Transfer learning with model producer control in a wireless communication network

Abstract

Various aspects of the present disclosure relate to a method (1000) performed by a first network function, comprising: determining (1002) a requirement for transfer learning for a learning model task; and discovering (1004) one or more second network functions for supporting transfer learning for the learning model task, based at least in part on one or more transfer learning capabilities of the one or more second network functions.

Description

TRANSFER LEARNING WITH MODEL PRODUCER CONTROL IN A WIRELESS COMMUNICATION NETWORKTECHNICAL FIELD

[0001] The subject matter disclosed herein relates generally to the field of implementing transfer learning (TL) with model producer control in a wireless communication network. In particular, this document defines a first network function for wireless communication, a second network function for wireless communication, a third network function for wireless communication, and methods performed by a first network function, second network function and third network function.BACKGROUND

[0002] A wireless communications system may include one or multiple network communication devices, such as base stations, which may support wireless communications for one or multiple user communication devices, which may be otherwise known as user equipment (UE), or other suitable terminology. The wireless communications system may support wireless communications with one or multiple user communication devices by utilizing resources of the wireless communication system (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers, or the like). Additionally, the wireless communications system may support wireless communications across various radio access technologies including third generation (3G) radio access technology, fourth generation (4G) radio access technology, fifth generation (5G) radio access technology, among other suitable radio access technologies beyond 5G (e.g., sixth generation (6G)).

[0003] In some cases, the wireless communication system may support network analytics and learning models. The learning models may include artificial intelligence (AI) / machine learning (ML) models. For example, one or more network entities or network functions of a core network, such as a 5G core network, may support various network analytics and learning models. For example, a network data analytics function (NWDAF) may support various analytics types as described in the 3 GPP Technical Specification TS23.288 titled, “Architecture enhancements for 5G system to support network data analytics services”. The NWDAF may support one or more analytics identifiers (IDs) and may be operable to, or configured to support inference and may referred to as an NWDAF analytics logical function (AnLF). The NWDAF may be operable to, or configured to support training and may be referred to as an NWDAF model training logical function (MTLF). In some cases, the NWDAF may support both inference and training functionality. An AnLF that supports a specific analytics ID inference may subscribe to a corresponding MTLF that supports training.SUMMARY

[0004] An article “a” before an element is unrestricted and understood to refer to “at least one” of those elements or “one or more” of those elements. The terms “a,” “at least one,” “one or more,” and “at least one of one or more” may be interchangeable. As used herein, including in the claims, “or” as used in a list of items (e.g., a list of items prefaced by a phrase such as “at least one of’ or “one or more of’ or “one or both of’) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). Also, as used herein, the phrase “based on” shall not be construed as a reference to a closed set of conditions. For example, an example step that is described as “based on condition A” may be based on both a condition A and a condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase “based on” shall be construed in the same manner as the phrase “based at least in part on. Further, as used herein, including in the claims, a “set” may include one or more elements.

[0005] There is provided a first network function for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first network function to: determine a requirement for transfer learning (TL) for a learning model task; and discover one or more second network functions for supporting TL for the learning model task, based at least in part on one or more TL capabilities of the one or more second network functions.

[0006] There is further provided a second network function for wireless communication, comprising: at least one memory; and at least one processor coupled withthe at least one memory and configured to cause the second network function to: transmit, to a third network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities.

[0007] There is further provided a third network function for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the third network function to: receive, from a second network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities; and store, as a network function profile for the second network function, the one or more fourth parameters.

[0008] There is further provided a method performed by a first network function, the method comprising: determining a requirement for TL for a learning model task; and discovering one or more second network functions for supporting TL for the learning model task, the discovering being based at least in part on one or more TL capabilities of the one or more second network functions.

[0009] There is further provided a method performed by a second network function for wireless communication, comprising transmitting, to a third network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities.

[0010] There is further provided a method performed by a third network function for wireless communication, comprising: receiving, from a second network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities; and storing, as a network function profile for the second network function, the one or more fourth parameters.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 illustrates an example of a wireless communications system in accordance with aspects of the present disclosure.

[0012] Figure 2 illustrates an example of NWDAF flavours including potential input data sources and output consumers in accordance with aspects of the present disclosure.

[0013] Figure 3 illustrates an example of the transfer learning (TL) concept in accordance with aspects of the present disclosure.

[0014] Figure 4 illustrates an example of AI / ML model storage for TL in an ADRF and the removal from the ADRF in accordance with aspects of the present disclosure.

[0015] Figure 5 illustrates an example of the registration and discovery of TL enabled NWDAFs containing MTLFs in accordance with aspects of the present disclosure.

[0016] Figure 6 illustrates an example of a procedure for TL AI / ML model retrieval via NWDAF MTLF in accordance with aspects of the present disclosure.

[0017] Figure 7 illustrates an example of a user equipment (UE) 700 in accordance with aspects of the present disclosure.

[0018] Figure 8 illustrates an example of a processor 800 in accordance with aspects of the present disclosure.

[0019] Figure 9 illustrates an example of a network equipment (NE) 900 in accordance with aspects of the present disclosure.

[0020] Figure 10 illustrates a flowchart of a method 1000 performed by a NE in accordance with aspects of the present disclosure.

[0021] Figure 11 illustrates a flowchart of a method 1100 performed by a NE in accordance with aspects of the present disclosure.

[0022] Figure 12 illustrates a flowchart of a method 1200 performed by a NE in accordance with aspects of the present disclosure.DETAILED DESCRIPTION

[0023] A network equipment (NE) such as a network entity or network function, may support an AI / ML technique, in which the NE trains an AI / ML model independently for a specific usage. A specific usage may comprise an action, a task, or an operation, for instance. The NE may train the AI / ML model without inputting training data, parameters, or constraints (i.e., prior training knowledge or experience). In other words, an instantiation of an AI / ML model might require training from scratch, assuming that the NE (such as a corresponding NWDAF MTLF) can collect sufficient input data. The amount of input data that is sufficient may be a quantity of input data satisfying a certain requirement, such as being greater than or equal to a threshold quantity of input data. If the collected input data is insufficient (such as the quantity of input data being less than or equal to the threshold quantity of input data), this may be reflected in a confidence value or parameter provided to an analytics consumer with a requested output data. The confidence value may indicate a quality or accuracy of the output data (i.e., mainly for a predictive data output) generated by the NE, taking into account the collected quantity of input data that is needed.

[0024] In some cases, collecting sufficient input data may be challenging, especially when handling a new analytics service. For example, collecting sufficient input data might be unfeasible or can result in high resource usage (e.g., processing resources) or power consumption, especially when a requested time for training an AI / ML model is limited.

[0025] In some cases, one or more of a NE or a UE may support transfer learning (TL), in which one or more of the NE or the UE may reuse an AI / ML model pre-trained on a specific usage (e.g., action, task, or operation) for a new usage (e.g., a new action, task, or operation), utilizing the prior knowledge gained from the previous usage. One or more of the NE or the UE may be equipped or configured with various protocol layers of a protocol stack. In some cases, at an application layer (as described with reference in 3 GPP Technical Report TR 23.700-82 titled “Study on application layer support for AI / ML services”), TL has been considered in where an AI / ML model characterised as a pre-trained candidate can be stored in a model registry considering: (i) an analytics service (i.e., analytics identifier (ID)) or an application service, (ii) an AI / ML task / job, (iii) user(s) (i.e., UE, or group of UEs), (iv) service profile including a geographic area, time information, environment,usage / intent and charging information, (v) context information for supporting a pre-trained model consumer which may include a set of features, training data requirements, required computing power and (vi) permissions which may include access, exposure and vendor specific permissions.

[0026] A TL enabler (such as an NWDAF MTLF) may be introduced to detect a condition for using TL for an AI / ML usage (e.g., an action, task, or operation) or Analytics ID. A subscriber to the TL enabler may provide context information including a set of features, dataset requirement, model information, a required confidence level, UE(s), service or service profile, geographical area, and time of interest. The TL enabler can discover and fetch (e.g., retrieve, obtain, or receive) an AI / ML model from a repository such as a database and determine whether the AI / ML model shall be used as a pre-trained AI / ML model based on a model request from the consumer, which may also be an enabler client residing (e.g., located) in a UE. The criteria to determine whether the AI / ML model shall be used as the pre-trained AI / ML model may be based on a comparison and similarity of features, data requirements, analytics service, model, or task to perform, whether the AI / ML model was used for a similar usage (e.g., a task, action, or operation) and related model rating or reputation of the AI / ML model or input data sources of the AI / ML model.

[0027] Some wireless communications systems, such as 5G networks and 5G core networks support a technique for retrieving (e.g., obtaining, fetching, or receiving) an AI / ML model wherein an NWDAF MTLF model producer and owner authenticates a candidate model consumer before providing requested model information. This technique makes the applicability of TL in the 5G core network different from the one introduced in the application layer as described above. Additionally, these wireless communications systems do not include TL adaptation considerations (i.e., how to adapt an AI / ML model used for one usage to another new usage). Instead, these wireless communications systems contain only a selection criteria. Employing a specific adaptation strategy depends on a target usage (e.g., a task, action, or operation), such as how similar or partially similar the target usage is compared to the source usage, and / or on related network environment conditions. Based on these considerations, it is important to gain an understanding of which part of the knowledge can be transferred, as well as where and when it makes sense totransfer knowledge to gain a performance improvement. In some cases, the adoption of a specific strategy would require interface enhancements by introducing new meta data parameters to describe how to perform knowledge transfer by employing an adaptation strategy, mechanisms to discover and provide the knowledge transfer and decision criteria to identify when such knowledge transfer needs to take place to be beneficial, while avoiding negative effects.

[0028] Various aspects of the present disclosure relate to an apparatus and method that facilitates TL among analytic services by providing TL related information of a model producer and pre-trained AI / ML model. This TL related information is referred to as TL capabilities and TL competencies. Specifically, the TL related information may be provided in the AI / ML model information and via a producer consumer exchange, which can characterize how fit an AI / ML model can be when transferred and used in a specified target analytics service and / or task (i.e., use case, usage). The disclosure herein facilitates transfer learning to model consumers by introducing said transfer learning capabilities of model producers and transfer learning competencies of respective pre-trained AI / ML models for new learning tasks. Potential model consumers tend to be able to discover model producers based on the transfer learning capabilities. Furthermore, the suitability of pre-trained AI / ML models for new learning tasks tends to be able to be determined based on the transfer learning competencies. The disclosure herein tends to enable features of pre-trained AI / ML models to be identified that can be adopted, modified or dropped through TL to convert the pre-trained AI / ML model for use in a new learning model task.

[0029] Aspects of the present disclosure are described in the context of a wireless communications system.

[0030] Figure 1 illustrates an example of a wireless communications system 100 in accordance with aspects of the present disclosure. The wireless communications system 100 may include one or more NE 102, one or more UE 104, and a core network (CN) 106. The wireless communications system 100 may support various radio access technologies. In some implementations, the wireless communications system 100 may be a 4G network, such as an LTE network or an LTE-advanced (LTE-A) network. In some other implementations, the wireless communications system 100 may be a NR network, such as a5G network, a 5G-advanced (5G-A) network, or a 5G ultrawideband (5G-UWB) network. In other implementations, the wireless communications system 100 may be a combination of a 4G network and a 5G network, or other suitable radio access technology including Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20. The wireless communications system 100 may support radio access technologies beyond 5G, for example, 6G. Additionally, the wireless communications system 100 may support technologies, such as time division multiple access (TDMA), frequency division multiple access (FDMA), or code division multiple access (CDMA), etc.

[0031] The one or more NE 102 may be dispersed throughout a geographic region to form the wireless communications system 100. One or more of the NE 102 described herein may be or include or may be referred to as a network node, a base station, a network element, a network function, a network entity, a radio access network (RAN), a NodeB, an eNodeB (eNB), a next-generation NodeB (gNB), or other suitable terminology. An NE 102 and a UE 104 may communicate via a communication link, which may be a wireless or wired connection. For example, an NE 102 and a UE 104 may perform wireless communication (e.g., receive signalling, transmit signalling) over a Uu interface.

[0032] An NE 102 may provide a geographic coverage area for which the NE 102 may support services for one or more UEs 104 within the geographic coverage area. For example, an NE 102 and a UE 104 may support wireless communication of signals related to services (e.g., voice, video, packet data, messaging, broadcast, etc.) according to one or multiple radio access technologies. In some implementations, an NE 102 may be moveable, for example, a satellite associated with a non-terrestrial network (NTN). In some implementations, different geographic coverage areas associated with the same or different radio access technologies may overlap, but the different geographic coverage areas may be associated with different NE 102.

[0033] The one or more UE 104 may be dispersed throughout a geographic region of the wireless communications system 100. A UE 104 may include or may be referred to as a remote unit, a mobile device, a wireless device, a remote device, a subscriber device, a transmitter device, a receiver device, or some other suitable terminology. In someimplementations, the UE 104 may be referred to as a unit, a station, a terminal, or a client, among other examples. Additionally, or alternatively, the UE 104 may be referred to as an Internet-of-Things (loT) device, an Internet-of-Everything (loE) device, or machine-type communication (MTC) device, among other examples.

[0034] A UE 104 may be able to support wireless communication directly with other UEs 104 over a communication link. For example, a UE 104 may support wireless communication directly with another UE 104 over a device-to-device (D2D) communication link. In some implementations, such as vehicle-to-vehicle (V2V) deployments, vehicle-to-everything (V2X) deployments, or cellular-V2X deployments, the communication link may be referred to as a sidelink. For example, a UE 104 may support wireless communication directly with another UE 104 over a PC5 interface.

[0035] An NE 102 may support communications with the CN 106, or with another NE102, or both. For example, an NE 102 may interface with other NE 102 or the CN 106 through one or more backhaul links (e.g., SI, N2, N2, or network interface). In some implementations, the NE 102 may communicate with each other directly. In some other implementations, the NE 102 may communicate with each other or indirectly (e.g., via the CN 106. In some implementations, one or more NE 102 may include subcomponents, such as an access network entity, which may be an example of an access node controller (ANC). An ANC may communicate with the one or more UEs 104 through one or more other access network transmission entities, which may be referred to as a radio heads, smart radio heads, or transmission-reception points (TRPs).

[0036] The CN 106 may support user authentication, access authorization, tracking, connectivity, and other access, routing, or mobility functions. The CN 106 may be an evolved packet core (EPC), or a 5G core (5GC), which may include a control plane entity that manages access and mobility (e.g., a mobility management entity (MME), an access and mobility management functions (AMF)) and a user plane entity that routes packets or interconnects to external networks (e.g., a serving gateway (S-GW), a Packet Data Network (PDN) gateway (P-GW), or a user plane function (UPF)). In some implementations, the control plane entity may manage non-access stratum (NAS) functions, such as mobility,authentication, and bearer management (e.g., data bearers, signal bearers, etc.) for the one or more UEs 104 served by the one or more NE 102 associated with the CN 106.

[0037] The CN 106 may communicate with a packet data network over one or more backhaul links (e.g., via an SI, N2, N2, or another network interface). The packet data network may include an application server. In some implementations, one or more UEs 104 may communicate with the application server. A UE 104 may establish a session (e.g., a protocol data unit (PDU) session, or the like) with the CN 106 via an NE 102. The CN 106 may route traffic (e.g., control information, data, and the like) between the UE 104 and the application server using the established session (e.g., the established PDU session). The PDU session may be an example of a logical connection between the UE 104 and the CN 106 (e.g., one or more network functions of the CN 106).

[0038] In the wireless communications system 100, the NEs 102 and the UEs 104 may use resources of the wireless communications system 100 (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers)) to perform various operations (e.g., wireless communications). In some implementations, the NEs 102 and the UEs 104 may support different resource structures. For example, the NEs 102 and the UEs 104 may support different frame structures. In some implementations, such as in 4G, the NEs 102 and the UEs 104 may support a single frame structure. In some other implementations, such as in 5 G and among other suitable radio access technologies, the NEs 102 and the UEs 104 may support various frame structures (i.e., multiple frame structures). The NEs 102 and the UEs 104 may support various frame structures based on one or more numerologies.

[0039] One or more numerologies may be supported in the wireless communications system 100, and a numerology may include a subcarrier spacing and a cyclic prefix. A first numerology (e.g., / r=0) may be associated with a first subcarrier spacing (e.g., 15 kHz) and a normal cyclic prefix. In some implementations, the first numerology (e.g., / r=0) associated with the first subcarrier spacing (e.g., 15 kHz) may utilize one slot per subframe. A second numerology (e.g., / r=l) may be associated with a second subcarrier spacing (e.g., 30 kHz) and a normal cyclic prefix. A third numerology (e.g., / r=2) may be associated with a third subcarrier spacing (e.g., 60 kHz) and a normal cyclic prefix or an extended cyclicprefix. A fourth numerology (e.g., / r=3) may be associated with a fourth subcarrier spacing (e.g., 120 kHz) and a normal cyclic prefix. A fifth numerology (e.g., / r=4) may be associated with a fifth subcarrier spacing (e.g., 240 kHz) and a normal cyclic prefix.

[0040] A time interval of a resource (e.g., a communication resource) may be organized according to frames (also referred to as radio frames). Each frame may have a duration, for example, a 10 millisecond (ms) duration. In some implementations, each frame may include multiple subframes. For example, each frame may include 10 subframes, and each subframe may have a duration, for example, a 1 ms duration. In some implementations, each frame may have the same duration. In some implementations, each subframe of a frame may have the same duration.

[0041] Additionally or alternatively, a time interval of a resource (e.g., a communication resource) may be organized according to slots. For example, a subframe may include a number (e.g., quantity) of slots. The number of slots in each subframe may also depend on the one or more numerologies supported in the wireless communications system 100. For instance, the first, second, third, fourth, and fifth numerologies (i.e., / r=0, jU=l, / r=2, jU=3, / r=4) associated with respective subcarrier spacings of 15 kHz, 30 kHz, 60 kHz, 120 kHz, and 240 kHz may utilize a single slot per subframe, two slots per subframe, four slots per subframe, eight slots per subframe, and 16 slots per subframe, respectively.# Each slot may include a number (e.g., quantity) of symbols (e.g., OFDM symbols). In some implementations, the number (e.g., quantity) of slots for a subframe may depend on a numerology. For a normal cyclic prefix, a slot may include 14 symbols. For an extended cyclic prefix (e.g., applicable for 60 kHz subcarrier spacing), a slot may include 12 symbols. The relationship between the number of symbols per slot, the number of slots per subframe, and the number of slots per frame for a normal cyclic prefix and an extended cyclic prefix may depend on a numerology. It should be understood that reference to a first numerology (e.g., / r=0) associated with a first subcarrier spacing (e.g., 15 kHz) may be used interchangeably between subframes and slots.

[0042] In the wireless communications system 100, an electromagnetic (EM) spectrum may be split, based on frequency or wavelength, into various classes, frequency bands, frequency channels, etc. By way of example, the wireless communications system 100 maysupport one or multiple operating frequency bands, such as frequency range designations FR1 (410 MHz - 7.125 GHz), FR2 (24.25 GHz - 52.6 GHz), FR3 (7.125 GHz - 24.25 GHz), FR4 (52.6 GHz - 114.25 GHz), FR4a or FR4-1 (52.6 GHz - 71 GHz), and FR5 (114.25 GHz - 300 GHz). In some implementations, the NEs 102 and the UEs 104 may perform wireless communications over one or more of the operating frequency bands. In some implementations, FR1 may be used by the NEs 102 and the UEs 104, among other equipment or devices for cellular communications traffic (e.g., control information, data). In some implementations, FR2 may be used by the NEs 102 and the UEs 104, among other equipment or devices for short-range, high data rate capabilities.

[0043] FR1 may be associated with one or multiple numerologies (e.g., at least three numerologies). For example, FR1 may be associated with a first numerology (e.g., / r=0), which includes 15 kHz subcarrier spacing; a second numerology (e.g., / r=l), which includes 30 kHz subcarrier spacing; and a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing. FR2 may be associated with one or multiple numerologies (e.g., at least 2 numerologies). For example, FR2 may be associated with a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing; and a fourth numerology (e.g., / r=3), which includes 120 kHz subcarrier spacing.

[0044] Figure 2 illustrates an example 200 of NWDAF flavours including potential input data sources and output consumers in accordance with aspects of the present disclosure.

[0045] The various NWDAF flavours and their respective input data and output result consumers are illustrated. Output result consumers may include 5G core network functions (NFs), application functions (AFs), 5G core repositories (e.g., network repository function (NRF), unified data management (UDM), etc.), and the operations, administration and maintenance (0AM) (management service (MnS) consumer or management function (MF)).

[0046] As illustrated in Figure 2, a first data collection coordination functionality (DCCF) 212 receives inputs from 5G core network functions 202, application functions 204, a network exposure function 206 (for use if, for instance, application functions 204 are untrusted), 5G core repositories 208, and 0AM data 210. The 5G core repositories 208may comprise an NRF, a binding support function (BSF), an analytics data repository Function (ADRF), a UDM, and / or a unified data repository (UDR). The OAM 210 may comprise a MnS Producer or an MF that provides performance measurements, key performance indicators, configuration management, and alarm information. Optionally, the first DCCF 212 may provide data to an NWDAF AnLF / MTLF 214, an NWDAF AnLF 216, and an NWDAF MTLF 218.

[0047] The NWDAF containing AnLF / MTLF 214, the NWDAF containing AnLF 216, and the NWDAF containing MTLF 218 may pass on data to the second DCCF 222. The second DCCF 222 may further provide data to the 5G core network functions 224, application functions 228, untrusted application functions 228 via a network exposure function 226, 5G core repositories 230, and OAM data 232. The 5G core repositories 230 may comprise an ADRF, a UDM, and / or a unified data repository (UDR). The OAM 232 may comprise a management services (MnS) consumer or a management function (MF).

[0048] In operation, MTLF 218 and AnLF 216 may exchange AI / ML models. Such exchange may be performed by means of serialization or containerization or via parameters or weights exchange. Optionally, DCCF and MFAF may be involved in the AI / ML model exchange to distribute and collect repeated data towards or from various data sources.

[0049] Figure 3 illustrates an example 300 of the TL concept in accordance with aspects of the present disclosure.

[0050] The example 300 shows a Data Set A 312 being provided to an AI / ML model 314 as part of an Analytics Task A 310. The AI / ML model 314 is shown as outputting Analytics A 316. The example 300 further shows a Data Set B 322 being provided to an AI / ML model 324 as part of Analytics Task B 320. The AI / ML model 324 is shown as outputting Analytics B 326.

[0051] The example 300 shows AI / ML model 314 and AI / ML model 324 being the same AI / ML model X. Specifically, the example 300 shows a knowledge transfer 330, i.e., transfer of AI / ML Model X, among the two different analytics tasks (e.g., Analytics Task A 310, Analytics Task B 320) that use different input data sets (e.g., Data Set A 312, Data Set B 322) to produce a different output (e.g., Analytics A 316, Analytics B 326). This is theTL principle. TL is a method that can reuse a pre-trained AI / ML model 314 or components of a pre- trained AI / ML model 314 from a first task 310: (i) as pre-trained, i.e., without further re-training, or (ii) as the starting point for further training with respect to a new task 320.

[0052] TL may use a pre-trained AI / ML model which employs common features applied on a similar network environment having common input data statistics (e.g., the same Analytics ID among urban environments with similar characteristics). TL may use a pre-trained AI / ML model which employs common features applied on a different network environment, i.e., different geographical area or density of RAN and / or encountering different input data statistics, for example, in terms of the data range / distribution, average / deviation or data quantity (e.g., the same Analytics ID among different cities with a different population). TL may use a pre-trained AI / ML model which employs partially common features with some additional or different features applied on the same network environment having common input data statistics, (e.g., different Analytics IDs employed in the same area of interest). TL may use a pre-trained AI / ML model which employs partially common features with some additional or different features applied on a different network environment and / or encountering different input data statistics, (e.g., different Analytics IDs employed among different areas of interest with a different population).

[0053] Ideally, the adoption of TL (i.e., registration of a network function and / or pretrained AI / ML models for TL and their discovery) shall be based on the TL capabilities that include the indicated permissions (e.g., per vendor and / or operator / vertical), slice and service area and on a degree of commonality among the involved analytics services, (i.e., AI / ML model features, input data characteristics used for pre-training, the network environment (e.g., topology connectivity, density, network capacity), expected computing including CPU, storage, and RAM, usage, i.e., use case, analytics target (e.g., same UE(s) or NF, etc.), and timestamp that indicates when training was performed on the pre-trained AI / ML model.

[0054] AI / ML model pre-training data characteristics may include a list of types of data sources, type of data, i.e., Event IDs, PM, KPIs of the local data for training, and may alsoinclude the dataset statistical properties (including the data distribution), the time window of the data samples and the minimum number of data samples, i.e., data quantity.

[0055] For the case of deep learning, TL capabilities may include a specific layer with respect to the AI / ML model architecture, (i.e., multiple connected layers can assist in learning different features, with initial layers focusing on higher-level features that narrow down to fine-grained features as the AI / ML model use more layers going deeper) and / or the task carried out (e.g., to extract higher-level features), to assist the AI / ML transfer.

[0056] There is also a need to indicate the TL model competence, i.e., the AI / ML model information with respect to the target Analytics IDs and / or task, (i.e., use case), related to the applicability of TL. The TL model competence (also referred to herein as TL competence or competence / competency for TL) may be based on feature information including the similarity / dis-similarity of the feature space, (i.e., common features in both tasks), the feature significance and the feature meaning for both analytics services and / or tasks. The TL model competence may be based on label information including the labels availability related to the target analytics service and / or task, the label space similarity / dis- similarity (i.e., common labels in both source and target analytics services and / or tasks) and the notion of labels, (e.g., whether labels have a common meaning). The TL model competence may be based on domain information including the similarity / dis-similarity of the type of available training data (i.e., data sources, data context, i.e., (a) data type (e.g., Event IDs, performance measurements (PMs), KPIs), (b) data object (e.g., per UE, cell, application), data statistics and network environment. The TL model competence may be based on accuracy information including self-assessment (i.e., using validation and testing), scores and / or rating provided by a previous usage.

[0057] The disclosure herein considers an NWDAF containing MTLF as the analytic service entities involved in the TL process. The NWDAF containing MTLF that produces or owns a pre-trained AI / ML model is referred to as MTLF TL model producer. The NWDAF containing MTLF that consumes a pre-trained AI / ML model referred to as MTLF TL model consumer.

[0058] Aspects of the present disclosure and certain terminology used herein will now be generally introduced.

[0059] A feature in broader sense can be a characteristic or a property / attribute related to collected raw data. It can be: (i) a numerical feature, i.e., can hold numerical values that can be measured on a specific scale (e.g., an Event ID on NF load or number of handovers), (ii) a categorical feature, i.e., discrete values that can be grouped in categories (e.g., quality of experience (QoE) levels), (iii) time series features, i.e., measurements taken over different times (e.g., peak and off-peak times) and (iv) text or evaluation features, i.e., represent words (e.g., reviews or feedback related to a service). Features can be either: included in the AI / ML model information as attributes of the AI / ML model and can be included in the registration process in the NRF; or exchanged among the MTLF TL model producer and MTLF TL model consumer as a part of a selection preparation process that aims to provide knowledge transfer to identify if a specific AI / ML model is fit for TL and / or as a part of a selected pre-trained AI / ML model transfer.

[0060] To characterize a feature, there is a need to have knowledge of the significance of a feature upon TL in the target Analytics ID and / or task (i.e., use case). Features that are common can be re-used, otherwise features that serve a different role and may not be significant for the target Analytics IDs and / or tasks can be ignored. Alternatively, a new feature can be derived from an inter-relation with the original feature upon TL, for example, network load in an area of interest is inter-related to the mobility rate towards that area of interest, or transformed when a feature has a different meaning in the target task, for example, a high network element load is desirable in energy saving analytics (since energy can be conserved) but serves as a negative indication in network performance analytics.

[0061] A label instruction in broader sense can be a guide on how to annotate the input data and can be conveyed via exchanging the labelling criteria or the rules / logic. Label instruction availability can be exchanged among the MTLF TL model producer and the MTLF TL model consumer as a part of a model selection preparation process that aims to indicate if labels can be handy and accessible in the target MTLF TL model consumer, which shall employ a pre-trained AI / ML model as part of TL. Label instructions can then be provided during the pre-trained AI / ML model provision phase.

[0062] Lack of local label availability means that a selected pre-trained AI / ML model needs to be ready for use without the need of re-training (due to limitation of local data).Otherwise, lack of local labels may be resolved by receiving labelling criteria or rules / logic exchanged in the model provision phase, which can assist in generating labelled data if the MTLF TL model producer authorization and respective permissions are satisfied. Alternatively, the target MTLF TL model consumer may re-train the received pre-trained AI / ML model using simulation or digital twin tools.

[0063] Domain rules can be a guide on how to inter-relate the knowledge gained in one domain to another. Typically, a higher-level common domain may determine the rules on how subdomains are related, for example, an end-to-end domain may relate radio and core network domains, or an aggregate analytics service may relate different area-based components (like the case of NWDAF as per clause 6.1A.2 of the 3GPP Technical Specification TS 23.288) or knowledge gained from application throughput in a UE may relate to the network load. The notion of domain may relate different technologies, or different administrative parts of a network, or different geographical areas, i.e., areas of interest.

[0064] Domain information, i.e., related to residing domain, can be included in the AI / ML model information registered in the NRF, while domain adaptation rules availability shall be exchanged among MTLF TL model producer and MTLF TL model consumer as a part of a model selection preparation process that aims to provide knowledge transfer among different domains. Domain rules can then be provided during the pre-trained AI / ML model provision phase.

[0065] TL related feature, label and domain information can be configured (e.g., by the 0AM configuration management procedures) in the MTLF TL model producer and MTLF TL model consumer that owns an AI / ML model eligible for TL with respect to a target Analytics ID and / or task (i.e., use case) and may include: feature information, for example, Event ID, Event IDs, PM, KPIs, etc.; a flag that indicates the significance of a feature; a weight, function / logic, or rule that inter-relates features or transforms one feature to another; a labelling criteria or rules / logic (if the source of the AI / ML model allows to share such information or following potential restrictions); a domain rule that inter-relates the knowledge gained in one domain to another; an accuracy information when a model was used from a producer task to a consumer task.

[0066] It shall be noted that feature information, including the significance of a feature and domain information, i.e., residing domain can be included in the NRF as a part of a registration process or alternatively can be exchanged in a newly introduced model selection preparation process.

[0067] The model selection preparation process may also include information with respect to the MTLF TL model consumer TL task (i.e., use case) regarding the availability of: (i) a weight, function / logic, or rule that inter-relates features or transforms one feature to another, (ii) a labelling criteria or rules / logic and (iii) interrelated domain rules. The actual weights, rules, function and / or logic related to the adaptation / transformation or interrelation of features, labelling criteria and domain knowledge shall be exchanged as a part of the pre-trained AI / ML model transfer.

[0068] The exchange of information in the model selection preparation process and AI / ML model exchange process is performed once the potential MTLF TL model consumer is authorized by the respective MTLF TL model producer.

[0069] Specific embodiments of the present disclosure will now be introduced and relate to: (i) TL model storage and removal, (ii) TL model registration and discovery and (iii) TL model selection and retrieval.

[0070] Figure 4 illustrates an example 400 of AI / ML model storage for TL in an ADRF and the removal from the ADRF in accordance with aspects of the present disclosure.

[0071] The example 400 shows a MTLF TL model producer 410 and a data / model repository in the form of an ADRF 420. The messaging flow and procedural steps between the MTLF TL model producer 410 and ADRF 420 will now be described.

[0072] In an initial step 400a, the MTLF TL model producer 410 determines to store a (or set of) AI / ML models in the ADRF 420 for sharing in TL scenarios based on MTLF policy.

[0073] In further step 401, the MTLF TL model producer 410 requests to store the (or the set of) AI / ML model(s) to the ADRF 420 providing the AI / ML model ID, an indicationof TL support, the TL capabilities, and the respective TL model competence by invoking the Nadrf_MLModelManagement_StorageRequest service.

[0074] In further step 402, if instead of the AI / ML model(s), the AI / ML model address(es) is / are included in the request at step 401, the ADRF 420 downloads the AI / ML model(s) as well as the respective AI / ML model ID, TL capabilities, and TL model competence using the provided address(es) and stores them locally.

[0075] In further step 403, the ADRF 420 sends a Nadrf_MLModelManagement_StorageResponse message to the NWDAF containing MTLF (i.e., the MTLF TL model producer 410 or owner) including the AI / ML Model storage result indication.

[0076] It should be noted that TL model updates can also be handled in a similar way indicating the TL support, the AI / ML Model ID and the corresponding updates in the TL capabilities and respective TL model competence.

[0077] In steps 404-406, if the MTLF TL model producer 410 decides to delete a previously stored version of an AI / ML model for TL use in ADRF 420, it issues a Nadrf_MLModelManagement_Delete request service operation (step 404). The AI / ML model ID may be included in the delete request. The ADRF 420 deletes the stored AI / ML model(s) and the related TL capabilities and TL model competence information (step 405) and then indicates the result (i.e., deleted, not found, etc.) to the MTLF TL model producer 410 using the Nadrf_MLModelManagement_Delete response service operation.

[0078] Figure 5 illustrates an example 500 of the registration and discovery of TL enabled NWDAFs containing MTLFs in accordance with aspects of the present disclosure.

[0079] The example 500 shows one or more MTLF TL model producers 510, an MTLF TL model consumer 530 and a NRF 540. The various messaging flows and procedural steps between the MTLF TL model producer 510, MTLF TL model consumer 530 and NRF 540 will now be described.

[0080] The steps 501 to 503 are the MTLF TL model producer registration procedure.

[0081] In a first step 501, the MTLF TL model producer 510 registers to NRF 540 with its NF profile indicating a TL support option or flag. The registration includes NWDAF NF Type (as per clause 5.2.7.2.2 of the 3GPP Technical Specification TS 23.502 titled “Procedures for the 5G System”), Analytics ID(s) corresponding to the pre-trained AI / ML model(s), model filter information including the S-NSSAI(s) and service area, AI / ML model interoperability or permission indicator (e.g., the vendor), TL capabilities, and NWDAF address information. In this step 501 the MTLF TL model producer 510 uses the Nnrf_NFManagement_NFRegister Request.

[0082] In step 502, the NRF 540 stores the received NWDAF MTLF TL capabilities.

[0083] In step 503, the NRF 540 provides a registration response to the MTLF TL model producer 510. The NRF 540 uses the Nnrf_NFManagement_NFRegister Response.

[0084] The steps 504 to 506 are the MTLF TL model consumer discovery procedure.

[0085] The MTLF TL model consumer 530 determines that an AI / ML model requires knowledge transfer based on TL according to the operator policy (e.g., pre-configured), Analytic ID, service area / DNAI or if the input data cannot be collected locally, i.e., lack of data, or if collecting data, it is too expensive.

[0086] In step 504, the MTLF TL model consumer 530 discovers a (or set of) MTLF TL model producer(s) 510 for the purpose of TL from NRF 540 by invoking the Nnrf_NFDiscovery_Request service operation. The following criteria might be used: Analytic ID of the AI / ML model required, indication of TL support, service area, S-NSSAI, TL capabilities, the time schedule information, (i.e., when the pre-trained AI / ML model is needed, or an indication of urgency), and the domain information (i.e., including the area or interest) in where the AI / ML model would be adopted.

[0087] In step 505, the NRF 540 authorises the discovery request.

[0088] In step 506, the NRF 540 provides a response to the MTLF TL model consumer530. The NRF 540 uses the Nnrf_NFDiscovery_Request Response. The NRF 540 may include information indicating one or more candidate NWDAF MTLF TL model producers 510.

[0089] Figure 6 illustrates an example 600 of a procedure for TL AI / ML model retrieval via NWDAF MTLF in accordance with aspects of the present disclosure.

[0090] The example 600 shows one or more MTLF TL model producers 610, a model repository in the form of an ADRF 620 and an MTLF TL model consumer 630.

[0091] In general, the MTLF TL model consumer 630 initially checks the candidate pre-trained AI / ML model details (i.e., in a selection preparation phase) related to the availability of features, label, and domain information with respect to the target Analytic ID, task (use case) and network environment considering the data characteristics, etc. Then after selecting the desired pre-trained AI / ML model, the MTLF TL model consumer 630 obtains the AI / ML model (i.e., as a file, containerized or as parameter weight exchange) also receiving the weight, function / logic, or rules that inter-relates or transforms one feature to another, the labelling criteria or rules / logic and domain rules that inter-relates the knowledge gained in one domain to another.

[0092] The various messaging flows and procedural steps between the MTLF TL model producers 610, the ADRF 620 and the MTLF TL model consumer 630, will now be described. The procedure for model retrieval assumes that the MTLF TL model consumer 630 has already discovered one or more candidate MTLF TL model producers 610 as per the example 500 of Figure 5.

[0093] In a first step 601, the MTLF TL model consumer 630 requests information regarding one or more pre-trained AI / ML model(s) indicating the purpose of TL related to one or more Analytics ID(s) by invoking the Nnwdaf MLModelTraininglnfo Request service operation. Such a request may contain an active flag or other TL indicator related to TL selection preparation to indicate that at this stage only pre-trained AI / ML model information describing whether and how a pre-trained model can be related to the needs of the target usage in the MTLF TL model consumer 630 shall be included for the purpose of selecting the optimal re-trained AI / ML model.

[0094] In step 601, the request may also include the following information related to pre-trained AI / ML model needs to assist the receiving MTLF TL model producer 610 to select the appropriate information to send back to the MTLF TL model consumer 630: (i)Analytics ID and task (i.e., use case), (ii) the desired model layer if the request focus on deep learning and the respective purpose (iii) label availability indicator, (iv) residing domain information, slice and / or area of interest, (v) data characteristics, (vi) network environment, (vii) required AI / ML model accuracy, and (viii) time schedule information (when the pre-trained mode shall be obtained).

[0095] In step 601, based on this information, the MTLF TL model producer 610 shall be able to derive the most suitable pre-trained AI / ML model and provide the AI / ML model inter-related information in terms of features, data labelling and domain specific information.

[0096] It should be noted that the TL selection preparation can be skipped if the MTLF model consumer 630 can decide, which MTLF model producer 610 shall be selected to share a pre-trained AI / ML model for TL, for example, based on information acquired from previous TL procedures or from an NRF, or based on local configuration.

[0097] In step 602, The MTLF TL model producer 610 authenticates the TL request from the MTLF TL model consumer 630 and determines whether the requested set of pretrained ML Model(s) information associated with one or more Analytics ID(s) are available locally or should be retrieved from the ADRF 620.

[0098] In step 602, if the MTLF TL model producer 610 determines that the set of requested TL AI / ML Model(s) information need to be retrieved from the ADRF 620 it can either: (i) issue a retrieve TL AI / ML model(s) information request towards the ADRF 620 (with the MTLF TL model consumer 630 being agnostic to where in the ADRF 620 the ML model(s) and the corresponding information is stored) or (ii) it may authorize the MTLF TL model consumer 630 to retrieve the TL AI / ML model(s) information stored in the ADRF 620 directly (i.e., skipping steps 603 and 604).

[0099] In step 602, if the MTLF TL model producer 610 issues a request to the ADRF 620 then steps 603 and 604 follow.

[0100] In step 603, the ADRF service consumer, i.e., MTLF TL model producer 610, requests information for the pre-trained TL AI / ML model stored in the ADRF 620 by invoking the Nadrf_MLModelManagement_Retrieve request service operation (i.e., usingthe storage transaction ID or one or more unique TL AI / ML model ID(s)) and using a flag to indicate model selection preparation, i.e., that the aim is to get only details about the availability of the model features, labels and domain information with respect to the MTLF TL model consumer 630 needs.

[0101] In step 604, the ADRF 620 sends an Nadrf MLModelManagement Retrieval response service operation to MTLF TL model producer 610 providing the TL AI / ML model(s) information.

[0102] If in step 602 the MTLF TL model producer 610 does not need to contact the ADRF 620, steps 603 and 604 are omitted.

[0103] In step 605, the MTLF TL model producer 610 provides a response to the MTLF TL model consumer 630. The MTLF TL model producer 610 uses the Nnwdaf MLModelTraininglnfo Request Response. Two options for the response are possible.

[0104] As a first option in step 605, if the pre- trained AI / ML model information is accessible locally, the MTLF TL model producer 610 can provide the TL pre-trained AI / ML model information availability per requested Analytics ID, including the availability of the following TL capabilities, (where and if they are applicable) potential feature adaptation / transformation information, labelling criteria or rules / logic and domain specific transfer knowledge rules and / or logic.

[0105] As a second option in step 605, if the pre-trained AI / ML model information is stored at the ADRF 620, the MTLF TL model producer 610 can provide the unique TL AI / ML Model ID and the file address (e.g., URL or FQDN) or the related TL AI / ML model storage transaction ID (when the ADRF (Set) ID is provided) to allow an authorized MTLF TL model consumer 630 to retrieve the TL AI / ML model(s) information stored in the ADRF 620 directly.

[0106] If the MTLF TL model consumer 630 receives details to retrieve the desired pre-trained AI / ML model information availability from the ADRF 620 then step 606 and 607 are performed, otherwise they are skipped.

[0107] In step 606, if the MTLF TL model consumer 630 receives details to retrieve the desired pre-trained AI / ML model information availability from the ADRF 620, it invokes the Nadrf_MLModelManagement_Retrieval request service operation, including a flag to indicate model selection preparation, the storage transaction ID or one or more unique TL AI / ML Model ID(s) and the corresponding address (e.g., URL or FQDN).

[0108] In step 607, the ADRF 620 sends a Nadrf MLModelManagement Retrieval response to the MTLF TL model consumer 630 including the information availability related to the pre- trained AI / ML model ID(s) stored in the ADRF 620, i.e., it provides the AI / ML model inter-related information availability in terms of features significance, feature adaptation / transformation availability, potential labelling criteria or rules / logic availability, and domain information and / or the availability of rules / logic that inter- relates the knowledge gained to the target domain.

[0109] In step 608, the MTLF TL model consumer 630 can then select the pre-trained AI / ML model with the optimal knowledge transfer based on the information received during the discovery and TL selection preparation phase.

[0110] Once the pre-trained AI / ML model is selected, the MTLF TL model consumer 630 obtains the AI / ML model (i.e., the model files or model containerization or the corresponding model parameter weights), either from the MTLF TL model producer 610 or from the ADRF 620 using a one-time request / response or by subscribing to receive regular TL model updates.

[0111] The MTLF TL model consumer 630 may also receive at this stage the pretrained AI / ML model adoption information including potential feature adaptation / transformation information, potential labelling criteria or rules / logic / function and domain specific transfer knowledge rules and / or logic.

[0112] It should be notated that the interaction with the ADRF 620 can also relate to obtaining a pre- trained AI / ML model at the address, for example, URL or FQDN, to retrieve the stored TL AI / ML model file(s) or other related information.

[0113] Figure 7 illustrates an example of a UE 700 in accordance with aspects of the present disclosure. The UE 700 may include a processor 702, a memory 704, a controller706, and a transceiver 708. The processor 702, the memory 704, the controller 706, or the transceiver 708, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0114] The processor 702, the memory 704, the controller 706, or the transceiver 708, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0115] The processor 702 may include an intelligent hardware device (e.g., a general- purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 702 may be configured to operate the memory 704. In some other implementations, the memory 704 may be integrated into the processor 702. The processor 702 may be configured to execute computer-readable instructions stored in the memory 704 to cause the UE 700 to perform various functions of the present disclosure.

[0116] The memory 704 may include volatile or non-volatile memory. The memory 704 may store computer-readable, computer-executable code including instructions when executed by the processor 702 cause the UE 700 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such the memory 704 or another type of memory. Computer-readable media includes both non- transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0117] In some implementations, the processor 702 and the memory 704 coupled with the processor 702 may be configured to cause the UE 700 to perform one or more of the functions described herein (e.g., executing, by the processor 702, instructions stored in the memory 704). For example, the processor 702 may support wireless communication at theUE 700 in accordance with examples as disclosed herein. The UE 700 may be configured to support a means for performing aspects of the disclosure herein.

[0118] The controller 706 may manage input and output signals for the UE 700. The controller 706 may also manage peripherals not integrated into the UE 700. In some implementations, the controller 706 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 706 may be implemented as part of the processor 702.

[0119] In some implementations, the UE 700 may include at least one transceiver 708. In some other implementations, the UE 700 may have more than one transceiver 708. The transceiver 708 may represent a wireless transceiver. The transceiver 708 may include one or more receiver chains 710, one or more transmitter chains 712, or a combination thereof.

[0120] A receiver chain 710 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 710 may include one or more antennas for receive the signal over the air or wireless medium. The receiver chain 710 may include at least one amplifier (e.g., a low- noise amplifier (LNA)) configured to amplify the received signal. The receiver chain 710 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 710 may include at least one decoder for decoding the processing the demodulated signal to receive the transmitted data.

[0121] A transmitter chain 712 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 712 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM). The transmitter chain 712 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 712 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0122] Figure 8 illustrates an example of a processor 800 in accordance with aspects of the present disclosure. The processor 800 may be an example of a processor configured to perform various operations in accordance with examples as described herein. The processor 800 may include a controller 802 configured to perform various operations in accordance with examples as described herein. The processor 800 may optionally include at least one memory 804, which may be, for example, an L1 / L2 / L3 cache. Additionally, or alternatively, the processor 800 may optionally include one or more arithmetic-logic units (ALUs) 806. One or more of these components may be in electronic communication or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces (e.g., buses).

[0123] The processor 800 may be a processor chipset and include a protocol stack (e.g., a software stack) executed by the processor chipset to perform various operations (e.g., receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) in accordance with examples as described herein. The processor chipset may include one or more cores, one or more caches (e.g., memory local to or included in the processor chipset (e.g., the processor 800) or other memory (e.g., random access memory (RAM), read-only memory (ROM), dynamic RAM (DRAM), synchronous dynamic RAM (SDRAM), static RAM (SRAM), ferroelectric RAM (FeRAM), magnetic RAM (MRAM), resistive RAM (RRAM), flash memory, phase change memory (PCM), and others).

[0124] The controller 802 may be configured to manage and coordinate various operations (e.g., signalling, receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) of the processor 800 to cause the processor 800 to support various operations in accordance with examples as described herein. For example, the controller 802 may operate as a control unit of the processor 800, generating control signals that manage the operation of various components of the processor 800. These control signals include enabling or disabling functional units, selecting data paths, initiating memory access, and coordinating timing of operations.

[0125] The controller 802 may be configured to fetch (e.g., obtain, retrieve, receive) instructions from the memory 804 and determine subsequent instruction(s) to be executedto cause the processor 800 to support various operations in accordance with examples as described herein. The controller 802 may be configured to track memory address of instructions associated with the memory 804. The controller 802 may be configured to decode instructions to determine the operation to be performed and the operands involved. For example, the controller 802 may be configured to interpret the instruction and determine control signals to be output to other components of the processor 800 to cause the processor 800 to support various operations in accordance with examples as described herein. Additionally, or alternatively, the controller 802 may be configured to manage flow of data within the processor 800. The controller 802 may be configured to control transfer of data between registers, arithmetic logic units (ALUs), and other functional units of the processor 800.

[0126] The memory 804 may include one or more caches (e.g., memory local to or included in the processor 800 or other memory, such RAM, ROM, DRAM, SDRAM, SRAM, MRAM, flash memory, etc. In some implementations, the memory 804 may reside within or on a processor chipset (e.g., local to the processor 800). In some other implementations, the memory 804 may reside external to the processor chipset (e.g., remote to the processor 800).

[0127] The memory 804 may store computer-readable, computer-executable code including instructions that, when executed by the processor 800, cause the processor 800 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as system memory or another type of memory. The controller 802 and / or the processor 800 may be configured to execute computer- readable instructions stored in the memory 804 to cause the processor 800 to perform various functions. For example, the processor 800 and / or the controller 802 may be coupled with or to the memory 804, the processor 800, the controller 802, and the memory 804 may be configured to perform various functions described herein. In some examples, the processor 800 may include multiple processors and the memory 804 may include multiple memories. One or more of the multiple processors may be coupled with one or more of the multiple memories, which may, individually or collectively, be configured to perform various functions herein.

[0128] The one or more ALUs 806 may be configured to support various operations in accordance with examples as described herein. In some implementations, the one or more ALUs 806 may reside within or on a processor chipset (e.g., the processor 800). In some other implementations, the one or more ALUs 806 may reside external to the processor chipset (e.g., the processor 800). One or more ALUs 806 may perform one or more computations such as addition, subtraction, multiplication, and division on data. For example, one or more ALUs 806 may receive input operands and an operation code, which determines an operation to be executed. One or more ALUs 806 be configured with a variety of logical and arithmetic circuits, including adders, subtractors, shifters, and logic gates, to process and manipulate the data according to the operation. Additionally, or alternatively, the one or more ALUs 806 may support logical operations such as AND, OR, exclusive-OR (XOR), not-OR (NOR), and not- AND (NAND), enabling the one or more ALUs 806 to handle conditional operations, comparisons, and bitwise operations.

[0129] The processor 800 may support wireless communication in accordance with examples as disclosed herein. The processor 800 may be configured to or operable to support a means for performing aspects of the disclosure herein.

[0130] Figure 9 illustrates an example of a NE 900 in accordance with aspects of the present disclosure. The NE 900 may include a processor 902, a memory 904, a controller 906, and a transceiver 908. The processor 902, the memory 904, the controller 906, or the transceiver 908, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0131] The processor 902, the memory 904, the controller 906, or the transceiver 908, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0132] The processor 902 may include an intelligent hardware device (e.g., a general- purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 902 may be configured to operate the memory 904. In some other implementations, the memory 904 may be integrated into the processor 902. The processor 902 may be configured to execute computer-readable instructions stored in the memory 904 to cause the NE 900 to perform various functions of the present disclosure.

[0133] The memory 904 may include volatile or non-volatile memory. The memory 904 may store computer-readable, computer-executable code including instructions when executed by the processor 902 cause the NE 900 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such the memory 904 or another type of memory. Computer-readable media includes both non- transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0134] In some implementations, the processor 902 and the memory 904 coupled with the processor 902 may be configured to cause the NE 900 to perform one or more of the functions described herein (e.g., executing, by the processor 902, instructions stored in the memory 904). For example, the processor 902 may support wireless communication at the NE 900 in accordance with examples as disclosed herein. The NE 900 may be configured to support a means for performing aspects of the disclosure herein. The NE 900 may be an MTLF TL model producer 410 or a data repository 420 of Figure 4. The NE 900 may be a MTLF TL Model Producer 510, MTLF TL model consumer 530 or NRF 540 of Figure 5. The NE 900 may be a MTLF TL model producer 610, ADRF 620 or MTLF TL Model Consumer 630 of Figure 6. The NE 900 may perform the methods 1000, 1100 or 1200 of Figures 10, 11 and 12, for example.

[0135] The controller 906 may manage input and output signals for the NE 900. The controller 906 may also manage peripherals not integrated into the NE 900. In some implementations, the controller 906 may utilize an operating system such as iOS®,ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 906 may be implemented as part of the processor 902.

[0136] In some implementations, the NE 900 may include at least one transceiver 908. In some other implementations, the NE 900 may have more than one transceiver 908. The transceiver 908 may represent a wireless transceiver. The transceiver 908 may include one or more receiver chains 910, one or more transmitter chains 912, or a combination thereof.

[0137] A receiver chain 910 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 910 may include one or more antennas for receive the signal over the air or wireless medium. The receiver chain 910 may include at least one amplifier (e.g., a low- noise amplifier (LNA)) configured to amplify the received signal. The receiver chain 910 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 910 may include at least one decoder for decoding the processing the demodulated signal to receive the transmitted data.

[0138] A transmitter chain 912 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 912 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM). The transmitter chain 912 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 912 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0139] Figure 10 illustrates a flowchart of a method 1000 in accordance with aspects of the present disclosure. The operations of the method may be implemented by a NE as described herein. In some implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the described functions.

[0140] At 1002, the method 1000 may include determining a requirement for TL for a learning model task. The operations of 1002 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1002 may be performed by a NE as described with reference to Figure 9.

[0141] At 1004, the method 1000 may include discovering one or more second network functions for supporting TL for the learning model task, the discovering being based at least in part on one or more TL capabilities of the one or more second network functions. The operations of 1004 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1004 may be performed by a NE as described with reference to Figure 9.

[0142] It should be noted that the method 1000 described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0143] Figure 11 illustrates a flowchart of a method 1100 in accordance with aspects of the present disclosure. The operations of the method 1100 may be implemented by a NE as described herein. In some implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the described functions.

[0144] At 1102, the method 1100 may include transmitting, to a third network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities. The operations of 1102 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1102 may be performed by a NE as described with reference to Figure 9.

[0145] Figure 12 illustrates a flowchart of a method 1200 in accordance with aspects of the present disclosure. The operations of the method 1200 may be implemented by a NE as described herein. In some implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the described functions.

[0146] At 1202, the method 1200 may include receiving, from a second network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities. The operations of 1202 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1202 may be performed by a NE as described with reference to Figure 9.

[0147] At 1204, the method 1200 may include storing, as a network function profile for the second network function, the one or more fourth parameters. The operations of 1204 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1204 may be performed by a NE as described with reference to Figure 9.

[0148] There is provided a first network function for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first network function to: determine a requirement for transfer learning (TL) for a learning model task; and discover one or more second network functions for supporting TL for the learning model task, based at least in part on one or more TL capabilities of the one or more second network functions.

[0149] Current artificial intelligence (AI) / machine learning (ML) mechanisms in the third-generation partnership project (3 GPP) adopt an isolated approach to AI / ML model training. Each AI / ML model is independently trained for a specific usage without considering prior training knowledge or experience. Current procedures for instantiation of an AI / ML model require training of the AI / ML model from scratch, assuming that the corresponding Network Data Analytics Function (NWDAF) Model Training Logical Function (MTLF) is capable to collect enough input data.

[0150] Collecting enough input data is itself challenging, especially when dealing with a new analytics service. Being capable to collect the required input data is not always feasible or can be too expensive (considering the required network resources), especially when the requested / available time for AI / ML model training is limited.

[0151] TL is a technique that reuses an AI / ML model that is pre-trained with respect to a certain task. The pre-trained AI / ML model is reused for a new task, exploiting the knowledge gained from the earlier task. In the application-layer, i.e., application data analytics enablement service (AD AES), a pre-trained ML model as a candidate for TL can be stored in a model registry. In addition, the notion of a TL enabler has been introduced for detecting the need for using TL for a particular AI / ML task or analytics based on context information. Such a TL enabler may be a network function such as a MTLF TL model producer or consumer, such as an NWDAF. Furthermore, the 5G core, as a difference to the application-layer, can include a mechanism for AI / ML model retrieval wherein a model producer and owner (i.e., a NWDAF MTLF) authenticates a potential model consumer (i.e., another NWDAF MTLF) before providing requested model information.

[0152] Despite the current procedures for instantiation of an AI / ML model and for TL, TL adaptation considerations are still not implemented when determining a pre-trained AI / ML model for TL. Current methods merely utilise selection criteria for determining the pre-trained AI / ML model for TL and do not consider how such an AI / ML model is to be adapted in order to be made suitable for a separate TL task. The adaptation strategy will depend on the new task itself and will determine which parts of the pre-trained AI / ML model knowledge can be transferred, where the transfer can occur, and when, in order to optimise the performance improvement.

[0153] For TL to consider an AI / ML model adaptation strategy, interface enhancements are required in the 5G core network which include the introduction of new meta data parameters to describe how to perform knowledge transfer by employing an adaptation strategy. Furthermore, the core network will require mechanisms to discover appropriate network functions with appropriate TL capabilities that can provide the knowledge transfer. Moreover, the core network will require decision criteria to identify when knowledge transfer should take place (i.e., when a pre- trained AI / ML model is actually suitable for adaptation for a new AI / ML model task).

[0154] Current proposals are introducing TL based on the use of an AI / ML model in similar tasks, but without considering the potential of adapting an AI / ML model employedin one task to solve another new problem. In many cases dissimilar tasks may still relate in terms of model components (e.g., features), dependencies among input data or similarity of network and user behaviour trends. This observation can relate even dissimilar problems if an appropriate transfer strategy is in place to assist how, where and when knowledge can be transferred among different problems to be beneficial, especially when TL is the only option instead of training a model from scratch due to data, time, and resource limitations.

[0155] Current proposals do not consider any mechanism to assist TL in such respects for the 5G core network,

[0156] The solutions described herein facilitate TL to consumers by enabling discovery of network functions supporting TL learning based on TL capabilities. Such TL capabilities may be introduced in the network function (NF) profile stored in a network repository function (NRF).

[0157] Additionally, the competency of a pre-trained AI / ML model for TL for an AI / ML model task can be determined using one or more competencies for TL (i.e., TL competence information) stored locally at a model producer or in an analytics data repository function (ADRF) to capture whether and how a pre-trained AI / ML model can be used to execute a new task. In general, TL competence examines a given model’s feature information, in terms of significance, inter-relation or adaptation from one to another, labelling availability and transfer methods as well as domain transfer knowledge rules and model accuracy from prior experience.

[0158] The solutions described herein more generally relate to an apparatus and method that facilitates TL among analytic services by introducing TL related details in the form of AI / ML model information and through a producer consumer exchange, which can characterize how fit a pretrained AI / ML model can be when transferred and used in a specified target analytics service and / or task, i.e., use case.

[0159] The first network function may determine the requirement for TL according to operator policy, an identifier for an analytics, a service area, or, based on a lack of training data for training a new AI / ML model locally (or if such training data would be too expensive to obtain locally).

[0160] Determining a requirement for TL may comprise determining that a model cannot be generated from scratch (for instance owing to a lack of training data) or that training a model from scratch is too expensive (in respect of the resources required). Determining a requirement for TL may comprise determining / selecting one or more components of a model that can benefit from TL. For instance, this may involve determining one or more components that can be adopted, modified, or dropped from a preexisting or pre-trained model via a TL process for resolving the model task.

[0161] The learning model task may be an AI / ML model task. The learning models described herein may comprise AI / ML models.

[0162] The at least one processor may be further configured to cause the first network function to: transmit, to a third network function, a first discovery request message for network functions supporting TL, wherein the first discovery request message comprises one or more first parameters indicating the requirement for TL and one or more required TL capabilities for the learning model task; and receive, from the third network function, a first discovery response message indicating the one or more second network functions supporting TL, the one or more second network functions supporting at least one of the one or more required TL capabilities for the learning model task.

[0163] The one or more first parameters may further comprise at least one of: an analytics service information; an analytics service environment information; and a required learning model information.

[0164] The analytics service information may include one or more required analytics service types and / or tasks. This may include, for instance, a required identifier for analytics, such as an analytics ID, or a required identifier for the learning model task. This may include a required service area or network slice. The analytics service information may alternatively or additionally include analytics filter information such as time scheduling capabilities (such as when a model is needed or an indication or urgency) or domain information (such as an area of interest where a model will be adopted). The analytics service information may alternatively or additionally include analytics target information.

[0165] The analytics service environment information may include one or more network characteristics. The analytics service environment information may alternatively or in addition include input data information.

[0166] The required learning model information may include model component information (i.e., one or more model features, deep learning layer, task), one or more computing requirements, pre-training time information, reputation information, and / or access permissions.

[0167] The first network function may be a model consumer. Each of the one or more second network functions may be a model producer. The third network function may be a network repository function. The model consumer and / or model producer may be respective NWDAF MTLFs.

[0168] The at least one processor may be further configured to cause the first network function to: transmit, to each of the one or more second network functions, a second request message for competencies for TL for the learning model task; and receive, from each of the one or more second network functions, a second response message comprising one or more second parameters indicating competencies of a pre-trained learning model for TL for the learning model task.

[0169] The competencies for TL may be referred to herein as TL competencies. The competencies for TL indicate a suitability of a pre-trained learning model for the learning model task, which may include how, when and where features or knowledge of the pretrained learning model can be adapted to the new learning model task. In particular, the competencies for TL may contain instructions regarding rules, and / or logic, and / or function, and / or criteria that inter-relates knowledge gained in an analytics service or task under a particular usage environment, to another analytics service (i.e., the learning model task) under the same or a different usage environment.

[0170] The one or more second parameters may comprise at least one of: a learning model information of the pre-trained learning model, the learning model information indicating a suitability of the pre-trained learning model for TL for the learning model task;and / or an information for accessing the learning model information from a fourth network function.

[0171] The learning model information is requested by the first network function. The second request message may be for learning model information that indicates an availability of the required TL capabilities for the learning model task. The second request message may be for learning model information that relates one or more pre-trained learning models of the one or more second network functions to the learning model task. The learning model information may be information that describes whether, how and when the pre- trained learning models can be related to the needs of the target usage (i.e., for the learning model task). This may include model inter-related information such as feature significance, feature adaptation / transformation ability / availability, potential labelling criteria, rules / logic availability, domain information, availability of rules / logic that interrelate knowledge to the target domain (i.e., to the learning model task).

[0172] If the learning model information is available locally at the second network functions, this information may be provided to the first network function in the second response messages. For instance, this information may be provided for a requested identifier for analytics or task based on locally stored information.

[0173] If, alternatively, the learning model information is not available locally at the second network functions, the second response messages may instead comprise an identifier for the pre-trained learning models and their address at, for instance, a fourth network function (i.e., a repository such as ADRF). The address may be a URL or FQDN. A storage transaction ID for the fourth network function may also be provided. The first network function may be authorized by the second network functions to obtain the learning model information directly from the fourth network function.

[0174] The learning model information may comprise at least one of: a feature information for the pre-trained learning model; a label information for the pre-trained learning model; a training data information for the pre-trained learning model; an accuracy information of prior use of the pre-trained learning model; and a domain information for the pre-trained learning model.

[0175] The feature information, label information, training data information and accuracy information may relate respective features, labelling, training data and accuracy of the pre-trained learning model, to the learning model task. The relating may be in respect of commonality, adaptability, suitability, significance and / or notion / meaning.

[0176] The feature information may comprise, for instance, information regarding the adaptability of one or more features of the pre-trained learning model to the learning model task.

[0177] The label information may comprise labelling criteria for the pre-trained learning model.

[0178] The training data information may relate to pre-training of the pre-trained learning model and suitability of the pre-training to the learning model task.

[0179] The accuracy information may relate to prior use of the pre-trained learning model with respect to an analytics service and / or task and / or with respect to a usage environment.

[0180] The domain information may indicate the network environment used for the pretrained learning model and any domain specific transfer rules may also be provided indicating how knowledge of the pre-trained learning model can be transferred to the domain for the learning model task.

[0181] The at least one processor may be further configured to cause the first network function to (based on the one or more second parameters comprising the information for accessing the learning model information from the fourth network function) transmit, to the fourth network function, a third request message for retrieving the learning model information; and receive, from the fourth network function, the learning model information.

[0182] The third request may comprise a flag to indicate the learning model information is required for learning model selection. The third request may comprise a storage transaction identifier. The third request may comprise an identifier for the pretrained learning model. The third request may comprise the information for accessing the learning model information (i.e., the address information such as URL or FQDN).

[0183] The second request message may comprise one or more third parameters for selecting the learning model information for TL for the learning model task, wherein the one or more third parameters comprise at least one of: a flag indicating learning model information for pre-trained learning models is required; one or more required identifiers for analytics or for the learning model task; a required usage for the learning model task; a required deep-learning layer for the learning model task; a required label information for the learning model task; a required domain information for the learning model task; one or more computing requirements for the learning model task; one or more targets for analytics for the learning model task; a required training data information for the learning model task; a required feature information of the pre-trained learning model; a required accuracy information for the learning model task; and a required time schedule information for the learning model task.

[0184] The at least one processor may be further configured to cause the first network function to: determine, based at least in part on the second response message, a pre-trained learning model associated with a respective second network function for TL for the learning model task; and obtain the determined pre-trained learning model.

[0185] The first network function may utilize the information obtained from the second response messages (i.e., the learning model information) to determine / select a pre-trained learning model associated with a respective second network function. The first network function may further utilize any information obtained in the first discovery response message for the selection (i.e., information received when discovering the one or more second network functions such as the capabilities for TL).

[0186] The first network function may obtain the determined pre-trained learning model from the respective second network function or a related repository function (i.e., an ADRF storing the selected pre-trained learning model).

[0187] The first network function may further subscribe for updates of the determined pre-trained learning model.

[0188] There is further provided a second network function for wireless communication, comprising: at least one memory; and at least one processor coupled withthe at least one memory and configured to cause the second network function to: transmit, to a third network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities.

[0189] The fourth request for registering the second network function may be a request to register a network function profile that indicates support for TL and the capabilities for TL.

[0190] The second network function may further receive a response from the third network function, indicating whether the request for registration has been successful.

[0191] The registration of the second network function tends to allow discovery of the second network function by including the indication for supporting TL and by including the respective TL capabilities.

[0192] The one or more TL capabilities may include at least one of: one or more permissions for TL; a network slice; a network service area; one or more features of the pre-trained learning model; one or more training data characteristics; an environment information; one or more computing requirements; a usage or use case; an intended target for analytics; a timestamp indicating when training was performed on the pre-trained learning model. The environment information may comprise network information such as topology connectivity, density or capacity, etc, where the pre-trained learning model was trained and used. The computing requirements may comprise a particular CPU type or configuration, a data storage requirement, a RAM requirement. The intended targets may comprise one or more UEs or NFs for example. The one or more training data characteristics may be referred to as pre-training data characteristics. Such characteristics may include a list of the types of data sources, the types of data (event IDs, PM, KPIs for example), statistical properties of the data (i.e., data distribution), a time window of the samples of the data and a minimum number of samples of the data (i.e., the quantity of data), used for training the pre-trained learning model. The one or more features of the pretrained learning model may include an indication of one or more deep learning layers associated with learning different features of a learning analytics task. In deep learning,multiple connected layers can assist in learning different features, with initial layers focusing on higher-level features that narrow down to fine-grained features as the learning model use more layers. This tends to assist in identifying aspects of the one or more pretrained learning models relevant for TL for the task being carried out, e.g., to extract higher-level features, to assist the learning model transfer.

[0193] The one or more fourth parameters may further comprise a network function type of the second network function, an identifier for analytics; a filter information; an address information for the pre-trained learning model.

[0194] The second network function may be a model producer. The third network function may be a network repository function.

[0195] The at least one processor may be further configured to cause the second network function to: receive, from a first network function, a second request message for competencies for TL for a learning model task; authorize, the first network function, to receive a second response message; and and transmit, to the first network function, the second response message, wherein the second response message comprises one or more second parameters indicating competencies of the pre-trained learning model for TL for the learning model task.

[0196] The one or more second parameters may comprise at least one of: a learning model information of the pre-trained learning model, the learning model information indicating a suitability of the pre-trained learning model for TL for the learning model task; and / or an information for accessing the learning model information from a fourth network function.

[0197] The learning model information may comprise at least one of: a feature information for the pre-trained learning model; a label information for the pre-trained learning model; a training data information for the pre-trained learning model; an accuracy information of prior use of the pre-trained learning model; and a domain information for the pre-trained learning model.

[0198] The second request message may comprise one or more third parameters for selecting the learning model information for TL for the learning model task, wherein theone or more third parameters comprise at least one of: a flag indicating learning model information for pre-trained learning models is required; one or more required identifiers for analytics or for the learning model task; a required usage for the learning model task; a required deep-learning layer for the learning model task; a required label information for the learning model task; a required domain information for the learning model task; one or more computing requirements for the learning model task; one or more targets for analytics for the learning model task; a required training data information for the learning model task; a required feature information of the pre-trained learning model; a required accuracy information for the learning model task; and a required time schedule information for the learning model task.

[0199] The first network function may be a model consumer.

[0200] There is further provided a third network function for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the third network function to: receive, from a second network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities; and store, as a network function profile for the second network function, the one or more fourth parameters.

[0201] The at least one processor may be further configured to cause the third network function to: receive, from a first network function, a first discovery request message for network functions supporting TL, wherein the first discovery request message comprises one or more first parameters indicating a requirement for TL and one or more required TL capabilities for a learning model task; and transmit, to the first network function, a first discovery response message indicating the second network function as supporting TL for the one or more TL capabilities.

[0202] There is further provided a method performed by a first network function, the method comprising: determining a requirement for TL for a learning model task; and discovering one or more second network functions for supporting TL for the learningmodel task, the discovering being based at least in part on one or more TL capabilities of the one or more second network functions.

[0203] The method may further comprise transmitting, to a third network function a first discovery request message for network functions supporting TL, wherein the first discovery request message comprises one or more first parameters indicating the requirement for TL and one or more required TL capabilities for the learning model task; and receiving, from the third network function, a first discovery response message indicating the one or more second network functions supporting TL, the one or more second network functions supporting at least one of the one or more required TL capabilities for the learning model task.

[0204] The one or more first parameters may further comprise at least one of: an analytics service information; an analytics service environment information; and a required learning model information.

[0205] The first network function may be a model consumer; each of the one or more second network functions may be a model producer; and / or the third network function may be a network repository function.

[0206] The method may comprise transmitting, to each of the one or more second network functions, a second request message for competencies for TL for the learning model task; and receiving, from each of the one or more second network functions, respective second response messages comprising one or more second parameters indicating respective competencies of a pre-trained learning model for TL for the learning model task.

[0207] The one or more second parameters may comprise at least one of: a learning model information of the pre-trained learning model, the learning model information indicating a suitability of the pre-trained learning model for TL for the learning model task; and / or an information for accessing the learning model information from a fourth network function.

[0208] The learning model information may comprise at least one of: a feature information for the pre-trained learning model; a label information for the pre-trained learning model; a training data information for the pre-trained learning model; an accuracyinformation of prior use of the pre-trained learning model; and a domain information for the pre-trained learning model.

[0209] The method may further comprise, based on the one or more second parameters comprises the information for accessing the learning model information from the fourth network function: transmitting, to the fourth network function, a third request for retrieving the learning model information; and receiving, from the fourth network function, the learning model information.

[0210] The second request message may comprise one or more third parameters for selecting the learning model information for TL for the learning model task, wherein the one or more third parameters comprise at least one of: a flag indicating learning model information for pre-trained learning models is required; one or more required identifiers for analytics or for the learning model task; a required usage for the learning model task; a required deep-learning layer for the learning model task; a required label information for the learning model task; a required domain information for the learning model task; one or more computing requirements for the learning model task; one or more targets for analytics for the learning model task; a required training data information for the learning model task; a required feature information of the pre-trained learning model; a required accuracy information for the learning model task; and a required time schedule information for the learning model task.

[0211] The method may comprise determining, based at least in part on the second response message, a pre-trained learning model associated with a respective second network function for TL for the learning model task; and obtaining the determined pretrained learning model.

[0212] There is further provided a method performed by a second network function for wireless communication, comprising transmitting, to a third network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities.

[0213] The second network function may be a network data analytics function model training logical function model producer; and / or the third network function may be a network repository function.

[0214] The method may further comprise receiving, from a first network function, a second request message for competencies for TL for a learning model task; authorizing, the first network function, to receive a second response message; and transmitting, to the first network function, the second response message, wherein the second response message comprises one or more second parameters indicating competencies of the pre-trained learning model for TL for the learning model task.

[0215] The one or more second parameters may comprise at least one of: a learning model information of the pre-trained learning model, the learning model information indicating a suitability of the pre-trained learning model for TL for the learning model task; and / or an information for accessing the learning model information from a fourth network function.

[0216] The learning model information may comprise at least one of: a feature information for the pre-trained learning model; a label information for the pre-trained learning model; a training data information for the pre-trained learning model; an accuracy information of prior use of the pre-trained learning model; and a domain information for the pre-trained learning model.

[0217] The second request message may comprise one or more third parameters for selecting the learning model information for TL for the learning model task, wherein the one or more third parameters comprise at least one of: a flag indicating learning model information for pre-trained learning models is required; one or more required identifiers for analytics or for the learning model task; a required usage for the learning model task; a required deep-learning layer for the learning model task; a required label information for the learning model task; a required domain information for the learning model task; one or more computing requirements for the learning model task; one or more targets for analytics for the learning model task; a required training data information for the learning model task; a required feature information of the pre-trained learning model; a required accuracyinformation for the learning model task; and a required time schedule information for the learning model task.

[0218] The first network function may be a model consumer.

[0219] There is further provided a method performed by a third network function for wireless communication, comprising: receiving, from a second network function, a fourth request for registering the second network function for being discoverable for TL, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports TL for a pre-trained learning model and one or more TL capabilities; and storing, as a network function profile for the second network function, the one or more fourth parameters.

[0220] The method may further comprise receiving, from a first network function, a first discovery request message for network functions supporting TL, wherein the first discovery request message comprises one or more first parameters indicating a requirement for TL and one or more required TL capabilities for a learning model task; and transmitting, to the first network function, a first discovery response message indicating the second network function as supporting TL for the one or more TL capabilities.

[0221] Further aspects of the disclosure herein provide a network function for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the network function to: transmit, to a repository function, a request message for storage of a pre-trained learning model for transfer learning, wherein the request message comprises one or more parameters indicating one or more capabilities for transfer learning, wherein the request message further comprises one or more parameters indicating one or more competencies for transfer learning. The network function may be a model producer such as a NWDAF MTLF. The repository function may be an ADRF. The network function may transmit, to the repository function a further message to remove or update the stored pre-trained learning model.

[0222] Further aspects of the disclosure herein provide method performed by a network function for wireless communication, comprising: transmitting, to a repository function, a request message for storage of a pre-trained learning model for transfer learning, whereinthe request message comprises one or more parameters indicating one or more capabilities for transfer learning, wherein the request message further comprises one or more parameters indicating one or more competencies for transfer learning. The network function may be a model producer such as a NWDAF MTLF. The repository function may be an ADRF. The method may further comprise transmitting, to the repository function a further message to remove or update the stored pre-trained learning model.

[0223] Further aspects of the disclosure herein provide a repository function for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the repository function to: receive, from a network function, a request message for storage of a pre-trained learning model for transfer learning, wherein the request message comprises one or more parameters indicating one or more capabilities for transfer learning, wherein the request message further comprises one or more parameters indicating one or more competencies for transfer learning. The at least one processor may be further configured to cause the repository function to store the one or more parameters indicating capabilities for transfer learning. The at least one processor may be further configured to cause the repository function to store the one or more parameters indicating competencies for transfer learning. The network function may be a model producer such as a NWDAF MTLF. The repository function may be an ADRF.

[0224] Further aspects of the disclosure herein provide a method in a repository function for wireless communication, comprising: receiving, from a network function, a request message for storage of a pre-trained learning model for transfer learning, wherein the request message comprises one or more parameters indicating one or more capabilities for transfer learning, wherein the request message further comprises one or more parameters indicating one or more competencies for transfer learning. The method may further comprise storing the one or more parameters indicating capabilities for transfer learning. The method may further comprise storing the one or more parameters indicating competencies for transfer learning. The network function may be a model producer such as a NWDAF MTLF. The repository function may be an ADRF.

[0225] Current 3 GPP proposals are introducing transfer learning based on the use of an AI / ML model in similar tasks, but without considering the potential of adapting an AI / ML model employed in one task to solve another new problem. In many cases dissimilar tasks may still relate in terms of model components (e.g., features), dependencies among input data or similarity of network and user behaviour trends. This observation can relate even dissimilar problems if an appropriate transfer strategy is in place to assist how, where and when knowledge can be transferred among different problems to be beneficial, especially when transfer learning is the only option instead of training a model from scratch due to data, time, and resource limitations.

[0226] The disclosure herein facilitates transfer learning to authorized consumers by introducing transfer learning capabilities in a NF profile stored in the NRF and additionally transfer learning model competence information stored locally or in the ADRF to capture whether and how a pre-trained AI / ML model can be used to execute a new task. Transfer learning competence examines the model feature information, in terms of significance, inter-relation or adaptation from one to another, labelling availability and transfer methods as well as domain transfer knowledge rules and model accuracy from prior experience.

[0227] Current 3 GPP proposals do not consider any mechanism to assist transfer learning in the 5G core network, while application layer transfer learning proposals are specific to the service enabler paradigm and are not applicable.

[0228] The disclosure herein provides for TL model storage and removal, where an MTLF TL producer can store its model in the ADRF including the TL model competence, i.e., the AI / ML model information enabling a model applicability with respect to the target Analytics IDs and / or task, (i.e., use case).

[0229] The disclosure herein further provides for TL model registration and discovery, in where an MTLF TL producer can register its TL capabilities in the NRF allowing potential MTLF TL consumers to discover MTLF producers of potential pre-trained models.

[0230] The disclosure herein further provides for TL model selection and retrieval, in where MTLF TL consumers check the potential of adopting a pre-trained model for a newproblem, after being authorized by the MTLF TL producer. To check the potential of adopting pre-trained models, MTLF TL consumers consider the TL capabilities and TL model competence, before selecting and obtaining the optimal pre-trained AI / ML model.

[0231] There is provided an apparatus and a method for enabling transfer learning between an authorized first analytics function (i.e., a MTLF TL model consumer) and a second analytics function (i.e., a MTLF TL model producer) by selecting the components of a pre-trained model that shall be adopted or modified or dropped to convert the said pretrained model for resolving a new problem in the same or different part of the network.

[0232] The apparatus and method may enable transfer learning by allowing a first analytics function (i.e., a MTLF TL model consumer) to: discover a second analytics function (i.e., a MTLF TL model producer) based on its transfer learning capabilities; check the transfer learning competencies of a pre-trained model produced by the second analytics function ( i.e., the MTLF TL model producer) to characterize how fit the said pre-trained model can be when transferred and used for a new task; select a suitable pre-trained model considering the transfer learning capabilities of a second analytics function (i.e., the MTLF TL model producer) in combination with the model competencies for solving a new task; and obtain the selected pre-model related to a second analytics function (i.e., the MTLF TL model producer).

[0233] The transfer learning capabilities that enable the discovery of a second analytics function (i.e., a MTLF TL model producer) rely on any combination of the of the following parameters related to: analytics service information including analytics service types and / or tasks, analytics filter information including time scheduling capabilities, analytics target information; analytics service environment related to network characteristics and / or input data information; and model information including model component information (i.e., features, deep learning layer and task), computing requirements, pre-training time information, reputation information, access permissions.

[0234] The transfer learning competencies may relate a pre-trained model information in a second analytics function (i.e., a MTLF TL model producer) with respect to a target analytics service and / or task at first analytics function (i.e., a MTLF TL model consumer).

[0235] The transfer learning competencies may rely on any combination of feature information and / or label information and / or label information and / or pre-training data information in terms of commonality and / or significance and / or notion / meaning.

[0236] The transfer learning competencies may rely on accuracy information from a prior usage with respect to an analytics service and / or task and / or usage environment.

[0237] The transfer learning capabilities of a second analytics function (i.e., a MTLF TL model producer) may be obtained from a network repository function; and / or the transfer learning competence information of a pre-trained model of a second analytics function (i.e., a MTLF TL model producer) may be obtained after interacting with the said second analytics function.

[0238] The transfer learning competencies may contain instructions regarding rules and / or logic and / or function and / or criteria that inter-relates knowledge gained in an analytics service and / or task under a usage environment to another analytics service and / or task under the same or different usage environment.

[0239] The apparatus and method may use a preparation process that allows the first analytics function (i.e., the MTLF TL model consumer) to check the transfer learning quality of the pre-trained model is indicated by a flag.

[0240] A second analytics function (i.e., MTLF TL model producer) may register its profile in a network repository function to allow discovery by including an indication for supporting transfer learning operations, and by including the respective transfer learning capability information.

[0241] A second analytics function (i.e., a MTLF TL model producer) may: receive a request for transfer learning from a first analytics function (i.e., a MTLF TL model consumer) that includes a flag for checking the transfer learning competency information; authorize a first analytics function (i.e., a MTLF TL model consumer) to obtain transfer learning information; provide transfer learning competency information related to a pretrained model to a first analytics function (i.e., a MTLF TL model consumer); and provision a transfer learning model to a first analytics function (i.e., a MTLF TL modelconsumer) when it receives a model provision request related to a selected pre-trained model.

[0242] The transfer learning competencies related to a pre-trained model of a second analytics function (i.e., a MTLF TL model producer), may be contained locally at the said second analytics function (i.e., the MTLF TL model producer); or may be stored at a data and / or mode repository function.

[0243] It should be noted that the method described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0244] The description herein is provided to enable a person having ordinary skill in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to a person having ordinary skill in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the disclosure is not limited to the examples and designs described herein but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.

[0245] The following abbreviations are relevant in the field addressed by this document: 5G, 5th Generation of Mobile Communication; AI / ML, Artificial Intelligence / Machine Learning; AD AES, Application Data Analytics Enabler Service; ADRF, Analytical Data Repository Function; AF, Application Function; AnLF, Analytics Logical Function; BSF, Binding Support Function; CPU, Central Processing Unit; DCAF, Data Collection AF; DCCF, Data Collection Coordination Functionality; DNAI, Data Network Access Identifier; FQDN, Fully Qualified Domain Name; KPI, Key Performance Indicator; MF, Management Function; MFAF, Messaging Framework Adaptor Function; MnS, Management Service; MTLF, Model Training Logical Function; NEF, Network Exposure Function; NF, Network Function; NRF, Network Repository Function ; NWDAF, Network Data Analytics Function; 0AM, Operations, Administration and Maintenance;PM, Performance Measurement; QoE, Quality of Experience; RAN, Radio Access Network ; TL, Transfer Learning; S-NSSAI, Single-Network Slice Selection Assistance Information; UDM, User Data Manager; UDR, User Data Repository; UE, User Equipment; and URL, Uniform Resource Locator.

Claims

CLAIMSWhat is claimed is:

1. A first network function for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first network function to: determine a requirement for transfer learning for a learning model task; and discover one or more second network functions for supporting transfer learning for the learning model task, based at least in part on one or more transfer learning capabilities of the one or more second network functions.

2. The first network function of claim 1, wherein the at least one processor is further configured to cause the first network function to: transmit, to a third network function, a first discovery request message for network functions supporting transfer learning, wherein the first discovery request message comprises one or more first parameters indicating the requirement for transfer learning and one or more required transfer learning capabilities for the learning model task; and receive, from the third network function, a first discovery response message indicating the one or more second network functions supporting transfer learning, the one or more second network functions supporting at least one of the one or more required transfer learning capabilities for the learning model task.

3. The first network function of claim 2, wherein the one or more first parameters further comprise at least one of: an analytics service information; an analytics service environment information; and a required learning model information.

4. The first network function of any one of claims 2-3, wherein: the first network function is a model consumer;each of the one or more second network functions is a model producer; and / or the third network function is a network repository function.

5. The first network function of any one of claims 2-4, wherein the at least one processor is further configured to cause the first network function to: transmit, to each of the one or more second network functions, a second request message for competencies for transfer learning for the learning model task; and receive, from each of the one or more second network functions, a second response message comprising one or more second parameters indicating competencies of a pretrained learning model for transfer learning for the learning model task.

6. The first network function of claim 5, wherein the one or more second parameters comprise at least one of: a learning model information of the pre-trained learning model, the learning model information indicating a suitability of the pre-trained learning model for transfer learning for the learning model task; and / or an information for accessing the learning model information from a fourth network function.

7. The first network function of claim 6, wherein the learning model information comprises at least one of: a feature information for the pre-trained learning model; a label information for the pre-trained learning model; a training data information for the pre-trained learning model; an accuracy information of prior use of the pre-trained learning model; and a domain information for the pre-trained learning model.

8. The first network function of any one of claims 6-7, wherein the at least one processor is further configured to cause the first network function to: based on the one or more second parameters comprising the information for accessing the learning model information from the fourth network function:transmit, to the fourth network function, a third request message for retrieving the learning model information; and receive, from the fourth network function, the learning model information.

9. The first network function of any one of claims 6-8, wherein the second request message comprises one or more third parameters for selecting the learning model information for transfer learning for the learning model task, wherein the one or more third parameters comprise at least one of: a flag indicating learning model information for pre-trained learning models is required; one or more required identifiers for analytics or for the learning model task; a required usage for the learning model task; a required deep-learning layer for the learning model task; a required label information for the learning model task; a required domain information for the learning model task; one or more computing requirements for the learning model task; one or more targets for analytics for the learning model task; a required training data information for the learning model task; a required feature information of the pre-trained learning model; a required accuracy information for the learning model task; and a required time schedule information for the learning model task.

10. The first network function of any one of claims 5-9, wherein the at least one processor is further configured to cause the first network function to: determine, based at least in part on the second response message, a pre-trained learning model associated with a respective second network function for transfer learning for the learning model task; and obtain the determined pre-trained learning model.

11. A second network function for wireless communication, comprising: at least one memory; andat least one processor coupled with the at least one memory and configured to cause the second network function to: transmit, to a third network function, a fourth request for registering the second network function for being discoverable for transfer learning, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports transfer learning for a pre-trained learning model and one or more transfer learning capabilities.

12. The second network function of claim 11, wherein: the second network function is a model producer; and / or the third network function is a network repository function.

13. The second network function of any one of claims 11-12, wherein the at least one processor is further configured to cause the second network function to: receive, from a first network function, a second request message for competencies for transfer learning for a learning model task; authorize, the first network function, to receive a second response message; and transmit, to the first network function, the second response message, wherein the second response message comprises one or more second parameters indicating competencies of the pre-trained learning model for transfer learning for the learning model task.

14. The second network function of claim 13, wherein the one or more second parameters comprise at least one of: a learning model information of the pre-trained learning model, the learning model information indicating a suitability of the pre-trained learning model for transfer learning for the learning model task; and / or an information for accessing the learning model information from a fourth network function.

15. The second network function of claim 14, wherein the learning model information comprises at least one of: a feature information for the pre-trained learning model; a label information for the pre-trained learning model; a training data information for the pre-trained learning model; an accuracy information of prior use of the pre-trained learning model; and a domain information for the pre-trained learning model.

16. The second network function of any one of claims 13-15, wherein the second request message comprises one or more third parameters for selecting the learning model information for transfer learning for the learning model task, wherein the one or more third parameters comprise at least one of: a flag indicating learning model information for pre-trained learning models is required; one or more required identifiers for analytics or for the learning model task; a required usage for the learning model task; a required deep-learning layer for the learning model task; a required label information for the learning model task; a required domain information for the learning model task; one or more computing requirements for the learning model task; one or more targets for analytics for the learning model task; a required training data information for the learning model task; a required feature information of the pre-trained learning model; a required accuracy information for the learning model task; and a required time schedule information for the learning model task.

17. The second network function of any one of claims 13-16, wherein the first network function is a model consumer.

18. A third network function for wireless communication, comprising: at least one memory; andat least one processor coupled with the at least one memory and configured to cause the third network function to: receive, from a second network function, a fourth request for registering the second network function for being discoverable for transfer learning, wherein the fourth request comprises one or more fourth parameters indicating the second network function supports transfer learning for a pre-trained learning model and one or more transfer learning capabilities; and store, as a network function profile for the second network function, the one or more fourth parameters.

19. The third network function of claim 18, wherein the at least one processor is further configured to cause the third network function to: receive, from a first network function, a first discovery request message for network functions supporting transfer learning, wherein the first discovery request message comprises one or more first parameters indicating a requirement for transfer learning and one or more required transfer learning capabilities for a learning model task; and transmit, to the first network function, a first discovery response message indicating the second network function as supporting transfer learning for the one or more transfer learning capabilities.

20. A method performed by a first network function, the method comprising: determining a requirement for transfer learning for a learning model task; and discovering one or more second network functions for supporting transfer learning for the learning model task, the discovering being based at least in part on one or more transfer learning capabilities of the one or more second network functions.

Images