Managing machine learning models

Abstract

The invention relates to managing such a machine learning model. In an example of the present disclosure, a user equipment (UE) is configured to receive a first message from a network, where the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model. The UE may then perform training and / or validation of the machine learning model using the received instructions, and prepare a second message to be sent to the network, where the second message includes information indicative of a result of the training and / or validation of the machine learning model.

Description

Technical Field

[0001] Examples of the present disclosure relate to managing machine learning models. Some examples relate to managing machine learning models in communication networks. Background Art

[0002] The machine learning model can be used in the communication network to perform various suitable tasks. For example, the machine learning model can be used to assist in handover-related decisions, such as predicting which cell the UE should handover to, predicting when the UE should handover, predicting the trajectory information of the UE, and / or any other suitable decisions and / or information. Summary of the invention

[0003] According to various but not necessarily all examples of the present disclosure, there is provided a user equipment (UE), including components for the following operations: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of the machine learning model; Using the received instructions to train and / or validate the machine learning model; and Prepare a second message to be sent to the network, wherein the second message includes information indicating results of training and / or validation of the machine learning model.

[0004] The first message may include at least one of the following: an indication of the parameters to be used to train the machine learning model; an indication of a training rule to be used to train the machine learning model; An indication of the validation rules to be used to validate the machine learning model; An indication of the exit criteria for the validation process for the machine learning model.

[0005] Rules to be used for the machine learning model can be configured to control one or more results of the machine learning model to be network-preferred results.

[0006] This component can be used to determine whether the validation of the machine learning model has been successful.

[0007] The second message may include an indication of whether validation of the machine learning model has been successful.

[0008] The component may be configured to receive a third message from the network, wherein the third message includes instructions indicating how the UE should perform reasoning using the trained machine learning model.

[0009] A third message may be received in the event that the second message indicates that verification of the machine learning has been successful.

[0010] The third message may include at least one of the following: an indication of parameters to be used for the trained machine learning model during inference; An indication of an inference rule to be used by the machine learning model during inference; Indications for counters inferred by machine learning models; an indication of the accuracy level of the revalidation for the trained machine learning model; An indication of the validation threshold for the trained machine learning model.

[0011] This component can be used to enable the UE to be configured into non-machine learning mode if validation of the machine learning model is not successfully performed.

[0012] The component can be used to receive an updated first message from the network, wherein the updated first message includes updated instructions indicating how the UE should perform training and / or validation of the machine learning model.

[0013] An updated first message may be received in which case the second message indicates that validation of the machine learning model has not been successful.

[0014] According to various but not necessarily all examples of the present disclosure, there is provided a method comprising: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of the machine learning model; Using the received instructions to train and / or validate the machine learning model; and Prepare a second message to be sent to the network, wherein the second message includes information indicating results of training and / or validation of the machine learning model.

[0015] According to various but not necessarily all examples of the present disclosure, there is provided a computer program comprising computer program instructions which, when executed by a processing circuit, cause: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of the machine learning model; Using the received instructions to train and / or validate the machine learning model; and Prepare a second message to be sent to the network, wherein the second message includes information indicating results of training and / or validation of the machine learning model.

[0016] According to various but not necessarily all examples of the present disclosure, there is provided a network device including components for the following operations: enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; and A second message is received from the UE, wherein the second message includes information indicating results of training and / or validation of the machine learning model by the UE.

[0017] This component can be used to implement further control of the machine learning model based on the results of training and / or validation of the machine learning model indicated by the UE.

[0018] The first message may include at least one of the following: an indication of the parameters to be used to train the machine learning model; an indication of a training rule to be used to train the machine learning model; An indication of the validation rules to be used to validate the machine learning model; An indication of the exit criteria for the validation process for the machine learning model.

[0019] Rules to be used for the machine learning model can be configured to control one or more results of the machine learning model to be network-preferred results.

[0020] The second message may include an indication of whether validation of the machine learning model has been successful.

[0021] The component can be used to enable a third message to be sent to the UE, wherein the third message includes instructions indicating how the UE should use the trained machine learning model for reasoning, wherein the third message is sent if the second message indicates that verification of the machine learning model has been successful.

[0022] The third message may include at least one of the following: an indication of parameters to be used for the trained machine learning model during inference; An indication of an inference rule to be used by the machine learning model during inference; Indications for counters inferred by machine learning models; an indication of the accuracy level of the revalidation for the trained machine learning model; An indication of the validation threshold for the trained machine learning model.

[0023] This component can be used to control the UE to be configured into non-machine learning mode if the verification of the machine learning model is not successfully performed.

[0024] The component can be used to update instructions indicating how the UE should perform training and / or validation of the machine learning model in response to one or more messages from the UE, and enable the updated instructions to be sent to the UE.

[0025] Instructions indicating how the UE should perform training and / or validation may be updated in response to receiving a second message by the UE indicating that validation of the machine learning has not been successful.

[0026] According to various but not necessarily all examples of the present disclosure, there is provided a method comprising: enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; and A second message is received from the UE, wherein the second message includes information indicating results of training and / or validation of the machine learning model by the UE.

[0027] According to various but not necessarily all examples of the present disclosure, there is provided a computer program comprising computer program instructions which, when executed by a processing circuit, cause; enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; and A second message is received from the UE, wherein the second message includes information indicating results of training and / or validation of the machine learning model by the UE.

[0028] Although the above examples and optional features of the present disclosure are described separately, it should be understood that their provision in all possible combinations and arrangements is included in the present disclosure. It should be understood that various examples of the present disclosure may include any or all features described with respect to other examples of the present disclosure, and vice versa. In addition, it should be understood that any one or more or all features in any combination may be implemented / included / may be executed by a device, method and / or computer program instruction as needed and appropriately by a device, method and / or computer program instruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Some examples will now be described with reference to the accompanying drawings, in which:

[0030] Figure 1 An example network is shown;

[0031] Figure 2 An example functional framework is shown;

[0032] Figure 3A and 3B An example method is shown;

[0033] Figure 4 An example method is shown;

[0034] Figure 5 An example signaling diagram is shown; and

[0035] Figure 6 An example controller is shown.

[0036] The drawings are not necessarily to scale. For the sake of clarity and brevity, certain features and views of the drawings may be schematically shown or enlarged to scale. For example, the size of some elements in the drawings may be enlarged relative to other elements to aid illustration. Corresponding reference numerals are used in the drawings to indicate corresponding features. For the sake of clarity, all reference numerals may not necessarily be shown in all drawings. definition AMF Access and Mobility Management Function gNB NR Base Station NR New Radio O-RAN Open Radio Access Network RAN Radio Access Network RSRP Reference Signal Received Power SIM Subscriber Identity Module UE User Equipment UL Uplink ULSRS Uplink Sounding Reference Signal UPF User Plane Function DETAILED DESCRIPTION

[0037] Figure 1 An example of a network 100 including multiple network entities is shown, including a terminal device 110, a node device 120, and one or more network devices 130. The terminal device 110 and the node device 120 communicate with each other. The one or more network devices 130 communicate with the access node 120. In some examples, the one or more network devices 130 communicate with the terminal device 110.

[0038] In some examples, one or more network devices 130 may communicate with each other. In some examples, one or more node devices 120 may communicate with each other.

[0039] The network 100 may be a cellular network comprising a plurality of cells 122, each cell being served by a node device 120. In this example, the interface between the terminal device 110 and the node device 120 that defines the cells 122 is a wireless interface 124.

[0040] The node device 120 comprises one or more cellular radio transceivers. The terminal device 110 comprises one or more cellular radio transceivers.

[0041] In the example shown, the cellular network 100 is a 3rd Generation Partnership Project (3GPP) network, wherein the terminal device 110 is a user equipment (UE) and the node device 120 may be an access node such as a base station.

[0042] The term "user device" is used to designate a mobile device that includes a smart card, such as a Subscriber Identity Module (SIM), for authentication / encryption, etc. In some examples, the term "user device" is used to designate a mobile device that includes circuitry embedded as part of the user device for authentication / encryption, such as a software SIM.

[0043] The node device 120 may be any suitable base station. A base station is an access node. It may be a network element that is responsible for radio transmission to or radio reception from the UE 110 in one or more cells. The node device 120 may be a network element in a radio access network (RAN), an open radio access network (O-RAN), or any other suitable type of network.

[0044] Network device 130 may be part of a core network. Network device 130 may be configured to manage functions related to the connection of UE 110. For example, network device 130 may be configured to manage functions such as connectivity, mobility, authentication, authorization, and / or other suitable functions. In some examples, network device 130 may include an access and mobility management function (AMF) and / or a user plane function (UPF) or any other suitable entity.

[0045] exist Figure 1 In the example of FIG. 1 , network device 130 is shown as a single entity. In some examples, network device 130 can be distributed across multiple entities. For example, network device 130 can be cloud-based or distributed in any other suitable manner.

[0046] For example, the network 100 may be a 4G or 5G network. For example, the network 100 may be a new radio (NR) network using a gNB or an eNB as an access node 120. New radio is the 3GPP name for 5G technology. In this case, the node device 120 may include a gNodeB (gNB) 120, which is configured to provide user plane and control plane protocol termination to the UE 110 and / or perform any other suitable functions. The gNBs 120 are interconnected to each other via an X2 / Xn interface 126. The gNBs are also connected to a network device 130 via an N2 interface 128. The gNB may be connected to an AMF or any other suitable network device 130. In other examples, other types of networks and interfaces may be used. Other types of networks may include next generation mobile and communication networks, such as 6G networks.

[0047] In some examples, network 100 can utilize machine learning models. For example, machine learning models can be used to implement processes such as predicting cells for handover, beam selection for transmission and / or reception, and / or any other suitable process.

[0048] Examples of the present disclosure relate to managing such machine learning models. In examples of the present disclosure, the machine learning model may be implemented by UE 110, but may be at least partially controlled by network device 130. Enabling network device 130 to maintain some control over the machine learning model implemented by UE 110 may provide a technical effect of keeping the performance of the machine learning model at UE 110 consistent with the requirements of network 100. This may prevent UE 110 from being biased toward results that are not expected or preferred by network 100.

[0049] Figure 2 An example functional framework 200 that can be used for RAN intelligence is shown. The functional framework 200 utilizes a machine learning model. The functional framework 200 can be used in a system such as Figure 1 The network is implemented within the network 100.

[0050] exist Figure 2 In the example of , the functional framework 200 includes a data collection box 202, a machine learning model training box 204, a machine learning model reasoning box 206, and an executor box 208. Other boxes and / or arrangements of boxes may be used in examples of the present disclosure.

[0051] The data collection block 202 is configured to collect data to be used by the machine learning model. The data collected by the data collection block 202 can be used as input to the machine learning model. The data collected by the data collection block 202 can be used by the machine learning model for training and / or reasoning.

[0052] The data collected by the data collection block 202 may include measurements from the UE 110, measurements from other network entities, feedback 220 from the performer block 208 output from a machine learning model, and / or any other suitable data.

[0053] The data collection block 202 may be configured to provide different outputs including different types of data. Figure 2 In the example of , the data collection block 202 provides an output including training data 210. The training data 210 is provided as input to the machine learning model training block 204. Figure 2 In the example of , the data collection block 202 also provides outputs including inference data 212. The inference data 212 is provided as input to the machine learning model inference block 206.

[0054] The data provided by the data collection block 202 may be provided in any suitable format. The data provided by the data collection block 202 need not be provided in a format specific to the machine learning model. For example, processes such as data preprocessing and cleaning, formatting, transformation, or other suitable processes need not be performed by the data collection block 202. These processes may be performed by other blocks in the framework 200.

[0055] The machine learning model training block 204 is configured to perform training and validation of the machine learning model. Validation may include testing and / or generating performance metrics of the machine learning model.

[0056] The machine learning model training box 204 is configured to receive training data 210 as input. In some examples, the machine learning model training box 204 can be configured to perform a process of preparing training data for use by the machine learning model. The process can include data preprocessing and cleaning, formatting, transformation, or other suitable processes.

[0057] The machine learning model training block 204 is configured to provide a model deployment output 214. The model deployment output 214 may be used to deploy the trained and validated machine learning model to the machine learning model inference block 206, or to deliver an updated machine learning model to the machine learning model inference block 206. The machine learning model training block 204 may also receive performance feedback 216 from the machine learning model inference block 206. The performance feedback 216 may include information about the performance of the machine learning model during the inference process.

[0058] The machine learning model inference block 206 is configured to perform inference using the machine learning model. The machine learning model inference block 206 may be configured to process the data collected by the data collection block 202 using the machine learning model to provide output 218. The output 218 may include a prediction or a decision.

[0059] The machine learning model inference block 206 is configured to receive inference data 212 as input. In some examples, the machine learning model inference block 206 can be configured to perform a process of preparing inference data for use by the machine learning model. The process can include data preprocessing and cleaning, formatting, transformation, or other suitable processes.

[0060] The machine learning model inference box 206 can also provide performance feedback 216 to the machine learning model training box 204.

[0061] The output 218 of the machine learning model reasoning block 206 is provided to an actor block 208. The actor block may be configured to perform or trigger actions based on the output 218 of the machine learning model reasoning block 206. The actor block 208 may perform functions itself and / or may trigger other entities to perform actions.

[0062] The performer block 208 may provide feedback 220 as an output. The feedback 220 may be provided to the data collection block 202. The feedback may include information that may be used to derive training data, inference data, or monitor the performance of the machine learning model and its impact on the network 100. The performance of the machine learning model may be monitored by updating key performance indicators and performance counters.

[0063] In examples of the present disclosure, the machine learning model may include a trainable model, such as a neural network (such as, but not limited to, a convolutional neural network, a long short-term memory (LSTM)), or any other suitable type of trainable model. The term "machine learning model" refers to artificial intelligence (AI), intelligence, or any other type of method that is trainable or tunable using data. The machine learning model may include a computer program. The machine learning model can be trained to perform a task, such as predicting which base station to switch to, without being explicitly programmed to perform the task. If the performance of the machine learning model at a task in T (as measured by P) improves as experience E increases, the machine learning model can be configured to learn about certain classes of tasks T and performance measures P from experience E. These example machine learning models can learn from previous channel metrics and configurations for the same or similar channel conditions. The machine learning model can also be a trainable computer program. Other types of machine learning models may be used in other examples.

[0064] Any suitable process may be used to train the machine learning model. Training of the machine learning model may be performed using real world and / or simulated data. In some examples, the real world data used to train the machine learning model may include data collected by UE 110 and / or any other suitable network entity. In some examples, the output of the machine learning model may be based at least in part on simulated data. An example of real world data in a mobile communication network may include a reference signal received power (RSRP) measured by UE 110 and reported to the network for a group of cells over a period of time. Another example may be an uplink sounding reference signal (ULSRS) measured by the network 100 due to UE transmissions to determine the uplink (UL) channel quality. Such data may be used as training data to train a machine learning model that learns to predict the switching of a mobile UE 110 when the mobile UE 110 moves from one cell to another.

[0065] The training data used to train the machine learning model may include historical data collected from the network 100 and / or real-time data collected by the UE 110 and the network. The training of the machine learning model may be repeated as appropriate until the machine learning model has achieved a sufficient level of stability. The machine learning model has a sufficient level of stability when the fluctuations in the predictions provided by the machine learning model are low enough that the machine learning model provides a response consistent with the test input. The stability of the machine learning model may be tested during the validation process.

[0066] In some examples, training of the machine learning model may be repeated as appropriate until one or more parameters of the output have reached a predefined threshold and / or until a predefined accuracy has been achieved and / or until any other suitable criteria are met.

[0067] The machine learning model can be executed using any suitable device, such as a CPU, GPU, ASIC, FPGA, in-memory computing, analog or digital or optical devices. The machine learning model can also be executed in a device that combines features from any number of these, such as a digital-optical or analog-digital mixer. In some examples, the weights and required calculations in these systems can be programmed to correspond to the machine learning model.

[0068] Figure 3A and 3B Methods are shown that may be performed by UE 110 and network devices or any other suitable entities.

[0069] Figure 3A 1 shows an example method that may be performed by UE 110. UE 110 may perform the following steps: Figure 1 In the illustrated network 100, or any other suitable type of network, UE 110 in these examples has machine learning capabilities.

[0070] The method includes receiving a first message from the network 100 at block 300. The first message includes instructions indicating how the UE 110 should perform training and / or validation of a machine learning model.

[0071] The first message may be sent from any suitable entity within network 100. In some examples, the first message may be sent by network device 130.

[0072] The first message may include any information that may be used to control the training and / or validation of the machine learning model at the UE 110. The training of the machine learning model may include determining parameters that should be used for the machine learning model. The validation of the machine learning model may include evaluating whether the training has been adequately performed. The validation may include using a test data set to determine whether the trained machine learning model generates the expected results with a sufficient level of reliability or may include any other suitable evaluation.

[0073] In some examples, the first message may include an indication of parameters to be used to train the machine learning model. The parameters may relate to the internal structure of the machine learning model. For example, if the machine learning model includes a neural network, the parameters may include sampling weights for input data of the neural network. Other types of machine learning models and / or parameters may be used.

[0074] The parameters included in the first message may include learnable parameters that can be adjusted when the machine learning model is trained. The parameters included in the first message may include hyperparameters that are not adjusted when the machine learning model is trained.

[0075] In some examples, the first message may include an indication of training rules to be used to train the machine learning model. The training rules may include instructions for training the machine learning model to be followed by UE 110. The training rules may include instructions for how the training data should be adjusted, how parameters should be adjusted, when to stop training and perform validation, and / or any other suitable rules.

[0076] In some examples, the first message may include an indication of a validation rule to be used to validate the machine learning model. The validation rule may include instructions for validation of the machine learning model to be followed by UE 110. The validation rule may include an indication of the data to be used for validation, where the data for validation should be obtained from, when the validation should be performed, and / or any other suitable rules.

[0077] In some examples, the first message may include an indication of an exit condition for the validation process for the machine learning model. The exit condition includes any requirements that may be met to exit the validation process. The exit condition may include a time threshold, a limit on the number of times the validation process may be performed, and / or any other suitable condition.

[0078] In some examples, the rules to be used for the machine learning model are configured to control one or more results of the machine learning model to be results preferred by the network 100. For example, the training rules and / or validation rules can be configured to bias the results of the machine learning model toward the results preferred by the network 100. As an example, if the machine learning model is used to predict a cell for handover, the training rules sent by the network 100 to the UE 110 may be biased toward a specific cell or set of cells, or may not be biased toward a specific cell or set of cells or beams within a given cell or set of cells. This control can help the network 100 manage flexible load distribution across cells / beams or perform any other suitable function, such as efficient resource scheduling for achieving better power savings.

[0079] At box 302, the method includes using the received instructions to train and / or validate the machine learning model.

[0080] UE 110 may be configured to determine whether validation of the machine learning model has succeeded. Requirements for determining whether validation has succeeded may be received in a first message by UE 110. The requirements for determining whether validation has succeeded may include a required level of accuracy for the results of the machine learning model, a time threshold for expiration, and / or any other suitable requirements.

[0081] At block 304, the method includes preparing a second message to be sent to the network 100. The message may be transmitted from the UE 110 to the network 100.

[0082] The second message includes information indicating the results of the training and / or validation of the machine learning model. In some examples, the second message includes an indication of whether the validation of the machine learning model has been successful.

[0083] Figure 3A The method shown in shows how the training and / or validation of the machine learning model UE 110 can be at least partially controlled by the network 100. In some examples, the UE 110 can also be configured so that the network 100 can at least partially control the inference process of the machine learning model at the UE 110. The inference process can be performed after the validation of the machine learning model has been successful. In an example of the present disclosure, the inference process can be performed after the network 100 receives validation feedback from the UE 110.

[0084] In order to enable the network 100 to at least partially control the inference process of the machine learning model at the UE 110, the UE 110 may be configured to receive a third message from the network 100. The third message may include instructions indicating how the UE 110 should perform inference using the trained machine learning model.

[0085] The third message may be received in the event that the second message sent by UE 110 indicates that verification of the machine learning has been successful. For example, the second message may indicate that a required level of accuracy has been achieved. In this case, the third message may be received in response to the indication that verification has been successful.

[0086] The third message may be sent from any suitable entity within network 100. In some examples, the third message may be sent from network device 130. The third message may be sent from the same network device 130 that sent the first message.

[0087] The third message may include any information that may be used to control the inference of the machine learning model at UE 110. The inference may include processing the acquired data to obtain one or more outputs. The outputs may be used to determine the actions to be taken by UE 110 and / or network 100.

[0088] In some examples, the third message may include an indication of parameters to be used for the trained machine learning model during inference. The parameters may include an indication of the machine learning model to be used, a desired accuracy of the output of the machine learning model, and / or any other suitable parameters.

[0089] In some examples, the third message may include an indication of an inference rule to be used by the machine learning model during inference. The inference rule may include instructions for inference of the machine learning model to be followed by UE 110. The inference rule may include an indication of input data to be obtained, updated requirements for revalidation of the machine learning model, and / or any other suitable rules.

[0090] In some examples, the third message may include an indication of a counter inferred by the machine learning model. The counter may set a time limit for when the machine learning model is considered valid. The counter may take into account changes outside of UE 110. For example, the environment around UE 110 may be changing. The change in the environment may be a change in the location of UE 110 and / or a change in the location of other UEs 110 around UE 110.

[0091] In some examples, the third message may include an indication of an accuracy level for a revalidation of the trained machine learning model. For example, an inference rule may indicate that a revalidation of the machine learning model is performed by UE 110 such that if the accuracy level of the machine learning model drops below a desired threshold, this may be reported back to network 100. The accuracy level of the machine learning model may change due to changes in the environment of UE 110 and / or changes in any other suitable factors.

[0092] In some examples, the third message may include an indication of a validation threshold for the trained machine learning model. The validation threshold may include a time limit, an accuracy requirement, and / or any other suitable validation requirement.

[0093] In some cases, validation of the machine learning model may not be successful. For example, a desired level of accuracy may not be achieved by the machine learning model, or a time threshold may be exceeded before the desired level of accuracy is achieved. In this case, if validation of the machine learning model is not successfully performed, the UE 110 may be configured for a non-machine learning mode. The switch from the machine learning mode of operation to the non-machine learning mode of operation may be at least partially controlled by the network 100. For example, the UE 110 may receive a message from the network 100 indicating that the UE 110 is to switch from the machine learning mode of operation to the non-machine learning mode of operation.

[0094] In some cases, if the validation of the machine learning model is unsuccessful, the UE 110 may be configured to update the training and / or validation process. For example, the UE 110 may be configured to receive an updated first message from the network 100. The updated first message may include updated instructions indicating how the UE 110 should perform the training and / or validation of the machine learning model. For example, the updated first message may include updated training rules, updated validation rules, and / or any other suitable information. The updated instructions may enable adjustments to the training and / or validation of the machine learning model to be implemented. In the event that the second message indicates that the validation of the machine learning model has not been successful, the updated first message may be received.

[0095] Figure 3B Shown with Figure 3A The method corresponding to the method. Figure 3B The example method may be performed by network device 130. Network device 130 may be any suitable device configured to control the functions of network 100. Network device 130 may be, for example, Figure 1 A single entity is shown, or may be distributed over multiple different entities.

[0096] At block 306, the method includes enabling a first message to be sent to UE 110. The first message may be sent from network device 130 to UE 110 via any suitable means. For example, base station 120 or other access node may be used to transmit the first message from network 100 to UE 110.

[0097] The first message includes instructions indicating how UE 110 should perform training and / or validation of the machine learning model. The instructions indicating how UE 110 should perform training and / or validation may be as described above and / or may include any other suitable instructions.

[0098] At block 308, the method includes receiving a second message. The second message may originate from UE 110. The second message may be sent from UE 110 to network device 130 via any suitable means.

[0099] The second message may include information indicating the results of training and / or validation of the machine learning model by UE 110. The information indicating the results of training and / or validation of the machine learning model by UE 110 may be as described above and / or may include any other suitable information.

[0100] Figure 3B The illustrated method shows how network device 130 at least partially controls the training and / or validation of the machine learning model by UE 110. In some examples, network device 130 may also be configured such that network device 130 may at least partially control the inference process of the machine learning model at UE 110. For example, network device 130 may be configured to implement further control of the machine learning model based on an indication result of the training and / or validation of the machine learning model by UE 110.

[0101] In some examples, network device 130 can be configured to enable a third message to be sent to UE 110 to enable network device 130 to at least partially control the inference process of the machine learning model at UE 110. The third message can include instructions indicating how UE 110 should perform inference using the trained machine learning model. The third message is sent if the second message indicates that validation of the machine learning model has succeeded.

[0102] In some examples, validation of the machine learning model may not be successful. In this case, the second message may include an indication that validation of the machine learning model has not been successful.

[0103] If the second message indicates that the validation of the machine learning model has not been successful, then in some cases, network device 130 may send a message to UE 110 indicating that UE 110 will switch from the machine learning mode of operation to the non-machine learning mode of operation. Then, based on the information in the message, UE 110 may be reconfigured to the non-machine learning mode of operation.

[0104] In some cases, if the validation of the machine learning model is unsuccessful, the network device 103 may update the training and / or validation process to be used by the UE 110. For example, the network device 103 may be configured to send an updated first message to the UE 110. The updated first message may include updated instructions indicating how the UE 110 should perform the training and / or validation of the machine learning model. For example, the updated first message may include updated training rules, updated validation rules, and / or any other suitable information. The updated instructions may enable adjustments to the training and / or validation of the machine learning model to be implemented. In the event that the second message indicates that the validation of the machine learning model has not been successful, the updated first message may be sent.

[0105] Figure 4 An example method according to an example of the present disclosure is shown. The method may be implemented by UE 110 or any other suitable entity utilizing network 100.

[0106] At block 400, a machine learning model is selected. In some examples, UE 110 may have the capability to run different machine learning models.

[0107] UE 110 may be configured in a non-machine learning mode or a machine learning mode. Figure 4 In the example of FIG. 4 , UE 110 is configured in a non-machine learning mode at block 402, or UE 110 is configured in a machine learning mode at block 406. Whether UE 110 is configured in the machine learning mode or the non-machine learning mode may be controlled by network 100. For example, network 100 may send one or more messages to UE 110 indicating whether UE 110 should be configured in the machine learning mode or the non-machine learning mode.

[0108] If UE 110 is configured in a non-machine learning mode, UE 110 does not perform any machine learning-assisted operations. Operations performed by UE 110 may be implemented using any suitable process that does not involve machine learning.

[0109] In non-machine learning mode, UE 110 may be configured to collect data that may be reported to other network entities to enable other network entities to perform machine learning operations. In non-machine learning mode, UE 110 will not process the collected data to make decisions or predict results.

[0110] exist Figure 4In the example of , at block 404, the method includes making measurements and reporting results of the measurements to the network 100. The measurements made by the UE 110 may include measurements related to one or more metrics of the environment in which the UE 110 is located. For example, the measurements may include reference signal received power (RSRP) measurements or any other suitable type of measurements.

[0111] Once the measurements are reported to the network 100, the network 100 can use the data from the measurements to make decisions or predict outcomes. The network 100 can do so using a machine learning model or any other suitable process.

[0112] If the UE 110 is configured in a machine learning mode, the UE 110 may perform machine learning assisted operations. In the machine learning mode, the UE 110 may be configured to collect data and process the collected data using a machine learning model to make a decision or predict an outcome. The machine learning assisted operations may be at least partially controlled by the network 100.

[0113] exist Figure 4 In an example of , at block 408, the method includes UE 110 taking measurements and saving data collected from the measurements for use in machine learning operations. The measurements may include measurements related to one or more metrics of the environment in which UE 110 is located. For example, the measurements may include reference signal received power (RSRP) measurements or any other suitable type of measurements.

[0114] At block 410, a determination is made as to whether sufficient data has been collected. The amount of data required may be determined by network 100. For example, network 100 may indicate to UE 110 the type and / or amount of data required to perform the machine learning operation. Network 100 may provide information such as Figure 3A and 3B This information is indicated to UE 110 in the first message shown or by any other suitable means.

[0115] If it is determined that sufficient data has not been collected, the method returns to block 408 and further measurements are taken by the UE 110 .

[0116] If it is determined that sufficient data has been collected, the method proceeds to block 412 where training and validation of the machine learning model is performed. The training and validation may be performed by the UE 110 according to the rules and policies set by the network 100. Figure 3A and 3B As shown, network 100 may inform UE 110 of these rules and policies in a first message. The first message may be sent when the machine learning mode is selected, or may be sent at any other suitable time.

[0117] If validation of the machine learning model determines that the machine learning model is valid, the method proceeds to box 414 and the machine learning model is used for reasoning by UE 110.

[0118] The inference may be performed by UE 110 according to rules and policies set by network 100. Network 100 may inform UE 110 of the rules and policies to be used for inference by sending a message to UE 110. The message may be sent in response to an indication from UE 110 that validation of the machine learning model has been successful, or the message may be sent at any other suitable time.

[0119] At block 416, the method also includes performing inference validation. Inference validation may assess whether the machine learning model is still valid during the inference phase. Rules and policies for determining whether the machine learning model is still valid during the inference phase may also be set by the network 100. These may be transmitted from the network 100 to the UE 110 in any suitable message.

[0120] At block 416, if the machine learning model is determined to be invalid, the method proceeds to block 418 and network policy is followed. The machine learning model may be determined to be invalid during the training phase at block 412 or during the inference phase at block 416. In this case, an indication that the machine learning model is invalid may be sent from UE 110 to network 100, and network 100 may make a decision on how to proceed based on network policy.

[0121] In this example, the network policy may require UE 110 to revert to non-machine learning mode or require machine learning policies and / or rules to be updated. If the network policy requires UE 110 to revert to non-machine learning mode, the method proceeds to block 402 and the non-machine learning mode is implemented.

[0122] If the network policy requires that the machine learning policies and / or rules be updated, the method proceeds to block 406 and the machine learning model is implemented again, but with updated rules and / or policies. To enable the machine learning to be updated, the network 100 may send a message to the UE 110 indicating the new policies and rules for training and validation of the machine learning model.

[0123] Figure 5 An example signaling diagram that may be used in some examples of the present disclosure is shown. In this example, signals are exchanged between UE 110 and network 100. Signals may be exchanged between UE 110 and any suitable portion of network 100, such as network device 130.

[0124] In this signal diagram, it is assumed that relevant information about UE 110 has been obtained by network 100. That is, network 100 has pre-acquired knowledge about UE 110. This knowledge may include the model type of UE 110, the machine learning capability of UE 110, or any other suitable information.

[0125] At box 500, the network 100 sends a first message 500 to the UE 110. The first message includes instructions indicating how the UE 110 should perform training and / or validation of the machine learning model. The instructions may include sampling weights for data for the machine learning model and / or any other suitable rules or instructions. In some examples, the weights for the machine learning model may be biased toward specific results. For example, if the machine learning model is used to predict handovers, the sampling weights for the input data may be biased toward preferred cells or beams for load balancing purposes. Other biases may be used in other examples. Other reasons may be used to bias or control the results of the machine learning model.

[0126] exist Figure 5 In the example of , the first message includes instructions indicating how the UE should perform training and / or validation of the machine learning model, but does not include any data to be used for training and / or validation. In this example, UE 110 is trusted to use its own data for training and / or validation. UE 110 can obtain data for training and / or validation by performing measurements and / or by any other suitable means.

[0127] At block 502, UE 110 performs training of the machine learning model. The training of the machine learning model is performed using data collected by UE 110 and rules and policies provided by network 100 in the first message. The training of the machine learning model may include determining sampling weights for input data or other parameters that should be used for the machine learning model. In addition to providing the rules and policies, network 100 does not perform any training of the machine learning model.

[0128] At block 504, validation of the machine learning model is performed. Validation of the machine learning model may include evaluating whether training has been adequately performed. Validation may include using a test data set to determine whether the trained machine learning model generates expected results with a sufficient level of reliability or may include any other suitable evaluation.

[0129] Validation is performed by the UE 110 according to the rules and policies that have been provided by the network 100. In addition to providing the rules and policies, the network 100 does not perform any validation of the machine learning model. For example, the network 100 may indicate a level of accuracy to be achieved by the machine learning model, but the UE 110 determines whether the level of accuracy is achieved. The level of accuracy may be a percentage such as 95%, or may include any other metric.

[0130] If the verification fails, the method proceeds to block 506, and the UE 110 counts the number of iterations of training and verification performed by the UE 110. The network 100 may indicate a threshold number of iterations to be performed. This indicates the maximum number of iterations to be performed before exiting the training and verification process. Other exit conditions for exiting the verification phase may be used in other examples, for example, a time threshold may be used instead of counting the number of iterations. The exit condition may be sent from the network 100 to the UE 110 in the first message at block 500 or in any other suitable message.

[0131] At block 506 , if it is determined that the threshold number of iterations has not been reached, the method returns to block 502 and training and validation are repeated.

[0132] At block 506, if it is determined that the threshold number of iterations has been reached, then at block 508, UE 110 sends a message to network 100. The message sent to network 100 at block 508 may be a second message. The second message indicates the result of the verification. In this case, the second message indicates that the verification has not been successful.

[0133] At block 510, network 100 sends another message to UE 110. The other message may be sent in response to the second message received at block 508 indicating that validation has not been successful. In this example, the other message is an updated first message. The updated first message may include updated instructions indicating how UE 110 should perform training and / or validation of the machine learning model. For example, the updated first message may include updated training rules, updated validation rules, updated accuracy levels for validation, and / or any other suitable information.

[0134] At box 512, UE 110 receives updated training and / or validation rules from network 100. The updated rules and policies may indicate whether UE 110 should continue to be in machine learning mode. After the updated training and / or validation rules are received, the method proceeds to box 532, and UE 110 determines whether UE 110 should remain in machine learning mode or whether UE 110 should switch to non-machine learning mode based on the information in the updated message. If the updated message indicates that UE 110 will remain in machine learning mode, the method returns to box 502, and training and validation are repeated using the updated rules and policies. If the updated message indicates that UE 110 will switch to non-machine learning mode, the method proceeds to box 534, and UE 110 stops using the machine learning model.

[0135] exist Figure 5 In the example of , the network 100 may decide whether the UE 110 continues to be in the machine learning mode or whether it should switch to the non-machine learning mode. For example, if the verification shows that the accuracy level of the machine learning model is close to the required level, the network 100 may decide that the UE 110 will continue to be in the machine learning mode with the adjusted parameters. The adjusted parameters may be changes to the sampling weights of the input data to the machine learning model, changes to the required level of accuracy, and / or any other suitable metric. If the verification shows that the accuracy level of the machine learning model is not close to the required level, the network 100 may decide that a sufficient level of accuracy is unlikely to be achieved by the UE 110, and may decide to switch the UE 110 to the non-machine learning mode.

[0136] At block 504, if the verification is passed, the method proceeds to block 514 and UE 110 sends a message to network 100. The message sent to network 100 at block 514 may be a second message. The second message indicates the result of the verification. In this case, the second message indicates that the verification has succeeded.

[0137] After receiving the message indicating that the validation has been successful, at block 516, the network 100 prepares inference rules. The inference rules may include instructions to be followed by the UE 110 during the inference phase of the machine learning model. The inference rules may include input data to be obtained, requirements for revalidation of the machine learning model, and / or indications of any other suitable rules. At block 514, the inference rules may be prepared based on the validation feedback received in the message.

[0138] At block 518, the network 100 sends a message to the UE 110. The message includes an indication of the inference rule. The message may be the third message as described above, or may be any other suitable message.

[0139] At block 520, UE 110 performs inference using the trained machine learning model. Figure 5 In some examples, UE 110 performs reasoning using data collected by UE 110. UE 110 may perform reasoning using rules and policies provided by network 100. In some examples, UE 110 may also perform reasoning using data collected from other network entities.

[0140] During inference, UE 110 may use a machine learning model to process input data to provide an output. The output may be used to predict an outcome or make a decision. The output may be used to control an actuator within UE 110 and / or to control an actuator within network 100. For example, inference may be used to predict a cell to use for handover, for beam selection, or for any other suitable purpose.

[0141] At box 520, validation is also performed during the reasoning phase. The validation performed during the reasoning phase can be used to evaluate whether the machine learning model is still valid. The validity of the machine learning model may change due to environmental changes around the UE 110 and / or due to any other suitable factors. In some examples, the validation performed during reasoning can include determining whether the machine learning model has achieved a required level of accuracy or meets some other metric. In some examples, the validation performed during reasoning can include determining whether some external criteria have been met, for example, it can be determined whether a time threshold has been reached. The time threshold can be set by the network 100. The time threshold can be indicated to the UE 110 in the message sent at box 518, or it can be indicated using any other suitable manner.

[0142] At box 522, if it is determined that the machine learning model is still valid, the method returns to box 520 and reasoning continues.

[0143] At block 522, if it is determined that the machine learning model is invalid or the inference counter expires, the method proceeds to block 524 and a message is sent from UE 110 to network 100. The message indicates that the machine learning model is no longer valid. In some examples, the message may also indicate why the machine learning model is no longer valid.

[0144] After receiving the message indicating that the machine learning model is no longer valid, the network 100 prepares updated training and / or validation rules at block 526. The updated training and / or validation rules may include instructions to be followed by the UE 110 during a new training and validation phase of the machine learning model. The updated training and / or validation rules may include updating the updated training and / or validation rules sent in the first message at block 500 and / or may be new training and / or validation rules.

[0145] At block 528 , updated training and / or validation rules are sent from the network 100 to the UE 110 .

[0146] At block 530, the UE receives updated training and / or validation rules from the network 100. At block 532, the UE 110 determines whether the UE 110 remains in the machine learning mode or whether the UE 110 is to switch to a non-machine learning mode based on the information in the updated training and / or validation rules. If the updated message indicates that the UE 110 is to remain in the machine learning mode, the method returns to block 502, and the training and validation are repeated using the updated rules and policies. If the updated message indicates that the UE 110 is to switch to a non-machine learning mode, the method proceeds to block 534, and the UE 110 stops using the machine learning model.

[0147] Variations of the signaling diagrams may be used in examples of the present disclosure. For example, in some cases, the network 100 may decide to switch the UE 110 out of the machine learning mode even if the machine learning model is determined to be effective. As an example, the UE 110 may make optimal or substantially optimal decisions, but these decisions may cause high levels of interference to neighboring network entities. In this case, the network policy may require the UE to switch to non-machine learning to reduce interference to other network entities.

[0148] The described methods and processes may be applied to achieve technical effects, such as enabling the network 100 to manage the training and validation phases at the UE 110. Examples of the present disclosure may also enable the network to control the inference phase of the machine learning model at the UE 110. This provides the benefit of enabling the UE 110 to perform machine learning operations. This may be beneficial because it enables the UE 110 to collect and process data itself, which may be more efficient. However, in examples of the present disclosure, the network 100 retains a sufficient level of control over the machine learning operations to reduce issues such as UE 110 causing interference by making decisions that negatively impact other entities within the network or making decisions that may cause performance degradation of the UE 110 or making decisions that may cause network performance degradation. Decisions that may cause performance degradation of the UE 110 or the network 100 may be switching too early, switching too late, switching to the wrong target, or any other related errors.

[0149] Figure 6 An example of a controller 600 is shown. The controller 600 may be provided within a device such as a UE 110 or a network device. An implementation of the controller 600 may be a controller circuit system. The controller 600 may be implemented in hardware only, with certain aspects of software (including firmware), or may be a combination of hardware and software (including firmware).

[0150] like Figure 6 As shown, the controller 600 can be implemented using instructions that implement hardware functions, for example, by using executable instructions of a computer program 606 in a general or special purpose processor 602, which executable instructions can be stored on a computer-readable storage medium (disk, memory, etc.) for execution by such a processor 602.

[0151] The processor 602 is configured to read from or write to the memory 604. The processor 602 may also include an output interface, which outputs data and / or commands via the processor 602, and an input interface, via which data and / or commands are input to the processor 602.

[0152] The memory 604 stores a computer program 606 including computer program instructions (computer program code) that controls the operation of the device when loaded into the processor 602. The computer program instructions of the computer program 606 provide logic and routines that enable the device to perform the methods shown in the drawings. The processor 602 can load and execute the computer program 606 by reading the memory 604.

[0153] In an example where the controller 600 is provided within the UE 110, the controller 600 thus comprises: at least one processor 602; and at least one memory 604 comprising computer program code 606, the at least one memory 604 storing instructions which, when executed by the at least one processor 602, cause the UE 110 to at least perform: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of the machine learning model; Using the received instructions to train and / or validate the machine learning model; and Prepare a second message to be sent to the network, wherein the second message includes information indicating results of training and / or validation of the machine learning model.

[0154] In an example where the controller 600 is provided within a network device, the controller 600 thus comprises: at least one processor 602; and at least one memory 604 comprising computer program code 606, the at least one memory 604 storing instructions which, when executed by the at least one processor 602, cause the network device to at least perform: enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; and A second message is received from the UE, wherein the second message includes information indicating results of training and / or validation of the machine learning model by the UE.

[0155] The computer program 606 may arrive at the device or network device via any suitable delivery mechanism 608. The delivery mechanism 608 may be, for example, a machine-readable medium, a computer-readable medium, a non-transitory computer-readable storage medium, a computer program product, a memory device, a recording medium such as a compact disk read-only memory (CD-ROM) or a digital versatile disk (DVD) or a solid-state memory, an article of manufacture including or tangibly embodying the computer program 606. The delivery mechanism may be a signal configured to reliably transmit the computer program 606. The device may propagate or transmit the computer program 606 as a computer data signal.

[0156] The computer program 606 may include computer program instructions for causing the UE 110 to perform at least the following operations or for performing at least the following operations: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of the machine learning model; Using the received instructions to train and / or validate the machine learning model; and Prepare a second message to be sent to the network, wherein the second message includes information indicating results of training and / or validation of the machine learning model.

[0157] The computer program 606 may include computer program instructions for causing the network device to perform at least the following operations or for performing at least the following operations: enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; and A second message is received from the UE, wherein the second message includes information indicating results of training and / or validation of the machine learning model by the UE.

[0158] The computer program instructions may be included in a computer program, a non-transitory computer readable medium, a computer program product, a machine readable medium. In some but not necessarily all examples, the computer program instructions may be distributed over more than one computer program.

[0159] Although memory 604 is shown as a single component / circuitry, it may be implemented as one or more separate components / circuitry, some or all of which may be integrated / removable and / or may provide permanent / semi-permanent / dynamic / cached storage.

[0160] Although processor 602 is shown as a single component / circuitry, it may be implemented as one or more separate components / circuitry, some or all of which may be integrated / removable.Processor 602 may be a single-core or multi-core processor.

[0161] References to "computer-readable storage medium," "computer program product," "tangibly embodied computer program," etc. or "controller," "computer," "processor," etc. should be understood to cover not only computers having different architectures, such as single / multi-processor architectures and sequential (von Neumann) / parallel architectures, but also special purpose circuits, such as field programmable gate arrays (FPGAs), application specific circuits (ASICs), signal processing devices, and other processing circuit systems. References to computer programs, instructions, code, etc. should be understood to cover software for a programmable processor or firmware, such as, for example, the programmable content of a hardware device, whether instructions for a processor or configuration settings for a fixed function device, gate array, or programmable logic device, etc.

[0162] As used in this application, the term "circuitry" may refer to one or more or all of the following: (a) hardware circuit implementation only (such as implementation in analog circuitry and / or digital circuitry only), and (b) a combination of hardware circuitry and software, such as (where applicable): (1) a combination of analog and / or digital hardware circuits and software / firmware, and (2) any portion of hardware processor(s) (including digital signal processor(s)), software, and memory(s) with software that work together to enable a device (such as a mobile phone or server) to perform various functions, and (c) Hardware circuit(s) and / or processor(s) such as microprocessor(s) or portion(s) of microprocessor(s) that require software (e.g., firmware) for operation, but the software may not be present when not required for operation.

[0163] This definition of circuitry applies to all uses of the term in this application, including any claims. As another example, as used in this application, the term "circuitry" also covers an implementation of hardware-only circuits, or a processor and its (or their) accompanying software and / or firmware. The term circuitry also covers (e.g., if applicable to a particular claim element) a baseband integrated circuit for a mobile device or a similar integrated circuit in a server, cellular network device, or other computing or network device.

[0164] Figures 2 to 5 The stages shown in the diagram may represent steps in a method and / or code segments in a computer program 606. The illustration of a particular order of blocks does not necessarily imply a required or preferred order for the blocks, and the order and arrangement of the blocks may vary. In addition, some blocks may be omitted.

[0165] Where a structural feature has been described, it may be replaced by a component for performing one or more of the functions of the structural feature, whether that or those functions are explicitly or implicitly described.

[0166] The term "comprising" is used herein in an inclusive, not exclusive, sense. That is, any reference to X comprising Y indicates that X may comprise only one Y or may comprise more than one Y. If "comprising" is intended to be used in an exclusive sense, the context will make it clear by reference to "comprising only one ..." or by use of "consisting of".

[0167] In this specification, the terms "connected", "coupled" and "in communication" and their derivatives mean operationally connected / coupled / in communication. It should be understood that any number or combination of intermediate components (including no intermediate components) may be present to provide direct or indirect connection / coupling / in communication. Any such intermediate components may include hardware and / or software components.

[0168] As used herein, the term "determine / determining" (and grammatical variations thereof) may include at least: computing, calculating, processing, deriving, measuring, investigating, identifying, searching (e.g., searching in a table, a database, or another data structure), ascertaining, etc. In addition, "determining" may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), obtaining, etc. Moreover, "determine / determining" may include resolving, selecting, choosing, establishing, etc.

[0169] In this specification, reference has been made to various examples. The description of features or functions of an example indicates that those features or functions are present in the example. The use of the term "example" or "for example" or "can" or "may" in the text indicates that, whether explicitly stated or not, such features or functions are present in at least the described example, whether described as an example or not, and they may but not necessarily be present in some or all other examples. Therefore, "example", "for example", "can" or "may" refers to a specific instance in a class of examples. The attributes of an instance may be attributes of only the instance or attributes of the class or attributes of a subclass of the class including some but not all instances in the class. Therefore, it is implicitly disclosed that a feature described with reference to one example rather than with reference to another example may be used in another example as part of a working combination where possible, but does not necessarily have to be used in another example.

[0170] Although the examples have been described in the preceding paragraphs with reference to various examples, it should be appreciated that modifications to the examples given can be made without departing from the scope of the claims.

[0171] Features described in the preceding description may be used in combinations other than the combinations explicitly described above.

[0172] Although functions have been described with reference to certain features, those functions may be performable by other features whether described or not.

[0173] Although features have been described with reference to certain examples, those features may also be present in other examples whether described or not.

[0174] The terms "a", "an", or "the" are used in this document in an inclusive, not exclusive, sense. That is, any reference to X that includes a / an / the Y indicates that X may include only one Y or may include more than one Y, unless the context clearly indicates otherwise. If "a", "an", or "the" is intended to be used in an exclusive sense, it will become clear from the context. In some cases, the use of "at least one" or "one or more" may be used to emphasize the inclusive meaning, but the absence of these terms should not be taken to infer any exclusive meaning.

[0175] The presence of a feature (or combination of features) in a claim relates to the feature or (combination of features) itself, and also to features that achieve substantially the same technical effect (equivalent features). Equivalent features include, for example, features that are variants and achieve substantially the same result in substantially the same way. Equivalent features include, for example, features that perform substantially the same function in substantially the same way to achieve substantially the same result.

[0176] In this specification, reference has been made to various examples using adjectives or adjective phrases to describe characteristics of examples. Such description of characteristics associated with examples indicates that the characteristics exist in some examples exactly as described, and in other examples substantially as described.

[0177] The above description describes some examples of the present disclosure, however, those of ordinary skill in the art will be aware of possible alternative structures and method features that provide equivalent functionality to the specific examples of such structures and features described above, and have been omitted from the above description for the sake of brevity and clarity. Nevertheless, unless such alternative structures or method features are explicitly excluded in the above description of examples of the present disclosure, the above description should be understood to implicitly include such alternative structures and method features that provide equivalent functionality.

[0178] While an effort has been made in the foregoing description to draw attention to those features regarded as important, it should be understood that the applicant may seek protection by way of the claims for any patentable feature or combination of features referenced above and / or shown in the accompanying drawings, whether or not emphasized.

Claims

1. A user equipment (UE), comprising components for: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; Using the received instructions to train and / or validate the machine learning model; as well as Preparing a second message to be sent to the network, wherein the second message includes information indicating a result of the training and / or the validation of the machine learning model.

2. The UE according to claim 1, wherein the first message comprises at least one of the following: an indication of parameters to be used to train the machine learning model; an indication of a training rule to be used to train the machine learning model; an indication of validation rules to be used to validate the machine learning model; An indication of an exit condition for the validation process for the machine learning model.

3. The UE according to claim 2, wherein the rules to be used for the machine learning model are configured to control one or more results of the machine learning model to be results preferred by the network.

4. A UE according to any preceding claim, wherein the component is used to determine whether the validation of the machine learning model has succeeded.

5. A UE according to any preceding claim, wherein the second message includes an indication of whether the validation of the machine learning model has been successful.

6. A UE according to any preceding claim, wherein the component is used to receive a third message from the network, wherein the third message includes instructions indicating how the UE should use the trained machine learning model for reasoning.

7. The UE of claim 6, wherein the third message is received if the second message indicates that the verification of the machine learning has succeeded.

8. The UE according to any one of claims 6 to 7, wherein the third message includes at least one of the following: an indication of parameters to be used for the trained machine learning model during inference; an indication of an inference rule to be used by the machine learning model during inference; an indication of a counter inferred by the machine learning model; an indication of a revalidated accuracy level for the trained machine learning model; An indication of a validation threshold for the trained machine learning model.

9. A UE according to any preceding claim, wherein the component is used to enable the UE to be configured into a non-machine learning mode if the validation of the machine learning model is not successfully performed.

10. A UE according to any preceding claim, wherein the component is used to receive an updated first message from the network, wherein the updated first message includes updated instructions indicating how the UE should perform training and / or validation of the machine learning model.

11. The UE of claim 10, wherein the updated first message is received if the second message indicates that the validation of the machine learning model has not been successful.

12. A method comprising: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; Using the received instructions to train and / or validate the machine learning model; as well as Preparing a second message to be sent to the network, wherein the second message includes information indicating a result of the training and / or the validation of the machine learning model.

13. A computer program comprising computer program instructions which, when executed by processing circuitry, cause: receiving a first message from a network, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; Using the received instructions to train and / or validate the machine learning model; as well as Preparing a second message to be sent to the network, wherein the second message includes information indicating a result of the training and / or the validation of the machine learning model.

14. A network device comprising components for: enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; and A second message is received from the UE, wherein the second message includes information indicating a result of the training and / or the validation of the machine learning model by the UE.

15. The network device according to claim 14, wherein the component is used to implement further control of the machine learning model based on the results of the training and / or the verification of the machine learning model indicated by the UE.

16. The network device according to any one of claims 14 to 15, wherein the first message comprises at least one of the following: an indication of parameters to be used to train the machine learning model; an indication of a training rule to be used to train the machine learning model; an indication of validation rules to be used to validate the machine learning model; An indication of an exit condition for the validation process for the machine learning model.

17. A network device according to any one of claims 14 to 16, wherein the component is used to enable a third message to be sent to the UE, wherein the third message includes instructions indicating how the UE should use the trained machine learning model for reasoning, and wherein the third message is sent if the second message indicates that the verification of the machine learning model has been successful.

18. The network device according to claim 17, wherein the third message comprises at least one of the following: an indication of parameters to be used for the trained machine learning model during inference; an indication of an inference rule to be used by the machine learning model during inference; an indication of a counter inferred by the machine learning model; an indication of a revalidated accuracy level for the trained machine learning model; An indication of a validation threshold for the trained machine learning model.

19. A method comprising: enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; as well as A second message is received from the UE, wherein the second message includes information indicating a result of the training and / or the validation of the machine learning model by the UE.

20. A computer program comprising computer program instructions which, when executed by processing circuitry, cause: enabling a first message to be sent to a UE, wherein the first message includes instructions indicating how the UE should perform training and / or validation of a machine learning model; and A second message is received from the UE, wherein the second message includes information indicating a result of the training and / or the validation of the machine learning model by the UE.

Images