Apparatus, method and computer program

Abstract

There is provided an apparatus, a method and a computer program for an apparatus of a first function, the computer causing the apparatus to perform: signaling to a second function a request for at least one decision made by a machine learning model and information of a respective context associated with making the at least one decision; and receiving the requested information from the second function.

Description

Technical Field

[0001] Various examples described herein relate generally to apparatus, methods, and computer programs, and more particularly (but not exclusively) to apparatus, methods, and computer programs for network devices. Background Art

[0002] In general, a communication system may be viewed as a facility for enabling communication sessions between two or more entities, such as user terminals, access nodes, and / or other nodes, by providing carrier waves between the various entities involved in the communication path. For example, a communication system may be provided by means of a communication network and one or more compatible communication devices. A communication session may include, for example, the communication of data for carrying communications such as voice, electronic mail (email), text messaging, multimedia, and / or content data. Content may be multicast or unicast to a communication device.

[0003] A user may access a communication system by means of an appropriate communication device or terminal. A user's communication device is generally referred to as a user equipment (UE) or user equipment. A communication device may access a carrier provided by an access node and transmit and / or receive communications on the carrier.

[0004] The communication system and the associated equipment usually operate according to the required standards or specifications, which specify what the various entities associated with the system are allowed to do and how it should be achieved. The communication protocols and / or parameters for the connection are usually also defined. An example of a communication system is UTRAN (3G radio). Another example of an architecture is the Long Term Evolution (LTE) or Universal Mobile Telecommunications System (UMTS) radio access technology. Another exemplary communication system is the so-called 5G system, which allows user equipment (UE) or user equipment to contact the 5G core via, for example, a new air interface (NR) access technology or via other access technologies such as non-trusted access to 5GC or wired access technology.

[0005] In 5G, the UE Registration Area (RA) includes a list of one or more Tracking Areas (TAs). A Tracking Area is a logical concept of an area where a UE can move around without updating the network. The network can assign a list with one or more TAs to a UE. Summary of the invention

[0006] According to a first aspect, a device for a first function is provided, the device comprising means for performing the following operations: signaling to a second function a request for information of at least one decision made by a machine learning model and a corresponding context associated with making the at least one decision; and receiving the requested information from the second function.

[0007] The apparatus may include means for signaling to a second function a request to modify at least one parameter included in the request; and receiving an indication that the second function has modified a report of the requested information in accordance with the modification of the at least one parameter.

[0008] The apparatus may include means for signaling to the second function a request to delete any reports associated with the request; and receiving an indication that the second function will no longer report the requested information to the first function.

[0009] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0010] According to a second aspect, a device for a second function is provided, the device comprising a means for performing the following operations: receiving a request from a first function for at least one decision made by a machine learning model and information of a corresponding context associated with making the at least one decision; obtaining the requested information; and signaling the requested information to the first function.

[0011] The means for obtaining the requested information may include means for: retrieving information related to the at least one decision from a storage function; and processing the retrieved information to form the requested information.

[0012] The apparatus may include means for: in response to the receiving, determining that the second function is currently maintaining a first reporting instance regarding the requested information; and updating the first instance to include instructions to signal the requested information to the first function.

[0013] The means for updating the first instance may include means for associating, in the first instance, an indication of a frequency with which the requested information is provided to the requesting first function with an identifier of the first function.

[0014] The apparatus may include means for: in response to the receiving, determining that the second function does not currently maintain a first reporting instance regarding the requested information; and instantiating a second instance including instructions to signal the requested information to the first function.

[0015] Means for performing a determination that a second function is not currently maintaining a first reporting instance regarding the requested information may include means for performing at least one of: determining that the received request includes an indication of a machine learning model that is different from any machine learning model that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a context that is different from any context that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a termination condition that is different from any termination condition that is the subject of the first reporting instance.

[0016] The apparatus may include means for providing an identifier of the first instance and / or an identifier of the second instance to the first function.

[0017] The means for obtaining the requested information may include means for: determining a filtering condition for obtaining the requested information; and applying the determined filtering condition to collect the requested information.

[0018] The apparatus may include means for receiving a request from a first function to modify at least one parameter included in the request; and signaling to the first function an indication that the second function has modified the reporting of the requested information in accordance with the modification of the at least one parameter.

[0019] The apparatus may include means for: receiving a request from the first function to delete any reports associated with the request; and signaling an indication that the second function will no longer report the requested information to the first function.

[0020] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0021] According to a third aspect, a device for a first function is provided, the device comprising: at least one processor; and at least one memory, the at least one memory comprising code, which, when executed by the at least one processor, causes the device to: signal to a second function a request for at least one decision made by a machine learning model and information of a corresponding context associated with making the at least one decision; and receive the requested information from the second function.

[0022] The apparatus may be caused to: signal to a second function a request to modify at least one parameter included in the request; and receive an indication that the second function has modified a report of the requested information in accordance with the modification of the at least one parameter.

[0023] The apparatus may be caused to: signal a request to a second function to delete any reports associated with the request; and receive an indication that the second function will no longer report the requested information to the first function.

[0024] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0025] According to a fourth aspect, a device for a second function is provided, the device comprising: at least one processor; and at least one memory, the at least one memory comprising code, which, when executed by the at least one processor, causes the device to: receive from a first function a request for at least one decision made by a machine learning model and information of a corresponding context associated with making the at least one decision; obtain the requested information; and signal the requested information to the first function.

[0026] Obtaining the requested information may include performing: retrieving information related to the at least one decision from a storage function; and processing the retrieved information to form the requested information.

[0027] The apparatus may be caused to perform: in response to the receiving, determining that the second function is currently maintaining a first reporting instance regarding the requested information; and updating the first instance to include instructions for signaling the requested information to the first function.

[0028] Updating the first instance may include performing, in the first instance, associating an indication of a frequency with which the requested information is provided to the requesting first function with an identifier of the first function.

[0029] The apparatus may be caused to perform: in response to the receiving, determining that a second function does not currently maintain a first reporting instance regarding the requested information; and instantiating a second instance including instructions for signaling the requested information to the first function.

[0030] Executing a determination that the second function is not currently maintaining a first reporting instance regarding the requested information may include performing at least one of: determining that the received request includes an indication of a machine learning model that is different from any machine learning model that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a context that is different from any context that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a termination condition that is different from any termination condition that is the subject of the first reporting instance.

[0031] The apparatus may be caused to perform: providing an identifier of the first instance and / or an identifier of the second instance to the first function.

[0032] Obtaining the requested information may include performing: determining a filter condition for obtaining the requested information; and applying the determined filter condition to collect the requested information.

[0033] The apparatus may be caused to receive a request from a first function to modify at least one parameter included in the request; and signal to the first function an indication that a second function has modified a report of the requested information in accordance with the modification of the at least one parameter.

[0034] The apparatus may be caused to: receive a request from the first function to delete any reports associated with the request; and signal an indication that the second function will no longer report the requested information to the first function.

[0035] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0036] According to a fifth aspect, a method for a device for a first function is provided, the method comprising executing: signaling to a second function a request for information of at least one decision made by a machine learning model and a corresponding context associated with making the at least one decision; and receiving the requested information from the second function.

[0037] The method may include performing: signaling to a second function a request to modify at least one parameter included in the request; and receiving an indication that the second function has modified a report of the requested information in accordance with the modification of the at least one parameter.

[0038] The method may include performing: signaling a request to the second function to delete any reports associated with the request; and receiving an indication that the second function will no longer report the requested information to the first function.

[0039] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0040] According to a sixth aspect, a method for a device for a second function is provided, the method comprising: receiving a request from a first function for information about at least one decision made by a machine learning model and a corresponding context associated with making the at least one decision; obtaining the requested information; and signaling the requested information to the first function.

[0041] Obtaining the requested information may include performing: retrieving information related to the at least one decision from a storage function; and processing the retrieved information to form the requested information.

[0042] The method may include performing: in response to the receiving, determining that a second function is currently maintaining a first reporting instance regarding the requested information; and updating the first instance to include instructions for signaling the requested information to the first function.

[0043] Updating the first instance may include performing, in the first instance, associating an indication of a frequency with which the requested information is provided to the requesting first function with an identifier of the first function.

[0044] The method may include performing: in response to the receiving, determining that a second function does not currently maintain a first reporting instance regarding the requested information; and instantiating a second instance including an instruction to signal the requested information to the first function.

[0045] Executing a determination that the second function is not currently maintaining a first reporting instance regarding the requested information may include performing at least one of: determining that the received request includes an indication of a machine learning model that is different from any machine learning model that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a context that is different from any context that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a termination condition that is different from any termination condition that is the subject of the first reporting instance.

[0046] The method may include performing: providing an identifier of the first instance and / or an identifier of the second instance to the first function.

[0047] Obtaining the requested information may include performing: determining a filter condition for obtaining the requested information; and applying the determined filter condition to collect the requested information.

[0048] The method may include performing: receiving a request from a first function to modify at least one parameter included in the request; and signaling to the first function an indication that the second function has modified the reporting of the requested information in accordance with the modification of the at least one parameter.

[0049] The method may include performing: receiving a request from a first function to delete any reports associated with the request; and signaling an indication that the second function will no longer report the requested information to the first function.

[0050] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0051] According to a seventh aspect, a device for a first function is provided, the device comprising: a signaling circuit for signaling a request for at least one decision made by a machine learning model and information of a corresponding context associated with making the at least one decision to a second function; and a receiving circuit for receiving the requested information from the second function.

[0052] The apparatus may comprise signalling circuitry for signalling to a second function a request to modify at least one parameter included in the request, and receiving circuitry for receiving an indication that the second function has modified a report of the requested information in accordance with the modification of the at least one parameter.

[0053] The apparatus may include signaling circuitry for signaling a request to the second function to delete any reports associated with the request, and receiving circuitry for receiving an indication that the second function will no longer report the requested information to the first function.

[0054] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0055] According to an eighth aspect, a device for a second function is provided, the device comprising: a receiving circuit for receiving a request for at least one decision made by a machine learning model and information of a corresponding context associated with making the at least one decision from a first function; an obtaining circuit for obtaining the requested information; and a signaling circuit for signaling the requested information to the first function.

[0056] The obtaining circuitry for obtaining the requested information may include: a retrieving circuitry for retrieving information related to the at least one decision from a storage function; and a processing circuitry for processing the retrieved information to form the requested information.

[0057] The apparatus may include determining circuitry for determining, in response to the receiving, that a second function is currently maintaining a first reporting instance regarding the requested information; and updating circuitry for updating the first instance to include instructions for signaling the requested information to the first function.

[0058] The updating circuitry for updating the first instance may include an associating circuit for associating, in the first instance, an indication of a frequency with which the requested information is provided to the requesting first function with an identifier of the first function.

[0059] The device may include: a determining circuit for determining, in response to the receiving, that a second function does not currently maintain a first reporting instance regarding the requested information; and an instantiating circuit for instantiating a second instance, which includes an instruction to signal the requested information to the first function.

[0060] The determination circuitry for determining that the second function is not currently maintaining a first reporting instance regarding the requested information may include execution circuitry for performing at least one of: determining that the received request includes an indication of a machine learning model that is different from any machine learning model that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a context that is different from any context that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a termination condition that is different from any termination condition that is the subject of the first reporting instance.

[0061] The apparatus may include providing circuitry for providing an identifier of the first instance and / or an identifier of the second instance to the first function.

[0062] The obtaining circuit for obtaining the requested information may include: a determining circuit for determining a filtering condition for obtaining the requested information; and an applying circuit for applying the determined filtering condition to collect the requested information.

[0063] The device may include: a receiving circuit for receiving a request from a first function to modify at least one parameter included in the request; and a signaling circuit for signaling to the first function an indication that the second function has modified the reporting of the requested information based on the modification of the at least one parameter.

[0064] The apparatus may include receiving circuitry for receiving a request from a first function to delete any reports associated with the request, and signaling circuitry for signaling an indication that the second function will no longer report the requested information to the first function.

[0065] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0066] According to a ninth aspect, a non-transitory computer-readable medium is provided, comprising program instructions for causing a device for a first function to perform at least the following operations: signaling to a second function a request for at least one decision made by a machine learning model and information about a corresponding context associated with making the at least one decision; and receiving the requested information from the second function.

[0067] The device may be caused to: signal to a second function a request to modify at least one parameter included in the request; and receive an indication that the second function has modified a report of the requested information in accordance with the modification of the at least one parameter.

[0068] The apparatus may be caused to: signal a request to the second function to delete any reports associated with the request; and receive an indication that the second function will no longer report the requested information to the first function.

[0069] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0070] According to the tenth aspect, a non-transitory computer-readable medium is provided, comprising program instructions for causing a device for a second function to perform at least the following operations: receiving a request from a first function for at least one decision made by a machine learning model and information about a corresponding context associated with making the at least one decision; obtaining the requested information; and signaling the requested information to the first function.

[0071] Obtaining the requested information may include performing: retrieving information related to the at least one decision from a storage function; and processing the retrieved information to form the requested information.

[0072] The device may be caused to perform: in response to the receiving, determining that the second function is currently maintaining a first reporting instance regarding the requested information; and updating the first instance to include instructions for signaling the requested information to the first function.

[0073] Updating the first instance may include performing, in the first instance, associating an indication of a frequency with which the requested information is provided to the requesting first function with an identifier of the first function.

[0074] The device may be caused to perform: in response to the receiving, determining that a second function does not currently maintain a first reporting instance regarding the requested information; and instantiating a second instance including instructions for signaling the requested information to the first function.

[0075] Executing a determination that the second function is not currently maintaining a first reporting instance regarding the requested information may include performing at least one of: determining that the received request includes an indication of a machine learning model that is different from any machine learning model that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a context that is different from any context that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a termination condition that is different from any termination condition that is the subject of the first reporting instance.

[0076] The apparatus may be caused to perform: providing the first function with an identifier of the first instance and / or an identifier of the second instance.

[0077] Obtaining the requested information may include performing: determining a filter condition for obtaining the requested information; and applying the determined filter condition to collect the requested information.

[0078] The apparatus may be caused to: receive from a first function a request to modify at least one parameter included in the request; and signal to the first function an indication that the second function has modified a report of the requested information in accordance with the modification of the at least one parameter.

[0079] The apparatus may be caused to: receive a request from the first function to delete any reports associated with the request; and signal an indication that the second function will no longer report the requested information to the first function.

[0080] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0081] According to an eleventh aspect, there is provided a computer program product stored on a medium, which can cause an apparatus to perform any method as described herein.

[0082] According to a twelfth aspect, there is provided an electronic device, which may comprise the device as described herein.

[0083] According to a thirteenth aspect, there is provided a chipset, which may include the apparatus as described herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Some examples will now be described, by way of illustration only, with reference to the accompanying drawings, in which:

[0085] Figure 1A and 1B A schematic diagram of a 5G system is shown;

[0086] Figure 2 A schematic diagram of a network device is shown;

[0087] Figure 3 A schematic diagram of a user equipment is shown;

[0088] Figure 4 A schematic diagram of a non-volatile storage medium storing instructions which, when executed by a processor, allow the processor to perform one or more of the steps of some example methods;

[0089] Figure 5 A schematic diagram of the network is shown;

[0090] Figures 6 to 10 shows an example architecture; and

[0091] Figures 11 to 20 Example signaling operations are shown. DETAILED DESCRIPTION

[0092] In the following description of examples, certain aspects are explained with reference to a mobile communication device capable of communicating via a wireless cellular system and a mobile communication system serving such a mobile communication device. For the sake of brevity and clarity, such aspects are described below with reference to a 5G wireless communication system. However, it should be understood that these aspects are not limited to a 5G wireless communication system and may, for example, be applied to other wireless communication systems (e.g., current 6G proposals).

[0093] Before describing the examples in detail, refer to Figure 1A and Figure 1B Briefly explain some general principles of 5G wireless communication systems.

[0094] Figure 1AA schematic diagram of a 5G system (5GS) 100 is shown. 5G may include a user equipment (UE) 102 (which may also be referred to as a communication device or terminal), a 5G access network (AN) (which may be a 5G radio access network (RAN) or any other type of 5G AN, such as a non-3GPP interworking function (N3IWF) / trusted non-3GPP gateway function (TNGF) for non-trusted / trusted non-3GPP access or a wired access gateway function (W-AGF) for wired access) 104, a 5G core (5GC) 106, one or more application functions (AF) 108, and one or more data networks (DN) 110.

[0095] A 5G RAN may include one or more gNodeB (gNB) distributed unit functions connected to one or more gNodeB (gNB) unit functions. A RAN may include one or more access nodes.

[0096] 5GC 106 may include one or more access and mobility management functions (AMF) 112, one or more session management functions (SMF) 114, one or more authentication server functions (AUSF) 116, one or more unified data management (UDM) functions 118, one or more user plane functions (UPF) 120, one or more unified data repository (UDR) functions 122, one or more network repository functions (NRF) 128 and / or one or more network exposure functions (NEF) 124. The role of NEF is to provide secure exposure of network services (e.g., voice, data connections, billing, subscriber data, etc.) to third parties. Although NRF 128 is not shown to have an interface, it should be understood that this is for clarity and that NRF 128 may have multiple interfaces with other network functions.

[0097] 5GC 106 also includes a network data analysis function (NWDAF) 126. NWDAF is responsible for providing network analysis information based on requests from one or more network functions or devices within the network. Network functions can also subscribe to NWDAF 126 to receive information from it. Therefore, NWDAF 126 is also configured to receive and store network information from one or more network functions or devices within the network. Data collection by NWDAF 126 can be performed based on at least one subscription to events provided by at least one network function.

[0098] The network may also include a management data analysis service (MDAS) producer or an MDAS management service (MnS) producer. The MDAS MnS producer may provide data analysis in the management plane taking into account parameters including, for example, load level and / or resource utilization. For example, an MDAS MnS producer for a network function (NF) may collect load-related performance data of the NF, such as the resource usage status of the NF. Analysis of the collected data may provide a prediction of resource usage information in a predefined future time window. The analysis may also recommend appropriate actions, such as scaling of resources, admission control, load balancing of services, etc.

[0099] Figure 1B A schematic diagram of a 5GC represented in the current 3GPP specification is shown. It should be understood that the architecture is intended to illustrate potential components that may be included in a core network, and the principles currently described are not limited to a core network comprising only the described components.

[0100] Figure 1B A 5GC 106' is shown including a UPF 120' connected to an SMF 114' via an N4 interface. The SMF 114' is connected to each of the UDM 122', NEF 124', NWDAF 126', AF 108', Policy Control Function (PCF) 130', AMF 112', and Charging Function 132' via an interconnect medium that also connects these network functions to each other. The 5G Core 106' also includes a Network Repository Function (NRF) 133' and Network Functions 134' connected to the interconnect medium.

[0101] 3GPP refers to a group of organizations that develop and publish different standardized communication protocols. 3GPP develops and publishes documents related to systems in "releases" (eg, Release 15, Release 16, and higher).

[0102] One of the current areas of 3GPP research involves cognitive autonomous networks (CANs). A cognitive autonomous network operates with little or no human intervention as it is able to configure, monitor and maintain itself independently. To achieve this, various machine learning (ML) models and applications and artificial intelligence systems are deployed within the network to make decisions about how to change network parameters in response to changing network conditions.

[0103] Cognitive autonomous networks promise to provide intelligence and autonomy in network operations, management and administration (OAM) and network processes to support the increased flexibility and complexity of radio networks. By using machine learning (ML) and artificial intelligence (AI) in cognitive functions, CANs may be able to: 1) take higher-level goals and derive appropriate performance targets, 2) learn from their environment and their individual or shared experiences therein, 3) learn to contextualize their operating conditions, and 4) learn their "best" behavior appropriate for the specific environment and context. The results of these four operations can be used by implementing at least one decision based on the information derived while performing them.

[0104] CAN can be deployed in any of a number of different use cases. An example use case for this cognitive automation is handover optimization.

[0105] The network automation functionality forming at least a part of the CAN may apply at least one machine learning model (ML model) to make appropriate inferences in different contexts, wherein a particular context represents a particular set of input parameters for the ML model.

[0106] Machine learning (ML) inference is the process of running data points (e.g., current network parameter values) into a machine learning algorithm (or "ML model") to compute an output (such as a single numerical score). This process is also known as "operating the ML model" or "putting the ML model into production." When an ML model is running in production, it is often described as artificial intelligence (AI) because it performs functions similar to human thinking and analysis.

[0107] The ML lifecycle can be broken down into two main, distinct parts. The first is the training phase, where an ML model is created or “trained” by running a specified subset of data into the model. ML inference is the second phase, where the model takes action on live data to produce executable output.

[0108] Depending on the context, an ML application (e.g., an ML model or ML-based function) may make different decisions when reasoning, resulting in correspondingly different results. However, the selected decision may be tracked for future reference. Tracking network parameters in response to implementing the selected decision may be used to evaluate whether the decision / reasoning that led to selecting the decision was appropriate for the context, and / or to evaluate degradation of the decision-making ability of the ML application. To this end, the network may not only have access to information about the reasoning ability of the ML application, but may also be able to track and enable use of the history of reasoning made by the ML application.

[0109] 3GPP has previously considered using reasoning history in the Management Data Analysis (MDA) role in the management loop.

[0110] For example, 3GPP TS28.104 relates to decision making for management actions for managed networks and services. If necessary, management actions are decided based on analysis reports (provided by the MDA and indicating, for example, root cause analysis of problems that are occurring, predictions of potential problems and corresponding related causes, and recommended actions for prevention, and / or predictions of network and / or service requirements) and other management data (e.g., historical decisions made previously). This historical data does not provide its context because it does not explain how these decisions were made. Instead, they indicate the decisions made (e.g., changes deployed in the network) and whether the changes lead to improvements. The decision can be made by a consumer of the MDA (in a closed management control loop) or by a human operator (in the case of an open management loop). The decision may include, for example, instructions on what action to take and when to take the action.

[0111] The historical decision can be used as part of the input data for decision making. However, the current 3GPP specification does not include a solution for managing the reasoning history in terms of information modeling and lifecycle management process of the reasoning history record.

[0112] Therefore, it would be useful for the network to have appropriate services and capabilities to support the management of inference history of ML models and ML-based functions.

[0113] In particular, it would be useful for the network function and / or its management system to have appropriate services and capabilities to support the management of the reasoning history of ML applications. The network function and / or its management system may model information objects for requesting and reporting ML reasoning history. The network function and / or management system may also be usefully configured to provide operations that support the process for requesting and reporting ML reasoning history.

[0114] The following relates to devices and methods for implementing ML reasoning history functionality (also referred to herein as ML reasoning history functionality) and related services for supporting request and delivery of ML reasoning history.

[0115] To help achieve this, for authorized consumers requesting reasoning history for a particular MLApp, the ML reasoning history function is configured to receive a request for a history of contextualized decisions made by the MLApp. Hereinafter, the received request is referred to as an ML reasoning history function request.

[0116] The ML reasoning history function may assume the existence of an ML reasoning history log that stores the decisions of the MLApp made for different contexts. A different context may be stored for each decision. The requested history may be written from these stored decisions.

[0117] The ML reasoning history function infers the requested history with relevant context from past decision events recorded in the ML reasoning history log.

[0118] In order for a producer of ML reasoning history (e.g., an ML reasoning history function) to provide a report on the reasoning history of a specific MLAppMLApp to authorized consumers, the ML reasoning history function may be configured to map a received ML reasoning history function request to a reporting process associated with the received ML reasoning history function request and the specified MLApp (hereinafter referred to as ML reasoning history function report).

[0119] In one example, the ML reasoning history function request can be mapped to an existing ML reasoning history function report. In this case, the requester is added to the recipients of the report provided by the existing ML reasoning history function report.

[0120] In another example, the request may not be mapped to any existing ML reasoning history function report. In this case, the ML reasoning history function may create a new ML reasoning history function report to associate with the received ML reasoning history function request and the MLApp specified in the request.

[0121] In both examples, the authorized consumer may include reporting properties for these reports in the ML reasoning history functionality request. These reporting properties may define at least one of: the type of information to be included in the report, an indication of where the report will be provided once it is formed; the frequency of the report and / or an indication of the MLApp to which the request relates.

[0122] The authorized consumers of these two examples may be configured to manage one or more requests regarding the reasoning history of a particular MLApp and receive at least one report regarding the reasoning history of the particular MLApp. At least a portion of the one or more requests may be managed simultaneously. The one or more requests may be managed at different times from each other.

[0123] To help store ML inference history, an information model for ML inference history function reporting can be defined, which provides data about the captured inference history. The information model can be a single ML inference record (MLInferencerecord). A process for requesting and reporting ML inference history functions can be defined.

[0124] Figure 6 An example architecture in which ML inference history functionality may be provided in the management plane is shown.

[0125] Figure 6An authorized consumer 601 is shown configured to request an ML model inference history about an MLApp from an ML model inference history producer 602 on a management plane 603. The ML model inference history producer 602 is configured to exchange signaling with an ML inference history function 604 using an ML model inference history control management service 605. The ML model inference history producer 602 may return a report about the ML model inference history to the ML model inference history consumer 601 via an interface 606.

[0126] Figure 7 An example architecture is shown in which ML reasoning history functionality can be provided over an air interface.

[0127] Figure 7 An ML inference history module 701 included in a UE 702 is shown. The ML inference history module 701 may receive a request for an ML model inference history from an access point 703 (e.g., a gNB). In response to the request, the ML inference history module 701 may report information to the ML model inference history through the access point 703.

[0128] The following provides the definitions of the Information Object Class (IOC) and data types used to implement the ML Reasoning History functionality.

[0129] Each Information Object Class definition includes a description of the purpose of the IOC and the attributes that define the IOC. An IOC does not include the values ​​of the attributes that make up its definition.

[0130] To simplify the IOC definition, the attributes of each information object class are divided, where similar attributes are grouped together. These attribute groupings are designated as independent modules and can be reused by other information object classes.

[0131] A data type (dataType) is a classification that specifies what type of value a variable has and what type of mathematical, relational, or logical operations can be applied to it without causing errors. For example, string is a data type used to classify text, and integer is a data type used to classify non-negative whole numbers (whole numbers).

[0132] A description of the relationship between these IOCs and data types in this case is provided further below.

[0133] Figure 8 , 9 and 10 illustrate at least one example for modeling ML reasoning history functionality.

[0134] Figure 8An information model used to control the history function of ML inference. In general, information models are abstract, formal representations of entity types, which may include their properties, relations, and operations that can be performed on them. Typically, they are used to model constrained domains that can be described by a closed set of entity types, properties, relations, and operations.

[0135] Figure 8 A managed entity proxy class 801 is shown providing a class name (such as, for example, Subnet ManagedFunction, ManagementFunction) to the ML Reasoning History Function IOC 802. A proxy class is a class that implements a list of interfaces specified at runtime when the class is created.

[0136] The ML Reasoning History Function IOC represents the properties of the ML Reasoning History Function. The ML Reasoning History Function is the function responsible for receiving requests and sending related reports. Therefore, it can be instantiated from the information object class and name contained in the subnet, managed function or management function. Therefore, the ML Reasoning History Function is a type of managed function (i.e., the ML Reasoning History Function can be considered a subclass of the managed function and inherits the capabilities of the managed function).

[0137] A managed function may be an IOC as defined in TS28.622 clause 4.3.4. A managed function IOC may represent a telecommunications function implemented by software running on dedicated hardware or implemented by software running on a network function virtualization infrastructure (NFVI). Each managed function instance communicates with the manager (directly or indirectly) via one or more management interfaces exposed by the managed entity instances it contains.

[0138] The ML reasoning history function IOC properties will now be described. The ML reasoning history function IOC may be associated with one or more ML models. For example, the ML reasoning history function IOC may be associated with ML models associated via a list of ML model identifiers (MLModelIdentifers). The list of ML model identifiers may be defined in a model of the IOC within an operational communication protocol (e.g., within a 3GPP specification). The list may be created each time the IOC is instantiated (e.g., by an operator).

[0139] MLModel (also referred to herein as MLApp or AIMLEntity) can be an IOC. In this case, the ML Model IOC can be contained by a managed entity ManagedEntity and created by the system or consumer upon request. Alternatively, MLModel can be a data type. In this case, MLModel can be an attribute member of the IOC AIMLTrainingFunction (AIML training function) (as defined in current TS28.105). The lifecycle of the MLModel data type can be maintained with the training request associated with the AIMLTrainingFunction IOC.

[0140] The ML reasoning history function may include one or more ML reasoning history function requests and one or more ML reasoning history function report instances responsible for sending ML reasoning history function reports.

[0141] The ML Reasoning History Function IOC may include attributes inherited from a Managed Function IOC (such as, for example, the Managed Function IOC defined in 3GPP TS 28.622

[30] ) and the following attributes:

[0142] Property Name Readable Writable Unchanged Notifiable ML Inference History Feature Identifier real Fake Fake Fake mlAppsList real Fake Fake Fake ML Inference History Feature Request real real Fake real Attributes related to roles ML Inference History Feature Report real Fake Fake Fake

[0143] The ML reasoning history IOC 802 outputs to the ML reasoning history report IOC 803 and the MLApp data type 804. The ML reasoning history IOC 802 further outputs the name to the ML reasoning history request IOC 805.

[0144] The following defines an IOC that represents the properties of the ML reasoning history function request IOC.

[0145] For each request to collect ML reasoning history, the consumer may create a new ML reasoning history function request for the ML reasoning history function. In other words, the ML reasoning history function request may be an information object class that is instantiated for each request for ML reasoning history.

[0146] Each ML reasoning history function request can be associated with exactly one MLModel or MLApp. Each ML reasoning history function request can include specific reporting requirements. For example, each ML reasoning history function request can include a reporting requirement for a reporting period (ReportingPeriod), which defines how often the ML reasoning history function can report on the ML reasoning history function request.

[0147] ML reasoning history function requests can be associated with ML reasoning history function reports. However, the association is not necessarily a one-to-one mapping between instances of ML reasoning history function requests and ML reasoning history function report instances, as multiple requests can be mapped to a single report instance. Therefore, ML reasoning history function requests can be managed independently of ML reasoning history function reports. After determining which ML reasoning history function report IOC will provide a report for a particular request, the association between the ML reasoning history function request and its corresponding ML reasoning history function report can be written into the request by the ML reasoning history function.

[0148] The ML reasoning history function request may include a source for identifying where it came from (e.g., for identifying the consumer requesting the report). For example, the source may be an enumeration defined for network functions, operator roles, and / or other functional differences. For example, the source may be implemented as a value associated with an index, where the value and index may be used to point to / identify where the request was generated from.

[0149] The ML reasoning history function request IOC may include attributes inherited from a top-level IOC (such as, for example, the top-level IOC defined in 3GPP TS 28.622). The ML reasoning history function request may include the following attributes:

[0150]

[0151] The following defines an IOC that represents the properties of the ML reasoning history function request IOC.

[0152] The ML reasoning history function report IOC may represent the properties of the ML reasoning history function report. The ML reasoning history function report may represent the ability to write and deliver reports and notifications about the ML reasoning history function or its associated ML reasoning history function requests. The ML reasoning history function may be associated with one or more instances of the ML reasoning history function report.

[0153] An ML reasoning history capability report may report (and thus may be associated with) one or more ML reasoning history capability requests.

[0154] For each ML reasoning history function request, the characteristics of the request and the report requirements can be mapped to an ML reasoning history function report instance. The ML reasoning history function report can include an ML reasoning history function report matrix that defines the frequency with which reports will be sent and the report recipients as specific entities to which each report can be sent at each time.

[0155] ML reasoning history function reports can be managed separately from ML reasoning history function requests. This is because an ML reasoning history function report instance can provide reports on more than one ML reasoning history function request. An ML reasoning history function report should be associated with one or more ML reasoning history function reports.

[0156] The ML Reasoning History Function Report IOC may include properties inherited from the top-level IOC (e.g., defined in TS 28.622). The ML Reasoning History Function Report IOC may include the following properties:

[0157]

[0158] The ML model data type (also referred to herein as the MLApp data type) may represent properties of an MLModel.

[0159] Multiple ML reasoning history function requests can be submitted on a single MLModel, but ML reasoning history function requests can be independent of MLModel. Relatedly, although ML reasoning history function requests can be associated with MLModel, MLModel does not have to be associated with any specific ML reasoning history function request. Relatedly, MLModel does not have to be associated with any specific ML reasoning history function report instance, because the data used to write the ML reasoning history function report can be collected from logs independent of MLModel.

[0160] MLModel may include properties inherited from the top-level IOC (e.g., as defined in 3GPP TS 28.622). In addition, the model may be represented in a repository so that its properties (including the current state) may be retrieved from repository functions. Although MLModel is modeled here as a data type, it should be understood that MLModel may be modeled as an information object class.

[0161] The ML reasoning history report IOC 803 and the ML reasoning history request IOC 805 are output to the MLApp data type 804. The ML reasoning history request IOC 805 is further output to the report context data type 806 and the ML reasoning history report IOC 803.

[0162] The ML reasoning history report IOC 803 is also output to an ML reasoning history request IOC 805 , a report context data type 806 , an ML report matrix data type 807 , and an ML reasoning history report data type 808 .

[0163] The MLReportingMatrix data type represents the properties of an MLReportingMatrix.

[0164] The MLReportingMatrix can define how often reports should be sent and the specific entity to which each report should be sent at each moment. The frequency at which reports are to be sent can be defined from the ReportingPeriod (the reporting period, e.g., the time in seconds between 2 consecutive reports for this MLTrainingReporting instance).

[0165] The MLReportingMatrix can include the following properties:

[0166] Property Name Readable Writable Unchanged Notifiable ReportingPeriod real Fake Fake Fake consumerList real Fake Fake Fake

[0167] The ConsumerList is a list of DistinguishedNames of managed functions that have requested the relevant reports and to which the reports will be delivered at the times defined by the ReportingPeriod.

[0168] The ML inference history function report data type represents the properties of the ML inference history function report.

[0169] The ML reasoning history function 802 may generate one or more ML reasoning history function reports. Each of the generated ML reasoning history function reports may be associated with one or more MLModels / MLApps, since the ML reasoning history function report is associated with an ML reasoning history function report instance, which itself is associated with the MLModels / MLApps requesting and / or reporting reasoning history.

[0170] The ML Inference History feature report can include the following properties:

[0171] Property Name Readable Writable Unchanged Notifiable ML Inference History Feature Report ID real Fake Fake Fake ML Inference History Feature Report ID real Fake Fake Fake mlAppList real Fake Fake Fake ML Inference History Feature Request real Fake Fake Fake

[0172] ReportingContext<<Data Type>> represents the properties of a context. The context describes a list of constraints or conditions that should be evaluated to be true when the report is executed.

[0173] ReportingContext<<Data Type>> can be defined by the triple <attribute, condition, value range> condition.

[0174] Wherein, an attribute may describe (or indicate) a specific property of or related to an object or use case that relates to the MLApp on which the report is to be performed. An "attribute" may also or alternatively refer to at least one characteristic of such an object (e.g., its control parameter, meter, counter, key performance indicator (KPI), weighted metric, etc.). An "attribute" may also or alternatively refer to a property related to an operating condition of an object or use case (such as, for example, weather conditions, load conditions, etc.).

[0175] Additionally, a condition may represent a restriction within which an attribute is allowed / assumed. Allowed values ​​for a condition may include: "equal to"; "less than"; "greater than"; "within range"; and / or "outside range".

[0176] In addition, a value range (ValueRange) may describe (or indicate) a range of values ​​applicable to a property subject to a condition.

[0177] A report may be generated when a value representing a property is obtained, which falls within a range of values ​​when a condition is satisfied.

[0178] The context can include the following properties:

[0179] Property Name Readable Writable Unchanged Notifiable contextAttribute real real Fake Fake contextCondition real real Fake Fake contextValueRange real real Fake Fake

[0180] Fig. 9 The inheritance relationship in the ML reasoning history function is shown. Inheritance involves the relationship between so-called parent classes and child classes. In particular, a child class inherits all the attributes of the parent class. For example, suppose that parent class A has 2 attributes (attribute_p_1; attribute_p_2). When a new class B is defined as a child class of class A, class B inherits two attributes from class A, and class B also has 3 of its own attributes (attribute_c_1; attribute_c_2; attribute_c_3). That is, due to the inheritance relationship, the final class B will have a total of 5 attributes: (attribute_p_1; attribute_p_2; attribute_c_1; attribute_c_2; attribute_c_3).

[0181] Fig. 9 An ML reasoning history function IOC 901 is shown providing output to a managed function IOC.

[0182] Fig.10 The inheritance relationship with ML inference history requests and reports is shown.

[0183] Fig.10 A “top-level” IOC 1001 is shown receiving output from an ML reasoning history request IOC 1002 and an ML reasoning history report IOC 1003 . Fig. 9 and Fig.10 These entities can correspond to Figure 8 Those equivalently marked IOCs in .

[0184] The following references Figures 11 to 15 Describes the processes that can be performed by the ML inference history function.

[0185] The ML reasoning history function may have the capability to instantiate an ML reasoning history function report instance based on receiving at least one request from at least one consumer function. Fig.11 shown.

[0186] Fig.11 At least one consumer 1101 is shown interacting with at least one ML inference history management service producer 1102 (also referred to herein simply as a “producer”).

[0187] At 11001, at least one consumer 1101 signals a request to a producer 1102. The signaling may be a request to create an instance of a managed object. The request may include a request for an ML reasoning history. The request may include an identification of an ML model related to the request. The request may include an identifier for the request, which may be used in communications including information related to the request.

[0188] A request for the ML reasoning history function from at least one consumer may be received using an ML reasoning history function provision management service implemented via a CRUD (create, read, update, delete) operation on an ML reasoning history function request object.

[0189] As described above, the request may include an indication of an ML Model that the request relates to. For example, this may be a unique identifier of a particular MLModel on which the consumer wishes to receive inference. The request may alternatively indicate the ML Model by including an identifier of an MLApp that includes the MLModel on which the consumer wishes to receive inference.

[0190] The request may include an indication of a reporting context. The reporting context is the conditions and environmental characteristics to be considered when reporting the ML inference history function.

[0191] The request may include an ML report frequency. The frequency indicates how often the consumer requests to be informed of the ML reasoning history function. The frequency is provided when at least one consumer wishes to receive a report on the reasoning history more than once.

[0192] At 11002 , the producer 1102 may instantiate an ML reasoning history request (referred to herein as Req1 ).

[0193] At 11003, the producer 1102 may identify a filter to filter historical data to identify data that falls within the scope of the request.

[0194] At 11004 , producer 1102 may apply the filter identified at 11003 in order to collect reporting data for satisfying the request.

[0195] At 11005 , the producer 1102 may check whether a report instance already exists for the applied filter. When a report instance does not exist for the applied function, the producer 1102 proceeds to 11006 .

[0196] At 11006 , the producer 1102 instantiates an ML reasoning history report instance for Req1 . From 11006 , the producer 1102 proceeds to 11008 .

[0197] If the check at 10005 determines that a report already exists for the applied function, the producer 1102 proceeds to 11007 .

[0198] During 11007 , producer 1102 adds the requirement associated with Req1 to the existing ML reasoning history report instance. Producer 1102 proceeds from 11007 to 11008 .

[0199] During 11008, the producer 1102 signals to send a notification to at least one consumer 1101. The notification may include an indication that the request of the ML reasoning history request (Req1) has been included as part of the ML reasoning history report instance. An identifier may be provided for the ML reasoning history report instance. For example, in this example, the identifier of the ML reasoning history report instance may be labeled "JBx".

[0200] Therefore, in Fig.11 In the example of , after receiving the ML reasoning history function request from the consumer, the ML reasoning history function may create an instance of the ML reasoning history function report, or may update an existing instance of the ML reasoning history function report with the characteristics of the ML reasoning history function request.

[0201] The requirement(s) included in the ML reasoning history function request may be added to an existing ML reasoning history function report instance with related characteristics. For example, when both the ML reasoning history function request and the existing ML reasoning history function report instance are on the same model and context but have different frequencies, the requirement(s) included in the ML reasoning history function request may be added to the existing ML reasoning history function report instance. For example, when there is an ML reasoning history function report instance with an ML report frequency of 1 hour, if a request A with an ML report frequency of 30 minutes is received, a new instance may be added to the existing instance to create a common ML reasoning history function report instance with different ML report frequency instances for the two recipients.

[0202] For alternative cases, an ML reasoning history function report may be instantiated for an ML reasoning history function request. Different ML reasoning history function report instances may be instantiated for ML reasoning history function requests with different expected runtime contexts and / or termination conditions.

[0203] After instantiating a new ML reasoning history report instance and ML reasoning history function, the consumer who initiated the request may also be notified of the corresponding action taken regarding the request, such as described with respect to 1108 .

[0204] Fig.12 Operations related to reporting ML reasoning history for a given ML reasoning history function request instance are shown.

[0205] Fig.12 Operations that can be performed between at least one consumer 1201 and a producer 1202 are shown. These entities may correspond to Fig.11 Entity.

[0206] During 12001, at least one consumer 1201 and producer 1202 exchange signaling for instantiating an ML reasoning history request instance. This can be as described in relation to Fig.11 as described in steps 11001 to 11007.

[0207] In other words, for a given ML reasoning history function request, the ML reasoning history function producer can instantiate and trigger the ML reasoning history function report as a process. The reasoning history function report process may include Fig.12 For example, the process shown in

[0208] Identify filter features for selecting historical data from any available logs

[0209] Apply filters to select data

[0210] · Compile the acquired data into a reasoning history report to share with the requesting consumer.

[0211] During 12002, producer 1202 may send a signal to at least one consumer 1201. The signal of 12002 may correspond to Fig.11 The signaling of 11008 of 12002 may notify at least one consumer 1201 of the creation of a managed object instance. The signaling of 12002 may include an identifier of the instance. The identifier of the instance may be a unique identifier of the report instance, which may be a distinguished name (DN) of the ML reasoning history report. The signaling of 12002 may include data. The data included in the signaling of 12002 may correspond to a complete representation of the ML reasoning history report, including data about the reasoning history to be reported.

[0212] During 12003, producer 1202 writes the ML inference history as required by Req1.

[0213] During 12004, the producer 1202 signals at least one consumer 1201. The signaling of 12004 may include an indication that data is available according to the requirements of Req1. The signaling of 12004 may include at least a portion of the authoring data. The signaling of 12004 may not include any authoring data. The signaling of 12004 may include an indication of where at least a portion of the authoring data may be downloaded from (or otherwise accessible). For example, the signaling of 12004 may include an address from which at least a portion (and possibly all) of the authoring data may be downloaded. The signaling of 12004 may include an identifier of the ML model.

[0214] In other words, after instantiating the ML reasoning history function report, the ML reasoning history function function can configure the report for the corresponding ML reasoning history function request. If there is an ML reasoning history function report instance with the same characteristics as those described in the ML reasoning history function request, the ML reasoning history function producer can append the new report requirements to the existing ML reasoning history function report instance. Alternatively, the ML reasoning history function producer can instantiate a new ML reasoning history function report instance with the requirements as defined in the ML reasoning history function request.

[0215] The ML Reasoning History function has capabilities and control interfaces that allow consumers (e.g., operators) to configure and manage one or more ML Reasoning History function reporting instances. The control interface can enable consumers to obtain the results of the ML Reasoning History function reporting process. This can be done using Fig.12 The 3GPP shown in the figure provides a notification procedure for management services.

[0216] Fig.13 Operations associated with reading characteristics of one of a plurality of ML inference report instances are shown.

[0217] Fig.13 1301 and producer 1302. The at least one consumer and producer may correspond to Fig.11 and / or Fig.12 at least one consumer and producer of , and can therefore perform at least one of the actions described therein.

[0218] During 13001 , at least one consumer 1301 and producer 1302 exchange signaling for instantiating an ML reasoning history request ( Req1 ) and a report instance.

[0219] During 13002, at least one consumer 1301 signals to a producer 1302. The signaling of 13002 may include a request to retrieve an indication of an attribute of a managed object instance. For example, the signaling may include a request to retrieve an indication of at least one attribute associated with a corresponding at least one ML reasoning history report instance.

[0220] During 13003, producer 1302 signals at least one consumer 1301. The signaling of 13003 may include a file and / or stream reporting service operation. The signaling of 13003 may include information related to at least one ML inference history reporting inference. For each ML inference history reporting inference provided, the signaling of 13003 may include an identifier of the ML inference history reporting inference and associated data. The associated data may include a value of an instance of MLInferenceHistoryReporting, where the type of the value may be as defined in the IOC of MLInferenceHistoryReporting.

[0221] In other words, Fig.13 In the example of , a consumer can read the characteristics of an ML reasoning history function report instance. For example, a consumer can determine to read the state of an ML reasoning history function report instance, and / or read the results of an ML reasoning history function report instance (e.g., a decision and / or a historical evaluation of past decisions). This can be implemented using the 3GPP getMOIAttributes procedure that provides a management service operation, but it should be understood that other service operations can be used.

[0222] The status may represent a determination of the extent to which a reporting instance has been completed. For example, the provisioning of status may be provided in response to triggered data collection, data being collected, results being reported (when only one report is being made), and / or instantiated continuous reporting (when the request involves repeated reports). The status may correspond to the assignment of values ​​for each attribute in MLInferenceHistoryReporting. Thus, there may be jobs with status attributes.

[0223] Fig.14 Operations that may be performed to modify at least one characteristic of one or more ML reasoning history report instances are shown.

[0224] Fig.14 14 shows signaling that can be performed by at least one consumer 1401 and at least one producer 1402. The at least one consumer and the producer can correspond to Fig.11 and / or Fig.12 and / or Fig.13 at least one consumer and producer of , and can therefore perform at least one of the actions described therein.

[0225] During 14001, at least one consumer 1401 and producer 1402 may exchange signaling for instantiating ML reasoning history requests (Req1) and reporting instances. This may be as described above with reference to Fig.11 as described.

[0226] During 14002, at least one consumer 1401 signals to a producer 1402. The signaling of 14002 may include a request to change a value associated with at least one attribute of at least one ML reasoning history report instance. The signaling of 14002 may include a request to modify at least one attribute of a managed object instance. The signaling of 14002 may include a modifyMOIAttributes service operation. The signaling of 14002 may include at least one identifier associated with at least one ML reasoning history report instance that is the subject of the signaling of 14002. The signaling of 14002 may identify at least one attribute to be modified.

[0227] During 14003, producer 1402 signals at least one consumer 1401. The signaling may include an indication that values ​​associated with attributes respectively associated with at least one ML reasoning history report instance have been modified from a previous value to a current value. The signaling of 14003 may include an indication of the current value of any modified attribute identified in the signaling of 14003. The signaling of 14003 may include at least one identifier associated with at least one ML reasoning history report instance that is the subject of the signaling of 14003. The signaling of 14003 may include a notifyMOIAttributeValueChanges service operation.

[0228] in other words, Fig.14 The signaling can be configured to enable consumers to configure new and ongoing ML reasoning history function report instances. For example, an operator can assign priorities to one or more ML reasoning history function report instances to indicate that in the case of resource constraints, some specific ML reasoning history function report instances with higher priorities should be executed before ML reasoning history function report instances with lower priorities. Fig.14 The 3GPP provisioning management service modifyMOIAttributes procedure is shown to achieve configuration, but it will be appreciated that alternative service operations may be used to achieve the same functionality.

[0229] After receiving the update request, the ML reasoning history function producer may notify the consumer about the success of the update performed on the ML reasoning history function report instance.

[0230] Fig.15 Operations that may be performed to delete one or more ML reasoning history report instances are shown.

[0231] Fig.15 1501 and at least one producer 1502. The at least one consumer and the producer may correspond to Fig.11 and / or Fig.12 and / or Fig.13 and / or Fig.14 at least one consumer and producer of , and can therefore perform at least one of the actions described therein.

[0232] During 15001, at least one consumer 1501 and producer 1502 may exchange signaling for instantiating ML reasoning history requests (Req1) and reporting instances. This may be as described above with reference to Fig.11 as described.

[0233] During 15002, at least one consumer 1501 may signal a producer 1502. The signaling of 15002 may include a request to delete a managed object instance. For example, the signaling may include a request to delete an ML reasoning history report instance. The signaling of 15002 may include an identifier of the ML reasoning history report instance that is the subject of the signaling of 15002.

[0234] During 15003, producer 1502 signals at least one consumer 1501. The signaling of 15003 may include an indication that the ML reasoning history report instance that was the subject of the signaling of 15002 has been deleted. The signaling of 15003 may include an identifier of the ML reasoning history report instance that has been deleted. The signaling of 15003 may include an explicit indication that the identified ML reasoning history report instance has been deleted. The signaling of 15003 may include a NotifyMOIDeletion service operation. The signaling of 15003 may be considered a response to the signaling of 15002 because it indicates that the signaling of 15002 has succeeded with respect to its requested goal.

[0235] In other words, Fig.15 In , consumers can request to delete unwanted ML inference history function report instances. This can be done using Fig.15 The 3GPP provided management service deleteMOI process is shown to be implemented, but it should be understood that other service operations can be used to implement the function. After the deletion request, the ML reasoning history function producer can notify the consumer about the success of the deletion of the execution of the ML reasoning history function report instance.

[0236] Figures 16 to 18 Shown is signaling that can be used to manage and control ML model inference history requests.

[0237] Fig.16 Operations that may be performed to read characteristics of one or more ML model inference history requests are shown.

[0238] Fig.16 16 shows signaling that can be performed by at least one consumer 1601 and at least one producer 1602. The at least one consumer and the producer can correspond to Fig.11 and / or Fig.12 and / or Fig.13 and / or Fig.14 and / or Fig.15 at least one consumer and producer of , and can therefore perform at least one of the actions described therein.

[0239] During 16001, at least one consumer 1601 and producer 1602 may exchange signaling for instantiating ML reasoning history requests (Req1) and reporting instances. This may be as described above with reference to Fig.11 as described.

[0240] During 16002, at least one consumer 1601 signals to a producer 1602. The signaling of 16002 may include a request for information related to at least one attribute of an ML reasoning history request. The signaling of 16002 may include a ProvMnS.getMOIAttributes service operation. The signaling of 16002 may include an identifier of the ML reasoning history request that is the subject of the signaling of 16002.

[0241] During 16003, producer 1602 responds to the signaling of 16002. The response may include an identifier of the ML Inference History Request associated with the signaling of 16002. The response may include data. The data may include a value associated with an instance of MLInferenceHistoryRequests. Thus, the type of the value may be as defined in the IOC of MLInferenceHistoryRequests. The response may be implemented by a file / stream report service operation.

[0242] In other words, Fig.16 In the example of , the ML reasoning history function has capabilities and a control interface that allow a consumer (e.g., an operator) to configure and manage one or more ML reasoning history function requests. Among other functions, the control interface can enable the consumer to read the characteristics of the submitted ML reasoning history function request. This can be done using e.g. Fig.16 The 3GPP management service provided by the getMOIAttributes procedure is implemented as shown. For example, a consumer can request to read the number of submitted ML reasoning history function requests and / or the characteristics of different ML reasoning history function requests (e.g., priority, source, etc.).

[0243] Fig.17 Example signaling that may be performed to modify at least one characteristic of one or more ML reasoning history requests is shown.

[0244] Fig.17 1701 and at least one producer 1702. The at least one consumer and the producer may correspond to Fig.11 and / or Fig.12 and / or Fig.13 and / or Fig.14 and / or Fig.15 and / or Fig.16at least one consumer and producer of , and can therefore perform at least one of the actions described therein.

[0245] During 17001, at least one consumer 1701 and producer 1702 may exchange signaling for instantiating ML reasoning history requests (Req1) and reporting instances. This may be as described above with reference to Fig.11 as described.

[0246] During 17002, at least one consumer 1701 signals to a producer 1702. The signaling may include an indication of requesting modification of at least one attribute of an ML reasoning history request. The signaling of 17002 may include an identifier of an ML reasoning history request, at least one attribute of which is to be modified. The signaling of 17002 may include an identifier of at least one attribute requested to be modified. The signaling of 17002 may include a ProvMnS.modify.MOIAttributes service operation.

[0247] During 17003, producer 1701 may respond to the signaling of 17002. The response may include an indication that the values ​​associated with the attributes respectively associated with the at least one ML reasoning history request instance have been modified from a previous value to a current value. The signaling of 17003 may include an indication of the current value of any modified attribute identified in the signaling of 17003. The signaling of 17003 may include at least one identifier associated with the at least one ML reasoning history request instance that is the subject of the signaling of 17003. The signaling of 17003 may include a notifyMOIAttributeValueChanges service operation.

[0248] exist Fig.18 In an example of , the control interface can be configured to enable a consumer to configure submitted ML reasoning history function requests, for example, an operator can change the priority of one or more ML reasoning history function requests to indicate those requests that need to be prioritized relative to the instantiation of the relevant ML reasoning history function report instance. Relatedly, the control interface can enable a consumer to update the priority of a ML reasoning history function request sent a priori by the consumer.

[0249] For example, when a consumer adjusts a request, the producer must decide whether the modified request can still be served by the same original job or whether a new job needs to be instantiated. Fig.18 The 3GPP provides management services for the modifyMOIAttributes procedure shown to implement the configuration. After the update request, the ML Reasoning History Function may notify the consumer about the success of the update requested to be performed by the ML Reasoning History Function.

[0250] Fig.18 15 shows signaling that can be performed by at least one consumer 1501 and at least one producer 1802. The at least one consumer and the producer can correspond to Fig.11 and / or Fig.12 and / or Fig.13 and / or Fig.14 and / or Fig.15 and / or Fig.16 and / or Fig.17 at least one consumer and producer of , and can therefore perform at least one of the actions described therein.

[0251] During 18001, at least one consumer 1801 and producer 1802 may exchange signaling for instantiating ML reasoning history requests (Req1) and reporting instances. This may be as described above with reference to Fig.11 as described.

[0252] During 18002, at least one consumer 1801 may signal a producer 1802. The signaling of 18002 may include a request to delete a managed object instance. For example, the signaling may include a request to delete an ML reasoning history request instance. The signaling of 18002 may include an identifier of the ML reasoning history request instance that is the subject of the signaling of 18002.

[0253] During 18003, producer 1802 signals at least one consumer 1801. The signaling of 18003 may include an indication that the ML reasoning history request instance that was the subject of the signaling of 18002 has been deleted. The signaling of 18003 may include an identifier of the deleted ML reasoning history request instance. The signaling of 18003 may include an explicit indication that the identified ML reasoning history request instance has been deleted. The signaling of 18003 may include a NotifyMOIDeletion service operation. The signaling of 18003 may be considered a response to the signaling of 18002 because it indicates that the signaling of 18002 has succeeded with respect to its requested goal.

[0254] exist Fig.18 In the example of , the control interface can be configured to enable consumers to delete unnecessary ML reasoning history function requests. This can be achieved using the 3GPP deleteMOI procedure that provides management services, such as Fig.18 shown.

[0255] Therefore, when a consumer requests to delete an ML inference history function request, the producer receiving the request may delete the corresponding job when the job serves only the one request, or the producer may modify the properties of the job to remove the requirement of the deleted request.

[0256] After the deletion request, the ML reasoning history function may notify the consumer about the success of the deletion requested by the executed ML reasoning history function.

[0257] Fig.19 and 20 Example aspects of the above-described apparatus are shown. It should therefore be appreciated that the features of the above-described examples may be implemented hereinafter without loss of generality.

[0258] Fig.19 Operations that may be performed by a device for a first function are shown. Fig.19 The operation can be performed by a consumer (e.g., an authorized consumer). Fig.19 The operation can be performed by the producer.

[0259] During 1901, the device signals to a second function a request for information about at least one decision made by the machine learning model and a corresponding context associated with making the at least one decision. The second function may be as described below with respect to Fig. 20 described.

[0260] During 1902, the device receives the requested information from the second function.

[0261] The device may signal to the second function a request to modify at least one parameter included in the request, and receive an indication that the second function has modified reporting of the requested information in accordance with the modification of the at least one parameter.

[0262] The device may signal a request to the second function to delete any reports associated with the request, and receive an indication that the second function will no longer report the requested information to the first function.

[0263] The device may use the received request information to determine whether to change at least one network parameter and / or machine learning decision. This may occur when it is determined to change at least one network parameter and / or machine learning decision based on the received information.

[0264] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0265] Fig. 20 Operations that may be performed by a device for a second function are shown. The second function may be, for example, an ML reasoning history function.

[0266] During 2001, the device receives a request from a first function for information about at least one decision made by a machine learning model and a corresponding context associated with making the at least one decision. The first function may be Fig.19 equipment.

[0267] During 2002, the device obtains the requested information.

[0268] During 2003, the device sends the requested information to the first function signaling.

[0269] Obtaining the requested information may include retrieving information related to the at least one decision from a storage function, and processing the retrieved information to form the requested information.

[0270] In response to the receiving, the device may determine that the second function is currently maintaining a first reporting instance regarding the requested information, and update the first instance to include instructions to signal the requested information to the first function.

[0271] Updating the first instance may include associating, in the first instance, an indication of a frequency with which the requested information is provided to the requesting first function with an identifier of the first function.

[0272] In response to the receiving, the device may determine that the second function is not currently maintaining a first reporting instance regarding the requested information, and instantiate a second instance including instructions to signal the requested information to the first function.

[0273] Determining that the second function does not currently maintain a first reporting instance regarding the requested information may include at least one of: determining that the received request includes an indication of a machine learning model that is different from any machine learning model that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a context that is different from any context that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a termination condition that is different from any termination condition that is the subject of the first reporting instance.

[0274] The device may provide the first function with an identifier of the first instance and / or an identifier of the second instance.

[0275] Obtaining the requested information may include: determining a filtering condition for obtaining the requested information; and applying the determined filtering condition to collect the requested information.

[0276] The filtering conditions can be obtained from the MLInferenceHistoryRequest. For example, a ReportingContext can be indicated, where the ReportingContext is a list with at least one entry, each entry in the list is represented / defined by three fields <attribute, condition, value range>. The attribute field can indicate a certain key performance indicator name, which the second function can use to determine where to apply the field. The second function can apply further filtering using the condition and value range fields. When all filtering has been performed, the second function can generate a complete report as requested (MLInferenceHistoryReporting).

[0277] The device may receive a request from the first function to modify at least one parameter included in the request, and signal to the first function an indication that the second function has modified the reporting of the requested information in accordance with the modification of the at least one parameter.

[0278] The device may receive a request from the first function to delete any reports associated with the request, and signal an indication that the second function will no longer report the requested information to the first function.

[0279] The request may include at least one of: an identifier of a machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the requesting first function.

[0280] The system described herein has many advantages. For example, the currently disclosed mechanism enables authorized consumers to request and receive a contextualized history of ML reasoning. For example, such a history can be used to determine the appropriateness of different decisions in different contexts and to determine whether to trigger ML model retraining. ML model retraining can be triggered when it is determined that multiple relatively poor decisions have been made. The determination of whether multiple decisions have been made can be performed using predetermined criteria. For example, a decision can be determined to be poor when it does not result in an improvement in its expected implementation when implemented and / or when any improvement in its expected implementation is minimal relative to a predetermined threshold. In addition, the historical information provided can be used, for example, to compare decisions made by different ML models or ML applications before selecting a model or application to be used in a given context.

[0281] Figure 2An example of a control device for a communication system is shown, which is coupled to and / or used to control a station of an access system, such as a RAN node, such as a base station, a gNB, a central unit of a cloud architecture, or a node of a core network such as an MME or S-GW, a scheduling entity such as a spectrum management entity, or a server or host, such as a device hosting an NRF, NWDAF, AMF, SMF, UDM / UDR, etc. The control device can be integrated with a node or module of the core network or RAN or external to it. In some examples, the base station includes a separate control device unit or module. In other examples, the control device can be another network element, such as a radio network controller or a spectrum controller. The control device 200 can be arranged to provide control of communications in a service area of ​​the system. The device 200 includes at least one memory 201, at least one data processing unit 202, 203, and an input / output interface 204. Via the interface, the control device can be coupled to a receiver and a transmitter of the device. The receiver and / or the transmitter can be implemented as a radio front end or a remote radio head. For example, the control device 200 or the processor 201 can be configured to execute appropriate software code to provide control functions.

[0282] Now refer to Figure 3 describing in more detail possible wireless communication devices, Figure 3 A schematic partial cutaway view of a communication device 300 is shown. Such a communication device is generally referred to as a user equipment (UE) or terminal. Suitable mobile communication devices can be provided by any device capable of sending and receiving radio signals. Non-limiting examples include a mobile station (MS) or a mobile device such as a mobile phone or a so-called "smart phone", a computer equipped with a wireless interface card or other wireless interface facilities (e.g., a USB dongle), a personal data assistant (PDA) or a tablet computer equipped with wireless communication capabilities, or any combination of these or the like. Mobile communication devices can provide, for example, communications for carrying data such as voice, e-mail (email), text messages, multimedia, etc. Therefore, many services can be provided and supplied to users via their communication devices. Non-limiting examples of these services include two-way or multi-way calls, data communications or multimedia services, or simply access to a data communication network system, such as the Internet. Broadcast or multicast data can also be provided to users. Non-limiting examples of content include downloads, television and radio programs, videos, advertisements, various alarms, and other information.

[0283] A wireless communication device may be, for example, a mobile device, i.e., a device that is not fixed to a particular location, or it may be a stationary device. A wireless device may require human interaction to communicate, or may not require human interaction to communicate. As described herein, the term UE or "user" is used to refer to any type of wireless communication device.

[0284] The wireless device 300 may receive signals over the air or radio interface 307 via appropriate means for receiving, and may transmit signals via appropriate means for sending radio signals. Figure 3 In the embodiment, the transceiver arrangement is schematically designated by block 306. For example, the transceiver arrangement 306 may be provided by means of a radio component and an associated antenna arrangement. The antenna arrangement may be arranged internally or externally to the wireless device.

[0285] The wireless device typically has at least one data processing entity 301, at least one memory 302 and possibly other components 303 for software and hardware assisted execution of the tasks it is designed to perform, including controlling access to and communication with access systems and other communication devices. The data processing, storage and other related control devices may be provided on appropriate circuit boards and / or in chipsets. This feature is indicated by reference numeral 304. The user may control the operation of the wireless device by means of a suitable user interface such as a keyboard 305, voice commands, a touch-sensitive screen or tablet, a combination thereof, etc. A display 308, a speaker and a microphone may also be provided. In addition, the wireless communication device may include appropriate connectors (wired or wireless) to other devices and / or for connecting external accessories (e.g., hands-free devices) to other devices.

[0286] Figure 4 A schematic diagram of non-volatile storage media 400a (e.g., a computer disk (CD) or digital compact disk (DVD)) and 400b (e.g., a universal serial bus (USB) memory stick) storing instructions and / or parameters 402 that, when executed by a processor, allow the processor to perform Fig.19 and / or Fig. 20 The method and / or one or more steps of a method otherwise described previously.

[0287] As provided herein, various aspects are described in the detailed description of the embodiments and the claims. Typically, some examples may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, but examples are not limited thereto. Although various examples may be illustrated and described as block diagrams, flow charts, or represented using some other diagrams, it is well understood that, as non-limiting examples, these boxes, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic, general hardware or controllers or other computing devices, or some combination thereof.

[0288] The examples may be implemented by computer software stored in a memory and executable by at least one data processor of the entity involved, or by hardware, or by a combination of software and hardware. Furthermore, in this regard, it should be noted that any process (e.g. Fig.19 and / or Fig. 20 The software may be stored on physical media such as memory chips or memory blocks implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and data variants thereof, CDs, etc.

[0289] The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. As non-limiting examples, the data processor may be of any type suitable for the local technical environment and may include one or more of a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a gate-level circuit, and a processor based on a multi-core processor architecture.

[0290] Additionally or alternatively, some examples may be implemented using circuits. The circuits may be configured to perform one or more of the previously described functions and / or method steps. The circuits may be provided in a base station and / or in a communication device and / or in a core network entity.

[0291] As used in this application, the term "circuitry" may refer to one or more or all of the following:

[0292] (a) implemented only in hardware circuits (such as implemented only in analog and / or digital circuits);

[0293] (b) A combination of hardware circuits and software, such as:

[0294] (i) a combination of analog and / or digital hardware circuitry and software / firmware, and

[0295] (ii) any portion of a hardware processor (including a digital signal processor) with software, software and memory that work together to enable a device such as a communication device or base station to perform the various functions previously described; and

[0296] (c) Hardware circuits and / or processors, such as a microprocessor or portion of a microprocessor, that require software (e.g., firmware) for operation, but where software is not required for operation, the software may not be present.

[0297] This definition of circuitry applies to all uses of this term in this application, including in any claims. As another example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or a portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example, an integrated device.

[0298] The foregoing description has provided a complete and informative description of some examples by way of non-limiting examples. However, various modifications and adaptations may become apparent to those skilled in the relevant arts in view of the foregoing description when read in conjunction with the accompanying drawings and claims. However, all such and similar modifications of the teachings will still fall within the scope of the claims.

[0299] In the above, different examples are described using radio access architectures based on Long Term Evolution Advanced (LTE-Advanced, LTE-A) or New Radio (NR, 5G) as examples of access architectures to which the described techniques may be applied, however, the examples are not limited to such architectures. By appropriately adjusting parameters and procedures, these examples may also be applied to other types of communication networks with suitable devices. Some examples of other options for suitable systems are Universal Mobile Telecommunications System (UMTS) Radio Access Network (UTRAN), Wireless Local Area Network (WLAN or WiFi), Worldwide Interoperability for Microwave Access (WiMAX), Personal Communications Service (PCS), Wideband Code Division Multiple Access (WCDMA), systems using Ultra-Wideband (UWB) technology, sensor networks, mobile ad hoc networks (MANETs), and Internet Protocol Multimedia Subsystem (IMS), or any combination thereof.

[0300] Figure 5 An example of a simplified system architecture is depicted showing only some elements and functional entities, which are logical units, the implementation of which may differ from what is shown. Figure 5 The connections shown are logical connections; the actual physical connections may be different. It will be apparent to those skilled in the art that a system will typically include other components besides Figure 5 Other functions and structures than those shown in .

[0301] However, the examples are not limited to the systems given as examples, but a person skilled in the art may apply the solution to other communication systems having the necessary properties.

[0302] Figure 5 The example of FIG. 4 shows a portion of an exemplary radio access network. For example, the radio access network may support sidelink communications as described in more detail below.

[0303] Figure 5 Devices 500 and 502 are shown. Devices 500 and 502 are configured to be in wireless connection with node 504 on one or more communication channels. Node 504 is also connected to core network 506. In one example, node 504 can be an access node, such as an (e / g)NodeB serving devices in a cell. In one example, node 504 can be a non-3GPP access node. The physical link from the device to the (e / g)NodeB is referred to as an uplink or reverse link, and the physical link from the (e / g)NodeB to the device is referred to as a downlink or forward link. It should be understood that (e / g)NodeB or their functions can be implemented by using entities such as any node, host, server or access point suitable for such use.

[0304] The communication system typically includes more than one (e / g)NodeB, in which case the (e / g)NodeBs may also be configured to communicate with each other via wired or wireless links designed for this purpose. These links may be used for signaling purposes. (e / g)NodeB is a computing device configured to control the radio resources of the communication system to which it is coupled. NodeB may also be referred to as a base station, an access point, or any other type of interface device including a relay station capable of operating in a wireless environment. (e / g)NodeB includes or is coupled to a transceiver. A connection is provided from the transceiver of the (e / g)NodeB to an antenna unit, which establishes a bidirectional radio link to the device. The antenna unit may include multiple antennas or antenna elements. (e / g)NodeB is further connected to a core network 506 (CN or next generation core NGC). Depending on the technology deployed, the (e / g)NodeB is connected to a Serving and Packet Data Network Gateway (S-GW+P-GW) or User Plane Function (UPF) for routing and forwarding user data packets and for providing connectivity of the device to one or more external packet data networks, and to a Mobility Management Entity (MME) or Access Mobility Management Function (AMF) for controlling access and mobility of the device.

[0305] Examples of apparatus are subscriber units, user devices, user equipment (UE), user terminals, terminal equipment, mobile stations, mobile devices, and the like.

[0306] Device generally refers to a mobile or static device (e.g., a portable or non-portable computing device), including wireless mobile communication devices operating with or without a Universal Subscriber Identity Module (USIM), including but not limited to the following types of devices: mobile phones, smart phones, personal digital assistants (PDAs), handhelds, devices using wireless modems (alarm or measurement devices, etc.), laptop and / or touch screen computers, tablets, game consoles, notebooks, and multimedia devices. It should be understood that a device can also be an almost exclusive uplink-only device, an example of which is a camera or video camera that loads images or video clips to a network. A device can also be a device with the ability to operate in an Internet of Things (IoT) network, which is a scenario in which objects are provided with the ability to transmit data on a network without the need for human-to-human or human-to-computer interaction, such as for use in smart grids and connected vehicles. The device can also make use of the cloud. In some applications, a device may include a user-portable device with a radio component (such as a watch, headphones, or glasses), and the calculation is performed in the cloud.

[0307] The device illustrates a type of device to which resources on the air interface are allocated and assigned, and therefore any features described herein using a device arrangement may be implemented using a corresponding device (such as a relay node). An example of such a relay node is a layer 3 relay (self-backhaul relay) toward a base station. The device (or, in some examples, a layer 3 relay node) is configured to perform one or more of the user equipment functions.

[0308] The various techniques described herein can also be applied to cyber-physical systems (CPS) (systems of cooperating computing elements that control physical entities). CPS can implement and utilize a large number of interconnected information and communication technologies, ICTs, devices (sensors, actuators, processors, microcontrollers, etc.) embedded in physical objects in different locations. Mobile cyber-physical systems are a subcategory of cyber-physical systems, where the physical system in question has inherent mobility. Examples of mobile physical systems include mobile robots and electronic devices transported by humans or animals.

[0309] Additionally, although the device has been depicted as a single entity, different units, processors and / or memory units may be implemented. Figure 5 Not all are shown).

[0310] 5G enables the use of multiple-input multiple-output (MIMO) antennas, many more base stations or nodes than LTE (the so-called small cell concept), including macro sites that operate in collaboration with smaller stations, and various radio technologies depending on service requirements, use cases and / or available spectrum. 5G mobile communications support a wide range of use cases and related applications, including video streaming, augmented reality, different ways of data sharing, and various forms of machine-type applications (e.g., (massive) machine-type communications (mMTC), including vehicle safety, different sensors, and real-time control). 5G is expected to have multiple radio interfaces, such as below 6 GHz or above 24 GHz, centimeter waves, and millimeter waves, and can also be integrated with existing traditional radio access technologies such as LTE. At least in the early stages, integration with LTE can be implemented as a system in which macro coverage is provided by LTE and 5G radio interface access comes from small cells by aggregation to LTE. In other words, 5G plans to support both inter-RAT operability (e.g., LTE-5G) and inter-RI operability (inter-radio interface operability, such as below 6 GHz-centimeter waves, 6 or above 24 GHz-centimeter waves and millimeter waves). One of the concepts being considered for use in 5G networks is network slicing, where multiple independent and dedicated virtual subnetworks (network instances) can be created within the same infrastructure to run services with different requirements for latency, reliability, throughput, and mobility.

[0311] LTE network architecture is fully distributed in the radio and fully centralized in the core network. Low latency applications and services in 5G require content close to the radio, which leads to local burst and multi-access edge computing (MEC). 5G enables analytics and knowledge generation to be done at the data source. This approach requires leveraging resources that may not be continuously connected to the network, such as laptops, smartphones, tablets, and sensors. MEC provides a distributed computing environment for application and service hosting. It also enables the storage and processing of content close to cellular users for faster response times. Edge computing covers a wide range of technologies such as wireless sensor networks, mobile data collection, mobile signature analysis, collaborative distributed peer-to-peer networking and processing, and can also be categorized as local cloud / fog computing and grid / mesh computing, dew computing, mobile edge computing, cloudlets, distributed data storage and retrieval, autonomous self-healing networks, remote cloud services, augmented and virtual reality, data caching, Internet of Things (massive connectivity and / or latency critical), critical communications (autonomous vehicles, traffic safety, real-time analytics, time-critical control, healthcare applications).

[0312] The communication system can also communicate with or utilize services provided by other networks 512, such as a public switched telephone network, a VoIP network, the Internet, or a private network. The communication network can also be capable of supporting the use of cloud services, for example, at least a portion of the core network operations can be performed as a cloud service (this is in the Figure 5 514). When performed away from the core network, this may also be referred to as edge computing. The communication system may also include a central control entity or the like, which provides facilities for networks of different operators to collaborate, for example in spectrum sharing.

[0313] Edge computing technology can be introduced into the radio access network (RAN) using network function virtualization (NFV) and software defined network (SDN). The use of edge cloud technology can mean that access node operations are at least partially performed in a server, host or node that is operably coupled to a remote radio head or base station including a radio component. Node operations may also be distributed among multiple servers, nodes or hosts. The application of cloudRAN architecture enables RAN real-time functions to be performed at or near the remote antenna site (in the distributed unit DU 508), and non-real-time functions can be performed in a centralized manner (in the centralized unit CU 510).

[0314] It should also be understood that the labor distribution between core network operations and base station operations may be different from that of LTE or even non-existent. Some other technological advances that may be used are big data and all-IP, which may change the way networks are built and managed. 5G (or New Radio, NR) networks are being designed to support multiple hierarchical structures where edge computing servers can be placed between the core and the base station or nodeB (gNB). An example of edge computing is MEC defined by the European Telecommunications Standards Institute. It should be understood that MEC (and other edge computing protocols) can also be applied in 4G networks.

[0315] 5G can also make use of satellite communications to enhance or complement the coverage of 5G services, for example by providing backhaul. Possible use cases are providing service continuity for machine-to-machine (M2M) or Internet of Things (IoT) devices or on-board passengers, mobile broadband (MBB) or ensuring service availability for critical communications and future rail / maritime / aeronautical communications. Satellite communications can make use of geostationary orbit (GEO) satellite systems, but also low Earth orbit (LEO) satellite systems, in particular mega-constellations (systems where hundreds of (nano)satellites are deployed). Each satellite in a mega-constellation can cover multiple satellite-enabled network entities creating ground cells. Ground cells can be created via ground relay nodes or by gNBs located on the ground or in satellites.

[0316] The depicted system is only an example of a part of a radio access system, and in practice, the system may include multiple (e / g)NodeBs, the device may have access to multiple radio cells, and the system may also include other devices, such as physical layer relay nodes or other network elements. At least one of the (e / g)NodeBs may be a Home (e / g)NodeB. In addition, in the geographical area of ​​the radio communication system, multiple different kinds of radio cells as well as multiple radio cells may be provided. The radio cell may be a macro cell (or umbrella cell), which typically has a diameter of up to tens of kilometers, or a smaller cell, such as a micro cell, a femto cell or a pico cell. Figure 5 The (e / g)NodeB can provide any type of these cells. A cellular radio system can be implemented as a multi-layer network comprising several types of cells. Typically, in a multi-layer network, an access node.

Claims

1. A device for a first function, the device comprising means for performing the following operations: signaling to a second function a request for information of at least one decision made by the machine learning model and a corresponding context associated with making the at least one decision; and The requested information is received from the second function.

2. The apparatus according to claim 1, comprising means for: sending a request to modify at least one parameter included in the request to the second function signaling; and An indication is received that the second function has modified the reporting of the requested information based on the modification of the at least one parameter.

3. Apparatus according to any one of the preceding claims, comprising means for: signaling to the second function a request to delete any reports associated with the request; and An indication is received that the second function will no longer report the requested information to the first function.

4. A device for a second function, the device comprising means for performing the following operations: receiving, from a first function, a request for information of at least one decision made by a machine learning model and a corresponding context associated with making the at least one decision; obtaining the requested information; and The requested information is sent to the first function signaling.

5. The device according to claim 4, wherein: The means for obtaining the requested information comprises means for performing the following operations: retrieving information related to the at least one decision from a storage function; and The retrieved information is processed to form the requested information.

6. The apparatus according to any one of claims 4 to 5, further comprising means for: In response to said receiving, determining that said second function is currently maintaining a first reporting instance regarding said requested information; and The first instance is updated to include instructions to signal the requested information to the first function.

7. The device according to claim 6, wherein: The means for updating the first instance comprises means for performing the following operations: In the first instance, an indication of a frequency with which the requested information is provided to the requesting first function is associated with an identifier of the first function.

8. The apparatus according to any one of claims 4 to 5, further comprising means for: In response to said receiving, determining that said second function is not currently maintaining a first reporting instance regarding said requested information; and A second instance is instantiated, which includes an instruction to signal the requested information to the first function.

9. The device according to claim 8, wherein: The means for performing the determination that the second function is not currently maintaining a first reporting instance regarding the requested information comprises means for performing at least one of the following operations: determining that the received request includes an indication of a machine learning model that is different from any machine learning model that is the subject of the first reporting instance; and / or determining that the received request includes an indication of a context different from any context that was the subject of the first reporting instance; and / or A determination is made that the received request includes an indication of a termination condition different from any termination condition that was the subject of the first reporting instance.

10. Apparatus according to any one of claims 6 to 9, comprising means for: The first function is provided with an identifier of the first instance and / or an identifier of the second instance.

11. The apparatus according to any one of claims 4 to 10, wherein: The means for obtaining the requested information comprises means for performing the following operations: determining filtering conditions for obtaining the requested information; and The determined filter conditions are applied to collect the requested information.

12. Apparatus according to any one of claims 4 to 11, comprising means for: receiving, from the first function, a request to modify at least one parameter included in the request; and An indication is signaled to the first function that the second function has modified the reporting of the requested information in accordance with the modification of the at least one parameter.

13. Apparatus according to any one of claims 4 to 12, comprising means for: receiving, from the first function, a request to delete any report associated with the request; and Signaling an indication that the second function will no longer report the requested information to the first function.

14. Apparatus according to any one of the preceding claims, wherein: The request includes at least one of: an identifier of the machine learning model, an identifier of a machine learning application that includes the machine learning model, an indication of a context for the second function to consider when providing the requested information, and / or an indication of a frequency with which the requested information is to be provided to the first function being requested.

15. A method for a device for a first function, the method comprising: signaling to a second function a request for information of at least one decision made by the machine learning model and a corresponding context associated with making the at least one decision; and The requested information is received from the second function.

16. A method for a device for a second function, the method comprising: receiving, from a first function, a request for information of at least one decision made by a machine learning model and a corresponding context associated with making the at least one decision; obtaining the requested information; and The requested information is sent to the first function signaling.

17. A computer program product, when executed on a device for a first function, causing the device to perform: signaling to a second function a request for information of at least one decision made by the machine learning model and a corresponding context associated with making the at least one decision; and The requested information is received from the second function.

18. A computer program product, when executed on a device for a second function, causes the device to perform: receiving, from a first function, a request for information of at least one decision made by a machine learning model and a corresponding context associated with making the at least one decision; obtaining the requested information; and The requested information is sent to the first function signaling.

Images