computer, cause a device to: provide a trained machine learning model to one or more other user devices and / or to an application function of an application service provider serving the device; or receive a response message from a network function indicating whether training of the machine learning model has been performed.

[0032] In some embodiments, the computer program includes instructions that, when executed by a computer, cause the apparatus to perform at least one or more methods of the second aspect or examples thereof.

[0033] According to a fifth aspect, an apparatus is provided, comprising: at least one processor; and at least one memory storing instructions for an application, which, when executed by the at least one processor, cause the apparatus to at least: receive from a second apparatus a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and provide a response message to the second apparatus indicating whether the training of the machine learning model has been performed.

[0034] In some examples, the training of the machine learning model in an environment where the service provider is private includes at least one of the following: training of a machine learning model on a shared server implemented by the service provider; or training of a machine learning model on a dedicated server implemented by the service provider.

[0035] In some examples, the request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available on the device; a request for the type of environment for training the machine learning model available on the device; information indicating which of a plurality of user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; configuration requirements for training the machine learning model; or information indicating which of the plurality of user devices the machine learning model will be deployed to.

[0036] In some examples, the response message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model has been validated; the availability of one or more inference instances of the machine learning model; the ability of the second device to deploy the trained machine learning model to one or more other user devices and / or application functions of an application service provider serving the second device; or the ability of the second device to deploy the availability of one or more inference instances to the one or more other user devices and / or the application function.

[0037] In some examples, when executed by at least one processor, the instruction causes the device to: provide a confirmation message to a second device acknowledging the request message, the confirmation message including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0038] In some examples, when executed by at least one processor, the instruction causes the device to: receive a discovery request message from a second device for discovering network functions and include at least one of the following: a first identifier value indicating an application service provider serving the device; or a second identifier value indicating the type of the device.

[0039] In some examples, when executed by at least one processor, the instruction causes the device to: provide the second device with at least one of the following: one or more types of artificial intelligence services available at the device; or one or more types of environments available at the device for training machine learning models.

[0040] In some examples, when executed by at least one processor, the instruction causes the device to perform: based on a request message received from a second device, to provide a network controller with resources for training the machine learning model.

[0041] In some examples, when executed by at least one processor, the instruction causes the device to: receive a complete message from the network controller indicating at least one of the following: whether training of the machine learning model has been performed; the trained machine learning model; the availability of one or more inference instances of the machine learning model; or the location where the trained machine learning model is stored.

[0042] In some examples, when executed by at least one processor, the instruction causes the device to perform: verifying a machine learning model that has been trained by the network controller.

[0043] In some examples, the request message includes configuration requirements for training the machine learning model, which include information indicating at least one of the following: one or more uses of the machine learning model; the accuracy of the machine learning model; training data for the machine learning model; information indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; training data for training the machine learning model; validation data for validating the trained machine learning model; one or more training strategies for training the machine learning model; or one or more energy consumption strategies indicating the energy consumption for training the machine learning model.

[0044] In some examples, when executed by at least one processor, the instruction causes the device to: save and / or update the configuration requirements for training the machine learning model.

[0045] In some examples, when executed by at least one processor, the instruction causes the device to: provide a request message to a network controller requesting the network controller to update the configuration requirements for training the machine learning model; when training of the machine learning model has been executed, receive a complete message from the network controller indicating that the requested service has been approved and / or configuration information of the requested service; and when training of the machine learning model has not been executed, receive from the network controller an indication that the requested service has been rejected and / or a reason why the requested service has been rejected.

[0046] In some examples, when the training of the machine learning model has been performed, the response message includes: an indication that the requested service has been approved and / or configuration information of the requested service; and when the training of the machine learning model has not yet been performed, the response message includes an indication that the requested service has been rejected and / or the reason why the requested service has been rejected.

[0047] In some embodiments, when executed by at least one processor, the instruction causes the device to perform: authorize the second device.

[0048] According to a sixth aspect, a method is provided, comprising: receiving from a second device a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and providing the second device a response message indicating whether the training of the machine learning model has been performed.

[0049] In some examples, training the machine learning model in an environment where the service provider is private includes at least one of the following: training a machine learning model on a shared server implemented by the service provider; or training a machine learning model on a dedicated server implemented by the service provider.

[0050] In some examples, the request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available at the network function; a request for the type of environment for training the machine learning model available at the network function; information indicating which one or more user devices among a plurality of user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; configuration requirements for training the machine learning model; or information indicating which one or more user devices among the plurality of user devices the machine learning model will be deployed to.

[0051] In some examples, the response message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model has been validated; the availability of one or more inference instances of the machine learning model; the ability of the second device to deploy the trained machine learning model to one or more other user devices and / or application functions of an application service provider serving the second device; or the ability of the second device to deploy the availability of one or more inference instances to the one or more other user devices and / or the application function.

[0052] In some examples, the method may include: providing a confirmation message to a second device to acknowledge the request message, the confirmation message including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0053] In some examples, the method may include: receiving a discovery request message from a second device for discovering the device, and including at least one of the following: a first identifier value indicating an application service provider serving the second device; or a second identifier value indicating the type of the second device.

[0054] In some examples, the method may include providing the second device with at least one of the following: one or more types of artificial intelligence services available at the device; or one or more types of environments available at the device for training machine learning models.

[0055] In some examples, the method may include: providing a request message to the network controller with resources for training the machine learning model based on a request message received from the second device.

[0056] In some examples, the method may include receiving a complete message from a network controller indicating at least one of the following: whether training of the machine learning model has been performed; the trained machine learning model; the availability of one or more inference instances of the machine learning model; or the location where the trained machine learning model is stored.

[0057] In some examples, the method may include: validating a machine learning model that has been trained by the network controller.

[0058] In some examples, the request message includes configuration requirements for training the machine learning model, which include information indicating at least one of the following: one or more uses of the machine learning model; the accuracy of the machine learning model; training data for the machine learning model; information indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; training data for training the machine learning model; validation data for validating the trained machine learning model; one or more training strategies for training the machine learning model; or one or more energy consumption strategies indicating the energy consumption for training the machine learning model.

[0059] In some examples, the method may include saving and / or updating the configuration requirements used for training the machine learning model.

[0060] In some examples, the method may include: providing a request message to a network controller requesting the network controller to update configuration requirements for training the machine learning model; receiving a complete message from the network controller when training of the machine learning model has been performed, the complete message indicating that the requested service has been approved and / or configuration information of the requested service; and receiving an indication from the network controller that the requested service has been rejected and / or a reason why the requested service has been rejected when training of the machine learning model has not yet been performed.

[0061] In some examples, when the training of the machine learning model has been performed, the response message includes: an indication that the requested service has been approved and / or configuration information of the requested service; and when the training of the machine learning model has not yet been performed, the response message includes an indication that the requested service has been rejected and / or the reason why the requested service has been rejected.

[0062] In some embodiments, the method may include: authorizing the second device.

[0063] According to a seventh aspect, a user equipment is provided, comprising: receiving from a second device a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and providing a response message to the second device indicating whether the training of the machine learning model has been performed.

[0064] In some examples, the apparatus includes at least one or more components for performing the method of the sixth aspect.

[0065] According to an eighth aspect, a computer program is provided, the computer program including instructions that, when executed by a computer, cause a device to: receive from a second device a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and provide the second device a response message indicating whether the training of the machine learning model has been performed.

[0066] In some embodiments, the computer program includes instructions that, when executed by a computer, cause the apparatus to perform at least one or more of the methods of the sixth aspect.

[0067] According to a ninth aspect, an apparatus is provided, the apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: receive from a network function a request message for a service for training a machine learning model in an environment in which the training of the machine learning model is private to the service provider; and provide a complete message to the network function indicating whether the training of the machine learning model has been performed.

[0068] In some examples, the training of the machine learning model in a private environment of the service provider includes at least one of the following: training of the machine learning model on a shared server implemented by the service provider; or training of the machine learning model on a dedicated server implemented by the service provider.

[0069] In some examples, the full message also indicates at least one of the following: a trained machine learning model; or the availability of one or more inference instances of the machine learning model.

[0070] In some examples, the request message includes a request for resources used to train a machine learning model.

[0071] In some examples, the instructions, when executed by at least one processor, cause the device to allocate resources for training a machine learning model to one or more network servers.

[0072] In some examples, when executed by at least one processor, the instruction causes the device to perform the following: train a machine learning model.

[0073] In some examples, the request message received from the network function includes: a request to update configuration requirements for training a machine learning model, wherein the configuration requirements include information indicating at least one of the following: one or more uses of the machine learning model; the accuracy of the machine learning model; training data for the machine learning model; information indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; training data for training the machine learning model; validation data for validating the trained machine learning model; one or more training strategies for training the machine learning model; or one or more energy consumption strategies indicating the energy consumption for training the machine learning model.

[0074] In some embodiments, when executed by at least one processor, the instructions cause the device to: provide a complete message to the network function when training of the machine learning model has been performed, the complete message indicating that the requested service has been approved and / or configuration information of the requested service; and provide an indication to the network function that the requested service has been rejected and / or the reason why the requested service has been rejected when training of the machine learning model has not yet been performed.

[0075] According to a tenth aspect, a method is provided, comprising: receiving from a network function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and providing the network function with a complete message indicating whether the training of the machine learning model has been performed.

[0076] In some examples, the training of the machine learning model in a private environment of the service provider includes at least one of the following: training of the machine learning model on a shared server implemented by the service provider; or training of the machine learning model on a dedicated server implemented by the service provider.

[0077] In some examples, the full message also indicates at least one of the following: a trained machine learning model; or the availability of one or more inference instances of the machine learning model.

[0078] In some examples, the request message includes a request for resources used to train a machine learning model.

[0079] In some examples, the method may include allocating resources for training a machine learning model to one or more network servers.

[0080] In some examples, the method may include: training a machine learning model.

[0081] In some examples, the request message received from the network function includes: a request to update configuration requirements for training a machine learning model, wherein the configuration requirements include information indicating at least one of the following: one or more uses of the machine learning model; the accuracy of the machine learning model; training data for the machine learning model; information indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; training data for training the machine learning model; validation data for validating the trained machine learning model; one or more training strategies for training the machine learning model; or one or more energy consumption strategies indicating the energy consumption for training the machine learning model.

[0082] In some embodiments, the method may include: when training of a machine learning model has been performed, providing a complete message to a network function indicating that a requested service has been approved and / or configuration information of the requested service; and when training of a machine learning model has not yet been performed, providing the network function with an indication that a requested service has been rejected and / or a reason why the requested service has been rejected.

[0083] According to an eleventh aspect, an apparatus is provided, comprising components for: receiving from a network function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and providing a complete message to the network function indicating whether the training of the machine learning model has been performed.

[0084] In some embodiments, the apparatus includes components for performing at least one or more of the methods in the tenth aspect.

[0085] According to a twelfth aspect, a computer program is provided, the computer program including instructions that, when executed by a computer, cause a device to perform: receiving from a network function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; and providing a complete message to the network function indicating whether the training of the machine learning model has been performed.

[0086] In some embodiments, the computer program includes instructions that, when executed by a computer, cause the apparatus to perform at least one or more of the methods in the tenth aspect.

[0087] According to a thirteenth aspect, a user device is provided, comprising: at least one processor; and at least one memory storing instructions for an application, which, when executed by the at least one processor, cause the device to at least: provide an AI as a Service (AIaaS) function with an AI service discovery request message for discovering the AIaaS function; receive an AI service discovery response message from the AIaaS function based on the AI ​​service discovery request; provide the AIaaS function with an AIaaS request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which of a plurality of user devices the machine learning model will be deployed to; and receive an AIaaS completion message from the AIaaS function indicating whether the training of the machine learning model has been performed.

[0088] In some examples, training a machine learning model in an environment where the service provider is private includes at least one of the following: training a machine learning model on a shared server implemented by the service provider; or training a machine learning model on a dedicated server implemented by the service provider.

[0089] In some examples, the user equipment includes a user equipment controller.

[0090] In some examples, the AIaaS request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available at the AIaaS function; a request for the type of environment available at the AIaaS function for training the machine learning model; information indicating which one or more user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; or configuration requirements for training the machine learning model.

[0091] In some examples, the AIaaS completion message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model has been validated; the availability of one or more inference instances of the machine learning model; the user device's ability to deploy the trained machine learning model to one or more other user devices and / or application functions of application service providers serving the user device; or the user device's ability to deploy the availability of one or more inference instances to one or more other user devices and / or application functions.

[0092] In some examples, when executed by at least one processor, the instruction causes the user device to: provide a trained machine learning model to one or more other user devices and / or to application functions of an application service provider serving the user device; or provide availability of one or more inference instances to one or more other user devices and / or to application functions.

[0093] In some examples, when executed by at least one processor, the instruction causes the user device to: receive an AIaaS response message acknowledging the AIaaS request message from the AIaaS function, the AIaaS response message including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0094] In some embodiments, the AI ​​service discovery request message includes at least one of the following: a first identifier value indicating the application service provider serving the user device; or a second identifier value indicating the type of the user device.

[0095] In some embodiments, the AI ​​service discovery response message includes at least one of the following: one or more AI service types available at the AIaaS feature; or one or more environment types available at the AIaaS feature for training machine learning models.

[0096] According to the fourteenth aspect, a method is provided, comprising: providing an Artificial Intelligence as a Service (AIaaS) function with an Artificial Intelligence (AIaaS) service discovery request message for discovering the AIaaS function; receiving an AI service discovery response message from the AIaaS function based on the AI ​​service discovery request; providing the AIaaS function with an AIaaS request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and receiving an AIaaS completion message from the AIaaS function indicating whether the training of the machine learning model has been performed.

[0097] In some examples, training a machine learning model in an environment where the service provider is private includes at least one of the following: training a machine learning model on a shared server implemented by the service provider; or training a machine learning model on a dedicated server implemented by the service provider.

[0098] In some examples, the user equipment includes a user equipment controller.

[0099] In some examples, the AIaaS request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available at the AIaaS function; a request for the type of environment available at the AIaaS function for training the machine learning model; information indicating which one or more user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; or configuration requirements for training the machine learning model.

[0100] In some examples, the AIaaS completion message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model has been validated; the availability of one or more inference instances of the machine learning model; the user device's ability to deploy the trained machine learning model to one or more other user devices and / or application functions of application service providers serving the user device; or the user device's ability to deploy the availability of one or more inference instances to one or more other user devices and / or application functions.

[0101] In some examples, the method may include: providing a trained machine learning model to one or more other user devices and / or to application functions of an application service provider serving the user device; or providing the availability of one or more inference instances to one or more other user devices and / or to application functions.

[0102] In some examples, the method may include: receiving an AIaaS response message from an AIaaS function that confirms an AIaaS request message, the AIaaS response message including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0103] In some embodiments, the AI ​​service discovery request message includes at least one of the following: a first identifier value indicating the application service provider serving the user device; or a second identifier value indicating the type of the user device.

[0104] In some embodiments, the AI ​​service discovery response message includes at least one of the following: one or more AI service types available at the AIaaS feature; or one or more environment types available at the AIaaS feature for training machine learning models.

[0105] According to a fifteenth aspect, a user equipment is provided, comprising: providing an Artificial Intelligence as a Service (AIaaS) function with an Artificial Intelligence as a Service (AIaaS) function for discovering the AIaaS function; receiving an AI service discovery response message from the AIaaS function based on the AI ​​service discovery request; providing the AIaaS function with an AIaaS request message for a service for a machine learning model trained in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and receiving an AIaaS completion message from the AIaaS function, the AIaaS completion message indicating whether the training of the machine learning model has been performed.

[0106] In some embodiments, the user equipment includes components for performing at least one or more methods of the fourteenth aspect or examples thereof.

[0107] According to a sixteenth aspect, a computer program is provided, the computer program including instructions that, when executed by a computer, cause a user device to: provide an Artificial Intelligence as a Service (AIaaS) function with an Artificial Intelligence (AIaaS) service discovery request message for discovering the AIaaS function; receive an AI service discovery response message from the AIaaS function based on the AI ​​service discovery request; provide the AIaaS function with an AIaaS request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which of a plurality of user devices the machine learning model will be deployed to; and receive an AIaaS completion message from the AIaaS function indicating whether the training of the machine learning model has been performed.

[0108] In some embodiments, the computer program includes instructions that, when executed by a computer, cause a user device to perform at least one or more of the methods of the fourteenth aspect or examples thereof.

[0109] According to a seventeenth aspect, an apparatus is provided, comprising: at least one processor; and at least one memory storing instructions for an application, which, when executed by the at least one processor, cause the apparatus to at least: receive from a user equipment an artificial intelligence (AI) service discovery request for discovering AIaaS functionality; provide the user equipment with an AI service discovery response message based on the AI ​​service discovery request; receive from the user equipment an AIaaS request message for a service for training a machine learning model in an environment, the request message including: selection criteria for indicating which of a plurality of user devices the machine learning model will be deployed to; and provide the user equipment with an AIaaS completion message indicating whether training of the machine learning model has been performed.

[0110] In some examples, training a machine learning model in an environment where the service provider is private includes at least one of the following: training a machine learning model on a shared server implemented by the service provider; or training a machine learning model on a dedicated server implemented by the service provider.

[0111] In some examples, the AIaaS request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available at the AIaaS function; a request for the type of environment available at the AIaaS function for training the machine learning model; information indicating which one or more user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; or configuration requirements for training the machine learning model.

[0112] In some examples, the AIaaS completion message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model has been validated; the availability of one or more inference instances of the machine learning model; the user device's ability to deploy the trained machine learning model to one or more other user devices and / or application functions of application service providers serving the user device; or the user device's ability to deploy the availability of one or more inference instances to one or more other user devices and / or application functions.

[0113] In some examples, when executed by at least one processor, the instruction causes the device to perform: providing an AIaaS response message acknowledging the AIaaS request message to a user device, the AIaaS response message further including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0114] In some examples, the AI ​​service discovery request message includes at least one of the following: a first identifier value indicating the application service provider serving the user device; or a second identifier value indicating the type of the user device.

[0115] In some examples, when executed by at least one processor, the instruction causes the device to: provide the user device with at least one of the following: one or more artificial intelligence service types available at the device; or one or more environment types available at the device for training machine learning models.

[0116] In some examples, when executed by at least one processor, the instruction causes the device to: provide an AI training request fulfillment demand message to the AIaaS controller, requesting resources for training a machine learning model based on an AIaaS request message received from the user device.

[0117] In some examples, when executed by at least one processor, the instruction causes the device to: receive a training completion message from the AIaaS controller, the training completion message indicating at least one of the following: whether the training of the machine learning model has been performed; the trained machine learning model; the availability of one or more inference instances of the machine learning model; or the location where the trained machine learning model is stored.

[0118] In some examples, when executed by at least one processor, the instruction causes the device to perform: verifying a machine learning model that has been trained by the AIaaS controller.

[0119] According to the eighteenth aspect, a method is provided, comprising: receiving from a user equipment an artificial intelligence (AI) service discovery request for discovering AIaaS functionality; providing the user equipment with an AI service discovery response message based on the AI ​​service discovery request; receiving from the user equipment an AIaaS request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and providing the user equipment with an AIaaS completion message indicating whether the training of the machine learning model has been performed.

[0120] In some examples, training a machine learning model in an environment where the service provider is private includes at least one of the following: training a machine learning model on a shared server implemented by the service provider; or training a machine learning model on a dedicated server implemented by the service provider.

[0121] In some examples, the AIaaS request message may also include at least one of the following: a machine learning model for training; a request for the type of artificial intelligence service available at the AIaaS function; a request for the type of environment available at the AIaaS function for training the machine learning model; information indicating which one or more user devices the machine learning model is valid for; training data for training the machine learning model; validation data for validating the trained machine learning model; or configuration requirements for training the machine learning model.

[0122] In some examples, the AIaaS completion message also includes information indicating at least one of the following: a trained machine learning model; the trained machine learning model has been validated; the availability of one or more inference instances of the machine learning model; the user device's ability to deploy the trained machine learning model to one or more other user devices and / or application functions of application service providers serving the user device; or the user device's ability to deploy the availability of one or more inference instances to one or more other user devices and / or application functions.

[0123] In some examples, the method may include: providing a user device with an AIaaS response message that confirms the AIaaS request message, the AIaaS response message further including at least one of the following: information on the start time of training the machine learning model; or information on the completion time of training the machine learning model.

[0124] In some examples, the AI ​​service discovery request message includes at least one of the following: a first identifier value indicating the application service provider serving the user device; or a second identifier value indicating the type of the user device.

[0125] In some examples, the method may include providing the user device with at least one of the following: one or more types of artificial intelligence services available at the device; or one or more types of environments available at the device for training machine learning models.

[0126] In some examples, the method may include: providing an AI training request fulfillment demand message to the AIaaS controller to request resources for training a machine learning model based on an AIaaS request message received from the user device.

[0127] In some examples, the method may include receiving a training completion message from an AIaaS controller, the training completion message indicating at least one of the following: whether training of the machine learning model has been performed; the trained machine learning model; the availability of one or more inference instances of the machine learning model; or the location where the trained machine learning model is stored.

[0128] In some examples, the method may include: validating a machine learning model that has been trained by the AIaaS controller.

[0129] According to a nineteenth aspect, a user equipment is provided, comprising: receiving from the user equipment an artificial intelligence (AI) service discovery request for discovering AIaaS functions; providing the user equipment with an AI service discovery response message based on the AI ​​service discovery request; receiving from the user equipment an AIaaS request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and providing the user equipment with an AIaaS completion message indicating whether the training of the machine learning model has been performed.

[0130] In some embodiments, the apparatus includes components for performing at least one or more of the methods in the eighteenth aspect.

[0131] According to a twentieth aspect, a computer program is provided, the computer program including instructions that, when executed by a computer, cause the device to perform: receiving from a user equipment an artificial intelligence (AI) service discovery request for discovering AIaaS functionality; providing an AI service discovery response message to the user equipment based on the AI ​​service discovery request; receiving from the user equipment an AIaaS request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which of a plurality of user devices the machine learning model will be deployed to; and providing the user equipment an AIaaS completion message indicating whether the training of the machine learning model has been performed.

[0132] In some embodiments, the computer program includes instructions that, when executed by a computer, cause the apparatus to perform at least one or more of the methods of the eighteenth aspect.

[0133] According to a twenty-first aspect, an apparatus is provided, comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: receive from an Artificial Intelligence as a Service (AIaaS) function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and provide a training completion message to the AIaaS function, the training completion message indicating whether the training of the machine learning model has been performed.

[0134] In some embodiments, training a machine learning model in an environment where the service provider is private includes at least one of the following: training a machine learning model on a shared server implemented by the service provider; or training a machine learning model on a dedicated server implemented by the service provider.

[0135] In some examples, the training complete message also indicates at least one of the following: the trained machine learning model; or the availability of one or more inference instances of the machine learning model.

[0136] In some examples, the request message includes: an AI training request implementation requirement message for requesting resources for training a machine learning model.

[0137] In some examples, when executed by at least one processor, the instruction causes the device to allocate resources for training a machine learning model to one or more network servers.

[0138] In some examples, when executed by at least one processor, the instruction causes the device to perform the following: train a machine learning model.

[0139] According to the twenty-second aspect, a method is provided, comprising: receiving from an Artificial Intelligence as a Service (AIaaS) function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and providing the AIaaS function with a training completion message indicating whether the training of the machine learning model has been performed.

[0140] In some embodiments, training a machine learning model in an environment where the service provider is private includes at least one of the following: training a machine learning model on a shared server implemented by the service provider; or training a machine learning model on a dedicated server implemented by the service provider.

[0141] In some examples, the training complete message also indicates at least one of the following: the trained machine learning model; or the availability of one or more inference instances of the machine learning model.

[0142] In some examples, the request message includes: an AI training request implementation requirement message for requesting resources for training a machine learning model.

[0143] In some examples, the method may include allocating resources for training a machine learning model to one or more network servers.

[0144] In some examples, the method may include: performing training on a machine learning model.

[0145] According to the twenty-third aspect, an apparatus is provided, comprising components for: receiving from an Artificial Intelligence as a Service (AIaaS) function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and providing a training completion message to the AIaaS function, the training completion message indicating whether the training of the machine learning model has been performed.

[0146] In some embodiments, the apparatus includes components for performing at least one or more of the methods in the twenty-second aspect.

[0147] According to the twenty-fourth aspect, a computer program is provided, the computer program including instructions that, when executed by a computer, cause a device to perform: receiving from an Artificial Intelligence as a Service (AIaaS) function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to; and providing a training completion message to the AIaaS function, the training completion message indicating whether the training of the machine learning model has been performed.

[0148] In some embodiments, the computer program includes instructions that, when executed by a computer, cause the apparatus to perform at least one or more of the methods in the twenty-second aspect.

[0149] According to another aspect, a computer-readable medium is provided, including program instructions stored thereon for performing at least one of the methods described above.

[0150] According to one aspect, a non-volatile computer-readable medium is provided, including program instructions stored thereon for performing at least one of the methods described above.

[0151] According to one aspect, a non-volatile physical memory medium is provided, including program instructions stored thereon for performing at least one of the methods described above.

[0152] According to one aspect, a program (e.g., a computer program) is provided, comprising instructions that, when executed by a device, cause the device to perform at least one of the methods described above.

[0153] Many different aspects have been described above. As previously stated, it should be understood that additional aspects can be provided through any combination of two or more of the above aspects (or their corresponding parts). Other features, aspects, and elements will become apparent below. Attached Figure Description

[0154] Some examples will now be described by way of non-limiting and illustrative examples, with reference to one or more accompanying drawings, in which:

[0155] Figure 1 and Figure 2 Examples of communication networks in which the examples disclosed herein can be applied are shown;

[0156] Figure 3 A schematic representation of a communication system is shown;

[0157] Figure 4 Examples of communication networks to which the examples disclosed herein can be applied are shown;

[0158] Figure 5 An example signaling diagram is shown;

[0159] Figure 6 An example signaling diagram is shown;

[0160] Figure 7 Example methods are shown;

[0161] Figure 8 Example methods are shown;

[0162] Figure 9 Example methods are shown;

[0163] Figure 10 Example methods are shown;

[0164] Figure 11 Example methods are shown;

[0165] Figure 12 Example methods are shown; and

[0166] Figure 13 A block diagram representation of an apparatus based on some examples is shown. Detailed Implementation

[0167] Some examples of this disclosure can be implemented in communication networks, such as any of the following radio access technologies (RATs): Global Microwave Access Interoperability (WiMAX), Global System for Mobile Communications (GSM, 2G), GSM EDGE Radio Access Network (GERAN), General Packet Radio Service (GRPS), Universal Mobile Telecommunications System based on Basic Wideband Code Division Multiple Access (W-CDMA) (UMTS, 3G), High-Speed ​​Packet Access (HSPA), Long Term Evolution (LTE), Advanced LTE and Enhanced LTE (eLTE), 5G (also known as NR), or any future RAT, such as 6G. Furthermore, communication within the communication network can utilize any suitable wireless communication technology, including but not limited to: Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Frequency Division Duplex (FDD), Time Division Duplex (TDD), Multiple-Input Multiple-Output (MIMO), Orthogonal Frequency Division Multiplexing (OFDM), and / or Discrete Fourier Transform Extended Orthogonal Frequency Division Multiplexing (DFT-s-OFDM).

[0168] As used herein, the term "network device" or "network node" can refer to a node in a communication network through which terminal devices can access the network and / or control wireless communication and manage wireless resources within a cell. Depending on the technology used, a network device can be, for example, a Node B (or NB), an evolved Node B (or eNodeB or eNB), an NR NB (also known as a gNB), a Remote Radio Unit (RRU), a Radio Head (RH), a Remote Radio Head (RRH), a relay, an Integrated Access and Backhaul (IAB) node, a low-power node, a non-terrestrial network (NTN) or non-terrestrial network device (such as satellite network devices, low Earth orbit (LEO) satellites, and geostationary orbit (GEO) satellites), or an aircraft network device.

[0169] The term "terminal device" can refer to any terminal device capable of wireless communication. By way of non-limiting and illustrative examples, a terminal device can be referred to as a communication device, user equipment (UE), subscriber station (SS), or mobile station (MS). Terminal devices can include mobile phones, cellular phones, smartphones, Voice over IP (VoIP) phones, wireless local loop phones, tablets, wearable terminal devices, personal digital assistants (PDAs), portable computers, desktop computers, image capture terminal devices (such as digital cameras), gaming terminal devices, music storage and playback devices, in-vehicle wireless terminal devices, USB dongles, Internet of Things (IoT) devices, watches or other wearables, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in the context of industrial and / or automated processing chains), consumer electronics devices, devices operating on commercial and / or industrial wireless networks, and so on.

[0170] As used herein, the term "resource" can refer to radio resources in the time domain, frequency domain, spatial domain, and / or code domain. Some examples of resources include, for example, physical resource blocks (PRBs), radio frames, subframes, time slots, subbands, frequency domains, subcarriers, beams, etc. The terms "transmission" and / or "reception" can refer to wireless transmission and / or reception on radio resources via a radio propagation channel.

[0171] Figure 1 An example of a communication network that can be implemented according to the examples disclosed herein is shown. This communication network, or cellular communication network, may include network node 118a and network node 118b. Network node 118a is configured to provide one or more cells (such as cell 116a), and network node 118b is configured to provide one or more other cells (such as cell 116b). Each cell may be a macrocell, microcell, femtocell, or picocell, etc. Each cell may define a coverage area or service area for a corresponding access node.

[0172] Network node 118a can provide user equipment (UE) 100a (one or more UEs) with radio access to a communication network. This radio access may include downlink (DL) communication from the network node to UE 100a and uplink (UL) communication from UE 100a to the network node. Examples of uplink channels include the Physical Uplink Control Channel (PUCCH) for transmitting control information and the Physical Uplink Shared Channel (PUSCH) for transmitting data to the network. Examples of downlink channels include the Physical Downlink Control Channel (PDCCH) for transmitting control information and the Physical Downlink Shared Channel (PDSCH) for transmitting data to the user equipment.

[0173] The system can have multiple UEs 100a and 100b. Each UE among the multiple UEs 100a and 100b can be served by the same or different network nodes 118a and 118b. UEs can be configured with dual connectivity (DC), where a UE (e.g., UE 100a) can be connected to multiple network nodes 118a and 118b. They can communicate with each other when a device-to-device (D2D) communication interface is established between UEs 100a and 100b via a so-called "side link" (SL). For example, such D2D communication can be referred to as machine-to-machine communication, peer-to-peer (P2P) communication, or vehicle-to-vehicle (V2V) communication.

[0174] In a communication network with multiple network nodes, these nodes can be interconnected via interfaces. The LTE specification refers to such interfaces as the X2 interface. The interface between an LTE node and a 5G node, or between two 5G nodes, can be referred to as the Xn interface. Network nodes 118a and 118b can also be connected to the core network 116 of the communication network via another interface.

[0175] In the following text, various example references to communication devices capable of communicating with a communication system are explained. Before explaining the various examples of this disclosure in detail, references to a fifth-generation communication system (5GS), its access network and core network (5GC), and communication devices are provided. Figure 2 Briefly explained.

[0176] Figure 2 A schematic representation of a communication system is shown. (Reference) Figure 2 A user equipment (UE) 200 is shown communicating with an application server (not shown) hosting third-party application functions (not shown) via a communication network and a data network 224. This communication network includes a radio access network 220 (e.g., a next-generation radio access network (NG-RAN)) and a core network 222 (e.g., a 5G core network (5GC)), which operate based on the fifth-generation radio access technology, for example, in the 3rd Generation Partnership Project (3GPP) standard for new radios. The core network 222 includes network functions (generally referred to as network functions, and collectively as network functions) that can be connected to a management system configured to manage the communication network, as described in detail below.

[0177] Radio access network 220 includes one or more radio access network (RAN) nodes (also referred to as base stations). A RAN node can be configured to provide one or more cells. For example, a cell can be a macro cell, micro cell, femtocell, or picocell. A cell defines the coverage area or service area of ​​the RAN node. RAN nodes can be, for example, Node B (NodeB or NB), evolved Node B (eNodeB or eNB), next-generation Node B (gNB), remote radio unit (RRU), remote radio head (RRH), relay, integrated access and backhaul (IAB) node, or low-power node. RAN nodes can be deployed in non-terrestrial network (NTN) equipment, such as satellites (e.g., low Earth orbit (LEO) satellites or geostationary orbit (GEO) satellites), aircraft, or drones, where these NTN equipment form a non-terrestrial network, such as a ground station. RAN nodes can also be deployed on the ground, in which case the RAN node can generally be referred to as a terrestrial network equipment. A RAN that includes terrestrial network equipment is generally referred to as a terrestrial network.

[0178] RAN nodes can have a split architecture, where the functionality of the RAN node (e.g., eNB or gNB) is separated among various entities. A RAN node with a split architecture can include Radio Units (RUs) (also known as Remote Radio Heads (RRHs)), Centralized Units (CUs), and one or more Distributed Units (DUs). DUs can be connected to RUs via fronthaul. DUs can be connected to CUs via midhaul or F1 interfaces. CUs can be connected to the core network (e.g., core network 108) via backhaul. In a RAN node with a split architecture, the operation of the RAN node can be performed, for example, by CUs and / or DUs. A CU can control one or more DUs.

[0179] The RU converts radio signals sent to and from the antenna into digital signals(s) for transmission over packet networks, handles the digital front-end (DFE) and lower PHY layers, and includes digital beamforming functionality. A DU is a logical entity (e.g., software) hosted and running on a server near the RU. A CU is a logical entity (e.g., software) hosted and running on a server. A CU can be hosted and running on its own server or on the same server as the DU located near the RU. A DU includes a subset of the functionality of a RAN node (e.g., eNB or gNB), depending on the functional partitioning; and a CU includes other functionalities in the RAN node that are not part of the DU's subset of functionality. A DU may include a subset of the layers of the RAN node's protocol stack, while a CU may include other layers in the protocol stack that are not part of the DU's subset of layers. For example, in some implementations, the DU may include the Radio Link Control (RLC) layer, Media Access Control (MAC) layer, and Physical (PHY) layer of the protocol stack for the RAN node, while the CU may include layers above the RLC layer in the RAN node's protocol stack, such as the Packet Data Convergence Protocol (PDCP) layer, Radio Resource Control (RRC) layer, and Internet Protocol (IP) layer. The operation of the DU can be controlled by the CU.

[0180] Core network 222 can have a service-based architecture. The network functions of core network 222 include Access and Mobility Function (AMF), Authentication Server Function (AUSF), Network Openness Function (NEF), Network Repository Function (NRF), Network Slice Selection Function (NSSF), Policy Control Function (PCF), Session Management Function (SMF), User Plane Function (UPF), Unified Data Repository (UDM), and Network Data Analysis Function (NWDAF). For simplicity, other network functions of core network 222, such as Binding Support Function (BSF) and Charging Function (CHF), are not listed here. Figure 2 It is shown in the middle.

[0181] AMF handles access, authorization, and authentication of user equipment (including UE 200), and manages the mobility of user equipment 200 when it moves between different radio access networks, cells, or locations.

[0182] The SMF is responsible for establishing, maintaining, and terminating Protocol Data Unit (PDU) sessions in core network 222. The SMF manages user plane resources and interacts with the UPF in core network 222 to ensure that data packets are correctly routed and forwarded.

[0183] The UDM performs the authentication process, stores and manages user data (including subscriber profiles, authentication credentials, and authorization policies), implements security mechanisms to protect user data and resources of the communication network (e.g., core network 222) from unauthorized access attacks and vulnerabilities, and interacts with other network functions of core network 222 (such as PCF) to enforce access control policies, Quality of Service (QoS) parameters, and service restrictions based on user profiles and subscription plans. The UDM is also responsible for managing the registration of network functions serving user equipment 200.

[0184] The Network Analysis Data Function (NWAF) is configured to collect or retrieve data about one or more Network Functions (NFs) in the core network, generate analysis results based on the data collected or retrieved by the NWAF from one or more NFs, and provide the generated analysis results to itself or other NFs that have requested analysis generated by the NWAF. The NWAF may include an Analysis Logic Function (AnLF) configured to generate analysis (e.g., generate statistics and / or generate predictions) based on data about one or more NFs collected and / or retrieved by the NWAF. The NWAF also includes an analysis service exposed by the NWAF to provide the analysis generated by the AnLF. The NWAF also includes a Model Training Logic Function (NWDAF (MTLF)) configured to train an AI / ML model that can be used by the AnLF to generate analysis based on data about one or more network functions collected or retrieved from one or more network functions and / or OAM entities.

[0185] The functions of other network functions of core network 222 are not described in further detail.

[0186] 3GPP TR 22.870 V0.1.0 (2024-11) discusses 6G network use cases and 6G infrastructure. It discusses the computing and storage infrastructure that wireless network operators are equipped with, potentially still underutilized, given advanced technologies such as Generative Artificial Intelligence (Gen AI), Large Language Models (LLM), and AI-accelerated xPUs. Computing resources can be hosted in a distributed manner at edge sites. Operators (e.g., mobile network operators (MNOs)) can open these idle resources to third-party entities or AFs of UE providers, thereby allowing, supporting, and / or facilitating the utilization of computing and AI capabilities from the operator when needed.

[0187] Figure 3 This example illustrates the use of network resources and their accessibility to the UE provider. Figure 3A communication network (e.g., a 6G network) is illustrated, comprising one or more network functions (NFs) 334, a resource openness mechanism 326, and a policy function 328. This network can communicate with the UE provider's AF 306, one or more UEs 300, edge data center 336, charging function 330 and / or xPU, accelerator, and other resources 332. The resource openness mechanism 326 can provide a mechanism for checking the availability, accessibility, and / or authorization of the UE 300, edge data center 336, and / or AF 306.

[0188] Operators can provide these idle resources by billing third parties for computing cycles, AI processing, and storage utilization. This approach can encourage improved use of existing infrastructure.

[0189] By combining standardized interfaces, robust security protocols, and advanced resource management frameworks, communication networks (such as 6G networks) can allow, enable, and / or facilitate improved and secure access to operator resources. This can then position the communication network as a platform for distributed computing and artificial intelligence services, and for providing resources as a service. Network-aware resource openness is enabled by the network's centralized policy framework.

[0190] Communication networks (such as 6G networks) can provide a set of use cases (e.g., focusing on AI services). AI services can be executed on the network operator's cloud (e.g., on the edge cloud or central cloud of the 6G operator).

[0191] A new type of AI service is called Artificial Intelligence as a Service (AIaaS). AIaaS is a cloud-based service that allows, enables, and / or facilitates access to AI for individuals or organizations through third-party operators or providers. AIaaS can supply users with a set of network resources, including application programming interfaces (APIs), enabling users to leverage AI capabilities by benefiting from (e.g., directly and / or indirectly) corresponding network data or network functions. This can be achieved through collaboration between the user and network resources. In some examples, AIaaS users may include user equipment (UE), apps, and network functions of communication networks (e.g., 6G networks).

[0192] AIaaS can be viewed as a new type of service that can be provided by network operators. AIaaS can open APIs to UE consumers for building machine learning (ML) solutions. The "network as a platform" approach of communication networks (e.g., 6G networks) can be used to allow, enable, and / or facilitate AIaaS services. In this way, the AIaaS provided is a form of "Platform as a Service" (PaaS). PaaS provides and manages all hardware and software resources used for application development via the cloud. ML Operations as a Service (MLOps) has also been considered, which can allow, enable, and / or facilitate Infrastructure as a Service (IaaS) provisioning. IaaS provides organizations with infrastructure resources such as compute, storage, networking, and virtualization on demand via the cloud.

[0193] Each communication network element (e.g., a 6G network element) can integrate communication, computing, and sensing capabilities, thereby allowing, enabling, and / or facilitating the evolution from centralized intelligence in the cloud to global intelligence at the network edge (e.g., known as Network for AI (NET4AI) or AIaaS). In other words, the decentralization of intelligence has been considered.

[0194] AIaaS is related to the concept of Lifecycle Management (LCM), which has been studied in the context of Advanced 5G. The concept of LCM can allow, enable, and / or facilitate the collaboration of ML implementations and algorithms from different application service providers (e.g., suppliers). In the context of Advanced 5G, a standardized ML framework has been considered, encompassing both one-sided solutions (e.g., solutions implemented only on the UE side or only on the network side) and two-sided solutions (e.g., (multiple) ML algorithms split between the UE and network sides) for control plane signaling between the network and devices. This ML framework can provide a standardized approach for the large-scale deployment and operation of ML solutions in radio networks. ML framework dimensions may include at least one of the following: ML management (e.g., LCM procedures); data collection; UE ML capabilities; testing, verification, and validation; or the impact on UE mobility.

[0195] This ML framework aims to provide the foundation for next-generation native AI air interfaces, such as radio MLOp. Radio MLOp, or LCM for MLOp, can involve processes such as data collection, development and testing, deployment, and operation and monitoring. This ML framework can provide equipment vendors, network vendors, and operators with tools to deploy and operate ML solutions at scale and interoperably. This enables continuous improvement of the underlying ML model based on data and monitoring, allowing it to adapt to changing environments or conditions (which is useful for AI / ML use cases and solutions).

[0196] The AI / ML discussions and research on the air interface in advanced 5G provide the following use cases: enhanced channel state information (CSI) feedback (e.g., CSI compression and CSI prediction (for UE-side models)); enhanced beam management (e.g., downlink transmit beam prediction in the time and / or spatial domains); and enhanced positioning accuracy (e.g., direct positioning via ML algorithms and assisted positioning via ML algorithms).

[0197] These use cases can be referred to as ML-enabled features, including single-sided and two-sided solutions. Depending on the use case, ML functionality can reside in network nodes (e.g., gNB), location management functions (LMF), or NWDAF.

[0198] As AI is increasingly used in air interface use cases and UE applications, AI / ML models can be continuously updated, and the demand for network data may also increase. However, end devices (e.g., UEs) may lack sufficient computing power or resources to train AI / ML models on the end devices themselves.

[0199] Currently, training AI / ML models on the UE side is a task for the UE provider (i.e., the UE AI / ML model provider). However, the UE provider cannot train AI / ML models using the resources of the communication network. If a third party (e.g., a public cloud provider) uses network data to participate in the training of UE-side AI / ML models, network data security and privacy issues arise. Network data used for AI / ML model training and monitoring, as well as fallback triggers and policies used on the UE side, cannot be shared outside the public cloud provider's training server. Furthermore, the UE provider does not want to share AI / ML models with network operators.

[0200] However, due to the practical requirements imposed by MLOP, the training and maintenance of UE-side models are typically not performed in isolation from the network. These practical requirements imposed by MLOP may include at least one of the following: a continuous data stream for detecting AI / ML models; evaluation of AI / ML models to detect performance degradation; and evaluation of relevant data to detect and mitigate any potential data / concept drift through retraining of the AI / ML models.

[0201] In this way, there is no environment that allows, enables, and / or facilitates the training of AI / ML models for UE providers, and the environment remains private to third parties (e.g., public cloud providers) involved in training AI / ML models on the UE side.

[0202] The various examples in this article are intended to address one or more of the problems mentioned above.

[0203] The various examples in this paper provide UE-side model / application providers (e.g., UE suppliers) with practical and / or secure and / or low-cost and / or computationally efficient training environments, ensuring that the training environment and network data used to train AI / ML models remain private to third parties, while allowing, enabling, and / or facilitating the actual training and deployment of UE-side AI / ML models.

[0204] Some examples in this article focus on one-sided training. The (UE-side) LCM solution supports AIaaS use cases by establishing a UE-side Training Environment as a Service (TEaaS). TEaaS is a type of AIaaS provider / service. In some examples, TEaaS can provide the MNO with opportunities to leverage available network resources and data.

[0205] Some examples in this paper provide solutions for private AI / ML model training environments for training AI / ML models by UEs or UE providers, where the operator (e.g., a 6G operator) as a service provider cannot see the AI / ML model being trained. In other words, some examples in this paper provide solutions for private AI / ML model training environments for training AI / ML models by UEs or UE providers, where the training of the AI / ML model is private to the operator (e.g., a 6G operator) as a service provider. In this way, the intellectual property rights of the AI / ML model being trained can remain protected.

[0206] Example use cases for using TEaaS include at least the following: location-based prediction; UE behavior modeling; IoT device management; emergency response systems; security analytics; device performance monitoring; and UE power management.

[0207] Location-based predictions can use geospatial data to predict user behavior, such as geographic region-based mobility patterns, location-based services (LBS) usage, or device adoption rates.

[0208] UE behavior modeling may include creating ML models to analyze and predict UE behavior in different geographic regions, including at least one of traffic, channel utilization, or energy consumption.

[0209] IoT device management can include developing predictive models for IoT devices deployed across different regions, thereby allowing, enabling, and / or facilitating more efficient device management, maintenance, and energy consumption prediction.

[0210] Emergency response systems can integrate AI / ML with geographic data to develop systems that can quickly identify areas of interest (e.g., areas of need), allocate resources, and provide targeted assistance.

[0211] Security analytics can include analyzing geospatial data to detect potential security threats, such as patterns of malicious activity or suspicious device behavior in a specific area.

[0212] Device performance monitoring can include developing models to monitor device performance across different geographic regions, enabling UE providers to identify areas for improvement and optimize device performance.

[0213] UE power management can include creating predictive models to optimize UE power consumption based on geospatial data, thereby reducing energy waste and improving overall device efficiency.

[0214] All examples in this article are about Figures 3 to 6 Described. In particular, Figures 3 to 6 Some principles are shown, and these principles will Figures 7 to 12 They are illustrated together in any of the examples.

[0215] Figure 4 A communication network (e.g., a 6G network) is schematically illustrated, comprising an AIaaS function 402, an AIaaS controller 404, one or more network functions (NFs) 434, and a resource openness mechanism 426. This network can communicate with the UE provider's AF 406, one or more UEs 400, an edge data center 436, one or more servers 412, a charging function 430 and / or an xPU, accelerators, and other resources 432. The resource openness mechanism 426 can provide a mechanism for checking the availability, accessibility, and / or authorization of the UE 400, the edge DC 436, and / or the AF 406.

[0216] In some examples, AIaaS function 402 may be a network function and / or provided by the device. For example, AIaaS function 402 may be part of NWDAF. In some examples, AIaaS function 402 may be or include a standalone network function as part of the core network (e.g., a 6G core network). In some examples, AIaaS controller 404 may be included within the device or provided by the device, wherein AIaaS controller 404 is implemented as a functional part of an existing cloud orchestration function or mechanism.

[0217] like Figure 4As shown, to use TEaaS, the UE provider's AF 406 and / or UE 400 can request training of a UE-side AI / ML model from the network. The UE provider or AF 406 can develop models that rely on network data to improve UE 400 performance. UE 400 may include a UE training controller for communicating with the network. AF 406 and / or UE 400 can provide the network with the UE-side AI / ML model and / or selection criteria (e.g., geographic location, cell ID, UE ID, UE type ID, or at least one UE feature availability) for selecting at least one UE to deploy the trained AI / ML model. Providing the selection criteria to the network can allow, enable, and / or facilitate training an AI / ML model for multiple UEs that meet the selection criteria. AF 406 and / or UE 400 can provide the network with proprietary data for training the AI / ML model.

[0218] Training of the AI / ML model can be performed at the AIaaS controller 404 by the server 412 (e.g., a TEaaS server). In some examples, training of the AI / ML model can be performed at the AIaaS controller 404 by a shared server or a dedicated server. Whether a shared server or a dedicated server is used may depend on the required level of privacy.

[0219] In one example, a training environment for the UE (e.g., for training a UE-side AI / ML model) can be provided, where the AI / ML model is provided to the network (e.g., a 6G network) to be trained together using configurations for training and validation, along with compilation instructions. Training of the AI / ML model can be performed / run on a shared server of the MNO (e.g., a shared server of the service provider). In some examples, this may provide a sufficient level of protection for most use cases. In this example, because the training of the AI / ML model is performed on a shared server in the operator's cloud (e.g., a shared server of the service provider), network data, UE provider data, and / or UE data cannot be physically accessed by the public cloud provider. The operator, as a service provider (e.g., a 6G operator), can physically store UE provider data and / or UE data and / or the AI / ML model, but all data is encrypted to enhance security.

[0220] In another example, a dedicated server within the operator's cloud (e.g., a 6G operator's cloud) can be provided for training UE-side AI / ML models acquired from the UE or UE provider. This example provides an opaque and more secure option for the UE training environment. In this example, the operator, as the service provider, cannot access the UE or UE provider's AI / ML models or associated data. For example, the operator's dedicated server may not be shared with any other cloud users (e.g., 6G cloud users), which provides an additional layer of security. Because the training is performed on a dedicated server in the operator's cloud, network data, UE data, and / or UE provider data are not physically accessible to the public cloud provider. The operator can still view and control the data collected from the network and fed into the training process.

[0221] Figure 5 An example is provided for establishing a connection between the UE provider and the operator's cloud infrastructure to provide a secure training environment (e.g., TEaaS) for the UE provider. Figure 5 The principle allows the UE provider to request services for training AI / ML models from the network through AF instance 506 in the core network.

[0222] Figure 5 The signaling between AF 506, AIaaS function 502, AIaaS controller 504, NWDAF 508, and NRF 510 is illustrated. AF 506 (or the provisioning AF) can be provided by the UE provider (e.g., an application service provider or AIaaS user). AIaaS function 502 can be a network function and / or provided by the device. For example, AIaaS function 502 can be part of NWDAF. In some examples, AIaaS function 502 can be a standalone network function as part of the core network (e.g., a 6G core network). In some examples, AIaaS controller 504 can be included in or provided by the device, wherein AIaaS controller 504 is implemented as a functional part of an existing cloud orchestration function or mechanism. Figure 5 As shown, AIaaS function 502 and AIaaS controller 504 can be implemented in a communication network (e.g., a 6G network).

[0223] At S501a, AIaaS function 502 can register with NRF 510. For example, AIaaS function 502 can register with NRF 510 using the standardized protocol Nnrf_NFmanagement_register. In some examples, during the registration, AIaaS function 502 can provide NRF 510 with information on the features and characteristics of AIaaS function 506 to allow, enable, and / or facilitate interaction with other network components (e.g., as part of the profile of AIaaS function 502). In some examples, these features and characteristics can then be used for discovery by the UE provider's AF 506. In some examples, the information on the features and characteristics of AIaaS function 506 may include at least one of the following: AIaaS function ID; supported UE type and provider ID; supported services, such as TEaaS; supported use cases (e.g., CSI compression or location); supported geographic regions; support for AIaaS management; or support for AIaaS optimization. In some examples, when AF 506 is located outside the trusted domain, the registration of AIaaS function 502 can be performed using NEF.

[0224] At S501b, the AIaaS controller 504 can register with the NRF 510. For example, the AIaaS controller 504 can register with the NRF 510 using the standardized protocol Nnrf_NFmanagement_register. In some examples, during the registration process, the AIaaS controller 504 can provide the NRF 510 with information about its characteristics and features to enable interaction with other network components (e.g., as part of the AIaaS controller 504's profile). In some examples, these characteristics and features can then be discovered by the UE provider's AF 506. In some examples, the information about the AIaaS controller 504's characteristics and features may include at least one of the following: support services and images (e.g., TEaaS); compute management; or data management. In some examples, if the AF 506 is located outside a trusted domain, the registration of the AIaaS controller 504 can be performed using the NEF.

[0225] At S501c, the NWDAF 508, which supports AIaaS analytics or data openness, can register with the NRF 510. For example, the AIaaS controller 504 can register with the NRF 510 using the standardized protocol Nnrf_NFmanagement_register. The NWDAF 508 can provide the NRF 510 with information indicating its AIaaS support.

[0226] At S502, NRF 510 can update the corresponding profile based on information received from AIaaS function 502 and / or AIaaS controller 504 in S501a and S501b. For example, if there are updates or changes to the features or characteristics provided by AIaaS function 504 and / or AIaaS controller 504, NRF 504 can use Nnrf_NFmanagement_update to create or update the corresponding configuration file.

[0227] At S503a, S503b, and S503c, the NRF 510 provides register response messages to the AIaaS function 502, the AIaaS controller 504, and / or the NWDAF 508.

[0228] At S504, AF 506 can provide a discovery request message to NRF 510. This discovery request message can be used to discover AIaaS function 502. In some examples, the discovery request message may include or request an AIaaS type or characteristic. In some examples, AF 506 can send a discovery request message to AIaaS function 502 for the purpose of discovering AIaaS function 502. For example, AF 506 can send an AI service discovery request message to AIaaS function 502.

[0229] At S505, NRF 510 can authorize the discovery request messages received in S504.

[0230] At S506, NRF 510 can send a discovery response message to AF 506. In some examples, AIaaS function 502 can send a discovery response message to AF 506. For example, AIaaS function 502 can send an AI service discovery response message to AF 506.

[0231] At S507, AF 506 may provide a request message (e.g., an AIaaS request message) to AIaaS function 502. For example, this request message may be used to request services for training AI / ML models in an environment private to the service provider (e.g., private to the network operator). In some examples, training AI / ML models in an environment private to the service provider includes at least one of the following: training an AI / ML model on a shared server implemented by the service provider; or training an AI / ML model on a dedicated server implemented by the service provider.

[0232] In some examples, a request message is sent to AIaaS function 502 to request the network operator to provide TEaaS to the UE provider. In some examples, the request message for TEaaS is sent by AF 506 based on AF 506 or the UE provider determining that a UE AI / ML model needs to be trained (e.g., when a requirement is identified for training). In some examples, the request message is sent as a dedicated message via the NAS protocol. In some examples, the request message includes the AI / ML model for training.

[0233] In some examples, the request message includes a list of configuration requirements and / or policies for AIaaS feature 506 for training one or more AI / ML models. For example, the request message may include one or more AI / ML models whose metadata includes at least one of the following: the use case of the AI / ML model; the accuracy of the AI / ML model; the F1 score, confusion matrix, and / or mean squared error of the AI / ML model; the training data for the AI / ML model; or information on the appropriate deployment region for the AI / ML model. In some examples, the information on the appropriate deployment region for the AI / ML model may include selection criteria for deploying the trained AI / ML model. For example, the selection criteria may indicate which UE or UE group or group the AI / ML model is effective for. For example, the selection criteria may indicate that the AI / ML model, once trained, is effective and / or will be deployed for UEs having at least one of the following: a specific geographic location; a specific cell ID; a specific UE ID; a specific UE type ID; or feature availability at the UE.

[0234] In some examples, the configuration requirements included in the request message may also include at least one of the following: one or more training strategies (e.g., triggering or stopping conditions) for training the AI / ML model; one or more validation strategies for validating the trained AI / ML model; training data for training the AI / ML model; validation data for validating the strategies (e.g., for validating the trained AI / ML model); or one or more energy consumption strategies indicating the energy consumption for training the AI / ML model. In some examples, the one or more energy consumption strategies may indicate at least one of the following: an energy consumption threshold for training the AI / ML model; a strategy for minimizing energy consumption during the training of the AI / ML model; or a strategy for ignoring energy consumption during the training of the AI / ML model (e.g., a strategy indicating that energy consumption is not considered during the training of the AI / ML model).

[0235] At S508, AIaaS function 502 can inspect and / or authorize request messages received from AF 506.

[0236] At S509, AIaaS function 502 can save and / or update the configuration requirements for training the AI / ML model received from AF 506 in S507.

[0237] At step S510, AIaaS function 502 may provide AIaaS controller 504 with a request message requesting AIaaS controller 504 to update the configuration requirements for training the AI / ML model. For example, the request message may be or include a service configuration request message.

[0238] At S511, AIaaS controller 504 can provide AIaaS function 502 with a complete message (e.g., a service configuration request / response message). In some examples, when training of the AI / ML model has been performed, the complete message can indicate that the requested service (from S507) has been approved. In some examples, when training of the AI / ML model has been performed, the complete message can provide configuration information for the requested service (from S507). In some examples, when training of the AI / ML model has not yet been performed, AIaaS controller 504 can provide AIaaS function 502 with an indication that the requested service (from S507) has been rejected and / or the reason why the requested service was rejected. In some examples, the complete message may include the trained AI / ML model. In some examples, the complete message may include the location where the trained AI / ML model is stored (e.g., GitHub or a private repository).

[0239] At S512, AIaaS function 502 sends a request completion message (e.g., an AIaaS completion message) to AF 506. This request completion message can notify AF 506 that the AI / ML training request (e.g., the request in S507) has been executed. In other words, the request completion message can indicate whether the training of the AI / ML model has been executed. For example, when the training of the AI / ML model has been executed, the request completion message includes: an indication (e.g., in S507) that the requested service has been approved, and / or provides configuration information for the requested service. In some examples, when the training of the AI / ML model has not yet been executed, the request completion message includes: an indication (e.g., in S507) that the requested service has been rejected, and / or the reason why the requested service was rejected. In some examples, the request completion message may include the trained AI / ML model. In some examples, the request completion message may include the location where the trained AI / ML model is stored (e.g., GitHub or a private code repository).

[0240] Figure 6An example is provided of establishing a connection between the UE 600 and the operator's cloud infrastructure to provide the UE 600 with a secure training environment (e.g., TEaaS). Figure 6 The principle behind this allows the UE 600 to directly request services from the network for training AI / ML models. This may require modifications to existing signaling between the UE and the network.

[0241] Figure 6 The signaling between UE 600, AIaaS function 602, and AIaaS controller 604 is illustrated. UE 600 may include, be included in, or implement a UE training controller. In some examples, the UE training controller is a logical or functional entity implemented at UE 600. In some examples, the UE training controller is installed at UE 600. In some examples, the UE training controller is vendor-specific (or application service provider-specific). For example, the UE training controller is associated with a specific UE vendor. In some examples, AIaaS function 602 may be a network function and / or provided by the device. For example, AIaaS function 602 may be part of NWDAF. In some examples, AIaaS function 602 may be or include a standalone network function as part of the core network (e.g., a 6G core network). In some examples, AIaaS controller 604 may be included within or provided by the device, wherein AIaaS controller 604 is implemented as a functional part of an existing cloud orchestration function or mechanism. Figure 4 As shown, AIaaS function 402 and AIaaS controller 404 can be implemented within a network (e.g., a 6G network).

[0242] As shown in Figure S601, UE 600 sends a discovery request message (e.g., an AI service discovery request message) to AIaaS function 602. In some examples, this discovery request message is used to discover the available AIaaS function 602 on the network, and / or to discover whether the TEaaS service is available at the network operator. In some examples, the discovery request message includes a UE provider ID and / or a UE type ID. The UE provider ID and / or UE type ID may be used by the network (e.g., or required) to identify which available AIaaS is applicable to UE 600 and / or the UE training controller. In some examples, the UE provider ID is a first identification value indicating the provider or application service provider serving UE 600. In some examples, the UE type ID is a second identification value indicating the type of UE 600. For example, the UE type ID may include at least one of the following: the ID of the UE 600 (e.g., the ID of a user in the network environment, such as the UE ID in the core network associated with a subscription); or an ID indicating the model and manufacturer of the UE 600 (and thus indicating the supported capabilities of the UE 600). In some examples, the discovery request message may be carried via the NAS protocol, where the UE type and supported capabilities are part of the NAS protocol design. The UE provider ID and UE type ID may be provided as information elements (IEs).

[0243] In some examples, the UEVendorID IE may include at least one of the following properties: Type: Enumerated(RegistredVendor1,RegistredVendor2,…); Description: Represents one of n possible states; or Format: Enumerated value.

[0244] In some examples, UETypeID IE may include at least one of the following properties: Type: Enumerated (UEType1, UEType2, ...); Description: Represents one of m possible states; or Format: Enumerated value.

[0245] At S602, UE 600 can receive a discovery response message (e.g., an AI service discovery response message) from AIaaS function 602. In some examples, the discovery response message may indicate the availability of TEaaS services at the network operator's cloud. In some examples, the discovery response message includes a list of one or more AIaaS types available at AIaaS function 602, and / or a list of one or more TEaaS types available (e.g., one or more environment types for training UE AI / ML models available at network function AIaaS function 602). In some examples, AIaaS types and TEaaS types may be provided as information elements (IEs).

[0246] In some examples, AIaaSType IE can include at least one of the following attributes: Type: Enumerated (TEaaS, Inference as a Service, Training as a Service, etc.); Description: Represents one of k possible states. Defines the type of possible AIaaS provided by a communication network (e.g., a 6G network); or Format: Enumerated value.

[0247] TEaaSType IE can include at least one of the following attributes: Type: Enumerated (shared, private); Description: Represents one of two possible states. Defines how AI / ML training is performed on the operator's network cloud server; or Format: Enumerated value.

[0248] At S603, UE 600 may send a request message (e.g., an AIaaS request message) to AIaaS function 602. For example, this request message may be used to request the training of an AI / ML model in an environment where the training of the AI / ML model is private to the service provider (e.g., private to the network operator). In some examples, the training of the AI / ML model in an environment where the training of the AI / ML model is private to the service provider includes at least one of the following: training of the AI / ML model on a shared server implemented by the service provider; or training of the AI / ML model on a dedicated server implemented by the service provider.

[0249] In some examples, a request message is sent to AIaaS function 602 to request the network operator to provide TEaaS to the network of UE 600. In some examples, the request message for TEaaS is sent by UE 600 based on UE 600 determining the need to train a UE AI / ML model (e.g., when a requirement is identified for training). In some examples, the request message is sent in a dedicated message via the NAS protocol. In some examples, the request message includes the AI / ML model for training.

[0250] In some examples, the request message includes a request for a specific AIaaS type or TEaaS type. In other words, the request message may include: a request for the type of AI service available at AIaaS function 602, and / or a request for the type of environment available at AIaaS function 602 for training AI / ML models. For example, UE 600 may request that the training of the AI / ML model be performed by a dedicated server, as described above. Figure 4 The above discussion. In other examples, UE 600 may request that the training of the AI / ML model be performed by a shared server, as mentioned above. Figure 4 The subject of discussion.

[0251] In some examples, the request message includes information on selection criteria for deploying the trained AI / ML model. For example, selection criteria might indicate which one or more UEs or groups of UEs the AI / ML model is valid for. For instance, selection criteria might indicate that the AI / ML model, once trained, is valid and / or will be deployed for UEs having at least one of the following: a specific geographic location; a specific cell ID; a specific UE ID; a specific UE type ID; or feature availability at the UE. In some examples, selection criteria might be provided as an information element (IE).

[0252] In some examples, Selection Criteria IE can include at least one of the following attributes: Type: String (geolocation, cell ID, UEID / type); Description: Information that identifies which UE / UE group the AI / ML model is valid for; or Format: string.

[0253] In some examples, the request message includes training data for training AI / ML models. For instance, the training data could be data used to train an AI / ML model of interest to the UE provider. The training data can be provided in a Training Data IE.

[0254] In some examples, TrainingData IE may include at least one of the following properties: Type: JSON (or other data formats) (a data container with training data); Description: Specific data of interest to UE providers for AI / ML training; or Format: JSON.

[0255] In some examples, the request message includes validation data for validating the trained AI / ML model. For instance, validation data is provided to AIaaS function 602 so that the AI / ML model can be validated as soon as it is trained. The validation data can be provided in a Validation Data IE.

[0256] In some examples, ValidationData IE may include at least one of the following properties: Type: JSON (can also be other data formats) (a data container with validation data); Description: Specific data used for AI / ML; or Format: JSON.

[0257] In some examples, the request message may include information indicating the configuration or training requirements for training an AI / ML model. For example, training requirements for performing training on an AI / ML model may include information on at least one of the following: number of epochs; validation criteria; required accuracy; required F1 score, confusion matrix, and / or mean squared error; or maximum allowed time. Training requirements may be provided in the Training Requirements IE.

[0258] In some examples, TrainingRequirement IE may include at least one of the following properties: Type: JSON (number of rounds, validation criteria, required accuracy, maximum allowed time, etc.); Description: The configuration used for the training process (e.g., the configuration required for the training process); or Format: JSON.

[0259] In some examples, the request message may include deployment configuration. For example, the deployment configuration may indicate which UEs the trained AI / ML model will be deployed to. For example, the deployment configuration may indicate that the trained AI / ML model will be deployed to UEs with matching selection criteria (e.g., deployed to UEs in a specific geographic area indicated by the selection criteria).

[0260] At S604, AIaaS function 602 sends a request-response message to UE 600. In some examples, the response message may include at least one of the following: confirmation of service provision, start time for training the AI / ML model, or estimated completion time for training the AI / ML model.

[0261] At S605, AIaaS function 602 can send a training request message (e.g., an AI / ML training request fulfillment requirement message) to AIaaS controller 604. This training request message can instruct AIaaS controller 604 to deploy one or more servers (e.g., TEaaS servers) to fulfill an AI / ML model training request requested by UE 600 (e.g., a request message from S603). AIaaS function 602 (in the training request message) can also send a request to AIaaS controller 604 for physical resources used for training the AI / ML model based on information indicated in the request message received from UE 600.

[0262] At S606, the AIaaS controller 604 can allocate resources for requests to train AI / ML models to one or more network servers (e.g., TEaaS servers). For example, the AIaaS controller 604 can allocate resources for requests to train AI / ML models to dedicated servers or shared servers. In some examples, the AIaaS controller 604 can deploy one or more servers allocated to AI / ML model training in the network operator's cloud by running deployed TEaaS server functionality.

[0263] At S607, the AIaaS controller 604 can perform training of the AI / ML model. For example, the AIaaS controller 604 can perform training of the AI / ML model using a deployed server (as discussed in S606). For example, the AIaaS controller 604 can train the AI / ML model while using network resources.

[0264] At S608, the AIaaS controller 604 can provide a complete message to the AIaaS function 602. In some examples, the complete message informs the AIaaS function 602 whether the training of the AI / ML model has been successfully executed. In some examples, the complete message includes information about the trained AI / ML model. For example, the complete message may include: the availability of the trained machine learning model, and / or one or more inference instances of the AI / ML model. In some examples, the complete message includes: information indicating the location where the trained AI / ML model is stored (e.g., GitHub or a private repository).

[0265] At S609, AIaaS function 602 can perform validation checks on the trained AI / ML model. For example, AIaaS function 602 can perform validation checks on the trained AI / ML model using validation data provided by UE 600 (e.g., in the request message at S603). AIaaS function 602 can also determine whether validation criteria and / or compilation requirements for the trained AI / ML model are met.

[0266] At S610, the AIaaS function sends a request completion message (e.g., an AIaaS completion message) to UE 600. This request completion message may notify UE 600 (e.g., at S603) that the AI / ML training request has been executed. The response completion message may include information indicating at least one of the following: the trained AI / ML model; the location of the AI / ML model file and / or inference instance (allowing UEs that have validated the AI / ML model to access the inference instance); the availability of one or more inference instances of the AI / ML model; the ability of UE 600 to deploy the trained AI / ML model to one or more other UEs and / or to the AF of a UE provider (e.g., an application service provider) serving UE 600; or, the availability of one or more inference instances deployed to one or more other UEs and / or to the AF of a UE provider (e.g., an application service provider) serving UE 600.

[0267] In some examples, the trained AI / ML model can be deployed to (e.g., directly to) the requesting UE600. In some examples, the trained AI / ML model can be deployed to the UE provider's AF (e.g., Figure 5 (AF 506). In some examples, a trained AI / ML model can be deployed to a UE that matches the selection criteria requested by UE 600 (e.g., in S603). For example, a trained AI / ML model can be deployed to a UE whose geographic location matches the geographic location indicated in the selection criteria.

[0268] In some examples, UE 600 may provide a trained AI / ML model to the AF of one or more other UEs and / or UE providers (e.g., application service providers) serving UE 600. In some examples, UE 600 may provide one or more inference instances to the AF of one or more other UEs and / or UE providers (e.g., application service providers) serving UE 600.

[0269] Figure 4 , Figure 5 as well as Figure 6Any of the principles shown can be used as a standalone embodiment, or combined and / or used simultaneously. For example, as Figure 6 As shown, UE 600 can request services (e.g., TEaaS) from the network and provide the information needed for training the AI / ML model. After training has been performed, the trained AI / ML model can be provided... Figure 5 AF 506 was deployed.

[0270] Figures 7 to 12 Various aspects of the above examples are illustrated. Therefore, it can be understood that the above description can provide features that functionally correspond to the features discussed below, and thus can be used to better understand the aspects currently described.

[0271] refer to Figure 7 The diagram illustrates some example methods. Figure 7 This method can be performed by a device. In some examples, the device may include, be included in, or be a UE. In some examples, the device may include, be included in, or be a UE training controller. In some examples, the device may include, be included in, or be a UE that includes a UE training controller. In some examples, the device may include or be configured to provide an application / model service provider (AF). In some examples, the device may include or be configured to provide an AF for an application / model service provider (e.g., a UE provider). In some examples, the device may be as described above. Figure 1 , 2 5, 6 and / or Figure 13 The device may include, be included in, or be described above. Figure 5 and Figure 6 UE 500 and / or UE600.

[0272] As referenced at S701, the method may include: providing a network function with a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider.

[0273] As referenced at S702, the method may further include: receiving a response message from a network function indicating whether training of the machine learning model has been performed.

[0274] See Figure 8 The diagram illustrates some example methods. Figure 8This method can be performed by a device. In some examples, the device may include or be configured to provide AIaaS functionality. For example, the device may include or be configured to provide network functionality, or be a device for network functionality (e.g., NWDAF (or a portion thereof) or a standalone network function as part of the core network). In some examples, the device may be combined as... Figure 1 , 2 5, 6 and / or Figure 13 The device may include or be configured to provide, for example, the aforementioned. Figure 5 and Figure 6 AIaaS function 502 and / or UE AIaaS function 602.

[0275] As referenced at S801, the method may include: receiving from a second device a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider.

[0276] As referenced at S802, the method may further include: providing a response message to a second device, the response message indicating whether training of the machine learning model has been performed.

[0277] See Figure 9 The diagram illustrates some example methods. Figure 9 This method can be performed by a device. In some examples, the device may include or be configured to provide an AIaaS controller. For example, the device may include or be configured to provide a network controller, or be a device for a network controller (e.g., a device implemented as part of an existing cloud orchestration function or mechanism). In some examples, the device may be as described regarding... Figure 1 , Figure 2 , Figure 5 , Figure 6 and / or Figure 13 As described above. For example, the device may be or be configured to provide... Figure 5 and Figure 6 AIaaS controller 504 and / or UE AIaaS controller 604.

[0278] As referenced at S901, the method may include: receiving from a network function a request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider.

[0279] As referenced at S902, the method may further include: providing a complete message to the network function indicating whether training of the machine learning model has been performed.

[0280] See Figure 10The diagram illustrates some example methods. Figure 10 This method can be performed by a device. In some examples, the device may include, be included in, or be a UE. In some examples, the device may include, be included in, or be a UE training controller. In some examples, the device may include, be included in, or be a UE including a UE training controller. In some examples, the device may include or be configured to provide an application / model service provider (AF). In some examples, the device may include or be configured to provide an AF for an application / model service provider (e.g., a UE provider). In some examples, the device may be related to... Figure 1 , 2 5, 6 and / or Figure 13 As described herein. For example, the device may include, be included in, or be used as... Figure 5 and 6 UE 500 and / or UE 600.

[0281] As referenced at S1001, the method may include: providing an AI as a Service (AIaaS) function with an AI service discovery request message for discovering the AIaaS function.

[0282] As referenced in S1002, the method may further include: receiving an AI service discovery response message from an AIaaS function based on the AI ​​service discovery request.

[0283] As referenced at S1003, the method may further include: providing an AIaaS request message to an AIaaS function for a service for training the machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to.

[0284] As referenced at S1004, the method may further include: receiving an AIaaS completion message from an AIaaS function, the AIaaS completion message indicating whether training of the machine learning model has been performed.

[0285] See Figure 11 The diagram illustrates some example methods. Figure 11 This method can be performed by a device. In some examples, the device may include or be configured to provide AIaaS functionality. For example, the AIaaS functionality may be a network function and / or provided by the device. For example, the AIaaS functionality may be part of NWDAF. In some examples, the AIaaS functionality may be a standalone network function as part of the core network (e.g., a 6G core network). In some examples, the device may be as described regarding... Figure 1 , 25, 6 and / or Figure 13 The device may include or be configured to provide, for example, the aforementioned. Figure 5 and Figure 6 AIaaS function 502 and / or UE AIaaS function 602.

[0286] As referenced at S1101, the method may include: receiving an artificial intelligence (AI) service discovery request from a user equipment for discovering AIaaS functions.

[0287] As referenced in S1102, the method may further include: providing an AI service discovery response message to the user device based on the AI ​​service discovery request.

[0288] As referenced at S1103, the method may further include: receiving from a user device an AIaaS request message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to.

[0289] As referenced at S1104, the method may further include: providing an AIaaS completion message to the user device, indicating whether the training of the machine learning model has been performed.

[0290] See Figure 12 The diagram illustrates some example methods. Figure 12 This method can be performed by a device. In some examples, the device may include an AIaaS controller. For example, the device may include a network controller or a network controller device (e.g., a device implemented as part of an existing cloud orchestration function or mechanism). In some examples, the device may be as described above. Figure 1 , Figure 2 , Figure 5 , Figure 6 and / or Figure 13 As described above. For example, the device could be... Figure 5 and 6 AIaaS controller 504 and / or UE AIaaS controller 604.

[0291] As referenced at S1201, the method may include: receiving from an AI as a Service (AIaaS) function an AI training request fulfillment requirement message for a service for training a machine learning model in an environment where the training of the machine learning model is private to the service provider; the AI ​​training request fulfillment requirement message includes selection criteria for indicating which one or more user devices among a plurality of user devices the machine learning model will be deployed to.

[0292] As referenced in S1202, the method may include: providing a trained complete message to the AIaaS function, the trained complete message indicating whether training of the machine learning model has been performed.

[0293] The technology in this disclosure can be associated with at least one of several different advantages.

[0294] The examples in this paper provide a private, secure, and computationally efficient training environment for the UE or UE-side model provider (e.g., UE supplier). For instance, the network operator (e.g., a 6G Operator-as-a-Service provider) cannot see the model being trained. In other words, the training of the UE-side AI / ML model remains private to the network operator. The examples in this paper provide a UE-side TEaaS that allows, enables, and / or facilitates the training of UE-side models utilizing the network. In some examples, the UE-side TEaaS is provided without compromising the privacy or security of UE data or UE supplier data.

[0295] Figure 13 A block diagram of apparatus 10 is shown by way of non-limiting and illustrative example. Apparatus 10 includes, for example, at least one processor 12 and at least one memory 14 storing instructions 15 that, when executed by the at least one processor, cause apparatus 10 to perform at least one or more methods disclosed herein. In some examples, the at least one memory and instructions (e.g., program code, computer program code, software, etc.) together with the at least one processor are configured to cause apparatus 10 to perform one or more methods disclosed herein. In some examples, the instructions may be those instructions for any functional or network function described in this disclosure.

[0296] Processor 12 may include, or be constituted by, one or more circuit systems configured to perform phases of the methods described herein. As used herein, the term “circuit system” may refer to one or more of the following: (a) an implementation of hardware circuitry alone, such as analog, digital, and / or quantum circuit systems; and (b) a combination of (multiple) hardware circuitry and software, such as (if applicable): (i) a combination of (multiple) analog, digital, and / or quantum hardware circuitry and software / firmware; and (ii) any or all portions of (multiple) hardware processors having software (including (multiple) digital and / or quantum processors), software, and (multiple) memories, which work together to enable a device (such as a mobile phone or server) to perform various functions; and (c) (multiple) hardware circuitry and / or (multiple) quantum processors (such as (multiple) microprocessors or portions thereof) that require software (e.g., firmware) to operate, but may be absent when operation does not require software. This definition of “circuit system” applies to all uses of this term in this application, including in any claim. As another example, as used herein, the term "circuit system" also covers only the implementation of hardware circuitry or a processor (or multiple processors) or portions thereof and their accompanying software and / or firmware. For example, if applicable to a particular claim element, the term "circuit system" also covers baseband integrated circuits or processor integrated circuits for mobile devices or similar integrated circuits in servers, cellular network devices or other computing or networking devices.

[0297] The memory 14 can be implemented using any suitable data storage technology. The memory may include a database for storing data. The memory 14 is at least partially located outside the device 10, but is accessible to the device 10.

[0298] Instruction 15 may be included in a computer-readable medium or a non-transitory computer-readable medium. As used herein, the term “non-transitory” refers to a limitation on the medium itself (e.g., a tangible medium, rather than a signal), rather than a limitation on the persistence of data storage (e.g., random access memory (RAM) and read-only memory (ROM).

[0299] As previously described, device 10 may include a UE, be included within a UE, or be a UE. Device 10 may include at least a portion of the UE as previously described, be included within at least a portion of the UE as previously described, or implement at least a portion of the UE as previously described. As another example, device 10 may be included in such a UE (e.g., as a chipset configured to control a terminal device). In some examples, device 10 may be or provide a UE training controller. Device 10 may include or implement at least a portion of the UE training controller as previously described. In some examples, device 10 may include or include a UE training controller. In some examples, device 10 may include or be configured to provide an application service provider (AF). In some examples, device 10 may include or be configured to provide an AF for an application service provider. Device 10 may include or be configured to implement at least a portion of the AF as previously described. Device 10 may be caused or configured to perform at least one or more of the methods described in the examples.

[0300] As another example, device 10 may include or be configured to provide network functionality (e.g., the AIaaS functionality described above). In another example, device 10 may be included in such network functionality, for example, as a chipset configured to control that network functionality. In some examples, device 10 may include or be configured to implement at least a portion of the network functionality described above. Device 10 may be controlled or configured to perform at least one or more of the methods described in the examples.

[0301] As another example, device 10 may include or be configured to provide a network controller (e.g., the AIaaS controller described above). In another example, device 10 may be included in such a network controller, for example, as a chipset configured to control the network functionality. In some examples, device 10 may include or be configured to implement at least a portion of the network controller described above. Device 10 may be caused or configured to perform at least one or more of the methods described in the examples.

[0302] The device may include any one or more entities at the protocol layer, such as a MAC entity, RRC entity, RLC entity, PDCP entity, or PHY entity. In some examples, the entity may be configured to perform at least the methods of the examples and / or any one or more of the examples.

[0303] Device 10 may include a radio interface 16. Radio interface 16 may provide communication capabilities to device 10. Radio interface 16 may include a receiver configured to receive information according to at least one cellular or non-cellular standard. Radio interface 16 may include a transmitter configured to transmit information according to at least one cellular or non-cellular standard. Receivers may include more than one receiver. Transmitters may include more than one transmitter. Radio interface 16 may include a transceiver configured to receive and transmit information according to at least one cellular or non-cellular standard. Transceivers may include more than one transceiver.

[0304] Device 10 may include an interface (e.g., a user interface) 18, which may include at least one of a keyboard, microphone, touchscreen, display screen, speaker, etc. Interface 18 can be used by a user to control the device. Interface 18 may be at least partially located outside of device 10. For example, device 10 may be connected to another device (such as a computer) via a wireless or wired connection, and device 10 may be controlled via the computer (e.g., by a user).

[0305] The terms “send,” “transmit,” and “provide” used in this article can be used interchangeably.

[0306] The terms “receive” and “acquire” used in this article can be used interchangeably.

[0307] While the terms "a," "an," or "some" may be used repeatedly in this disclosure, this does not necessarily mean that each reference refers to the same (or multiple) examples, or that a particular feature applies only to a single example. Individual features of different examples may also be combined to provide other examples. Furthermore, when a particular feature, structure, or characteristic is described with respect to an example, whether explicitly stated or not, applying such a feature, structure, or characteristic to other examples should be within the knowledge of someone skilled in the art. It should be understood that although the terms "first," "second," etc., may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another.

[0308] It should be understood that the various network functions mentioned above (e.g., AMF, SMF, etc.) can be implemented by means of performing at least some of the functions associated with these network functions. Furthermore, means configured to implement network functions can also be configured as virtual network function instances that implement the network functions.

[0309] It should be understood that these devices may include or be coupled to other units or modules, such as radio components or radio heads for transmission and / or reception. Although these devices have been described as a single unit, different modules and memories may be implemented in one or more physical or logical entities.

[0310] It should be noted that while some examples have been described for 5G networks, similar examples can be applied to other networks and communication systems (e.g., 6G networks). Therefore, although some example references to certain example architectures for wireless networks, technologies, and standards have been described above by way of non-limiting and illustrative examples, other examples can be applied to any other suitable form of communication system besides those shown and described herein.

[0311] Furthermore, it should be noted that there are several variations and modifications that can be made to the various examples described herein without departing from the scope of this disclosure.

[0312] As used herein, “at least one of A or B”, “at least one of A and B”, and “A and / or B” all refer to (A), (B), or (A and B). For the purposes of this disclosure, “A, B, and / or C” refers to (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C).

[0313] As used herein, the term “or” means non-exclusive “or” unless otherwise stated (e.g., “otherwise” or “or in the alternative”).

[0314] As used herein, unless otherwise expressly stated, “responding to A” execution of a step does not indicate that the step is executed immediately after “A” occurs, and one or more intermediate steps may be included. Similarly, “based on A” execution of a step or function does not indicate that the step or function is executed solely based on “A”, and may be included as one or more additional conditions.

[0315] In some embodiments, at least some of the processes described herein may be performed by an apparatus including components for performing at least some of the processes. Components for performing the methods disclosed herein may include software and / or hardware components of apparatus 10. For example, at least one processor 12, memory 14, and computer program code form components for performing one or more methods disclosed herein and any embodiments. As used herein, the term “component” should be interpreted as either singular (i.e., a single element) or plural (i.e., a combination of multiple single elements). Therefore, the term “component for [performing A, B, C]” should not be interpreted as encompassing only one component for performing A, B, and C, or having multiple separate components for performing A, B, and C, or having partially or completely overlapping components for performing A, B, and C. Furthermore, the terms “component for performing A, component for performing B, component for performing C” should be interpreted as encompassing only one component for performing A, B, and C, or having multiple separate components for performing A, B, and C, or having partially or completely overlapping components for performing A, B, and C.

[0316] The scope of protection sought by the various embodiments of this disclosure is defined by the independent claims. Examples and features (if any) described in this disclosure that do not fall within the scope of the independent claims should be interpreted as examples that help to understand the various embodiments of this disclosure.

[0317] Although examples of this disclosure have been described above with reference to the accompanying drawings, it is obvious that these examples are not limited thereto, but can be modified in several ways within the scope of this disclosure. Therefore, all words and expressions should be interpreted broadly and are intended to illustrate the examples, not limit them. It should be understood that the scope of this disclosure can be implemented and adapted in various ways as technology advances. Furthermore, those skilled in the art will understand that the described examples can, but are not necessarily, combined with other examples in various ways.

Claims

1. A user equipment comprising components for: Provide an AI service discovery request message to the AIaaS function for discovering the AIaaS function; Based on the AI ​​service discovery request, receive the AI ​​service discovery response message from the AIaaS function; The AIaaS function provides an AIaaS request message for a service used to train a machine learning model in an environment where the training of the machine learning model is private to the service provider. The request message includes: Selection criteria used to indicate which one or more user devices the machine learning model will be deployed to; as well as Receive an AIaaS completion message from the AIaaS function, the AIaaS completion message indicating whether the training of the machine learning model has been performed.

2. The user equipment of claim 1, wherein the training of the machine learning model in an environment where the service provider is private comprises at least one of the following: Training of the machine learning model on a shared server implemented by the provider of the service; or Training of the machine learning model on a dedicated server implemented by the provider of the service.

3. The user equipment according to claim 1 or claim 2, wherein the user equipment includes a user equipment controller.

4. The user equipment according to claim 3, wherein the AIaaS request message further comprises at least one of the following: The machine learning model used for training; Requests for the type of artificial intelligence service available at the AIaaS function; A request for the type of training environment available at the AIaaS feature for the machine learning model; Information indicating which one or more user devices among a plurality of user devices the machine learning model is valid for; Training data used for training the machine learning model; Validation data used for validating a trained machine learning model; or Configuration requirements for training the machine learning model.

5. The user equipment according to claim 3 or claim 4, wherein the AIaaS completion message further includes information indicating at least one of the following: A trained machine learning model; The trained machine learning model has been validated; The availability of one or more inference instances of the machine learning model; The user equipment's ability to deploy the trained machine learning model to one or more other user equipment and / or application functions of the application service provider serving the user equipment; or The user equipment's ability to deploy the availability of the one or more inference instances to the one or more other user equipment and / or application functions.

6. The user equipment according to any one of claims 3 to 5, wherein the component is configured to: Provide trained machine learning models to one or more other user devices and / or to application functions of application service providers serving said user devices; or To provide the availability of the one or more inference instances to the one or more other user devices and / or to the application functions.

7. The user equipment according to any one of claims 3 to 6, wherein the component is configured to: receive an AIaaS response message acknowledging an AIaaS request message from the AIaaS function, the AIaaS response message comprising at least one of the following: Information on the start time of training for the machine learning model; or Information on the completion time of the training used for the machine learning model.

8. The user equipment according to any one of claims 3 to 7, wherein the AI ​​service discovery request message includes at least one of the following: The first identifier value indicates the application service provider currently serving the user equipment; or A second identifier value indicating the type of the user equipment.

9. The user equipment according to any one of claims 3 to 8, wherein the AI ​​service discovery response message includes at least one of the following: One or more AI service types available at the AIaaS functionality; or One or more environment types available at the AIaaS function for training the machine learning model.

10. A communication apparatus comprising components for: Receive AI service discovery requests from user devices to discover AI as a Service (AIaaS) functionalities. Based on the AI ​​service discovery request, an AI service discovery response message is provided to the user equipment; The user equipment receives an AIaaS request message for a service used to train a machine learning model in an environment where the training of the machine learning model is private to the service provider, the request message including: Selection criteria used to indicate which one or more user devices the machine learning model will be deployed to; as well as The user equipment is provided with an AIaaS completion message, which indicates whether the training of the machine learning model has been performed.

11. The apparatus of claim 10, wherein the component is configured to provide an AI training request fulfillment requirement message to an AIaaS controller for requesting resources for training the machine learning model based on the AIaaS request message received from the user equipment.

12. The apparatus of claim 10 or claim 11, wherein the component is configured to: receive a training completion message from an AIaaS controller, the training completion message indicating at least one of the following: Has the training of the machine learning model been performed? A trained machine learning model; The availability of one or more inference instances of the machine learning model; or The location where the trained machine learning model is stored.

13. The apparatus according to any one of claims 10 to 12: wherein the component is configured to verify a machine learning model that has been trained by the AIaaS controller.

14. A communication apparatus comprising components for: The AIaaS (Artificial Intelligence as a Service) functionality receives an AI training request fulfillment message for training a machine learning model in an environment where the training of the machine learning model is private to the service provider. The AI ​​training request fulfillment message includes: Selection criteria used to indicate which one or more user devices the machine learning model will be deployed to; as well as Provide the AIaaS function with a complete training message, which indicates whether the training of the machine learning model has been performed.

15. The apparatus of claim 14, wherein the component is configured for at least one of: Allocate the resources used for training the machine learning model to one or more network servers; or Perform training on the machine learning model.

Images