System and apparatus for model training in a network and a method in association thereto

Abstract

System (100), device (104) and a method (300) for model training in a network are disclosed. The method (300) includes obtaining a user device capability status; generating at least one signal for a local model training based on the user device capability status; and training the local model based on the at least one signal.

Description

SYSTEM AND APPARATUS FOR MODEL TRAINING IN A NETWORK AND A METHOD IN ASSOCIATION THERETOField Of Invention

[0001] The present disclosure generally relates to one or both of a system and a device for model training in a network in association with, for example, a User Equipment (UE) usable for communication. The present disclosure further relates a method which can be associated with the system and / or the apparatus.Background of Invention

[0002] Generally, energy efficiency and power saving would be helpful in communication networks, for example, a 3rd Generation Partnership Project (3GPP) 5G (fifth generation) New Radio (NR) standard-based telecommunications network.

[0003] Current techniques may not address the issue of reducing overhead dataset transmission during model training in a communication network. Specifically, datasets in a model training may be large and large overhead may be caused by the transfer of these datasets. This may lead to problems such as excessive overhead signaling and unnecessary energy consumption when training a model, for example an Artificial Intelligence / Machine Learning (AIML) model. Thus, the current techniques may not facilitate energy efficiency and power saving in an optimal manner.

[0004] The present disclosure contemplates that it would be helpful to address or at least mitigate one or more issues in relation to conventional techniques for facilitating energy efficiency and power saving when training a model in a network.Summary of the Invention

[0005] According to a first aspect of the present invention, there is provided a method for model training in a network, the method comprising: obtaining a user devicecapability status; generating at least one signal for a local model training based on the user device capability status; and training the local model based on the at least one signal.

[0006] Advantageously, the method as described herein can reduce overhead signaling for a model training. In particular, no dataset is transferred therefore overhead signaling can be saved and data privacy can be maintained. In addition, parallel processing of multiple models at different entities can be achieved as the model training for each model happens independently without the need of global convergence.

[0007] In an embodiment, obtaining the user device capability status comprises receiving data associated with split learning capability of the user device.

[0008] In an embodiment, generating the at least one signal for a local model training comprises: generating an indication for model split training; and determining one or more sub-models of the local model for model split training.

[0009] In an embodiment, the method further comprises transmitting the one or more sub-models and the indication for model split training.

[0010] In an embodiment, the one or more sub-model comprises at least one of: Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), and / or K- Nearest Neighbors (KNN).

[0011] In an embodiment, transmitting the one or more sub-models and the indication for model split training comprises transmitting via at least one of: User Equipment (UE) specific message, Radio Resource Control (RRC) reconfiguration and / or a signal configured for a model.

[0012] In an embodiment, the method further comprises generating one or more datasets from the local model training; and transmitting the one or more datasets from the local model training to a global model.

[0013] In an embodiment, the one or more datasets comprises activations having corresponding labels.

[0014] In an embodiment, the method further comprises receiving the one or more datasets; updating the global model based on the one or more datasets; generating computed gradients based on the updated global model; and transmitting the computed gradients to the local model to update the local model based on the computed gradients.

[0015] In an embodiment, the method further comprises configuring one or more user devices into groups based on a computational capability associated with each of the one or more user devices.

[0016] In an embodiment, the computational capability comprises at least one of: mobility, latency, hardware capabilities, and / or battery capabilities.

[0017] In an embodiment, the local model and the global model is an Artificial Intelligence / Machine Learning (AIML) model.

[0018] In an embodiment, there is provided a computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of the first aspect.

[0019] In an embodiment, there is provided a computer readable storage medium having data stored therein representing software executable by a computer, the software including instructions, when executed by the computer, to carry out the method of the first aspect.

[0020] In an embodiment, there is provided a device for model training in a network comprising: a first module configured to receive at least one signal associated with a user device capability status; a second module configured to at least one of processand facilitate the method of the first aspect to generate at least one output signal; and a third module configured to communicate at least one output signal, wherein the output signal corresponds to a control signal for a local model training based on the user device capability status.

[0021] In an embodiment, the device corresponds to a base station communicable with an apparatus corresponding to a User Equipment (UE), and wherein the base station corresponds to a Next generation Node B (gNB) configured to receive the at least one input signal from the UE.

[0022] In an embodiment, there is provided a system comprising: at least one apparatus(es); and at least one device(s), wherein the apparatus(es) and the device(s) are capable of being coupled via at least one of wired coupling and wireless coupling.

[0023] Advantageously, the system as described herein can provide fundamental mechanisms of interworking and data information flow in radio access network collaboration for AIML support. Accordingly, AIML performance for wireless communication can be improved.Brief Description of the Drawings

[0024] Embodiments of the disclosure are described hereinafter with reference to the following drawings, in which:

[0025] Fig. 1A shows a schematic diagram illustrating a system for model training in a network which can include at least one device, according to an embodiment of the invention.

[0026] Figs. 1 B shows an example scenario in association with the system of Fig. 1A, according to an embodiment of the invention.

[0027] Fig. 2 shows a schematic diagram illustrating the device of Fig. 1A in further detail, according to an embodiment of the invention.

[0028] Fig. 3 shows a method in association with the system of Fig. 1A, according to an embodiment of the invention.

[0029] Fig. 4A to Fig. 4D show schematic diagrams illustrating the flow of information in association with the method of Fig. 3, according to an embodiment of the inventionDetailed Description

[0030] The present specification discloses apparatus and / or device for performing the operations of the methods. Such apparatus and / or device may be specially constructed for the required purposes, or may comprise a computer or other device selectively activated or reconfigured by a computer program stored in the computer. The algorithms and displays presented herein are not inherently related to any particular computer or other apparatus. Various machines may be used with programs in accordance with the teachings herein. Alternatively, the construction of more specialized apparatus to perform the required method steps may be appropriate. The structure of a computer will appear from the description below.

[0031] In addition, the present specification also implicitly discloses a computer program, in that it would be apparent to the person skilled in the art that the individual steps of the method described herein may be put into effect by computer code. The computer program is not intended to be limited to any particular programming language and implementation thereof. It will be appreciated that a variety of programming languages and coding thereof may be used to implement the teachings of the disclosure contained herein. Moreover, the computer program is not intended to be limited to any particular control flow. There are many other variants of the computer program, which can use different control flows without departing from the spirit or scope of the disclosure.

[0032] Furthermore, one or more of the steps of the computer program may be performed in parallel rather than sequentially. Such a computer program may be stored on any computer readable medium. The computer readable medium may include storage devices such as magnetic or optical disks, memory chips, or other storage devices suitable for interfacing with a computer. The computer readable medium may also include a hard-wired medium such as exemplified in the Internet system, or wireless medium such as exemplified in the mobile telephone system. The computer program when loaded and executed on such a computer effectively results in an apparatus and / or a device that implements the steps of the preferred method.

[0033] The present disclosure generally contemplates the facilitation and optimization of a network (for example in association with 3GPP based standard / specification etc.) and / or user equipment (UE) efficiency (for example energy efficiency or power saving), in accordance with an embodiment of the invention. Specifically, the present disclosure contemplates the possibility of model training in connection with 3GPP standard(s).

[0034] The present disclosure generally contemplates that an air-interface for Artificial Intelligence / Machine Learning (AIML) model can be implemented in a network (for example in association with 3GPP based standard / specification etc.). Specifically, UE-sided model training data may be collected at Radio Access Network (RAN), for example RAN2 / RAN1. Corresponding contents of UE data collection may be identified and the UE data collection mechanisms may be analyzed along with the implications and limitations of each of the methods.

[0035] The present disclosure also contemplates the UE data collection may be or may not be transparent. Moreover, UE data collection may require numerous computing cycles and memories from the UE. For example, UE data collection for AIML-based channel state information (CSI) compression and prediction may require numerous Central Processing Units (CPUs).

[0036] The present disclosure further contemplates there are three use cases for consideration in a current network. For example, the three use-cases can be CSI compression and prediction, beam management and AIML Assisted Positioning. It can be appreciated that more than three use-cases are possible and the aforementioned use-cases may not be exhaustive. The number of use-cases for AIML may increase and data collection may become an important aspect of model training. Therefore, the present disclosure contemplates that methods may be required where the dataset transfer is minimal or not needed for model training.

[0037] The present disclosure further contemplates a distributed training method can be implemented in a network where entities may have different AIML models, and the edge entities can generate and use their own dataset without the need to transfer the dataset for model training to a global model.

[0038] In the above manner, each UE may or may not perform the local model update to the server and overhead signaling can thus be reduced to minimize signaling of inference. Power saving and energy consumption efficiency can possibly be facilitated in the network, in accordance with an embodiment of the invention.

[0039] The foregoing will be discussed in further detail with reference to Fig. 1 to Fig. 4 hereinafter.

[0040] Referring to Fig. 1A, a schematic diagram illustrating a system 100 for model training in a network is shown, according to an embodiment of the invention. The system 100 can, for example, be suitable for facilitating energy and improve power efficiency, in accordance with an embodiment of the invention.

[0041] As shown, the system 100 can include one or more apparatuses 102, at least one device 104 and, optionally, a communication network 106, in accordance with an embodiment of the invention.

[0042] The apparatus(es) 102 can be coupled to the device(s) 104. Specifically, the apparatus(es) 102 can, for example, be coupled to the device(s) 104 via the communication network 106, in accordance with an embodiment of the invention.

[0043] In one embodiment, the apparatus(es) 102 can be coupled to the communication network 106 and the device(s) 104 can be coupled to the communication network 106. Coupling can be by manner of one or both of wired coupling and wireless coupling. The apparatus(es) 102 can, in general, be configured to communicate with the device(s) 104 via the communication network 106, according to an embodiment of the invention.

[0044] The apparatus(es) 102 can, for example, be associated with or correspond to or include one or more user equipment (UE) which can carry one or more computers, in accordance with an embodiment of the invention. For example, an apparatus 102 can correspond to a UE (or user device) carrying at least one computer (e.g. an electronic device or module having computing capabilities such as an electronic mobile device which can be carried into a vehicle or an electronic module which can be installed in a vehicle, in accordance with an embodiment of the invention) which can be configured to perform one or more processing tasks in association with the UE (or user device), in accordance with an embodiment of the invention.

[0045] The device(s) 104 can, for example, be associated with / correspond to at least one base station, where the at least one base station can be a Next Generation Node B (gNB). Moreover, the device(s) 104 can, for example, be configured to carry / be associated with / include one or more computers (e.g., an electronic device / module having computing capabilities) which can, for example, be configured to perform one or more processing tasks in association with the base station. The device(s) 104 can be configured to generate one or more output signals which can be communicated to the apparatus(es) 102, in accordance with an embodiment of the invention. This will be discussed later in further detail in the context of an example scenario, in accordance with an embodiment of the invention.

[0046] In an embodiment, the device(s) 104 can, for example, be configured to receive one or more input signals and perform at least one processing task based on the input signal(s) in a manner to generate one or more output signals. The input signal(s) can, for example, be communicated from the apparatus(es) 102 and received by the device(s) 104, in accordance with an embodiment of the invention. The input signal can be a signal associated with a user device (or UE) capability status. As a possible option, the output signal(s) can, for example, be communicated from the device(s) 104, in accordance with an embodiment of the invention. The output signal may correspond to a control signal for a local model training. The device(s) 104 will be discussed later in further detail with reference to Fig. 2, according to an embodiment of the invention.

[0047] The communication network 106 can, for example, correspond to an Internet communication network, a cellular-based communication network, a wired-based communication network, a Global Navigation Satellite System (GNSS) based communication network, a wireless-based communication network, or any combination thereof. Communication (e.g., between the apparatuses 102 and / or between the apparatus(es) 102 and the device(s) 104) via the communication network 106 can be by manner of one or both of wired communication and wireless communication.

[0048] As mentioned, the device(s) 104 can, for example, be configured to receive at least one input signal and perform at least one processing task in association with dynamic / adaptive / gradual control on the input signal(s) in a manner so as to generate at least one output signal. Moreover, the apparatus(es) 102 can, for example, be configured to generate (and communicate) the input signal(s) to the device(s) 104, in accordance with an embodiment of the invention. Accordingly, the apparatus(es) 102 can determine the user device capability status and communicate or transmit at least one input signal associated with the user device capability status to the device(s) 104. This will be discussed, in accordance with an embodiment of the invention, in the context of an example scenario with reference to Fig. 1 B, hereinafter.

[0049] Fig. 1 B shows an example scenario in association with the system of Fig. 1A, according to an embodiment of the invention. Specifically, Fig. 1 B shows an example of a wireless network for model training, for example an Artificial Intelligence / Machine Learning (AIML) Federated Learning (FL) model. As shown in the Figure, a global model, such as a Machine Learning (ML) model (e.g. an AIML model), may be located at a base station (e.g. gNB) and is in communication with multiple user devices or User Equipment (UE). Each of the user device (or UE) may include a local model (e.g. ML model) that may be a part of the global model at the base station (or gNB). The global model may obtain activations from each of the UE (or user device) for global model training and inference. The UE-side activations may include trained weights and labelling related to the AIML model. The global model may also transmit computed gradients to each of the UE (or user device) for model update.

[0050] In such communications, the UE data collection may be or may not be transparent and UE data collection may require numerous computing cycles and memories from the UE. For example, UE data collection for AIML-based channel state information (CSI) compression and prediction may require numerous Central Processing Units (CPUs).

[0051] The above-described aspect(s) of the system 100 of the present invention can also apply analogously (all) the aspect(s) of a below described device 104 of the present invention. Likewise, all below described aspect(s) of the device 104 of the invention can also apply analogously (all) the aspect(s) of above-described system 100 of the invention.

[0052] The aforementioned device(s) 104 or base station (or gNB) will be discussed in further detail with reference to Fig. 2 hereinafter.

[0053] Referring to Fig. 2, a schematic diagram illustrating a device 104 is shown in further detail in the context of an example implementation 200, according to an embodiment of the invention.

[0054] In the example implementation 200, the device 104 can correspond to an electronic module 200a. The electronic module 200a can, in one example, correspond to a base station or a cell (or gNB), in accordance with an embodiment of the invention. In another example, the electronic module 200a can correspond to an electronic device which can be installed / mounted in the base station (or gNB), in accordance with an embodiment of the invention.

[0055] It is contemplated that the electronic module 200a can be capable of performing one or more processing tasks in association with adaptive / dynamic / gradual control related processing, in accordance with an embodiment of the invention.

[0056] The electronic module 200a can, for example, include a casing 200b. Moreover, the electronic module 200a can, for example, carry any one of a first module 202, a second module 204, a third module 206, or any combination thereof.

[0057] In one embodiment, the electronic module 200a can carry a first module 202, a second module 204 and / or a third module 206. In a specific example, the electronic module 200a can carry a first module 202, a second module 204 and a third module 206, in accordance with an embodiment of the invention.

[0058] In this regard, it is appreciable that, in one embodiment, the casing 200b can be shaped and dimensioned to carry any one of the first module 202, the second module 204 and the third module 206, or any combination thereof.

[0059] The first module 202 can be coupled to one or both of the second module 204 and the third module 206. The second module 204 can be coupled to one or both of the first module 202 and the third module 206. The third module 206 can be coupled to one or both of the first module 202 and the second module 204. In one example, the first module 202 can be coupled to the second module 204 and the second module 204 can be coupled to the third module 206, in accordance with an embodiment of the invention. Coupling between the first module 202, the secondmodule 204 and / or the third module 206 can, for example, be by manner of one or both of wired coupling and wireless coupling. Each of the first module 202, the second module 204 and the third module 206 can correspond to one or both of a hardware-based module and a software-based module, according to an embodiment of the invention.

[0060] In one example, the first module 202 can correspond to a hardware-based receiver which can be configured to receive one or more input signals. The input signal(s) can, for example, be communicated from the apparatus(es) 102 (or user device or User Equipment UE), in accordance with an embodiment of the invention.

[0061] The second module 204 can, for example, correspond to a hardware-based processor which can be configured to perform one or more processing tasks (e.g., in a manner so as to generate one or more output signals) as will be discussed later in further detail with reference to Fig. 3, in accordance with an embodiment of the invention.

[0062] The third module 206 can correspond to a hardware-based transmitter which can be configured to communicate one or more output signals from the electronic module 200a. The output signal(s) can, for example, include one or more instructions / commands / control signals in association with the aforementioned dynamic / adaptive / gradual control configuration / determination strategy so as to facilitate efficiency (e.g., power / energy efficiency and / or communication efficiency), in accordance with an embodiment of the invention. For example, the output signal(s) can be a control signal(s) for a local model training based on a user device (or UE) capability status.

[0063] The present disclosure contemplates the possibility that the first and second modules 202, 204 can be an integrated software-hardware based module, for example, an electronic part which can carry a software program or algorithm in association with receiving and processing functions or an electronic module programmed to perform the functions of receiving and processing. The present disclosure further contemplates the possibility that the first and third modules 202,206 can be an integrated software-hardware based module, for example an electronic part which can carry a software program or algorithm in association with receiving and transmitting functions or an electronic module programmed to perform the functions of receiving and transmitting. The present disclosure yet further contemplates the possibility that the first and third modules 202, 206 can be an integrated hardware module, for example a hardware-based transceiver, capable of performing the functions of receiving and transmitting.

[0064] The device 104 (or base station or UE) can, for example, be further configured to process the input signal(s), as will be discussed later in further detail with reference to Fig. 3, in a manner so as to generate one or more output signals in a manner so as to facilitate efficiency, for example power efficiency or energy efficiency, in accordance with an embodiment of the invention. In one specific example, the output signal(s) can include one or more control signals to facilitate some form of dynamic / adaptive / gradual control configuration / determination strategy so as to facilitate efficiency, for example power efficiency or energy efficiency, in accordance with an embodiment of the invention. For example, the output signal(s) can be a control signal(s) for local model training based on the user device (or UE) capability status.

[0065] The above-described aspect(s) of the device 104 of the present invention can also apply analogously (all) the aspect(s) of a below described processing / communication method of the present invention. Likewise, all below described aspect(s) of the method of the invention can also apply analogously (all) the aspect(s) of above described device 104 of the invention. It is to be appreciated that these remarks apply analogously to the earlier discussed system 100 of the present disclosure.

[0066] Referring to Fig. 3, a method 300 (or a communication method) for model training, for example an Artificial Intelligence / Machine Learning (AIML) model, in association with the system 100 is shown, according to an embodiment of the invention.

[0067] The method 300 can, for example, be suitable for facilitating energy efficiency, network optimization and power saving in accordance with an embodiment of the invention.

[0068] The method 300 can include any one of an input step 302, a processing step 304 and an output step 306, or any combination thereof, in accordance with an embodiment of the invention.

[0069] In an embodiment, the processing method 300 can include the input step 302. In another embodiment, the processing method 300 can include the input step 302 and the processing step 304. In another embodiment, the processing method 300 can include the input step 302, the processing step 304 and the output step 306. In yet another embodiment, the processing method 300 can include the processing step 304 and one or both of the input step 302 and the output step 306. In yet a further embodiment, the processing method 300 can include the input step 302, the processing step 304 and the output step 306. In yet a further additional embodiment, the processing method 300 can include the processing step 304. In yet another further additional embodiment, the processing method 300 can include any one of or any combination of the input step 302, the processing step 304 and the output step 306 (i.e. , the input step 302, the processing step 304 and / or the output step 306).

[0070] With regard to the input step 302, one or more input signal(s) can be received. For example, the input signal(s) can be communicated from the apparatus 102 and can be received by the device 104, in accordance with an embodiment of the invention.

[0071] The input step 302 can include receiving at least one input signal associated with a user device capability status, which may include receiving data associated with split learning capability of the user device.

[0072] With regard to the processing step 304, at least a processing task can be performed in association with the received input signal(s) in a manner so as togenerate one or more output signals, in accordance with an embodiment of the invention.

[0073] The processing step 304 may include at least one of: generating at least one signal for a local model training based on the user device capability status and training the local model based on the at least one signal. The processing step 304 may further include generating an indication for model split training; determining one or more sub-models of the local model for model split training; and transmitting the one or more sub-models and the indication for model split training. The one or more sub-model may include at least one of: Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), and / or K-Nearest Neighbors (KNN) and transmitting the one or more sub-models and the indication for model split training may include transmitting via at least one of: User Equipment (UE) specific message, Radio Resource Control (RRC) reconfiguration and / or a signal configured for a model (e.g. a AIML model).

[0074] The processing step 304 may also include generating one or more datasets from the local model training and transmitting the one or more datasets from the local model training to a global model. The one or more datasets may include activations having corresponding labels, for example smashed data with computed weights.

[0075] The processing step 304 may further include receiving the one or more activations (or datasets); updating the global model based on the one or more activations (or datasets); generating computed gradients based on the updated global model; and transmitting the computed gradients to the local model to update the local model based on the computed gradients.

[0076] The processing step 304 may also include configuring one or more user devices into groups based on a computational capability associated with each of the one or more user devices, whereby the computational capability may be at least one of: mobility, latency, hardware capabilities (or hardware specification) and / or batterycapabilities. The local model and the global model may be an Artificial Intelligence / Machine Learning (AIML) model.

[0077] In an example embodiment, the processing step 304 may be performed by the apparatus 102 (user device or UE) and may further include determining the split learning capability of the apparatus 102 and transmitting at least one signal (or input signals) for a local model training based on the split learning capability to the device 104 (or base station).

[0078] In an embodiment, the UE (or user device) can provide the signal to the gNB (or base station) for split learning in a UE capability message. For example, the message can be a 1 -bit indication if the UE (or user device) supports split learning functionality, and if the UE message indicates 1 , it signifies the UE (or user device) supports split learning. On the other hand, if the UE message indicates 0, it signifies the UE (or user device) does not support split learning.

[0079] In an alternate embodiment, the message may be a 2-bit indication which notifies the computational capability of the UE (or user device), where the 2-bit indication may be present on the condition that the UE (or user device) indicates support for split learning. For example, a “00” 2-bit indication may indicate the UE (or user device) may be least capable for split learning, a “01” 2-bit indication may indicate the UE (or user device) may be fairly capable for split learning, a “10” 2-bit indication may indicate the UE (or user device) may be moderately capable for split learning and a “11” 2-bit indication may indicate the UE (or user device) may be highly capable for split learning.

[0080] With regards to the output step 306, the output signal(s) can, for example, be communicated, as an option, in accordance with an embodiment of the invention. For example, the output signal(s) can optionally be communicated from the device 104. In a more specific example, the output signal(s) can optionally be communicated from the device 104 to one or both of at least one apparatus 102, in accordance with an embodiment of the invention. In an embodiment, the apparatus102 (or UE) may also perform the input step 302, the processing step 304 and the output step 306.

[0081] The present disclosure further contemplates a computer program (not shown) which can include instructions which, when the program is executed by a computer (not shown), cause the computer to carry out the input step 302, the processing step 304 and / or the output step 306 as discussed with reference to the method 300. For example, the computer program can include instructions which, when the program is executed by a computer, cause the computer to carry out the input step 302 and / or the processing step 304, in accordance with an embodiment of the invention.

[0082] The present disclosure yet further contemplates a computer readable storage medium (not shown) having data stored therein representing software executable by a computer (not shown), the software including instructions, when executed by the computer, to carry out the input step 302, the processing step 304 and / or the output step 306 as discussed with reference to the method 300. For example, the computer readable storage medium can have data stored therein representing software executable by a computer, the software including instructions, when executed by the computer, cause the computer to carry out the input step 302 and / or the processing step 304, in accordance with an embodiment of the invention.

[0083] Further in view of the foregoing, it is appreciable that the present disclosure generally contemplates a device 104 for model training in a network which can include a first module 202, a second module 204 and / or a third module 206.

[0084] The first module 202 can be configured to receive one or more input signals. The input signal(s) can, for example, be associated with a user device (or UE) capability status.

[0085] The second module 204 can be configured to process and / or facilitate processing of the input signal(s) according to the method 300 as discussed earlier to generate one or more output signals.

[0086] The third module 206 can be configured to communicate one or more output signals. The output signal(s) can, for example, correspond to one or more control signals for a local model training based on the user device capability status.

[0087] In one embodiment, the apparatus 102 can correspond to a User Equipment (UE) which can communicate with a device 104 corresponding to a base station. The base station can, for example, correspond to a Next generation Node B (gNB) which can be configured to communicate one or more signals (e.g., output signal(s)) to the UE.

[0088] Yet further in view of the foregoing, it is appreciable that the present disclosure generally contemplates a system 100 which can include one or more apparatuses 102 and one or more devices 104. The apparatus(es) 102 and the device(s) 104 can, for example, be capable of being coupled via wired coupling and / or wireless coupling.

[0089] It should be appreciated that the embodiments described above can be combined in any manner as appropriate (e.g., one or more embodiments as discussed in the “Detailed Description” section can be combined with one or more embodiments as described in the “Summary of the Invention” section).

[0090] It should be further appreciated by the person skilled in the art that variations and combinations of embodiments described above, not being alternatives or substitutes, may be combined to form yet further embodiments.

[0091] In one example, the possibility of the output signal(s) being communicated from the apparatus(es) 102 was discussed. It is appreciable that the output signal(s) need not necessarily be communicated from the apparatus(es) 102. Specifically, the possibility that the output signal(s) need not necessarily be communicated outside of the apparatus(es) 102 is contemplated, in accordance with an embodiment of the invention. More specifically, the output signal(s) can, for example, correspond to internal command(s) / instruction(s) (e.g., communicated only within an apparatus 102)for adaptively controlling operational configuration of an apparatus 102, in accordance with an embodiment of the invention.

[0092] In another example, application(s) of the present disclosure in association with / in the context of low power wake up radio and / or ambient loT (Internet of Things) type device(s) can be possible, in accordance with an embodiment of the invention.

[0093] Fig. 4A to Fig. 4D show schematic diagrams illustrating the flow of information in association with the method of Fig. 3, according to an embodiment of the invention.

[0094] In the example context as shown in Fig. 4A, a User Equipment UE (or user device) can, for example, be configured to provide a user device (or UE) capability status such as split learning capability information and / or communicate one or more input signals which can correspond to or be associated with or include the user device capability status to a gNB (or base station), in accordance with an embodiment of the invention. In Fig. 4B, the gNB (or base station) can, for example, be configured to receive the UE (or user device) capability information and performs a model split based on the UE capability information, in accordance with an embodiment of the invention.

[0095] In the example context as shown in Fig. 4C, a split model framework is illustrated. In this example, the gNB (or base station), upon receiving the indication that the UE (or user device) is capable of model split, performs model split and provides the split model to one or more UEs (or user devices) based on their capability. The model split can include one or more sub-models which can be based on the number of neural layers trained at the UE (or user device). For example, a UE having a higher capability may have a greater number of layers of the split model for the UE side. The gNB (or base station) may signal the split model via Radio Resource Control (RRC) Reconfiguration message, a UE specific message or a signal configured for a model. In a specific implementation, the model split with one or more sub-models at the gNB (or base station) may include a single model or may be further split into smaller sub-models based on the network load and thecomputation the gNB (or base station) wants to use, in accordance with an embodiment of the invention.

[0096] Moreover, the gNB (or base station) may signal or share information to one or more UEs (or user devices) for split model. The information may include sub-models such as a Long Short-Term Memory (LSTM) model having two layers and a connected layer, a Recurrent Neural Network (RNN) having one layer and a normalization layer and / or a K-Nearest Neighbors (KNN) model having four layers and a dropout layer. The one or more UEs (or user devices) may perform model training for the sub-model based on the received or shared information.

[0097] In an embodiment, the framework may include a Life Cycle Management (LCM) process, which may not require having a data collection process. Instead, the UE (or user device) may perform a local model training based on its own dataset after receiving the model split. The local model training may generate one or more datasets and the UE may then provide the one or more datasets including activations having corresponding labels, for example smashed data including computed weights, to the global model at the gNB (or base station). The gNB (or base station) may use the activations (or smashed data) to train or update its own model (global model) to generate computed gradients. The gNB (or base station) subsequently may provide the computed gradients in the backward direction to the UE (or user device) for the UE to update its model and (local model) and converge the AIML model.

[0098] Fig. 4D shows another example embodiment of model training in association with the method of Fig. 3. In the example context of Fig. 4D, multiple UEs (or user devices) may be capable for split learning. In this embodiment, the gNB (or base station) may indicate group signaling for some UEs (or user devices) with a particular model and another group of UEs (or user devices) with another model. Specifically, the gNB (or base station) may configure the one or more UEs (or user devices) into groups based on UE (or user device) capabilities, such that the gNB (or base station) may create groups where the one or more UEs (or user devices) have the samecomputational capability. For example, the one or more UEs (or user devices) may be grouped based on at least one of: mobility, latency, hardware capabilities (or hardware specification) and / or battery capabilities. Such an example may also be implementation specific. In this way, multiple models can be trained and inference may be obtained simultaneously, in accordance with an embodiment of the invention

[0099] In the foregoing manner, various embodiments of the disclosure are described for addressing at least one of the foregoing disadvantages. Such embodiments are intended to be encompassed by the following claims and are not to be limited to specific forms or arrangements of parts so described and it will be apparent to one skilled in the art in view of this disclosure that numerous changes and / or modification can be made, which are also intended to be encompassed by the following claims.

Claims

Claim(s)1 . A method (300) for model training in a network, the method comprising: obtaining a user device capability status; generating at least one signal for a local model training based on the user device capability status; and training the local model based on the at least one signal.

2. The method (300) according to claim 1 , wherein obtaining the user device capability status comprises receiving data associated with split learning capability of the user device.

3. The method (300) according to claim 1 , wherein generating the at least one signal for a local model training comprises: generating an indication for model split training; and determining one or more sub-models of the local model for model split training.

4. The method (300) according to claim 3, further comprising transmitting the one or more sub-models and the indication for model split training.

5. The method (300) according to claim 3, wherein the one or more sub-model comprises at least one of: Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), and / or K-Nearest Neighbors (KNN).

6. The method (300) according to claim 4, wherein transmitting the one or more sub-models and the indication for model split training comprises transmitting via at least one of: User Equipment (UE) specific message, Radio Resource Control (RRC) reconfiguration and / or a signal configured for a model.

7. The method (300) according to claim 1 , further comprising: generating one or more datasets from the local model training; and transmitting the one or more datasets from the local model training to a global model.

8. The method (300) according to claim 7, wherein the one or more datasets comprises activations having corresponding labels.

9. The method (300) according to claim 7, further comprising: receiving the one or more datasets; updating the global model based on the one or more datasets; generating computed gradients based on the updated global model; and transmitting the computed gradients to the local model to update the local model based on the computed gradients.

10. The method (300) according to claim 1 , further comprising configuring one or more user devices into groups based on a computational capability associated with each of the one or more user devices.11 . The method (300) according to claim 10, wherein the computational capability comprises at least one of: mobility, latency, hardware capabilities and / or battery capabilities.

12. The method (300) according to claim 7, wherein the local model and the global model is an Artificial Intelligence / Machine Learning (AIML) model.

13. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method (300) according to any of the preceding claims.

14. A computer readable storage medium having data stored therein representing software executable by a computer, the software including instructions, when executed by the computer, to carry out the method (300) according to any one of claims 1-12.

15. A device (104) for model training in a network comprising: a first module (202) configured to receive at least one signal associated with a user device capability status;a second module (204) configured to at least one of process and facilitate the method (300) of claim 1 to claim 12 to generate at least one output signal; and a third module (206) configured to communicate at least one output signal, wherein the output signal corresponds to a control signal for a local model training based on the user device capability status.

16. The device (104) according to claim 15, wherein the device (104) corresponds to a base station communicable with an apparatus (102) corresponding to a User Equipment (UE), and wherein the base station corresponds to a Next generation Node B (gNB) configured to receive the at least one input signal from the UE.

17. A system (100) comprising: at least one device (104) according to any of claims 15 and 16; and at least one apparatus (102) according to claim 16, wherein the apparatus (102) and the device (104) are capable of being coupled via at least one of wired coupling and wireless coupling.

Images