Continual learning

The described solution optimizes continual learning in wireless networks by determining and configuring machine learning models within user equipment based on capability information, ensuring efficient learning and adaptation, and enabling efficient model transfers, addressing the challenges of existing technologies in managing UE learning processes.

WO2024245557A1PCT designated stage expired Publication Date: 2024-12-05NOKIA TECHNOLOGIES OY
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2023/064642
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-06-01
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Existing technologies face challenges in enhancing continual learning capabilities of machine learning models in user equipment (UE) within wireless networks, particularly in determining optimal architectures and managing performance to ensure efficient learning and adaptation.

Method used

The solution involves an apparatus and method that enable the transmission of continual learning capability information from UE to network nodes, determining suitable machine learning model architectures, configuring models accordingly, and managing performance to maintain or adapt learning processes, including latent replay information architectures and sidelink transfers.

Benefits of technology

This approach optimizes machine learning model configurations based on UE capabilities, ensuring efficient continual learning, adapting to performance levels, and facilitating the transfer of trained models across UEs with similar capabilities, thereby enhancing network performance and resource management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2023064642_05122024_PF_FP_ABST
    Figure EP2023064642_05122024_PF_FP_ABST
Patent Text Reader

Abstract

An apparatus (164) comprising means for enabling transmission (302) of continual learning capability information (166) of a user equipment, UE, (164) towards a network node (168) to enable determining of a machine learning model architecture for continual learning (170). The apparatus (164) receiving, from the network node (168), an indication of the determined machine learning model architecture for continual learning (172) and configuring (174), at the UE (164), at least one machine learning model according to the determined machine learning model architecture for continual learning.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] TITLE

[0002] CONTINUAL LEARNING

[0003] TECHNOLOGICAL FIELD

[0004] Examples of the disclosure relate to continual learning. Some examples relate to configuring at least one machine learning model for continual learning in a user equipment.

[0005] BACKGROUND

[0006] A wireless network comprises a plurality of network nodes including terminal nodes and access nodes. Communication between the terminal nodes and access nodes is wireless.

[0007] In some circumstances it may be desirable to improve or enhance continual learning of at least one machine learning model at a terminal node.

[0008] BRIEF SUMMARY

[0009] According to various, but not necessarily all, embodiments there is provided an apparatus comprising means for: enabling transmission of continual learning capability information of a user equipment, UE, towards a network node to enable determining of a machine learning model architecture for continual learning; receiving, from the network node, an indication of the determined machine learning model architecture for continual learning; and configuring, at the UE, at least one machine learning model according to the determined machine learning model architecture for continual learning. In some examples, continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures.

[0010] In some examples, the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE.

[0011] In some examples, enabling transmission of continual learning capability information comprises enabling transmission of at least one UE capability message and / or at least one message indicating applicable conditions.

[0012] In some examples, the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

[0013] In some examples, the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

[0014] In some examples, the means are configured to: determine performance information for the configured at least one machine learning model; and enable transmission of the determined performance information towards the network node to enable determining of control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model.

[0015] In some examples, the performance information comprises at least one key performance indicator for wireless network function.

[0016] In some examples, the means are configured to: receive the control information; and maintain or adapt continual learning using the at least one machine learning model based, at least in part, on the received control information.

[0017] In some examples, adapting continual learning using the at least one machine learning model comprises at least one of: exiting continual learning and configuring the at least one machine learning model according to a different machine learning model architecture for continual learning.

[0018] In some examples, the means are configured to: receive a request to train the at least one machine learning model; receive a request to transfer the at least one trained machine learning model to at least one other UE using at least one sidelink connection; and transfer the at least one trained machine learning model to the at least one other UE using at least one sidelink connection.

[0019] In some examples, the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

[0020] In some examples, the apparatus comprises: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the performance of the apparatus.

[0021] In some examples, the apparatus is or is comprised in the UE.

[0022] According to various, but not necessarily all, embodiments there is provided a method comprising: enabling transmission of continual learning capability information of a user equipment, UE, towards a network node to enable determining of a machine learning model architecture for continual learning; receiving, from the network node, an indication of the determined machine learning model architecture for continual learning; and configuring, at the UE, at least one machine learning model according to the determined machine learning model architecture for continual learning. In some examples, the continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures.

[0023] In some examples, the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE.

[0024] In some examples, enabling transmission of continual learning capability information comprises enabling transmission of at least one UE capability message and / or at least one message indicating applicable conditions.

[0025] In some examples, the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

[0026] In some examples, the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

[0027] In some examples, the method comprises: determining performance information for the configured at least one machine learning model; and enabling transmission of the determined performance information towards the network node to enable determining of control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model.

[0028] In some examlpes, the performance information comprises at least one key performance indicator for wireless network function.

[0029] In some examples, the method comprises: receiving the control information; and maintaining or adapting continual learning using the at least one machine learning model based, at least in part, on the received control information.

[0030] In some examples, adapting continual learning using the at least one machine learning model comprises at least one of: exiting continual learning and configuring the at least one machine learning model according to a different machine learning model architecture for continual learning.

[0031] In some examples, the method comprises: receiving a request to train the at least one machine learning model; receiving a request to transfer the at least one trained machine learning model to at least one other UE using at least one sidelink connection; and transferring the at least one trained machine learning model to the at least one other UE using at least one sidelink connection.

[0032] In some examples, the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

[0033] According to various, but not necessarily all, embodiments there is provided an apparatus comprising means for: receiving continual learning capability information of a user equipment, UE; determining based, at least in part, on the received continual learning capability information a machine learning model architecture for continual learning; and enabling transmission of an indication of the determined machine learning model architecture for continual learning towards the UE to configure, at the UE, at least one machine learning model according to the machine learning model architecture for continual learning.

[0034] In some examples, the continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures. In some examples, the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE.

[0035] In some examples, receiving continual learning capability information comprises receiving at least one UE capability message and / or at least one message indicating applicable conditions.

[0036] In some examples, the means are configured to determine the machine learning model architecture for continual learning based, at least in part, on at least one of machine learning model architecture use information and machine learning model architecture performance information.

[0037] In some examples, determining a machine learning model architecture for continual learning comprises selecting a machine learning model architecture for continual learning from a plurality of machine learning model architectures for continual learning.

[0038] In some examples, the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

[0039] In some examples, the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

[0040] In some examples, the means are configured to: receive, from the UE, performance information for the configured at least one machine learning model; determine, based, at least in part, on the received performance information, control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model; and enable transmission of the determined control information towards the UE. In some examples, the performance information comprises at least one key performance indicator for wireless network function.

[0041] In some examples, determining control information comprises determining whether performance of the configured at least one machine learning model at the UE is at an acceptable level.

[0042] In some examples, the means are configured to: if it is determined that performance of the configured at least one machine learning model at the UE is at an acceptable level, determine control information to control the UE to maintain continual learning using the configured at least one machine learning model.

[0043] In some examples, the means are configured to: if it is determined that performance of the configured at least one machine learning model at the UE is not at an acceptable level, determine if UE capabilities for continual learning are fully utilized; and if it is determined that the UE capabilities for continual learning are fully utilized, determine control information to control the UE to exit continual learning; and if it is determined that the UE capabilities for continual learning are not fully utilized, determine control information to configure the at least one machine learning model, at the UE, according to a different machine learning model architecture for continual learning.

[0044] In some examples, the means are configured to store information of the UE based, at least in part, on the received continual learning capability information of the user equipment.

[0045] In some examples, the means are configured to: enable transmission towards the UE of a request to train the at least one machine learning model; and enable transmission towards the UE of a request to transfer the at least one trained machine learning model to the at least one other UE using at least one sidelink connection.

[0046] In some examples, the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

[0047] In some examples, the apparatus comprises: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the performance of the apparatus.

[0048] According to various, but not necessarily all, embodiments there is provided a method comprising: receiving continual learning capability information of a user equipment, UE; determining based, at least in part, on the received continual learning capability information a machine learning model architecture for continual learning; and enabling transmission of an indication of the determined machine learning model architecture for continual learning towards the UE to configure, at the UE, at least one machine learning model according to the machine learning model architecture for continual learning.

[0049] In some examples, the continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures.

[0050] In some examples, the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE. In some examples, receiving continual learning capability information comprises receiving at least one UE capability message and / or at least one message indicating applicable conditions.

[0051] In some examples, determining the machine learning model architecture for continual learning based, at least in part, on at least one of machine learning model architecture use information and machine learning model architecture performance information.

[0052] In some examples, determining a machine learning model architecture for continual learning comprises selecting a machine learning model architecture for continual learning from a plurality of machine learning model architectures for continual learning.

[0053] In some examples, the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

[0054] In some examples, the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

[0055] In some examples, the method comprises: receiving, from the UE, performance information for the configured at least one machine learning model; determining, based, at least in part, on the received performance information, control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model; and enabling transmission of the determined control information towards the UE.

[0056] In some examples, the performance information comprises at least one key performance indicator for wireless network function. In some examples, determining control information comprises determining whether performance of the configured at least one machine learning model at the UE is at an acceptable level.

[0057] In some examples, the method comprises: if it is determined that performance of the configured at least one machine learning model at the UE is at an acceptable level, determining control information to control the UE to maintain continual learning using the configured at least one machine learning model.

[0058] In some examples, the method comprises: if it is determined that performance of the configured at least one machine learning model at the UE is not at an acceptable level, determining if UE capabilities for continual learning are fully utilized; and if it is determined that the UE capabilities for continual learning are fully utilized, determining control information to control the UE to exit continual learning; and if it is determined that the UE capabilities for continual learning are not fully utilized, determining control information to configure the at least one machine learning model, at the UE, according to a different machine learning model architecture for continual learning.

[0059] In some examples, the method comprises storing information of the UE based, at least in part, on the received continual learning capability information of the user equipment.

[0060] In some examples, the method comprises: enabling transmission towards the UE of a request to train the at least one machine learning model; and enabling transmission towards the UE of a request to transfer the at least one trained machine learning model to the at least one other UE using at least one sidelink connection. In some examples, the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

[0061] According to various, but not necessarily all, embodiments there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus at least to perform at least part of one or more methods described herein.

[0062] According to various, but not necessarily all, examples there is provided examples as claimed in the appended claims.

[0063] According to various, but not necessarily all, embodiments there is provided an apparatus comprising at least one processor; and at least one memory including computer program code; the at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform at least a part of one or more methods disclosed herein.

[0064] According to various, but not necessarily all, embodiments there is provided an apparatus comprising means for performing at least part of one or more methods disclosed herein.

[0065] The description of any function and / or action should additionally be considered to also disclose any means suitable for performing that function and / or action.

[0066] Functions and / or actions described herein can be performed in any suitable way using any suitable method.

[0067] While the above examples of the disclosure and optional features are described separately, it is to be understood that their provision in all possible combinations and permutations is contained within the disclosure. It is to be understood that various examples of the disclosure can comprise any or all of the features described in respect of other examples of the disclosure, and vice versa. Also, it is to be appreciated that any one or more or all of the features, in any combination, may be implemented by / comprised in / performable by an apparatus, a method, and / or computer program instructions as desired, and as appropriate.

[0068] BRIEF DESCRIPTION

[0069] Some examples will now be described with reference to the accompanying drawings in which:

[0070] FIG. 1 shows an example of the subject matter described herein;

[0071] FIG. 2 shows another example of the subject matter described herein;

[0072] FIG. 3A shows another example of the subject matter described herein;

[0073] FIG. 3B shows another example of the subject matter described herein;

[0074] FIG. 4 shows another example of the subject matter described herein;

[0075] FIG. 5 shows another example of the subject matter described herein;

[0076] FIG. 6 shows another example of the subject matter described herein;

[0077] FIG. 7 shows another example of the subject matter described herein;

[0078] FIG. 8 shows another example of the subject matter described herein;

[0079] FIG. 9 shows another example of the subject matter described herein;

[0080] FIG. 10 shows another example of the subject matter described herein;

[0081] FIG. 11 A shows another example of the subject matter described herein; and FIG. 11 B shows another example of the subject matter described herein.

[0082] The figures are not necessarily to scale. Certain features and views of the figures can be shown schematically or exaggerated in scale in the interest of clarity and conciseness. For example, the dimensions of some elements in the figures can be exaggerated relative to other elements to aid explication. Similar reference numerals are used in the figures to designate similar features. For clarity, all reference numerals are not necessarily displayed in all figures.

[0083] DETAILED DESCRIPTION

[0084] FIG 1 illustrates an example of a network 100 comprising a plurality of network nodes including terminal nodes 110, access nodes 120 and one or more core nodes 129.

[0085] The terminal nodes 110 and access nodes 120 communicate with each other. The one or more core nodes 129 communicate with the access nodes 120. The network 100 is in this example a radio telecommunications network, in which at least some of the terminal nodes 110 and access nodes 120 communicate with each other using transmission / reception of radio waves / signals.

[0086] The one or more core nodes 129 may, in some examples, communicate with each other. The one or more access nodes 120 may, in some examples, communicate with each other.

[0087] The network 100 may be a cellular network comprising a plurality of cells 122 each served by an access node 120. In this example, the interface between the terminal nodes 110 and an access node 120 defining a cell 122 is a wireless interface 124.

[0088] The access node 120 is a cellular radio transceiver. The terminal nodes 110 are cellular radio transceivers.

[0089] In the example illustrated the cellular network 100 is a third generation Partnership Project (3GPP) network in which the terminal nodes 110 are user equipment (UE) 164 (see, for example, FIG. 3A) and the access nodes 120 are base stations.

[0090] In examples the network 100 is an Evolved Universal Terrestrial Radio Access network (E-UTRAN). The E-UTRAN consists of E-UTRAN NodeBs (eNBs) 120, providing the E-UTRA user plane and control plane (RRC) protocol terminations towards the UE 164. The eNBs 120 are interconnected with each other by means of an X2 interface 126. The eNBs are also connected by means of the S1 interface 128 to the Mobility Management Entity (MME) 129.

[0091] In other examples the network 100 is a Next Generation (or New Radio, NR) Radio Access network (NG-RAN). The NG-RAN consists of gNodeBs (gNBs) 120, providing the user plane and control plane (RRC) protocol terminations towards the UE 164. The gNBs 120 are interconnected with each other by means of an Xn interface 126. The gNBs are also connected by means of the N2 interface 128 to the Access and Mobility management Function (AMF). In examples, the network 100 can comprise a combination of E-UTRAN and NG- RAN.

[0092] A user equipment 164 (UE) can comprise a mobile equipment. Where reference is made to user equipment that reference includes and encompasses, wherever possible, a reference to mobile equipment.

[0093] A terminal node 110, for example a UE 164, can train one or more machine learning models 174 (see, for example, FIG. 2) for use by the UE 164 and / or other UEs. The machine learning model(s) 174 can be used for any suitable purpose and / or function in a UE 164, for example the machine learning model(s) 174 can be used in wireless communication.

[0094] In examples, a machine learning model 174 can be configured on a UE 164 for continual learning to enable the machine learning model to continuously learn and / or evolve based on the input of increasing amounts of data while accounting for previously learned knowledge.

[0095] Accordingly, in examples, a UE 164 can train a machine learning model using continual learning to allow the machine learning model to continuously learn and / or adapt.

[0096] The ability of a UE 164 to train a machine learning model 174 using continual learning can be affected and / or constrained by hardware and / or software capabilities of the UE 164.

[0097] The ability to continually learn over time by accommodating new knowledge while retraining previously learned model(s) can be considered continual learning, and / or continuous learning, and / or life long learning, and / or stateful training, and / or stateful retraining.

[0098] A machine learning model 174 can be configured according to a machine learning model architecture 169. A machine learning model architecture 169 can be considered to define the configuration, and / or set-up, and / or parameters of a machine learning model 174.

[0099] A machine learning model architecture 169 can configure a machine learning model 174, for example with an appropriate set-up and / or parameters, for continual learning / to continually learn. For example, a machine learning model 174 configured according to a machine learning model architecture for continual learning 170 can continually learn and / or adapt.

[0100] A machine learning model architecture 169 that configures a machine learning model 174 for continual learning / to continually learn can be considered a machine learning model architecture for continual learning 170.

[0101] FIG. 2 schematically illustrates an example of a machine learning model architecture 169. FIG. 2 can be considered to illustrate a machine learning model 174 configured according to a machine learning model architecture 169.

[0102] In the example of FIG. 2, the machine learning model architecture 169 is a machine learning model architecture for continual learning 170. That is, in the example of FIG. 2, the machine learning model architecture 169 is configured to enable a machine learning model 174 to continuously learn and / or adapt.

[0103] In the illustrated example, the machine learning model architecture for continual learning 170 is configured to enable a machine learning model 174 to use latent replay information 178 in continual learning.

[0104] A machine learning model architecture for continual learning 170 that is configured to enable a machine learning model 174 to use latent replay information 178 in continual learning can be considered a latent replay information continual learning architecture 176. The example illustrated in FIG. 2 is an example of a latent replay information continual learning architecture 176. In the example of FIG. 2, the latent replay information continual learning architecture 176 comprises a frozen stage 198, a latent replay layer 188, and an adaptive stage 200.

[0105] In examples, a latent replay information continual learning architecture 176 comprises information to configure a machine learning model 174 with a frozen stage 198, an adaptive stage 200, and a latent replay layer 188.

[0106] The frozen stage 198 comprises a number of frozen layers 180 that have been trained and do not change or change slowly.

[0107] The adaptive stage 200 comprises a number of adaptive layers that have been trained but are free to learn and / or adapt and / or evolve.

[0108] The latent replay layer 188 is a layer between the frozen layers 180 and the adaptive layers 182. In examples, the frozen layers 180 are the layers below the latent replay layer 188 and the adaptive layers 182 are the layers above the latent replay layer 188.

[0109] Information of and / or representing the latent replay layer 178, such as past raw (input) data, or weights / activation functions representing features, is stored. Information of and / or representing the latent replay layer 178 can be considered latent replay information 178.

[0110] That is, in examples, latent replay information 178 can be considered information that is representative of a layer of a machine learning model 174.

[0111] In examples, latent replay information 178 can be considered a representation of previous data to mix with new data when updating the machine learning model 174.

[0112] The number of frozen layers 180, and / or the number of adaptive layers 182, and / or latent replay information 178 supported by a UE 164 can depend on the capabilities of the UE 164, for example hardware, and / or software, and / or radio resource capabilities of the UE 164. Examples of the disclosure relate to apparatus, methods, and / or computer programs for and / or involved in configuring at least one machine learning model for continual learning in a UE 164.

[0113] Additionally, or alternatively, examples of the disclosure relate to apparatus, methods, and / or computer programs for and / or involved in maintaining or adapting a configured machine learning model architecture at a UE 164.

[0114] Additionally, or alternatively, examples of the disclosure relate to apparatus, methods, and / or computer programs for and / or involved in transferring a trained machine learning model from a UE 164 to at least one other UE 164.

[0115] FIG. 3A illustrates an example of a method 300.

[0116] In examples, FIG. 3A can be considered to illustrate a plurality of methods. For example, FIG. 3A illustrates one or more actions at a plurality of actors / entities and, in examples, FIG. 3A can be considered to illustrate a plurality of methods performed by the individual actors / entities.

[0117] One or more of the features discussed in relation to FIG. 3A can be found in one or more of the other FIGs.

[0118] In the example of FIG. 3A, a plurality of apparatuses transmit and / or receive one or more signals and / or one or more messages across and / or via and / or using a network. In examples, any suitable form of communication in any suitable network can be used. For example, at least a portion of the network 100 of FIG. 1 can be used.

[0119] Accordingly, in examples, the plurality of apparatuses in FIG. 3A form at least a portion of network 100 as described in relation to FIG, 1.

[0120] In the illustrated example, a UE 164 and a network node 168 transmit and / or receive one or more signals and / or one or more messages. The network node 164 can comprise an access node 120, such as a gNodeB (gNB). In some examples, the network node 164 can comprise a core node 129.

[0121] In examples, communications and / or transmissions between elements illustrated in FIG. 3A can proceed via any number of intervening elements, including no intervening elements.

[0122] Although one UE 164 is illustrated in the example of FIG. 3A, in examples any suitable number of UEs 164 can be included, see, for example, FIG. 3B. Similarly, in examples, any suitable number of network nodes 168 can be included.

[0123] As used herein, a description of an action should also be considered to disclose enabling, and / or causing, and / or controlling that action. For example, a description of transmitting information should also be considered to disclose enabling, and / or causing, and / or controlling transmitting / transmission of information.

[0124] The above paragraphs regarding FIG. 3A also apply to, at least, FIG. 3B.

[0125] For example, a description of an apparatus, such as a UE 164, transmitting information should also be considered to disclose at least one controller of the apparatus enabling, and / or causing, and / or controlling the apparatus to transmit the information.

[0126] In examples, method 300 and / or at least a part of method 300 can be considered a method of configuring at least one machine learning model for continual learning in a UE 164.

[0127] In examples, method 300 and / or at least a part of method 300 can be considered a method of determining a machine learning model architecture for continual learning for use by a UE 164.

[0128] In examples, method 300 and / or at least a part of method 300 can be considered a method of maintaining and / or adapting continual learning by a machine learning model at a UE 164. In examples, method 300 and / or at least a part of method 300 can be considered a method of transferring a trained machine learning model from a UE 164 to at least one other UE 164.

[0129] In the illustrated example, the location of blocks indicates the entity performing the action(s). For example, block 304 is performed by the network node 168, for example an access node 120 such as a gNB.

[0130] At block 302, method 300 comprises transmitting continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170.

[0131] As FIG. 3A illustrates one or more actions / features of transmitting, FIG. 3A also illustrates the corresponding transmitting / enabling and / or causing transmitting action(s) / feature(s). For example, from the point of view of the network node 168, at block 302, method 300 comprises receiving continual learning capability information of a UE 164.

[0132] In examples, continual learning capability information 166 of a UE 164 can be considered information that indicates the capability of the UE 164 to support and / or use a machine learning model architecture for continual learning 170.

[0133] In examples, continual learning capability information 166 of a UE 164 can be considered information that indicates the capability of the UE 164 to support and / or use and / or train a machine learning model configured for continual learning according to a machine learning model architecture for continual learning 170.

[0134] The continual learning capability information 166 can comprise information of the capability of the UE 164 and / or information to allow a determination of the capability of the UE 164.

[0135] In examples, an indication of the capability of the UE 164 to support a machine learning model architecture for continual learning 170 can be considered an indication of and / or to enable a determination of the configuration(s) of a machine learning model architecture for continual learning 170 that can be used by the UE 164.

[0136] In examples, method 300 can be considered to comprise determining, for example by the UE 164, the continual learning capability information 166. The continual learning capability information 166 can be determined based, at least in part, on the hardware and / or software, and / or radio capabilities of the UE 164, and / or conditions at the UE 164.

[0137] For example, the continual learning capability information 166 can be determined based, at least in part, on processing capability, and / or memory capability, and / or available radio resources at the UE 164, and / or ongoing tasks. For example, a UE 164 may have an ongoing high priority data transfer that is consuming resources and leaves fewer resources for other things.

[0138] In examples, the continual learning capability information 166 comprises information indicating the capability of the UE 164 to support latent replay information continual learning architectures 176 (see, for example, FIG. 2).

[0139] For example, the continual learning capability information 166 can comprise information indicating the capability of the UE 164 to support and / or use latent replay information continual learning architectures 176.

[0140] For example, the continual learning capability information 166 can comprise information indicating the capability of the UE 164 to support and / or use and / or train a machine learning model 174 configured according to a latent replay information continual learning architecture 176. See, for example, FIG. 2.

[0141] In some examples, the information indicating the capability of the UE 164 to support latent replay information continual learning architectures 176 comprises information indicating at least latent replay information 178, a number of frozen layers 180 supported, and / or a number of adaptive layers 182 supported by the UE 164. In examples, transmitting continual learning capability information 166 comprises transmitting at least one capability message and / or at least one message indicating applicable conditions.

[0142] For example, the continual learning capability information 166 can be transmitted in a UE capability RRC message and / or any communication by the UE 164 containing applicable conditions.

[0143] At block 304, method 300 comprises determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170.

[0144] In examples, block 304 can be considered to comprise determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170 that can be supported and / or used by the UE 164.

[0145] Block 304 can comprise considering a plurality of different machine learning model architectures for continual learning 170, and the received continual learning capability information 166, to determine a machine learning model for continual learning 170.

[0146] In some examples, determining a machine learning model architecture for continual learning 170 comprises selecting a machine learning model architecture for continual learning 170 from a plurality of machine learning model architectures for continual learning 170.

[0147] In some examples, block 304 comprises determining the machine learning model architecture for continual learning 170 based, at least in part, on at least one of machine learning model architecture use information and machine learning model architecture performance information.

[0148] In examples, machine learning model architecture use information can be considered information of and / or related to use of at least one machine learning model architecture by, for example, at least one UE 164. Machine learning model architecture use information can comprise information indicating how many UEs 164 use a machine learning model architecture 169.

[0149] Accordingly, in some examples, block 304 comprises determining which machine learning model architecture for continual learning 170 is used most by UEs 164 out of machine learning model architectures for continual learning 170 that are supported by the UE 164.

[0150] This can be useful, for example, when most of the UEs 164 share similar capabilities (for example smartphones).

[0151] Machine learning model architecture use information can be determined by the network node in any suitable way using any suitable method.

[0152] In examples, machine learning model architecture use information can be received, for example, from at least one other UE 164 and / or determined based, at least in part, on information received from, for example, at least one other UE 164.

[0153] For example, at least one other UE 164 can directly or indirectly inform the network node 168 which machine learning model architecture for continual learning 170 is used by the at least one other UE 164.

[0154] Additionally, or alternatively, the network node 168 can keep track of which machine learning model architecture for continual learning 170 is determined for use by different UEs 164.

[0155] In examples, machine learning model architecture performance information can be considered information of and / or related to performance of at least one machine learning model architecture 169. For example, machine learning model architecture performance information can comprise information indicating how well at least one machine learning model architecture is performing in use by, for example, a UE 164.

[0156] In examples, the network node 168 can receive performance information 190 for a plurality of machine learning model architectures 169 from a plurality of UEs 164 providing feedback regarding how well the machine learning model architecture 169 configured at a UE 164 is performing (see, for example, block 310).

[0157] In examples, performance information 190 comprises at least one key performance indicator (KPI), for example, throughput improvement and / or handover failure / success rate.

[0158] Accordingly, in examples, block 304 comprises determining which machine learning model architecture for continual learning 170 maximizes or at least improves functional performance based, for example, on at least one KPI received from at least one other UE 164, out of machine learning model architectures for continual learning 170 that are supported by the UE 164.

[0159] For example, the network node 168 can select a machine learning model architecture for continual learning 170 that maximizes or at least improves throughput performance or minimized handover failures and so on.

[0160] In some examples, the determined machine learning model architecture for continual learning 170 comprises a latent replay information continual learning architecture 176. For example, the machine learning model architecture for continual learning 170 can comprise at least one of: a number of frozen layers 180, a number of adaptive layers 182, an index of a latent replay layer 188, or a total number of supported layers. See, for example, FIG. 2.

[0161] In examples, the determined machine learning model for continual learning 170 does not use the full capabilities of the UE 164, as indicated in the continual learning capability information 166, because of the architecture that is selected.

[0162] At block 306, method 300 comprises transmitting an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0163] Consequently, FIG. 3 illustrates a method 300 comprising: receiving continual learning capability information 166 of a UE 164; determining based, at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170; and enabling transmission of an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0164] The indication 172 of the determined machine learning model architecture for continual learning 170 can comprise any suitable information and / or have any suitable form. For example, the indication 172 can comprise any suitable information and / or have any suitable form to enable the UE 164 to configure at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0165] In examples where the determined machine learning model architecture for continual learning 170 comprises or is a latent replay information continual learning architecture 176, the indication 172 can comprise any suitable information and / or have any suitable form to enable the UE 164 to configure at least one machine learning model 174 with feature(s) for use in latent replay information continual learning. See, for example, FIG. 2.

[0166] In examples, the indication 172 of the determined machine learning model architecture for continual learning 170 comprises information indicating at least one of: a number of frozen layers 180, a number of adaptive layers 182, an index of a latent replay layer 188, or a total number of supported layers.

[0167] In some examples, the indication 172 comprises information indicating information that should be used as latent replay information 178.

[0168] From the point of view of the UE 164, block 306 comprises receiving, from the network node 168 an indication 172 of the determined machine learning model architecture for continual learning 170. In some examples, the indication 172 may be transmitted and received as one or more messages. For example, a number of frozen layers 180, a number of adaptive layers 182, an index of a latent replay layer 188, and / or or a total number of supported layers may be transmitted using one or more messages from the network node 168 to the UE 164. For example, the indication 172 may be transmitted in one of said one or more messages or in a separate message.

[0169] At block 308, method 300 comprises configuring, at the UE 164, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170. According to an example, UE 164 configures the at least one machine learning model 174 based on the indication 172.

[0170] Consequently, FIG. 3 illustrates a method 300 comprising: enabling transmission of continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170; receiving, from the network node 168, an indication 172 of the determined machine learning model architecture for continual learning 170; and configuring, at the UE, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0171] In examples, block 308 can be considered to comprise configuring at least one parameter, and / or feature, and / or characteristic of a machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0172] In examples, block 308 can be considered to comprise configuring at least one machine learning model 174 to have the indicated machine learning model architecture for continual learning 170.

[0173] In examples where the determined machine learning model architecture for continual learning 170 comprises or is a latent replay information continual learning architecture 176, block 308 can comprise configuring at least one machine learning model 174 with at least one parameter, and / or at least one characteristic, and / or at least one feature for use in latent replay information continual learning. See, for example, FIG. 2.

[0174] For example, block 308 can comprise configuring at least one of: a number of frozen layers 180, a number of adaptive layers 182, an index of a latent replay layer 188, or a total number of supported layers.

[0175] In some examples, block 308 can comprise configuring and / or storing information to be used as latent replay information 178.

[0176] The at least one machine learning model 174, that is configured at block 308, can comprise any suitable machine learning model 174 configured and / or trained for any suitable purpose and / or function at the UE 164.

[0177] For example, the at least one machine learning model 174 can comprise at least one of a deep neural network (DNN).

[0178] In examples, the at least one machine learning model 174 can be configured and / or trained to be used as at least part of any suitable function and / or to provide any suitable functionality at / in the UE 164.

[0179] In some examples, the at least one machine learning model 174 comprises or is a model used in wireless communication at the UE 174.

[0180] In some examples, the at least one machine learning model 174 comprises or is a model configured to enable wireless communication at the UE 174.

[0181] In some examples, the at least one machine learning model 174 comprises or is a model configured to be used in and / or to enable at least one 3rd generation partnership project (3GPP) standard defined behavior at the UE 164.

[0182] For example, the at least one machine learning model 174 can comprise or be a model used in transmission and / or reception of wireless signals at the UE 164. For example, the at least one machine learning model 174 can comprise or be a model used in coding and / or decoding information.

[0183] For example, the at least one machine learning model 174 can comprise or be a model used in compression or decompression of information, such as channel state information (CSI) compression.

[0184] For example, the at least one machine learning model 174 can comprise or be a model used in positioning measurements and / or mobility decisions.

[0185] Examples of the disclosure are advantageous and / or provide technical benefits. For example, examples of the disclosure provide for configuring of at least one machine learning model for continual learning at a UE 164 that is known to be suitable for UE capabilities, such as hardware and / or software capabilities, and / or conditions of the UE.

[0186] For example, examples of the disclosure provide for a best known performing machine learning model architecture for continual learning to be selected and configured at the UE, while ensuring the selected architecture is supported by, and therefore can be used by, the UE.

[0187] In some examples, the UE 164 can use and / or continually train the configured at least one machine learning model 174 and test how well the configured at least one machine learning model 174 performs.

[0188] At block 310, method 300 comprises determining performance information 190 for the configured at least one machine learning model 174.

[0189] Performance information 190 can comprise any suitable information indicative of how well the configured at least one machine learning model is performing. For example, the performance information 190 can comprise any suitable information indicative of how well the UE 164 is performing the function(s) and / or process(es) and / or task(s) in which the configured machine learning model 174 is used and / or involved. In examples, the performance information 190 can comprise any suitable information to enable determining whether performance of the configured at least one machine learning model 174 at the UE 164 is acceptable / at an acceptable level.

[0190] In examples, performance information 190 comprises at least one key performance indicator (KPI). For example, the performance information 190 can comprise at least one KPI for wireless network function.

[0191] In examples, performance information 190 can comprise information indicating throughput performance and / or ratio of handover failures / successes.

[0192] In examples, the performance information 190 can comprise collected performance KPIs based on the determined machine learning model for continual learning 170.

[0193] At block 312, method 300 comprises transmitting the determined performance information 190 towards the network node 168 to enable determining of control information 192 to control the UE 164 to maintain or adapt continual learning using the configured at least one machine learning model 174.

[0194] Consequently, in examples, method 300 comprises determining performance information 190 for the configured at least one machine learning model 174; and enabling transmission of the determined performance information 190 towards the network node 168 to enable determining of control information 192 to control the UE to maintain or adapt continual learning using the configured at least one machine learning model 174.

[0195] In examples, block 312 can comprise transmitting the determined performance information 190 towards the network node 168 to enable determining of control information 192 to decide either to maintain or adapt continual learning using the configured at least one machine learning model 174.

[0196] From the point of view of the network node 168 block 312 comprises receiving, from the UE 164, performance information 190 for the configured at least one machine learning model 174. At block 314, method 300 comprises determining, based at least in part, on the received performance information 190, control information 192 to control the UE 164 to maintain or adapt continual learning using the configured at least one machine learning model 174.

[0197] Control information can comprise any suitable information to control and / or instruct and / or command, and / or cause the UE 164 to perform at least one action and / or function.

[0198] Accordingly, in examples, block 314 can comprise determining, based at least in part, on the received performance information 190, control information 192 to control and / or instruct, and / or command, and / or cause the UE 164 to perform at least one action and / or function.

[0199] The control information 192 can have any suitable form. For example, the control information 192 can comprise or be a medium access control-control element (MAC- CE) or an information element (IE) in a radio resource control (RRC) message.

[0200] In examples, determining control information 192 comprises determining whether performance of the configured at least one machine learning model 174 at the UE 164 is at an acceptable level / is acceptable.

[0201] Determining whether performance of the configured at least one machine learning model 174 at the UE 164 is at an acceptable level can comprise processing and / or assessing, and / or considering the received performance information 190. For example, determining whether performance of the configured at least one machine learning model 174 at the UE 164 is at an acceptable level can comprise determining if at least one KPI is at an acceptable level.

[0202] Determining whether performance is at an acceptable level can comprise determining whether performance is within a tolerance margin of an acceptable level. For example, determining whether performance is at an acceptable level can comprise determining whether at least one KPI is within a tolerance margin of an acceptable level.

[0203] For example, determining whether performance is at an acceptable level can comprise determining whether throughput performance is lower than a predefined target or if handover failure rate is higher than a predefined target and so on.

[0204] In examples, block 314 comprises if it is determined that performance of the configured at least one machine learning model 174 at the UE 164 is at an acceptable level, determining control information 192 to control the UE 164 to maintain continual learning using the configured at least one machine learning model 174.

[0205] Accordingly, in examples, if the network node 168 determines that performance of the configured machine learning model 174 at the UE 164 is acceptable, the network node can instruct the UE 164 to continue with the machine learning model architecture for continual learning 170.

[0206] In examples, block 314 comprise if it is determined that performance of the configured at least one machine learning model 174 at he UE 164 is not at an acceptable level, determining if UE capabilities for continual learning are fully utilized and if it is determined that the UE capabilities for continual learning are fully utilized, determining control information 192 to control the UE 164 to exit continual learning, and if it is determined that the UE capabilities for continual learning are not fully utilized, determining control information 192 to configure the at least one machine learning model 174, at the UE 164, according to a different machine learning model architecture for continual learning 170.

[0207] Determining if UE capabilities for continual learning are fully utilized can be considered determining if the machine learning model architecture for continual learning that is configured at the UE 164 has fully utilized the capabilities of the UE 164 indicated in the continual learning capability information 166.

[0208] For example, determining if UE capabilities for continual learning are fully utilized can comprise determining if available processing and / or memory capacity is fully utilized by the machine learning model architecture for continual learning that is configured at the UE 164.

[0209] In examples where the determined / configured machine learning model architecture for continual learning 170 comprises or is a latent replay information continual learning architecture 176 determining if UE capabilities for continual learning are fully utilized can comprise determining if a different configuration of latent replay layer 188, and / or latent replay information 178, and / or frozen layers 180, and / or adaptive layers 182 can be used that more fully utilize the available capacity / capability at the UE 164.

[0210] Accordingly, in examples, if the network node 168 determines if performance of the configured at least one machine learning model 174 at the UE 164 is unacceptable, the network node 168 can determine if it is possible to change the determined / configured machine learning model architecture for continual learning 170 at the UE 164 to more fully utilize the UE’s capabilities for continual learning, and if not can control the UE 164 to stop continual learning to, for example, stop the unacceptable performance and protect functionality in the network.

[0211] In examples, the control information can comprise an indication 172 of a different and / or modified machine learning model architecture for continual learning 170.

[0212] At block 316, method 300 comprises transmitting the determined control information 192 towards the UE 164.

[0213] Consequently, in examples, method 300 comprises receiving, from the UE 164, performance information 190 for the configured at least one machine learning model 174; determining, based, at least in part, on the received performance information 190, control information 192 to control the UE 164 to maintain or adapt continual learning using the configured at least one machine learning model 174; and enabling transmission of the determined control information 192 towards the UE 164.

[0214] From the point of view of the UE 164, block 316 comprises receiving the control information 192. At block 318, method 300 comprises maintaining or adapting continual learning using the at least one machine learning model 174, based, at least in part, on the received control information 192.

[0215] Consequently, in examples, method 300 comprises receiving the control information 192; and maintaining or adapting continual learning using the at least one machine learning model 174 based, at least in part, on the received control information 192.

[0216] In examples, block 318 comprises maintaining or adapting continual learning using the at least one machine learning model 174 as instructed by the network node 168 in the control information 192.

[0217] In examples, maintaining continual learning using the at least one machine learning model 174 comprises continuing to use and / or train the at least one machine learning model 174 using the machine learning model architecture for continual learning determined at block 170.

[0218] In examples, adapting continual learning using the at least one machine learning model 174 comprises altering, and / or changing, and / or adapting, and / or varying continual learning using the at least one machine learning model in any suitable way.

[0219] In some examples, adapting continual learning using the at least one machine learning model 174 comprises at least one of: exiting continual learning and configuring the at least one machine learning model according to a different machine learning model architecture for continual learning 170.

[0220] Exiting continual learning can comprise at least one of: falling back to a legacy machine learning solution without continual learning, falling back to a legacy solution without machine learning, or falling back to a legacy solution if the solution is available and / or required, otherwise disabling the functionality / feature provided by the machine learning model 174. Accordingly, exiting continual learning can comprise switching to a non-machine learning solution.

[0221] Examples of the disclosure are advantageous and / or provide technical benefits.

[0222] For example, examples of the disclosure enable modification of continual learning at a UE if performance of a configured machine learning model is unacceptable.

[0223] For example, examples of the disclosure provide a mechanism to control adverse effects on system performance coming from UEs having lower capability for continual learning.

[0224] At block 320, method 300 comprises storing information of the UE 164 based, at least in part, on the received continual learning capability information 166 of the UE 164.

[0225] This allows, for example, the network node 168 to keep track of the capabilities of UEs 164 for continual learning. For example, this can allow / enable the network node 168 to group UEs having similar capabilities for continual learning. Similar capabilities can be considered capabilities within a given range, for example a given range for processing and / or memory capability.

[0226] Grouping UEs can allow the network node 168 to efficiently train and / or transfer machine learning models, such as machine learning models configured with a machine learning model architecture for continual learning 170, which can be considered continual learning machine learning models.

[0227] FIG. 3B illustrates an example of a method 300a.

[0228] FIG. 3B is similar to the example of FIG. 3A, however in the example of FIG. 3B there are a three different UEs 164, 164a, 164b. In some, but not necessarily all, examples FIG. 3B can be at least a part of method 300. Accordingly, in some examples, the UE 164 and network node 168 of the example of FIG. 3B can be those illustrated in the example of FIG. 3A.

[0229] However, in some examples, method 300a can be separate from and independent to the example of FIG. 3A.

[0230] At block 320, method 300a comprises transmitting towards the UE 164 a request 194 to train the at least one machine learning model 174.

[0231] The request 194 can have any suitable form. In examples, the request 194 can be considered control information.

[0232] At block 322, method 300a comprises transmitting towards the UE 164 a request 196 to transfer the at least one trained machine learning model 202 to at least one other UE 164a, 164b using at least one sidelink connection.

[0233] The request 196 can have any suitable form. In examples, the request 196 can be considered control information.

[0234] In examples, method 300a comprises determining, by the network node 168, the at least one other UE 164a, 164b to which the trained machine learning model 202 is to be transferred.

[0235] In some examples, the at least one other UE 164a, 164b is selected based, at least in part, on continual learning capability information 166 of the at least one other UE 164a, 164b.

[0236] For example, the network node 168 has received continual learning capability information 166 from the UEs 164, 164a, and 164b and can determine that the UEs 164, 164a, and 164b have similar continual learning capabilities and can therefore group the UEs 164, 164a, and 164b together. The network node 168 can use the grouping to determine that the machine learning model trained by the UE 164 can be transferred to UEs 164a and 164b that share similar continual learning capabilities with UE 164.

[0237] From the point of view of the UE 164, blocks 320 and 322 comprise receiving a request 194 to train the at least one machine learning model 174 and receiving a request to transfer the at least one trained machine learning model 202 to at least one other UE 164a, 164b using at least one sidelink connection.

[0238] At block 324 method 300a comprises training the machine learning model 174 to provide a trained machine learning model 202.

[0239] At block 326, method 300a comprises transferring the at least one trained machine learning model 202 to the at least one other UE 164 using at least one sidelink connection.

[0240] Examples of the disclosure are advantageous and / or provide technical benefits.

[0241] For example, examples of disclosure provide for efficient training and transfer of machine learning models at UEs.

[0242] For example, examples of the disclosure provide for efficient management of UEs based on continual learning capabilities.

[0243] FIG. 4 illustrates an example of a method 400.

[0244] In examples, method 400 can be performed by any suitable apparatus comprising any suitable means for performing method 400. For example, an apparatus as described in relation to FIG. 11 A and / or 11 B.

[0245] In examples, method 400 can be performed by a UE 164. At block 402, method 400 comprises enabling transmission of continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170.

[0246] At block 404, method 400 comprises receiving, from the network node 168, an indication 172 of the determined machine learning model architecture for continual learning 170.

[0247] At block 406, method 400 comprises configuring, at the UE 164, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0248] Consequently, FIG. 4 illustrates a method 400 comprising: enabling transmission of continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170; receiving, from the network node 168, an indication 172 of the determined machine learning model architecture for continual learning 170; and configuring, at the UE 164, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0249] FIG. 5 illustrates an example of a method 500.

[0250] In examples, method 500 can be performed by any suitable apparatus comprising any suitable means for performing method 500. For example, an apparatus as described in relation to FIG. 11 A and / or 11 B.

[0251] In examples, method 500 can be performed by a network node 168.

[0252] At block 502, method 500 comprises receiving continual learning capability information 166 of a UE 164. At block 504, method 500 comprises determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170.

[0253] At block 506, method 500 comprises enabling transmission of an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0254] Consequently, FIG. 5 illustrates a method 500 comprising: receiving continual learning capability information 166 of a UE 164; determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170; and enabling transmission of an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0255] Examples of the disclosure relate to a framework for continual learning application in wireless communication where different adaptive and fixed / frozen part may be selected for different UEs having different (computational and storage) capabilities for processing adaptive and fixed / frozen stages of a machine learning (ML) model.

[0256] FIG. 6 illustrates an example of a method 600.

[0257] In examples, method 600 can be performed by any suitable apparatus comprising any suitable means for performing method 600.

[0258] At block 602 UE informs the NW about its capabilities regarding supported latent replay information in the context of Continual Learning.

[0259] At block 604 network selects an ML architecture that can be supported for the given UE, as extracted from the latent replay information provided by the UE. At block 606 UE collects its performance KPIs based on selected architecture and reports it back to the NW.

[0260] At block 608, if the performance is within limited tolerance margin, network confirms the UE that it can keep using the suggested architecture. Otherwise, NW commands the UE to exit CL.

[0261] In examples, network will add this UE in the group of UEs with similar UE capabilities. The objective of this is to avoid having too many different architectures at the NW side for all different UEs on the field.

[0262] FIG. 6 can be considered to be a high-level illustration of examples of the disclosure.

[0263] FIG. 7 illustrates an example of a method 700. FIG. 7 is similar to the example of FIG. 3A.

[0264] In examples, FIG. 7 can be considered to illustrate a plurality of methods. For example, FIG. 7 illustrates one or more actions at a plurality of actors / entities and, in examples, FIG. 7 can be considered to illustrate a plurality of methods performed by the individual actors / entities.

[0265] Error! Reference soui t found.Rawe-EfGv 7 can be considered to show core signalling steps for implementing method 700 for performing continual learning (CL) with one frozen part and a distinct set of adaptive parts where selection of adaptive and frozen parts is based on the UE capabilities regarding latent replay information.

[0266] In examples, this information is either communicated via a UE Capability message i.e. a new field in the UE capability RRC message or by any other communication by the UE containing “applicable conditions’. Examples of the disclosure provide a framework in order to select a set of distinct adaptive parts (Aps) for CL, which is supportable by different UEs depending upon their capabilities and applicable conditions.

[0267] Examples of the disclosure also provide a mechanism to not badly impact network (NW) performance due to low capable UEs. Furthermore, examples of the disclosure provide a mechanism to reduce overhead in case more UEs share similar UE capabilities.

[0268] Method 700 comprises the following blocks as shown in Error! Reference source not found.RGv 7, where UE is performing local inference on ML model. This ML model can be assumed to be fully trained at NW side or locally at the UE side.

[0269] Block 0: Starting CL, the network request UEs to send their capabilities for CL (not illustrated).

[0270] 1. (block 702) UEs send feedback on its capabilities regarding Continual learning. This can be latent replay information in either the UE Capabilities message or the “applicable conditions” message currently under discussion in RAN1 for AI / ML. One example of latent replay information can be total number of supported layers and the index of the latent replay layer.

[0271] Block 704: NW considers UE capabilities for selecting an ML architecture for the given UE. This ML architecture must be supported for the given UE, as extracted from the latent replay information provided by the UE.

[0272] An example for such a criterion could be selecting an architecture out of supported architectures which is used for the most UEs. This criterion is useful when most of the UEs share similar capabilities (smartphone).

[0273] Another example is selecting an architecture out of supported architectures that maximizes functional performance based on a KPI. For instance, based on different UEs feedback, NW can select an architecture that maximizes throughput performance or minimized handover failures. In this stage, the suggested architecture may not be necessarily using full UE capabilities since the suggested architecture is based upon either of the above criteria.

[0274] 2. (block 706) NW sends the selected architecture to the UE

[0275] Block 708 UE configures the architecture sent by the UE and starts the training and inference phase.

[0276] 3. (block 710) NW requests the UE to send its performance KPI with the configured architecture.

[0277] 4. (block 712) UE collects its performance KPIs with the selected architecture and reports it to the network

[0278] Block 714 Network evaluates the performance based on a preselected criterion, with a tolerance margin 8 for UEs not capable of using optimal architecture, which is selected based on the feedback of a number of UEs from the field.

[0279] Block 716 Network determines if performance is acceptable.

[0280] 5. (blocks 718 and 720) If the performance is within limited tolerance margin 8, network confirms that the UE can keep using the suggested architecture.

[0281] Block 722 If the suggested architecture performance is not acceptable, the next action is based on the criterion used for suggested architecture selection:

[0282] 6a If the full UE capabilities have been used in the suggested architecture, then there is nothing much NW can do network asks the UE to stop using the suggested architecture, and UE would exit CL in this case.

[0283] 6b Otherwise, suggest an architecture to the UE using full UE capabilities. And go back to block 706. Network will keep track of this UE and will add them in the group of UEs with similar UE capabilities who are able to perform within the tolerated margin range. The objective of this step is to avoid having too many different architectures at the NW side for all different UEs on the field.

[0284] This group of UEs can further be used to optimally transfer trained ML models to different UEs in the field, as explained in the additional embodiment.

[0285] FIG. 8 illustrates a method 800. FIG. 8 can be considered to illustrate a flowchart of the method 700 illustrated in the example of FIG. 7.

[0286] Once the network verifies that KPI performance of the suggested model is within acceptable range, it sends the confirmation to the UE. At the same time, the network adds this UE in the group of UE supporting same latent replay information, this is necessary to avoid too much different architecture (for each UE) at the network side. Furthermore, network will select one / some UEs from each group for retraining / model transfer purposes.

[0287] FIG. 9 illustrates an example of a method 900. FIG. 9 is similar to FIG 3B.

[0288] In examples, FIG. 9 can be considered to illustrate a plurality of methods. For example, FIG. 9 illustrates one or more actions at a plurality of actors / entities and, in examples, FIG. 9 can be considered to illustrate a plurality of methods performed by the individual actors / entities.

[0289] In this example, UE1 164a, UE2 164b and UE3 164c are within the same group supporting same or similar latent replay information.

[0290] UE2 164b is selected (block 902) by to perform model training. At 2,3 (blocks 904, 906), UE2 164b receives from the gNB a request to realize the model training and transfer the trained model to both UE1 164a and UE3 164c through sidelink (blocks 910 and 916). The signaling can be over the existing ProSe PC3a and PC5 interfaces.

[0291] The received ML model from UE2 164b is then checked at both UE1 164a and UE3 164c (respectively 6 (block 912) and 9 (block 918)).

[0292] The performance is then shared with gNB to confirm that the model trained by UE2 164b is well suited for UE1 164a and UE3 164c (7 (block 914) and 10 (block 920)).

[0293] FIG. 10 illustrates an example of a method 1000. FIG. 10 is similar to FIG. 9.

[0294] As shown in the example of FIG. 10, sidelink can also be used among different UEs 164a, 164b, 164c, 164d with same or similar latent replay information for model transfer.

[0295] Fig 11A illustrates an example of an apparatus 130. The apparatus 130 may be a controller of an apparatus or device such as a terminal node 110, for example a UE 164, or network node 168 such as an access node 120. The apparatus 130 may be considered a controller.

[0296] Implementation of a 130 may be as controller circuitry. The controller 130 may be implemented in hardware alone, have certain aspects in software including firmware alone or can be a combination of hardware and software (including firmware).

[0297] In some examples, the apparatus 130 is or is comprised in the UE 164.

[0298] In some examples, the apparatus 130 is or is comprised in the network node 168.

[0299] As illustrated in Fig 11A the controller 130 may be implemented using instructions that enable hardware functionality, for example, by using executable instructions of a computer program 136 in a general-purpose or special-purpose processor 132 that may be stored on a computer readable storage medium (disk, memory etc) to be executed by such a processor 132. The processor 132 is configured to read from and write to the memory 134. The processor 132 may also comprise an output interface via which data and / or commands are output by the processor 132 and an input interface via which data and / or commands are input to the processor 132.

[0300] The memory 134 stores a computer program 136 comprising computer program instructions (computer program code) that controls the operation of the apparatus 130 when loaded into the processor 132. The computer program instructions, of the computer program 136, provide the logic and routines that enables the apparatus to perform the methods illustrated in the accompanying Figs. The processor 132 by reading the memory 134 is able to load and execute the computer program 136.

[0301] In examples, the apparatus 130 comprises: at least one processor 132; and at least one memory 134 including computer program code the at least one memory 134 and the computer program code configured to, with the at least one processor 132, cause the apparatus 130 at least to perform: enabling transmission of continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170; receiving, from the network node 168, an indication 172 of the determined machine learning model architecture for continual learning 170; and configuring, at the UE 164, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0302] In examples, the apparatus 130 comprises: at least one processor 132; and at least one memory 134 including computer program code the at least one memory 134 and the computer program code configured to, with the at least one processor 132, cause the apparatus 130 at least to perform: receiving continual learning capability information 166 of a UE 164; determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170; and enabling transmission of an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0303] In examples, the apparatus 130 comprises: at least one processor 132; and at least one memory 134 including computer program code, the at least one memory storing instructions that, when executed by the at least one processor 132, cause the apparatus at least to: enabling transmission of continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170; receiving, from the network node 168, an indication 172 of the determined machine learning model architecture for continual learning 170; and configuring, at the UE 164, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0304] In examples, the apparatus 130 comprises: at least one processor 132; and at least one memory 134 including computer program code, the at least one memory storing instructions that, when executed by the at least one processor 132, cause the apparatus at least to: receiving continual learning capability information 166 of a UE 164; determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170; and enabling transmission of an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0305] As illustrated in Fig 11 A, the computer program 136 may arrive at the apparatus 130 via any suitable delivery mechanism 162. The delivery mechanism 162 may be, for example, a machine readable medium, a computer-readable medium, a non- transitory computer-readable storage medium, a computer program product, a memory device, a record medium such as a Compact Disc Read-Only Memory (CD- ROM) or a Digital Versatile Disc (DVD) or a solid-state memory, an article of manufacture that comprises or tangibly embodies the computer program 136. The delivery mechanism may be a signal configured to reliably transfer the computer program 136. The apparatus 130 may propagate or transmit the computer program 136 as a computer data signal.

[0306] Computer program instructions for causing an apparatus to perform at least the following or for performing at least the following: enabling transmission of continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170; receiving, from the network node 168, an indication 172 of the determined machine learning model architecture for continual learning 170; and configuring, at the UE 164, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0307] Computer program instructions for causing an apparatus to perform at least the following or for performing at least the following: receiving continual learning capability information 166 of a UE 164; determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170; and enabling transmission of an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0308] The computer program instructions may be comprised in a computer program, a non- transitory computer readable medium, a computer program product, a machine readable medium. In some but not necessarily all examples, the computer program instructions may be distributed over more than one computer program. Although the memory 134 is illustrated as a single component / circuitry it may be implemented as one or more separate components / circuitry some or all of which may be integrated / removable and / or may provide permanent / semi-permanent / dynamic / cached storage.

[0309] In examples the memory 134 comprises a random access memory 158 and a read only memory 160. In examples the computer program 136 can be stored in the read only memory 158. See, for example, Fig. 11 B

[0310] Although the processor 132 is illustrated as a single component / circuitry it may be implemented as one or more separate components / circuitry some or all of which may be integrated / removable. The processor 132 may be a single core or multi-core processor.

[0311] References to ‘computer-readable storage medium’, ‘computer program product’, ‘tangibly embodied computer program’ etc. or a ‘controller’, ‘computer’, ‘processor’ etc. should be understood to encompass not only computers having different architectures such as single / multi- processor architectures and sequential (Von Neumann) / parallel architectures but also specialized circuits such as field- programmable gate arrays (FPGA), application specific circuits (ASIC), signal processing devices and other processing circuitry. References to computer program, instructions, code etc. should be understood to encompass software for a programmable processor or firmware such as, for example, the programmable content of a hardware device whether instructions for a processor, or configuration settings for a fixed-function device, gate array or programmable logic device etc.

[0312] As used in this application, the term ‘circuitry’ may refer to one or more or all of the following:

[0313] (a) hardware-only circuitry implementations (such as implementations in only analog and / or digital circuitry) and

[0314] (b) combinations of hardware circuits and software, such as (as applicable):

[0315] (i) a combination of analog and / or digital hardware circuit(s) with software / firmware and (ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory or memories that work together to cause an apparatus, such as a mobile phone or server, to perform various functions and (c) hardware circuit(s) and or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (for example, firmware) for operation, but the software may not be present when it is not needed for operation.

[0316] This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit for a mobile device or a similar integrated circuit in a server, a cellular network device, or other computing or network device.

[0317] The blocks illustrated in the accompanying Figs may represent steps in a method and / or sections of code in the computer program 136. The illustration of a particular order to the blocks does not necessarily imply that there is a required or preferred order for the blocks and the order and arrangement of the block may be varied. Furthermore, it may be possible for some blocks to be omitted.

[0318] Where a structural feature has been described, it may be replaced by means for performing one or more of the functions of the structural feature whether that function or those functions are explicitly or implicitly described.

[0319] Thus, the apparatus 130 can, in examples, comprise means for: enabling transmission of continual learning capability information 166 of a UE 164 towards a network node 168 to enable determining of a machine learning model architecture for continual learning 170; receiving, from the network node 168, an indication 172 of the determined machine learning model architecture for continual learning 170; and configuring, at the UE 164, at least one machine learning model 174 according to the determined machine learning model architecture for continual learning 170.

[0320] Thus, the apparatus 130 can, in examples, comprise means for: receiving continual learning capability information 166 of a UE 164; determining, based at least in part, on the received continual learning capability information 166 a machine learning model architecture for continual learning 170; and enabling transmission of an indication 172 of the determined machine learning model architecture for continual learning 170 towards the UE 164 to configure, at the UE 164, at least one machine learning model 174 according to the machine learning model architecture for continual learning 170.

[0321] In examples, an apparatus 130 can comprise means for performing one or more methods, or at least part of one or more methods, as disclosed herein.

[0322] In examples, an apparatus 130 can be configured to perform one or more methods, or at least a part of one or more methods, as disclosed herein.

[0323] The systems, apparatus, methods and computer programs may use machine learning which can include statistical learning. Machine learning is a field of computer science that gives computers the ability to learn without being explicitly programmed. The computer learns from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E. The computer can often learn from prior training data to make predictions on future data. Machine learning includes wholly or partially supervised learning and wholly or partially unsupervised learning. It may enable discrete outputs (for example classification, clustering) and continuous outputs (for example regression). Machine learning may for example be implemented using different approaches such as cost function minimization, artificial neural networks, support vector machines and Bayesian networks for example. Cost function minimization may, for example, be used in linear and polynomial regression and K-means clustering. Artificial neural networks, for example with one or more hidden layers, model complex relationship between input vectors and output vectors. Support vector machines may be used for supervised learning. A Bayesian network is a directed acyclic graph that represents the conditional independence of a number of random variables.

[0324] The above-described examples find application as enabling components of: automotive systems; telecommunication systems; electronic systems including consumer electronic products; distributed computing systems; media systems for generating or rendering media content including audio, visual and audio visual content and mixed, mediated, virtual and / or augmented reality; personal systems including personal health systems or personal fitness systems; navigation systems; user interfaces also known as human machine interfaces; networks including cellular, non-cellular, and optical networks; ad-hoc networks; the internet; the internet of things; virtualized networks; and related software and services.

[0325] The apparatus can be provided in an electronic device, for example, a mobile terminal, according to an example of the present disclosure. It should be understood, however, that a mobile terminal is merely illustrative of an electronic device that would benefit from examples of implementations of the present disclosure and, therefore, should not be taken to limit the scope of the present disclosure to the same. While in certain implementation examples, the apparatus can be provided in a mobile terminal, other types of electronic devices, such as, but not limited to: mobile communication devices, hand portable electronic devices, wearable computing devices, portable digital assistants (PDAs), pagers, mobile computers, desktop computers, televisions, gaming devices, laptop computers, cameras, video recorders, GPS devices and other types of electronic systems, can readily employ examples of the present disclosure. Furthermore, devices can readily employ examples of the present disclosure regardless of their intent to provide mobility.

[0326] The term ‘comprise’ is used in this document with an inclusive not an exclusive meaning. That is any reference to X comprising Y indicates that X may comprise only one Y or may comprise more than one Y. If it is intended to use ‘comprise’ with an exclusive meaning then it will be made clear in the context by referring to “comprising only one...” or by using “consisting”.

[0327] In this description, the wording ‘connect’, ‘couple’ and ‘communication’ and their derivatives mean operationally connected / coupled / in communication. It should be appreciated that any number or combination of intervening components can exist (including no intervening components), i.e. , so as to provide direct or indirect connection / coupling / communication. Any such intervening components can include hardware and / or software components.

[0328] As used herein, the term "determine / determining" (and grammatical variants thereof) can include, not least: calculating, computing, processing, deriving, measuring, investigating, identifying, looking up (for example, looking up in a table, a database or another data structure), ascertaining and the like. Also, "determining" can include receiving (for example, receiving information), accessing (for example, accessing data in a memory), obtaining and the like. Also, " determine / determining" can include resolving, selecting, choosing, establishing, and the like.

[0329] In this description, reference has been made to various examples. The description of features or functions in relation to an example indicates that those features or functions are present in that example. The use of the term ‘example’ or ‘for example’ or ‘can’ or ‘may’ in the text denotes, whether explicitly stated or not, that such features or functions are present in at least the described example, whether described as an example or not, and that they can be, but are not necessarily, present in some of or all other examples. Thus ‘example’, ‘for example’, ‘can’ or ‘may’ refers to a particular instance in a class of examples. A property of the instance can be a property of only that instance or a property of the class or a property of a sub-class of the class that includes some but not all of the instances in the class. It is therefore implicitly disclosed that a feature described with reference to one example but not with reference to another example, can where possible be used in that other example as part of a working combination but does not necessarily have to be used in that other example.

[0330] Although examples have been described in the preceding paragraphs with reference to various examples, it should be appreciated that modifications to the examples given can be made without departing from the scope of the claims.

[0331] Features described in the preceding description may be used in combinations other than the combinations explicitly described above.

[0332] Although functions have been described with reference to certain features, those functions may be performable by other features whether described or not. Although features have been described with reference to certain examples, those features may also be present in other examples whether described or not.

[0333] The term ‘a’, ‘an’ or ‘the’ is used in this document with an inclusive not an exclusive meaning. That is any reference to X comprising a / an / the Y indicates that X may comprise only one Y or may comprise more than one Y unless the context clearly indicates the contrary. If it is intended to use ‘a’, ‘an’ or ‘the’ with an exclusive meaning then it will be made clear in the context. In some circumstances the use of ‘at least one’ or ‘one or more’ may be used to emphasis an inclusive meaning but the absence of these terms should not be taken to infer any exclusive meaning.

[0334] The presence of a feature (or combination of features) in a claim is a reference to that feature or (combination of features) itself and also to features that achieve substantially the same technical effect (equivalent features). The equivalent features include, for example, features that are variants and achieve substantially the same result in substantially the same way. The equivalent features include, for example, features that perform substantially the same function, in substantially the same way to achieve substantially the same result.

[0335] In this description, reference has been made to various examples using adjectives or adjectival phrases to describe characteristics of the examples. Such a description of a characteristic in relation to an example indicates that the characteristic is present in some examples exactly as described and is present in other examples substantially as described.

[0336] The above description describes some examples of the present disclosure however those of ordinary skill in the art will be aware of possible alternative structures and method features which offer equivalent functionality to the specific examples of such structures and features described herein above and which for the sake of brevity and clarity have been omitted from the above description. Nonetheless, the above description should be read as implicitly including reference to such alternative structures and method features which provide equivalent functionality unless such alternative structures or method features are explicitly excluded in the above description of the examples of the present disclosure.

[0337] Whilst endeavoring in the foregoing specification to draw attention to those features believed to be of importance it should be understood that the Applicant may seek protection via the claims in respect of any patentable feature or combination of features hereinbefore referred to and / or shown in the drawings whether or not emphasis has been placed thereon. l / we claim:

Claims

CLAIMS1. An apparatus comprising means for: enabling transmission of continual learning capability information of a user equipment, UE, towards a network node to enable determining of a machine learning model architecture for continual learning; receiving, from the network node, an indication of the determined machine learning model architecture for continual learning; and configuring, at the UE, at least one machine learning model according to the determined machine learning model architecture for continual learning.

2. An apparatus as claimed in claim 1, wherein the continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures.

3. An apparatus as claimed in claim 2, wherein the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE.

4. An apparatus as claimed in any preceding claim, wherein enabling transmission of continual learning capability information comprises enabling transmission of at least one UE capability message and / or at least one message indicating applicable conditions.

5. An apparatus as claimed in any preceding claim, wherein the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

6. An apparatus as claimed in claim 5, wherein the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

7. An apparatus as claimed in any preceding claim, wherein the means are configured to: determine performance information for the configured at least one machine learning model; and enable transmission of the determined performance information towards the network node to enable determining of control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model.

8. An apparatus as claimed in claim 7, wherein the performance information comprises at least one key performance indicator for wireless network function.

9. An apparatus as claimed in claim 7 or 8, wherein the means are configured to: receive the control information; and maintain or adapt continual learning using the at least one machine learning model based, at least in part, on the received control information.

10. An apparatus as claimed in claim 9, wherein adapting continual learning using the at least one machine learning model comprises at least one of: exiting continual learning and configuring the at least one machine learning model according to a different machine learning model architecture for continual learning.

11. An apparatus as claimed in any preceding claim, wherein the means are configured to: receive a request to train the at least one machine learning model; receive a request to transfer the at least one trained machine learning model to at least one other UE using at least one sidelink connection; and transfer the at least one trained machine learning model to the at least one other UE using at least one sidelink connection.

12. An apparatus as claimed in claim 11, wherein the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

13. An apparatus as claimed in any preceding claim, wherein the apparatus comprises: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the performance of the apparatus.

14. An apparatus as claimed in any preceding claim, wherein the apparatus is or is comprised in the UE.

15. A method comprising: enabling transmission of continual learning capability information of a user equipment, UE, towards a network node to enable determining of a machine learning model architecture for continual learning; receiving, from the network node, an indication of the determined machine learning model architecture for continual learning; and configuring, at the UE, at least one machine learning model according to the determined machine learning model architecture for continual learning.

16. A method as claimed in claim 15, wherein the continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures.

17. A method as claimed in claim 16, wherein the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE.

18. A method as claimed in any of claims 15 to 17, wherein enabling transmission of continual learning capability information comprises enabling transmission of atleast one UE capability message and / or at least one message indicating applicable conditions.

19. A method as claimed in any of claims 15 to 18, wherein the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

20. A method as claimed in claim 19, wherein the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

21. A method as claimed in any of claims 15 to 20, comprising: determining performance information for the configured at least one machine learning model; and enabling transmission of the determined performance information towards the network node to enable determining of control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model.

22. A method as claimed in claim 21 , wherein the performance information comprises at least one key performance indicator for wireless network function.

23. A method as claimed in claim 21 or 22, comprising: receiving the control information; and maintaining or adapting continual learning using the at least one machine learning model based, at least in part, on the received control information.

24. A method as claimed in claim 23, wherein adapting continual learning using the at least one machine learning model comprises at least one of: exiting continual learning and configuring the at least one machine learning model according to a different machine learning model architecture for continual learning.

25. A method as claimed in any of claims 15 to 24, comprising:receiving a request to train the at least one machine learning model; receiving a request to transfer the at least one trained machine learning model to at least one other UE using at least one sidelink connection; and transferring the at least one trained machine learning model to the at least one other UE using at least one sidelink connection.

26. A method as claimed in claim 25, wherein the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

27. An apparatus comprising means for: receiving continual learning capability information of a user equipment, UE; determining based, at least in part, on the received continual learning capability information a machine learning model architecture for continual learning; and enabling transmission of an indication of the determined machine learning model architecture for continual learning towards the UE to configure, at the UE, at least one machine learning model according to the machine learning model architecture for continual learning.

28. An apparatus as claimed in claim 27, wherein the continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures.

29. An apparatus as claimed in claim 28, wherein the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE.

30. An apparatus as claimed in any of claims 27 to 29, wherein receiving continual learning capability information comprises receiving at least one UE capability message and / or at least one message indicating applicable conditions.

31. An apparatus as claimed in any of claims 27 to 30, wherein the means are configured to determine the machine learning model architecture for continual learning based, at least in part, on at least one of machine learning model architecture use information and machine learning model architecture performance information.

32. An apparatus as claimed in any of claims 27 to 31 , wherein determining a machine learning model architecture for continual learning comprises selecting a machine learning model architecture for continual learning from a plurality of machine learning model architectures for continual learning.

33. An apparatus as claimed in any of claims 27 to 32, wherein the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

34. An apparatus as claimed in claim 33, wherein the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

35. An apparatus as claimed in any of claims 27 to 34, wherein the means are configured to: receive, from the UE, performance information for the configured at least one machine learning model; determine, based, at least in part, on the received performance information, control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model; and enable transmission of the determined control information towards the UE.

36. An apparatus as claimed in claim 35, wherein the performance information comprises at least one key performance indicator for wireless network function.

37. An apparatus as claimed in claim 35 or 36, wherein determining control information comprises determining whether performance of the configured at least one machine learning model at the UE is at an acceptable level.

38. An apparatus as claimed in claim 37, wherein the means are configured to: if it is determined that performance of the configured at least one machine learning model at the UE is at an acceptable level, determine control information to control the UE to maintain continual learning using the configured at least one machine learning model.

39. An apparatus as claimed in claim 37 or 38, wherein the means are configured to: if it is determined that performance of the configured at least one machine learning model at the UE is not at an acceptable level, determine if UE capabilities for continual learning are fully utilized; and if it is determined that the UE capabilities for continual learning are fully utilized, determine control information to control the UE to exit continual learning; and if it is determined that the UE capabilities for continual learning are not fully utilized, determine control information to configure the at least one machine learning model, at the UE, according to a different machine learning model architecture for continual learning.

40. An apparatus as claimed in any of claims 27 to 39, wherein the means are configured to store information of the UE based, at least in part, on the received continual learning capability information of the user equipment.

41. An apparatus as claimed in any of claims 27 to 40, wherein the means are configured to: enable transmission towards the UE of a request to train the at least one machine learning model; and enable transmission towards the UE of a request to transfer the at least one trained machine learning model to the at least one other UE using at least one sidelink connection.

42. An apparatus as claimed in claim 41 , wherein the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

43. An apparatus as claimed in any of claims 27 to 42, wherein the apparatus comprises: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the performance of the apparatus.

44. A method comprising: receiving continual learning capability information of a user equipment, UE; determining based, at least in part, on the received continual learning capability information a machine learning model architecture for continual learning; and enabling transmission of an indication of the determined machine learning model architecture for continual learning towards the UE to configure, at the UE, at least one machine learning model according to the machine learning model architecture for continual learning.

45. A method as claimed in claim 44, wherein the continual learning capability information comprises information indicating the capability of the UE to support latent replay information continual learning architectures.

46. A method as claimed in claim 45, wherein the information indicating the capability of the UE to support latent replay information continual learning architectures comprises information indicating at least latent replay information, a number of frozen layers supported by the UE, and / or a number of adaptive layers supported by the UE.

47. A method as claimed in any of claims 44 to 46, wherein receiving continual learning capability information comprises receiving at least one UE capability message and / or at least one message indicating applicable conditions.

48. A method as claimed in any of claims 44 to 47, comprising determining the machine learning model architecture for continual learning based, at least in part, on at least one of machine learning model architecture use information and machine learning model architecture performance information.

49. A method as claimed in any of claims 44 to 48, wherein determining a machine learning model architecture for continual learning comprises selecting a machine learning model architecture for continual learning from a plurality of machine learning model architectures for continual learning.

50. A method as claimed in any of claims 44 to 49, wherein the determined machine learning model architecture for continual learning comprises a latent replay information continual learning architecture.

51. A method as claimed in claim 50, wherein the indication of the determined machine learning model architecture for continual learning comprises information indicating at least one of: a number of frozen layers, a number of adaptive layers, an index of a latent replay layer, or a total number of supported layers.

52. A method as claimed in any of claims 44 to 51 , comprising: receiving, from the UE, performance information for the configured at least one machine learning model; determining, based, at least in part, on the received performance information, control information to control the UE to maintain or adapt continual learning using the configured at least one machine learning model; and enabling transmission of the determined control information towards the UE.

53. A method as claimed in claim 52, wherein the performance information comprises at least one key performance indicator for wireless network function.

54. A method as claimed in claim 52 or 53, wherein determining control information comprises determining whether performance of the configured at least one machine learning model at the UE is at an acceptable level.

55. A method as claimed in claim 54, comprising: if it is determined that performance of the configured at least one machine learning model at the UE is at an acceptable level, determining control information to control the UE to maintain continual learning using the configured at least one machine learning model.

56. A method as claimed in claim 54 or 55, comprising: if it is determined that performance of the configured at least one machine learning model at the UE is not at an acceptable level, determining if UE capabilities for continual learning are fully utilized; and if it is determined that the UE capabilities for continual learning are fully utilized, determining control information to control the UE to exit continual learning; and if it is determined that the UE capabilities for continual learning are not fully utilized, determining control information to configure the at least one machine learning model, at the UE, according to a different machine learning model architecture for continual learning.

57. A method as claimed in any of claims 44 to 56, comprising storing information of the UE based, at least in part, on the received continual learning capability information of the user equipment.

58. A method as claimed in any of claims 44 to 57, comprising: enabling transmission towards the UE of a request to train the at least one machine learning model; and enabling transmission towards the UE of a request to transfer the at least one trained machine learning model to the at least one other UE using at least one sidelink connection.

59. A method as claimed in claim 58, wherein the at least one other UE is selected based, at least in part, on continual learning capability information of the at least one other UE.

60. A computer program comprising instructions which, when executed by an apparatus, cause the apparatus at least to perform a method as claimed in at least one of claims 15 to 26 and / or a method as claimed in at least one of claims 44 to 59.

Citation Information

Patent Citations

  • Information reporting method, apparatus and device, and storage medium

    US20220342713A1