Task-specific model for wireless networks

Abstract

According to an example embodiment, a method may include receiving, by a user device, a request for a subtask-specific model from a network node, the request including configuration parameters for the subtask-specific model, where the subtask-specific model is to perform or assist in performing machine learning enabled subtasks; validating the request for the sub-task specific model; modifying, by the user device, the subtask-specific model based on the trained generic model, and configuration parameters of the subtask or subtask-specific model; executing or completing, by the user device, the machine learning enabled subtask based on or using the modified subtask specific model; and sending, by the user equipment, a subtask-specific model output to the network node based on executing or completing the machine learning enabled subtask in accordance with or using the modified subtask-specific model.

Description

Technical Field

[0001] This specification relates to wireless communications. Background Art

[0002] A communication system may be a facility that enables communication between two or more nodes or devices, such as fixed or mobile communication devices. Signals may be transmitted over wired or wireless carriers.

[0003] An example of a cellular communication system is an architecture standardized by the Third Generation Partnership Project (3GPP). The latest development in this field is generally referred to as the Long Term Evolution (LTE) of the Universal Mobile Telecommunications System (UMTS) radio access technology. E-UTRA (Evolved UMTS Terrestrial Radio Access) is the air interface of the 3GPP Long Term Evolution (LTE) upgrade path for mobile networks. In LTE, a base station or access point (AP) (which is called an enhanced node AP (eNB)) provides wireless access within a coverage area or cell. In LTE, a mobile device or mobile station is referred to as a user equipment (UE). LTE has included many improvements or developments. Various aspects of LTE are also constantly improving.

[0004] 5G New Radio (NR) development is part of the ongoing mobile broadband evolution process to meet 5G requirements, similar to the earlier evolution of 3G and 4G wireless networks. In addition, 5G targets emerging use cases beyond mobile broadband. One goal of 5G is to significantly improve wireless performance, which can include new levels of data rates, latency, reliability, and security. 5G NR can also be extended to efficiently connect massive Internet of Things (IoT) and can provide new types of mission-critical services. For example, ultra-reliable and low-latency communication (URLLC) equipment may require high reliability and very low latency. Summary of the invention

[0005] According to an example embodiment, a method may include: receiving, by a first model training unit, from a second model training unit, a trigger indication that triggers or causes modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; receiving, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors; modifying, by the first model training unit, the subtask-specific model based on the general model, one or more parameters of the subtask or subtask-specific model and the training data received from one or more subtask-specific model collectors; and sending, by the first model training unit, the modified subtask-specific model to a first wireless node.

[0006] According to an example embodiment, a device may include: at least one processor; and at least one memory including computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: receive, by a first model training unit, from a second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; receive, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors; modify the subtask-specific model by the first model training unit based on the general model, one or more parameters of the subtask or subtask-specific model and the training data received from one or more subtask-specific model collectors; and send the modified subtask-specific model to a first wireless node by the first model training unit.

[0007] According to an example embodiment, a device may include: a component for receiving, by a first model training unit, from a second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; a component for receiving, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors; a component for modifying the subtask-specific model by the first model training unit based on the general model, one or more parameters of the subtask or subtask-specific model and the training data received from one or more subtask-specific model collectors; and a component for sending, by the first model training unit, the modified subtask-specific model to a first wireless node.

[0008] According to an example embodiment, a non-transitory computer-readable storage medium may include instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: receive, by a first model training unit, from a second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; receive, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors; modify, by the first model training unit, the subtask-specific model based on the general model, one or more parameters of the subtask or subtask-specific model and the training data received from one or more subtask-specific model collectors; and send, by the first model training unit, the modified subtask-specific model to a first wireless node.

[0009] According to an example embodiment, a method may include: receiving a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determining a need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; sending, by a second model training unit, to a first model training unit a trigger indication that triggers or causes the first model training unit to modify the subtask-specific model, an indication of a general model trained by the second model training unit for the general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; configuring, by the second model training unit, one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and receiving, by the second model training unit, from the first model training unit, the modified subtask-specific model modified by the first model training unit.

[0010] A device may include: at least one processor; and at least one memory, including computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: receive a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determine the need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; a second model training unit sends a trigger indication to trigger or cause the first model training unit to modify the subtask-specific model, an indication of the general model trained by the second model training unit for the general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model to the first model training unit; the second model training unit configures one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and the second model training unit receives from the first model training unit the modified subtask-specific model modified by the first model training unit.

[0011] According to an example embodiment, an apparatus may include: a component for receiving a request to modify a subtask-specific model for a first wireless node based on a general model or otherwise determining a need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; a component for sending, by a second model training unit, to a first model training unit a trigger indication that triggers or causes the first model training unit to modify the subtask-specific model, an indication of a general model trained by the second model training unit for the general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; a component for configuring, by the second model training unit, one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and a component for receiving, by the second model training unit, from the first model training unit, a modified subtask-specific model modified by the first model training unit.

[0012] According to an example embodiment, a non-transitory computer-readable storage medium may include instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: receive a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determine a need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; send, by a second model training unit, to a first model training unit a trigger indication that triggers or causes the first model training unit to modify the subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; configure, by the second model training unit, one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and receive, by the second model training unit, from the first model training unit the modified subtask-specific model modified by the first model training unit.

[0013] According to an example embodiment, a method may include: determining, by a user device, a trained general model to perform or assist in performing a general task that enables machine learning; determining, based on the trained general model, one or more general model-based outputs according to one or more signals or inputs; sending, by the user device, the one or more general model-based outputs to a network node; receiving, by the user device, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, the request including configuration parameters of the subtask-specific model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; verifying the request for the subtask-specific model; modifying, by the user device, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or subtask-specific model; and performing or completing a subtask that enables machine learning based on or using the modified subtask-specific model by the user device; and sending the subtask-specific model outputs to the network node by the user device based on performing or completing the subtask that enables machine learning according to or using the modified subtask-specific model.

[0014] A device may include: at least one processor; and at least one memory, including computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: determine a trained general model by a user device to perform or assist in performing a general task that enables machine learning; determine one or more general model-based outputs based on the trained general model according to one or more signals or inputs; send the one or more general model-based outputs by the user device to a network node; receive a request for a subtask-specific model from the network node by the user device at least in part based on the one or more general model-based outputs, the request including configuration parameters of the subtask-specific model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; verify the request for the subtask-specific model; modify the subtask-specific model by the user device based on the trained general model and the configuration parameters of the subtask or subtask-specific model; and perform or complete the subtask that enables machine learning based on or using the modified subtask-specific model by the user device; and send the subtask-specific model output to the network node by the user device based on performing or completing the subtask that enables machine learning according to or using the trained subtask-specific model.

[0015] According to an example embodiment, an apparatus may include: a component for determining, by a user device, a trained general model to perform or assist in performing a general task that enables machine learning; a component for determining, based on the trained general model, one or more general model-based outputs according to one or more signals or inputs; a component for sending, by the user device, the one or more general model-based outputs to a network node; a component for receiving, by the user device, a request for a subtask-specific model from a network node based at least in part on the one or more general model-based outputs, including configuration parameters of the subtask-specific model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; a component for verifying the request for the subtask-specific model; a component for modifying, by the user device, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or subtask-specific model; and a component for performing or completing a subtask that enables machine learning by the user device based on or using the modified subtask-specific model; and a component for sending the subtask-specific model outputs to the network node by the user device based on performing or completing a subtask that enables machine learning according to or using the trained subtask-specific model.

[0016] According to an example embodiment, a non-transitory computer-readable storage medium may include instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: determine, by a user device, a trained general model to perform or assist in performing a general task that enables machine learning; determine, based on the trained general model, one or more general model-based outputs according to one or more signals or inputs; send, by the user device, the one or more general model-based outputs to a network node; receive, by the user device, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, the request including configuration parameters of the subtask-specific model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; verify the request for the subtask-specific model; modify, by the user device, the subtask-specific model based on the trained general model, and the configuration parameters of the subtask or subtask-specific model; and perform or complete a subtask that enables machine learning based on or using the modified subtask-specific model by the user device; and send the subtask-specific model output to the network node by the user device based on performing or completing the subtask that enables machine learning according to or using the trained subtask-specific model.

[0017] According to an example embodiment, a method may include: determining, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; providing, by the network node, the trained general model to a user device; receiving, by the network node, a request for a subtask-specific model from the user device; verifying the request for the subtask-specific model; sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device; receiving, by the network node, at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device; modifying, by the network node, the subtask-specific model based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and sending, by the network node, the modified subtask-specific model to the user device.

[0018] A device may include: at least one processor; and at least one memory, including computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: determine a trained general model by a network node to perform or assist in performing a general task that enables machine learning; provide the trained general model to a user device by the network node; receive a request for a subtask-specific model from the user device by the network node; verify the request for the subtask-specific model; send a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device by the network node; receive at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device by the network node; modify the subtask-specific model by the network node based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and send the modified subtask-specific model to the user device by the network node.

[0019] According to an example embodiment, an apparatus may include: a component for determining, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; a component for providing, by the network node, a trained general model to a user device; a component for receiving, by the network node, a request for a subtask-specific model from a user device; a component for verifying the request for the subtask-specific model; a component for sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to a user device; a component for receiving, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from a user device; a component for modifying, by the network node, a subtask-specific model based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and a component for sending, by the network node, a modified subtask-specific model to a user device.

[0020] According to an example embodiment, a non-transitory computer-readable storage medium may include instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: determine, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; provide, by the network node, the trained general model to a user device; receive, by the network node, a request for a subtask-specific model from the user device; verify the request for the subtask-specific model; send, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device; receive, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device; modify, by the network node, the subtask-specific model based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and send, by the network node, the modified subtask-specific model to the user device.

[0021] According to an example embodiment, a method may include: determining, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; providing, by the network node, the trained general model to a user device; receiving, by the network node, a request for a subtask-specific model from the user device; verifying the request for the subtask-specific model; sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device; receiving, by the network node, at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device; modifying, by the network node, the subtask-specific model based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and sending, by the network node, the modified subtask-specific model to the user device.

[0022] A device may include: at least one processor; and at least one memory, including computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: determine a trained general model by a network node to perform or assist in performing a general task that enables machine learning; provide the trained general model to a user device by the network node; receive a request for a subtask-specific model from the user device by the network node; verify the request for the subtask-specific model; send a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device by the network node; receive at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device by the network node; modify the subtask-specific model by the network node based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and send the modified subtask-specific model to the user device by the network node.

[0023] According to an example embodiment, an apparatus may include: a component for determining, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; a component for providing, by the network node, a trained general model to a user device; a component for receiving, by the network node, a request for a subtask-specific model from a user device; a component for verifying the request for the subtask-specific model; a component for sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to a user device; a component for receiving, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from a user device; a component for modifying, by the network node, a subtask-specific model based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and a component for sending, by the network node, a modified subtask-specific model to a user device.

[0024] According to an example embodiment, a non-transitory computer-readable storage medium may include instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: determine, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; provide, by the network node, the trained general model to a user device; receive, by the network node, a request for a subtask-specific model from the user device; verify the request for the subtask-specific model; send, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device; receive, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device; modify, by the network node, the subtask-specific model based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and send, by the network node, the modified subtask-specific model to the user device.

[0025] The details of one or more examples of embodiments are set forth in the accompanying drawings and the description below. Other features will be apparent from the description and drawings, and from the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a block diagram of a wireless network according to an example embodiment.

[0027] Figure 2 is a flowchart illustrating the operation of a model training unit according to an example embodiment.

[0028] Figure 3 is a flowchart illustrating the operation of the second model training unit according to example embodiments.

[0029] Figure 4 is a flow chart illustrating the operation of a user device according to an example embodiment.

[0030] Figure 5 is a flow chart illustrating the operation of a network node (e.g., a gNB) according to an example embodiment.

[0031] Figure 6 is a diagram illustrating operations of a general model training unit (GMTU) and a meta-learning model training unit (MMTU) or a specific model training unit according to example embodiments.

[0032] Figure 7 is a diagram of a network in which general model (GM) training and subtask specific model (SSM) training are performed at network nodes according to an example embodiment.

[0033] Figure 8is a diagram of a network in which general model (GM) training and subtask specific model (SSM) training are performed at a UE or user equipment.

[0034] Fig. 9 is a block diagram of a wireless station or wireless node (eg, a network node, a user node or UE, a relay node, or other node). DETAILED DESCRIPTION

[0035] Figure 1 is a block diagram of a wireless network 130 according to an example embodiment. Figure 1 In a wireless network 130, user devices 131, 132, 133, and 135 (which may also be referred to as mobile stations (MS) or user equipment (UE)) may be connected (and communicate) with a base station (BS) 134, which may also be referred to as an access point (AP), an enhanced Node B (eNB), a gNB, or a network node. The terms user device and user equipment (UE) may be used interchangeably. The BS (or network node) may also include or may be referred to as a RAN (radio access network) node, and may include a portion of a BS or a portion of a RAN node, such as (for example, a centralized unit (CU) and / or a distributed unit (DU) in the case of a split BS or a split gNB). At least a portion of the functionality of a BS (e.g., an access point (AP), a base station (BS), or an (e) Node B (eNB), a gNB, a RAN node) may also be performed by any node, server, or host that may be operably coupled to a transceiver (such as a remote radio head). BS (or AP) 134 provides wireless coverage within cell 136, including to user equipment (or UE) 131, 132, 133 and 135. Although only four user equipment (or UE) are shown as connected or attached to BS 134, any number of user equipment may be provided. BS 134 is also connected to core network 150 via S1 interface 151. This is just a simple example of a wireless network, and other wireless networks may be used. Wireless nodes may include, for example, BSs, gNBs, eNBs, APs, RAN nodes, CUs and / or DUs (or other network nodes), relay nodes, user equipment, UEs, or other nodes with wireless communication capabilities, etc.

[0036] A base station (e.g., such as BS 134) is an example of a radio access network (RAN) node within a wireless network. A BS (or RAN node) may be or may include (or may alternatively be referred to as) an access point (AP), a gNB, an eNB or a portion thereof (such as a centralized unit (CU) and / or a distributed unit (DU) in the case of a split BS or split gNB), or other network node.

[0037] According to an illustrative example, a BS node or other network node (e.g., BS, eNB, gNB, CU / DU, transmission reception point (TRP), etc.) or a radio access network (RAN) may be part of a mobile telecommunication system. The RAN (radio access network) may include one or more BSs or RAN nodes that implement a radio access technology, for example, to allow one or more UEs to access a network or a core network. Thus, for example, a RAN (RAN node, such as a BS or gNB) may reside between one or more user equipment or UEs and a core network. According to an example embodiment, each RAN node (e.g., BS, eNB, gNB, CU / DU, etc.) or BS may provide one or more wireless communication services for one or more UEs or user equipment, for example, to allow a UE to access a network wirelessly via a RAN node. Each RAN node or BS may perform or provide wireless communication services, for example, such as to allow a UE or user equipment to establish a wireless connection to a RAN node, and to send data to one or more UEs and / or receive data from one or more UEs. For example, after establishing a connection to a UE, a RAN node or network node (e.g., BS, eNB, gNB, CU / DU, etc.) may forward data received from the network or core network to the UE, and / or forward data received from the UE to the network or core network. A RAN node or network node (e.g., BS, eNB, gNB, CU / DU, etc.) may perform a variety of other wireless functions or services, such as, for example, broadcasting control information (e.g., such as system information or on-demand system information) to the UE, paging the UE when there is data to be delivered to the UE, assisting the UE in switching between cells, scheduling resources for uplink data transmission from (multiple) UEs and downlink data transmission to (multiple) UEs, issuing control information for configuring one or more UEs, etc. These are several examples of one or more functions that a RAN node or BS may perform.

[0038] A user device or user node (user terminal, user equipment (UE), mobile terminal, handheld wireless device, etc.) may refer to a portable computing device, including a wireless mobile communication device that operates with or without a subscriber identity module (SIM), for example, including but not limited to the following types of devices: a mobile station (MS), a mobile phone, a cell phone, a smart phone, a personal digital assistant (PDA), a handset, a device using a wireless modem (alarm or measurement device, etc.), a laptop and / or touch screen computer, a tablet, a tablet phone, a game console, a notebook computer, a vehicle, a sensor, a multimedia device, or any other wireless device. It should be understood that a user device may also be (or may include) an almost exclusive uplink-only device, an example of which is a camera or video camera that loads images or video clips to a network. In addition, a user node may include a user equipment (UE), a user device, a user terminal, a mobile terminal, a mobile station, a mobile node, a subscriber device, a subscriber node, a subscriber terminal, or other user nodes. For example, a user node may be used to communicate wirelessly with one or more network nodes (e.g., gNB, eNB, BS, AP, CU, DU, CU / DU) and / or with one or more other user nodes, regardless of the technology or radio access technology (RAT). In LTE (as an illustrative example), the core network 150 may be referred to as an evolved packet core (EPC), which may include a mobility management entity (MME), one or more gateways, and other control functions or blocks. The MME may handle or assist in the movement / handover of user equipment between BSs, and the gateway may forward data and control signals between the BS and a packet data network or the Internet. Other types of wireless networks, such as 5G (which may be referred to as new radio (NR)), may also include a core network.

[0039] In addition, the techniques described herein may be applied to various types of user equipment or data service types, or may be applied to user equipment on which multiple applications may be running, which may be different data service types. New radio (5G) developments may support many different applications or many different data service types, such as: machine type communications (MTC), enhanced machine type communications (eMTC), Internet of Things (IoT) and / or narrowband IoT user equipment, enhanced mobile broadband (eMBB), and ultra-reliable low latency communications (URLLC). Many of these new 5G (NR) related applications may require higher performance than previous wireless networks.

[0040] IoT can refer to the growing group of objects that can have Internet or network connectivity so that they can send and receive information to other network devices. For example, many sensor-type applications or devices can monitor physical conditions or states and can report to servers or other network devices, for example, when events occur. For example, machine-type communications (MTC or machine-to-machine communications) can be characterized by fully automatic data generation, exchange, processing, and actuation between intelligent machines, with or without human intervention. Enhanced mobile broadband (eMBB) can support higher data rates than currently available in LTE.

[0041] Ultra-Reliable Low Latency Communication (URLLC) is a new data service type or new usage scenario that can be supported for the new radio (5G) system. This enables emerging new applications and services such as industrial automation, autonomous driving, vehicle safety, e-health services, etc. As an illustrative example, 3GPP aims to provide a 5G-capable network with the same performance as 10G. -5 Connectivity with reliability corresponding to a block error rate (BLER) of 100% and a U-plane (user / data plane) latency of up to 1 ms. Thus, for example, a URLLC user equipment / UE may require a significantly lower block error rate and low latency than other types of user equipment / UE (with or without a requirement for simultaneous high reliability). Thus, for example, a URLLC UE (or a URLLC application on a UE) may require shorter latency than an eMBB UE (or an eMBB application running on a UE).

[0042] The techniques described herein may be applied to a variety of wireless technologies or wireless networks, such as LTE, LTE-A, 5G (New Radio (NR)), cmWave and / or mmWave band networks, 6G, IoT, MTC, eMTC, eMBB, URLLC, etc., or any other wireless network or wireless technology. These exemplary networks, technologies, or data service types are provided only as illustrative examples.

[0043] According to example embodiments, a machine learning (ML) model may be used within a wireless network to perform (or assist in performing) one or more tasks. Typically, one or more nodes within a wireless network (e.g., a BS, gNB, eNB, RAN node, user node, UE, user equipment, relay node, or other wireless node) may use or employ an ML model, such as, for example, a neural network model (e.g., which may be referred to as a neural network, an artificial intelligence (AI) neural network, an AI neural network model, an AI model, a machine learning (ML) model, or an algorithm, model, or other terminology) to perform or assist in performing one or more ML-enabled tasks. Other types of models may also be used. ML-enabled tasks may include tasks that may be performed (or assisted in performing) by an ML model, or tasks that an ML model has been trained to perform or assist in performing.

[0044] ML-based algorithms or ML models may be used to perform and / or assist in performing various wireless and / or radio resource management (RRM) functions or tasks to improve network performance, such as, for example, in a UE for antenna panel or beam steering, RRM measurements and feedback (channel state information (CSI) feedback), link monitoring, transmit power control (TPC), etc. In some cases, the use of ML models may be used to improve the performance of a wireless network in one or more aspects, or as measured by one or more performance indicators or performance criteria.

[0045] For example, a model (e.g., a neural network or ML model) can be or can include a computational model used in machine learning composed of hierarchically organized nodes. Nodes are also called artificial neurons, or simply neurons, and perform functions on the inputs provided to produce a certain output value. Neural networks or ML models typically require a training period to learn the parameters, i.e., weights, for mapping inputs to desired outputs. The mapping is performed via a function. Therefore, the weights are the weights of the mapping function of the neural network. Each neural network model or ML model can be trained for a specific task.

[0046] In order to provide an output given an input, a neural network model or ML model should be trained, which may involve learning appropriate values ​​for a large number of parameters (e.g., weights) of a mapping function. These parameters are also commonly referred to as weights because they are used to weight the terms in the mapping function. Such training may be an iterative process in which the values ​​of the weights are adjusted in multiple rounds (e.g., thousands of rounds) of training until the best or most accurate value (or weight) is reached. In the context of a neural network (neural network model) or ML model, the parameters may be typically initialized using random values, and a training optimizer iteratively updates the parameters (weights) of the neural network to minimize the error of the mapping function. In other words, in each round or step of iterative training, the network updates the parameter values ​​so that the parameter values ​​eventually converge to the optimal values.

[0047] For example, a neural network model or ML model can be trained in a supervised or unsupervised manner. In supervised learning, training examples are provided to a neural network model or other machine learning algorithm. The training examples include inputs and expected or previously observed outputs. The training examples are also called labeled data because the inputs are labeled with the expected or observed outputs. In the case of a neural network, the network learns the values ​​of the weights used in the mapping function that most often result in the expected output given the training inputs. In unsupervised training, the neural network model learns to identify structure or patterns in the inputs provided. In other words, the model identifies implicit relationships in the data. Unsupervised learning is used for many machine learning problems and typically requires large amounts of unlabeled data.

[0048] According to example embodiments, the learning or training of a neural network model or ML model can be divided into (or can include) two major categories (supervised and unsupervised), depending on whether the model has a learning "signal" or "feedback" available. Thus, for example, in the field of machine learning, there can be two main types of learning or training of models: supervised and unsupervised. The main difference between these two types is that supervised learning is performed using known or prior knowledge, i.e., what the output values ​​of certain data samples should be. Thus, the goal of supervised learning can be to learn a function that, given a data sample and a desired output, most closely approximates the relationship between the input and output observable in the data. On the other hand, unsupervised learning has no labeled outputs, so its goal is to infer the natural structure that exists in a set of data points.

[0049] Supervised learning: A computer is presented with example inputs and their expected outputs, and the goal may be to learn general rules for mapping inputs to outputs. For example, supervised learning may be performed in the context of classification, where a computer or learning algorithm attempts to map inputs to output labels, or in the context of regression, where a computer or algorithm may map (multiple) inputs to (multiple) continuous outputs. Common algorithms in supervised learning may include, for example, logistic regression, naive Bayes, support vector machines, artificial neural networks, and random forests. In both regression and classification, the goal may include finding specific relationships or structures in the input data that allow correct output data to be efficiently produced. As a special case, the input signal is only partially available, or is limited to special feedback: Semi-supervised learning: The computer is only given an incomplete training signal: a training set that lacks some (usually many) target outputs. Active learning: The computer can only obtain training labels for a limited set of instances (based on a budget), or it can optimize its selection of objects for which it obtains labels. When used interactively, these may be presented to the user for labeling. Reinforcement learning: Training data (in the form of rewards and penalties) is given only as feedback on the behavior of the program in a dynamic environment, such as using real-time data.

[0050] Unsupervised learning: No labels are given to the learning algorithm, leaving it to find structure in the input. Some example tasks in unsupervised learning can include clustering, representation learning, and density estimation. In these cases, the computer or learning algorithm attempts to learn the inherent structure of the data without the use of explicitly provided labels. Some common algorithms include k-means clustering, principal component analysis, and autoencoders. Since no labels are provided, there may be no specific way to compare model performance in most unsupervised learning methods.

[0051] It may be advantageous to provide techniques that enable the use of ML models, coordinate the use of ML models, allow configuration of ML models, allow modification or training of ML models, and / or facilitate the distribution or communication of ML models or ML assistance functions over wireless networks (e.g., between gNB or RAN nodes and UEs) for various wireless network related tasks. Some example tasks that ML models may be used for may include, for example, channel state information (CSI) feedback enhancement, e.g., reducing overhead, improving accuracy, prediction; beam management, e.g., beam prediction in the time domain and / or spatial domain to reduce overhead and latency, improve beam selection accuracy; and / or UE positioning accuracy enhancement for different scenarios, including, for example, scenarios with severe NLOS (non-line-of-sight) conditions. These are just a few examples, and ML models may be applied to or used to perform or assist in performing a variety of tasks within wireless networks.

[0052] In general, meta-learning may include or may refer to modifying, training, or customizing a general model (e.g., which may have been trained using features or data extracted from heterogeneous sources or different UEs or different wireless nodes) for a specific type of entity and / or a specific task. For example, meta-learning may include the process of setting the knobs (or adjustable parameters) of the learning process and / or the process of modifying the weights or training of a model for a specific task. In addition, meta-learning may include the case where an ML model is trained for a general (or more general) task, and then (partially, possibly using transfer learning) the trained ML model (which has been trained for the general task) is trained or retrained for a specific subtask. In addition, in some cases, meta-learning may include the process in which the machine learning algorithm itself proposes its own task assignment and / or in which the knobs of the learning process are set (e.g., weights are adjusted or other modifications or training are performed) via optimization. An algorithm that automatically performs such optimization and / or training of an ML model may be referred to as a meta-learning algorithm. A meta-learning algorithm may take data and / or algorithms into account in many tasks. In wireless communications, meta-learning (e.g., which may include modifying or training ML models for specific tasks) may be used to develop, modify, and / or train ML models for various use cases or applications, such as radio resource management and / or receiver design (as illustrative examples).

[0053] Various example embodiments are described, for example, in which a model (e.g., an ML model) is trained for a first task (e.g., a general task), transmitted or sent from a first wireless node to a second wireless node, and then the ML model may be modified, trained, or customized by the second wireless node for a second task (e.g., a subtask, which may be a task different from the general task but may be related to the general task in some aspects, such as, for example, a subtask may belong to the same category as a subtask, for example, both the general task and the subtask may be related to positioning, or both the general task and the subtask may be related to CSI-RS measurement).

[0054] Figure 2 2 is a flowchart illustrating the operation of a model training unit according to an example embodiment. Operation 210 includes receiving, by a first model training unit (e.g., a meta-learning model training unit or a specific model training unit) from a second model training unit (e.g., a general model training unit), a trigger indication to trigger or cause modification (e.g., training) of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify (e.g., train) the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning. Operation 220 includes receiving, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors (e.g., UEs, gNBs, or other wireless nodes). Operation 230 includes modifying (e.g., training or retraining, adjusting weights, configuring or updating, or other modifications to the subtask-specific model) the subtask-specific model based on the general model, one or more parameters of the subtask or subtask-specific model, and the training data received from the one or more subtask-specific model collectors by the first model training unit. And, operation 240 includes sending, by the first model training unit, the modified (e.g., trained) sub-task-specific model to a first wireless node (e.g., UE or gNB).

[0055] exist Figure 2 In an example embodiment of the method, the modification may include at least one of the following: modifying one or more weights of the subtask-specific model; training the subtask-specific model; retraining the subtask-specific model; configuring or updating one or more weights or parameters of the subtask-specific model; and / or upgrading (e.g., increasing complexity, or increasing the number of inputs and / or outputs, or other upgrades) or downgrading (e.g., reducing complexity, reducing the number of inputs and / or outputs, or other downgrading) the subtask-specific model.

[0056] according to Figure 2In an example embodiment of the present invention, receiving a trigger indication may include: receiving, by the first model training unit from the second model training unit, one or more of the following: information about a general model, including one or more of the following: an architecture of the general model, weights of the general model, a loss function and / or an activation function of the general model, and / or an output type of the general model; subtask parameterization, including constraints of subtask-specific models, or subtask-specific cost functions of the subtask-specific models; and / or identifiers of one or more subtask-specific model collectors.

[0057] according to Figure 2 In an example embodiment, the modification of the subtask-specific model by the first model training unit may include performing one or more of the following based on one or more constraints of the general model and the subtask-specific model: pruning or reducing the size of the general model so that the subtask-specific model will adapt to the maximum allowed subtask-specific model constraints; deactivating one or more inputs of the general model so that the input of the subtask-specific model depth or size will adapt to the format, size, or depth of the training data received from one or more subtask-specific model collectors; replacing the general model activation function with an activation function specific to the subtask-specific model; or defining a cost function specific to the subtask-specific model.

[0058] according to Figure 2 In an example embodiment of the present invention, the method may further include sending, by the first model training unit, the modified sub-task specific model to the second model training unit.

[0059] according to Figure 2 In an example embodiment of the present invention, the method may further include sending, by the first model training unit, the modified subtask specific model to at least one subtask specific model collector of the one or more subtask specific model collectors.

[0060] according to Figure 2 An example embodiment, wherein any one of the following: the first model training unit and the second model training unit are set in the second wireless node; or the first model training unit is set in the second wireless node, and the second model training unit is set in the third wireless node.

[0061] according to Figure 2 In an example embodiment of the present invention, the method may further include sending, by the first model training unit to the second model training unit, a request for sub-task specific model training or meta-learning of the sub-task.

[0062] according to Figure 2An example embodiment: the second model training unit may include a general model training unit configured to modify a general model for a general task; and the first model training unit may include a meta-learning model training unit or a specific model training unit configured to modify (e.g., train) a specific model or a subtask-specific model for a subtask.

[0063] according to Figure 2 In an example embodiment, one or more of the first wireless node, the second wireless node, or the third wireless node may include at least one of the following: a user equipment, a user device, a base station, or a gNB.

[0064] Figure 3 310 includes receiving a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determining a need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning. Operation 320 includes sending, by the second model training unit, to the first model training unit a trigger indication that triggers or causes the first model training unit to modify the subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model. Operation 330 includes configuring, by the second model training unit, one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit. And, operation 340 includes receiving, by the second model training unit, from the first model training unit, a modified subtask-specific model modified by the first model training unit.

[0065] according to Figure 3 In an example embodiment of the method, the sending may include sending one or more of the following by the second model training unit to the first model training unit: information of the general model, including one or more of the following: the architecture of the general model, the weights of the general model, the loss function and / or activation function of the general model, and / or the output type of the general model; subtask parameterization, including constraints of subtask-specific models, or subtask-specific cost functions of subtask-specific models; and / or identifiers of one or more subtask-specific model collectors.

[0066] according to Figure 3 An example embodiment of the method, any one of the following: the first model training unit and the second model training unit are set in the second wireless node; or the first model training unit is set in the second wireless node, and the second model training unit is set in the third wireless node.

[0067] according to Figure 3In an example embodiment of the method, receiving a request for modification of a subtask-specific model or otherwise determining the need for it may include: receiving, by a second model training unit from a first model training unit, a request to modify or train the subtask-specific model for the subtask; and verifying the request to modify or train the subtask-specific model for the subtask.

[0068] according to Figure 3 In an example embodiment of the method, the second model training unit may include a general model training unit configured to modify or train a general model for a general task; and the first model training unit may include a meta-learning model training unit or a specific model training unit configured to modify or train a specific model or a subtask specific model for a subtask.

[0069] according to Figure 3 An example embodiment of the method, wherein one or more of the first wireless node, the second wireless node, or the third wireless node includes at least one of the following: user equipment, user equipment, base station, or gNB.

[0070] Figure 4 4 is a flow chart illustrating the operation of a user equipment according to an example embodiment. Operation 410 includes determining, by a user equipment (e.g., a UE or user device), a trained general model to perform or assist in performing a general task that enables machine learning. Operation 420 includes determining, based on the trained general model, one or more general model-based outputs based on one or more signals or inputs. Operation 430 includes sending, by the user equipment, the one or more general model-based outputs to a network node (e.g., a gNB). Operation 440 includes receiving, by the user equipment, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, the request including configuration parameters of the subtask-specific model, wherein the subtask-specific model is to perform or assist in performing a subtask that enables machine learning. Operation 450 includes validating the request for the subtask-specific model. Operation 460 includes modifying, by the user equipment, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or subtask-specific model. Operation 470 includes performing or completing, by the user equipment, the subtask that enables machine learning based on or using the modified subtask-specific model. And, operation 480 includes sending, by the user device, the subtask-specific model output to the network node based on performing or completing the machine learning-enabled subtask according to or using the modified subtask-specific model.

[0071] according to Figure 4In an example embodiment of the method, the modification may include at least one of the following: modifying one or more weights of the subtask specific model; training the subtask specific model; retraining the subtask specific model; configuring or updating one or more weights or parameters of the subtask specific model; and / or upgrading or downgrading the subtask specific model.

[0072] according to Figure 4 In an example embodiment of the method, verifying a request for a subtask-specific model may include: verifying at least one of the following for the subtask-specific model: the requested subtask-specific model is on a list of allowed subtask-specific models; a threshold amount of training data and / or input signals are available for training the subtask-specific model; and / or a threshold amount of processor resources and / or memory resources are available for training and / or using the subtask-specific model.

[0073] according to Figure 4 An example embodiment of the method receives a request for a subtask-specific model, the request including configuration parameters of the subtask-specific model, and may include receiving: a trigger indication for triggering or causing meta-learning or training of the subtask-specific model, and one or more parameters of the subtask or subtask-specific model to be used to train the subtask-specific model based on the general model, including receiving subtask parameterization, including constraints of the subtask-specific model, or a subtask-specific cost function of the subtask-specific model.

[0074] according to Figure 4 In an example embodiment of the method, configuration parameters of a subtask or sub-specific task model may include one or more constraints of the subtask-specific model, and wherein modification of the subtask-specific model by a user device based on the general model may include performing one or more of the following based on the one or more constraints of the general model and the subtask-specific model: pruning or reducing the size of the general model so that the subtask-specific model will adapt to the maximum allowed subtask-specific model depth or size; deactivating one or more inputs of the general model so that the input of the subtask-specific model depth or size will adapt to the format, size or depth of training data received from one or more subtask-specific model collectors; replacing the general model activation function with an activation function specific to the subtask-specific model; or defining a cost function specific to the subtask-specific model.

[0075] according to Figure 4 In an example embodiment of the method, determining one or more general model-based outputs may include: receiving, by a user device, a request from a network node to train a general model for a general task that enables machine learning; training the general model by the user device based on a configuration or input received from the network node; and using the trained general model to perform or complete the general task that enables machine learning to obtain one or more general model-based outputs.

[0076] according to Figure 4 In an example embodiment of the method, in response to the user equipment sending one or more general model-based outputs of the trained general model to the network node, a request for a sub-task specific model is received by the user equipment.

[0077] Figure 5 5 is a flow chart illustrating the operation of a network node (e.g., a gNB) according to an example embodiment. Operation 510 includes determining, by the network node, a trained general model to perform or assist in performing a general task that enables machine learning. Operation 520 includes providing, by the network node, the trained general model to a user equipment (e.g., a UE or user device). Operation 530 includes receiving, by the network node, a request for a subtask-specific model from the user equipment. Operation 540 includes validating the request for the subtask-specific model. Operation 550 includes sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user equipment. Operation 560 includes receiving, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from the user equipment. Operation 570 includes modifying, by the network node, the subtask-specific model based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data. And, operation 580 includes sending, by the network node, the modified subtask-specific model to the user equipment.

[0078] according to Figure 5 In an example embodiment of the method, the modification may include at least one of the following: modifying one or more weights of the subtask specific model; training the subtask specific model; retraining the subtask specific model; configuring or updating one or more weights or parameters of the subtask specific model; and / or upgrading or downgrading the subtask specific model.

[0079] according to Figure 5 In an example embodiment of the method, verifying a request for a subtask-specific model may include: verifying at least one of the following for the subtask-specific model: the requested subtask-specific model is on a list of allowed subtask-specific models; a threshold amount of training data and / or input signals are available for training the subtask-specific model; and / or a threshold amount of processor resources and / or memory resources are available for training and / or using the subtask-specific model.

[0080] according to Figure 5In an exemplary embodiment of the method, modification of the subtask-specific model by the network node may include performing one or more of the following based on one or more constraints of the general model and the subtask-specific model: pruning or reducing the size of the general model so that the subtask-specific model will adapt to the maximum allowed subtask-specific model depth or size; deactivating one or more inputs of the general model so that the input of the subtask-specific model depth or size will adapt to the format, size, or depth of training data received from one or more subtask-specific model collectors; replacing the general model activation function with an activation function specific to the subtask-specific model; or defining a cost function specific to the subtask-specific model.

[0081] According to example embodiments, a technique is provided that employs meta-learning (e.g., modification or training) of ML models for (multiple) ML-enabled functions or tasks, such as radio resource management (RRM)-related tasks for wireless networks. A framework for meta-learning (RRM) of ML-enabled functions is described. The framework may include and / or describe one or more operations and / or steps that may be used to port (i.e., generate and signal or send) a general ML model to another wireless node, and then use meta-learning (e.g., training and / or modification) of the general ML model (which was trained for the general task) for a sub-task-specific model that was trained to perform or assist in performing a specific sub-task. New training data may be received and used to modify or train the sub-task-specific model (e.g., develop the sub-task-specific model based on the general ML model and the new training data or sub-task-specific training data). Various example embodiments may provide or describe radio domain specific methods for: i) modifying and / or training (e.g., tuning / refining) a (e.g., general) ML model (which may have been trained to perform a general task or a first task) into (multiple) sub-task specific models (which may be trained to perform a second task or a specific sub-task, which may be different from the general task or the first task), and ii) related signaling between the network nodes involved.

[0082] The described framework, messages, or signaling may enable identification, communication, and implementation of decomposition of a generic task into subtasks, and / or modification and / or training of a generic model to create subtask-specific models that may be used to perform a specific subtask.

[0083] Example definition:

[0084] Definition 1: A subtask specific model (SSM) may include or may be defined as any of the following:

[0085] 1) A model that solves the same task as the generic model but is implemented by different types of NR (new radio) elements (e.g., wireless nodes) (compared to the NR elements used to train the generic model). In one embodiment, the training data for the generic model does not include training observations collected from specific element types (e.g., for specific types of wireless nodes). For example, the generic task may be UL (uplink) power control parameter optimization for UEs, while the subtask may be tuning UL power control to different UE mobility categories (or to specific UE mobility levels) (e.g., pedestrian UEs versus vehicle UEs, as identified by another algorithm). Thus, for example, a subtask may be the same or similar to the generic task, but it may be applied to different types of wireless nodes or a subtask-specific model may be applied to different types of wireless nodes (compared to a generic task or a generic ML model). Thus, for example, the generic task may generally be UL power control parameter optimization for UEs, while the specific subtask (for which the subtask-specific model is trained) is UL power control for a specific category (or subcategory) of UEs (e.g., pedestrian UEs, vehicle UEs, airborne UEs). In another example, the general task for UE energy saving may be performed by a NR UE, and the subtask may be UE energy saving performed in a reduced capability (RedCap) UE.

[0086] 2) A model that solves a task that is similar (but different) to the general model (thus the subtask is similar to the general task but different in at least one aspect). Thus, for example, a subtask may be different from the general (or general) task but may be related to the general task, for example, the subtask may be a task that belongs to the same category of tasks as the general (or general) task (e.g., UL power control), (for example, where for example, the general task may be for the general task and the subtask is for a different task that is in the same category of tasks as the general task (e.g., UL power control); and / or the subtask may be used for or applied to a subset (or a different set) of conditions or devices compared to the general task. For example, the general task may be optimal beam selection for a UE, while a similar (but different in at least one aspect) subtask may be optimal panel (or optimal antenna) selection. Furthermore, in some cases, the subtask-specific model may be implemented, for example, by another / specific type of NR element (or by a different type of NR element) than the NR element that has been used to train the general model.

[0087] Definition 2: A Generic Model Training Unit (GMTU) is a network (NW) node resident entity or a UE resident entity that trains a model to solve a generic task (e.g., a generic radio resource management (RRM) task). Therefore, a GMTU is an ML model training unit that trains a generic model to solve or perform a generic task.

[0088] Definition 3: Meta-learning model training unit (MMTU) is a NW (network) node resident entity or a UE resident entity that trains a model to solve a subtask. Therefore, the MMTU trains a subtask specific model; the MMTU and GMTU can be in different entities (e.g., UE or network node) or provided in the same entity (e.g., in the same network node or UE). They may require or use (or depend on) processing and data collection sources. The MMTU can be configured by the GMTU, and wherein the GMTU can transmit the GM (generic model) and SSM (subtask specific model) characteristics to the MMTU, and both the MMTU and the GMTU can reside in the same NR element, or in different NR elements. For example, the GMTU-MMTU interface can be Xn, F1 (when both are located on the NW side) or RRC / MAC (when one of the two is located in the NR UE).

[0089] Definition 4: In order to train subtask specific models, the MMTU may need assistance from other NR elements (residing on the UE side and / or the NW side), which are called SSM (subtask specific model) collectors, which collect and provide additional (or new or subtask specific) training data to refine the model for a specific subtask. The MMTU can interact with the SSM collector directly or indirectly via the GMTU. The interaction has the effect of refining the subtask model. For example: the MMTU can receive training data from an SSM collector (e.g., a UE or other wireless node), which is activated by the GMTU / MMTU to collect and provide training data. Then, the MMTU trains a new subtask specific model. The MMTU may first receive constraints on the SSM collector related to data acquisition and preprocessing (sampling resolution, periodicity of feature extraction and reporting, etc.), or model constraints related to the size of the trained SSM that the SSM collector can deploy (after the MMTU has trained it). Note that the MMTU and SSM collector can reside in the same NR element, for example, the NR UE can be responsible for (or can perform) both collecting training data and training SSM (subtask specific model). When the MMTU and SSM collector are not collocated, their collaboration can be achieved through Xn / F1 (for example, when both are NW elements or different wireless nodes), or if the MMTU is collocated with the SSM collector (located in the same wireless node, for example, in the same UE or the same network node), the RRC (Radio Resource Control) / MAC (Media Access Control) interface can be used.

[0090] Figure 6610 and the meta-learning model training unit (MMTU) 612 or the specific model training unit according to an example embodiment. The GMTU and the MMTU can be set in the same wireless node (e.g., collocated in a UE or a network node) or in different wireless nodes. Figure 6 Also shown in the network is a new radio element 616 (NR element or wireless node, which may be a UE or a network node, for example) and an SSM collector 614.

[0091] exist Figure 6 In step 1, the GMTU 610 collects features from different NR element types (different types of UEs, or devices that can use the model being trained, such as UE, TRP (transmission reception point), RSU roadside unit), and trains a general model (GM) for a given RRM task (e.g., for a general task). The GMTU 610 can maintain a list of RRM subtasks that are candidates for meta-learning (for which subtask-specific ML models can be trained).

[0092] exist Figure 6 In step 2, modification or learning / refinement of a subtask specific model (SSM) may be triggered by either: 1) on-demand / passive by (one or more of a set of) NR elements, e.g., if performance degradation is observed at such NR elements. Or, 2) periodically and / or proactively by the GMTU 610 or MMTU 612. The request may also contain or include the reason for the SSM training or refinement. For example, the gNB may have a generic model for calculating AoA (angle of arrival, used for UE positioning), and the UE wishes to use a subtask specific model to determine ToA (time of arrival, used for UE positioning). Thus, the subtask (e.g., determining ToA) and the generic task (e.g., such as determining AoA) may be similar, e.g., in that they may belong to the same category of tasks (e.g., category of positioning or determining arrival information for positioning), but the subtask is slightly different from the generic task (in this example, ToA compared to AoA).

[0093] exist Figure 6In step 3, the GMTU confirms / verifies the SSM learning request (e.g., but not necessarily verifies or confirms the SSM model itself). For example, verifying a request for a subtask-specific model (or a request for meta-learning or training of a subtask-specific model) may include, for example, verifying at least one of the following of the subtask-specific model: the requested subtask-specific model is on a list of allowed subtask-specific models; a threshold amount of training data and / or input signals are available for training the subtask-specific model; and / or a threshold amount of processor resources and / or memory resources are available for training and / or using the subtask-specific model. In addition, for example, to verify the SSM learning or training request, the GMTU 610 may check the age of the latest SSM (or a date stamp compared to the current date / time), the number, reason, and source of the SSM request. Then, after verification, the GMTU 610 may select a host MMTU 612 (meta-learning model training unit for modifying or training subtask-specific models) to manage model tuning, modification, or training, and generate a list of NR elements (e.g., a list of UEs, network nodes, or other wireless nodes) to act as an SSM collector. Each NR element used as an SSM collector may have (and / or may be identified by) a network identifier (ID). Using such an ID, each SSM collector may be contacted by other network elements (e.g., by the MMTU 612). In one example, the network ID may be an IP (Internet Protocol) address and a port number. The SSM collector may collect data upon request (e.g., for or related to a specific subtask or for a subtask-specific model), and share data generated thereon and / or data obtained from other sources via standardized (e.g., RRC, F1) and / or implementation-specific interfaces. If the confirmation / validation fails, the GMTU 610 indicates to the requesting node that the SSM cannot be generated.

[0094] exist Figure 6In step 4, the GMTU 610 triggers subtask-specific model tuning, modification, or training at the MMTU 612. The trigger message may include, for example: a trained general model (the architecture of the model - for example, the weights of the neural network, the loss function, the activation function, the output type generated by the GM; and the subtask parameterization - additional constraints - the activation function of the subtask may be different, and the input may be different (for example, only a subset of the GM input may be available, and new labels for the training data may be required). The trigger message may include: an indication / identification of the trained general model (GM), subtask-specific parameterization (for example, additional model constraints and subtask-specific cost functions), and / or an SSM collector ID. For example, the MMTU 612 may receive a trigger indication for triggering or causing the MMTU to modify or train a subtask-specific model. The trigger message may include a trigger indication, an indication of a general model trained by a second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify a subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning.

[0095] Thus, for example, modification or training of the subtask-specific model by the first MMTU based on the general model may include performing one or more of the following based on one or more constraints in the general model and the subtask-specific model: pruning or reducing the size of the general model so that the subtask-specific model will adapt to the maximum allowed subtask-specific model constraints; deactivating one or more inputs of the general model so that the input of the subtask-specific model depth or size will adapt to the format, size, or depth of the training data received from one or more subtask-specific model collectors; replacing the general model activation function with an activation function specific to the subtask-specific model; or defining a cost function specific to the subtask-specific model.

[0096] In addition, for example, modification or training of the subtask-specific model includes at least one of the following: modifying one or more weights of the subtask-specific model; training the subtask-specific model; retraining the subtask-specific model; configuring or updating one or more weights or parameters of the subtask-specific model; and / or upgrading or downgrading the subtask-specific model.

[0097] exist Figure 6In step 5, the GMTU 610 configures the SSM collector 614 to listen to MMTU requests and transmit any combination of the following: 1) its training data. The SSM collector may collect and send measurements or other training data, which will be processed at the MMTU to extract input features, or will be processed by the SSM collector and forwarded to the MMTU, or may simply send measurements to the MMTU 612; this training data may allow the MMTU 612 to train subtask-specific models. The type of training data may be defined by the GMTU 610 and explicitly requested when needed by the GMTU 610 or MMTU 612. The type of training data may be defined by any combination of the following: feature definition; the processes required for feature cleaning and standardization; the data collection period, i.e., the frequency with which training data is acquired; and the reporting frequency, i.e., the frequency with which training data is cleaned and transmitted, etc. 2) ML model limitations, such as the maximum depth of the NN (neural network or ML model); and 3) other SSM collector-specific constraints, such as power limits, mobility levels, etc.

[0098] exist Figure 6 In step 6, the MMTU 610 may issue a request to the SSM collector 614 to activate the selected SSM collector 614. Activation may include or involve issuing a request to initiate: 1) transmission of training data at a time characterized by an offset relative to receiving an activation signal; and 2) collection, cleaning, and transmission of training data.

[0099] exist Figure 6 In step 7, the SSM collector 614 may respond to the MMTU with the content requested in step 5 (including the above information).

[0100] exist Figure 6 In step 8, the subtask specific model is refined, modified or trained by MMTU 612. During this step, MMTU 612 uses the constraints associated with GM (general model) and SSM, and initializes SSM (subtask specific model). For example, MMTU 612 can: prune the GM model to adapt to the maximum allowed SSM depth; deactivate part / no input of GM to adapt to the training data format that can be provided by the SSM collector; replace the GM activation function with an SSM-specific activation function, for example, changing the regression problem solved by GM to a classification problem that SSM should solve; define an SSM-specific cost function, for example, by constraining the initial function with SSM-specific constraints, etc.

[0101] exist Figure 6 In step 9, the subtask specific model (SSM) may be transmitted (eg, sent or transferred) by the MMTU 612 to the GMTU 610.

[0102] exist Figure 6 At step 10, the SSM may be simultaneously or subsequently transferred to the subtask specific model, which may be transferred to the SSM collector.

[0103] exist Figure 6 In step 11, the SSM may be transmitted (sent or communicated) to the NR element 616 for use or reasoning, for example, so that the NR element (e.g., UE or gNB) may apply the subtask specific model (SSM) to solve or perform or assist in performing the subtask.

[0104] based on Figure 6 The general signaling flow in Figure 7 and Figure 8 is a diagram illustrating network operations, where: Figure 7 is a diagram of a network in which general model (GM) training and subtask specific model (SSM) training are performed at a network node (e.g., gNB, AP, BS, core network node, CU and / or DU, RAN node, TRP (Transmission Reception Point)); and, Figure 8 is a diagram of a network in which general model (GM) training and subtask specific model (SSM) training are performed at a UE or user equipment.

[0105] Figure 7 is a diagram of a network in which general model (GM) training and subtask specific model (SSM) training are performed at network nodes (e.g., gNB, AP, BS, core network node, CU and / or DU, RAN node, TRP (Transmission Reception Point)). Thus, for example, both GMTU and MMTU may be present or set in Figure 7 The network node 710 is within the network node 710. Figure 7 , the network node 710 may communicate with the UE 712. Meta-learning or training / modification of sub-task specific models (e.g., based on the general model, training data, and / or SSM constraints or parameters) may be performed by the network node 710. Figure 7 In the example, the network node 710 can manage ML models (e.g., general models and sub-task specific models), and these models can be used, applied, or implemented by the UE 712.

[0106] exist Figure 7 In step 1, when a new general model needs to be deployed (or provided to UE or other nodes for use), the network node 710 requests corresponding training data from one or more (selected) UEs. Figure 7 In step 1 of , the network node 710 may issue a request to the UE 712 (and possibly to other UEs) for task-specific training data and constraints.

[0107] exist Figure 7In step 2, UE 712 (which received the request in step 1) may provide the requested task-specific training data and / or constraints to network node 710.

[0108] exist Figure 7 In step 3, the network node 710 modifies or trains an ML model, which may be a general task model or a general model (GM) that is designed or trained to perform or solve a task, such as a general task. Some explanations and examples of general (or general) task models and subtask-specific models have been provided above. For example, in general mathematical terms, a general ML model (or general task model) may find or obtain a solution x of a function f(x)=0, while a subtask-specific model finds a solution x of a function g(x)=0, where there is a known relationship between the functions g() and f(), such as: g(x)=alpha*f(x)+(possibly one or more) constraints on the value of x. In addition, for example, a subtask-specific model may have a set of inputs that are a subset of the inputs used by the general task model x, so g(x'). Similarly, for example, a subtask-specific model may have a set of outputs that are a subset of the outputs of the general task model.

[0109] exist Figure 7 In step 4 , the network node 710 deploys (eg, sends or transmits) a GM (generic model) to the selected UE(s), such as UE 712 .

[0110] exist Figure 7 In step 5, the UE 712 uses the deployed GM (generic ML model) to perform or complete ML-enabled tasks or functions based on the GM. During the use or implementation of the GM, the UE 712 may identify or determine that adaptation of the general ML model will be beneficial or necessary, for example, due to changes in input distribution, reduction in accuracy, reduction in performance, etc., or the need to apply the model to a slightly different (e.g., but related) task (e.g., sub-task). Note that for this capability, if the dataset is well balanced (e.g., label distribution), the UE may need to obtain details about the context of the general model, i.e., how the GM training dataset was generated. Alternatively, the network node can also detect the need for ML model adaptation in the UE based on feedback (traditional or ML-related) from the UE, for example, the need to adapt (retrain) the general model to perform a different task or sub-task.

[0111] exist Figure 7In step 6 of , if the UE has detected in step 6 that ML adaptation (or retraining) of the general model is required (e.g., adapting or retraining the GM to perform a different task, such as a subtask), the UE 712 may generate and issue a request to the network node 710, the request including an indication of a subtask-specific change, or an indication of a requested subtask-specific model (SSM), or an indication or request to perform retraining / adaptation of the general model to perform the new task or subtask. In addition, for example, when the network node (e.g., gNB) determines or detects a need for ML adaptation of the general model, in sending a request to the GMTU (in Figure 7 This request (from the gNB) is still required in the case where the need for a subtask specific model is indicated within the network node 710.

[0112] exist Figure 7 In step 7, the network node (e.g., a GMTU within network node 710) may verify the request received in step 6 (e.g., verifying a request to adapt or retrain the general model to perform a different task or subtask, or a request for a subtask-specific model based on the general model. The verification may be performed based on a provided subtask description, an existing database of possible subtasks, and / or available computing power for such training tasks. Thus, for example, verifying the request for a subtask-specific model may include verifying at least one of the following for the subtask-specific model: the requested subtask-specific model is on a list of allowed subtask-specific models; at least a threshold amount of training data and / or input signals are available for training the subtask-specific model; and / or at least a threshold amount of processor resources and / or memory resources are available for training and / or using the subtask-specific model.

[0113] exist Figure 7 In step 8, the network node may trigger (or cause) the generation of a subtask specific model (SSM) and its required configuration parameters (training data type / source, etc.). The task of generating / training the SSM may be passed to the MMTU. Figure 7 In step 8, the network node 710 may trigger or cause subtask-specific meta-learning or training of subtask-specific models by the MMTU, and transmission of, for example, subtask parameterizations (from the GMTU to the MMTU, within the network node 710).

[0114] exist Figure 7In step 9, the network node 710 (e.g., which may be or include one or both of a GMTU and / or an MMTU) may request SSM training data and / or constraints from the UE 712 (or from one or more UEs including the UE 712). Note that the SSM training data need not be exactly the same as the GM training data, but may be a different set of data or a subset of the GM training data. For example, the SSM training data may be a subtask-specific subset of the GM training data, and the SSM and GM training data may have different granularities, reporting frequencies, etc. Thus, for example, in step 9, the network node 710 may send a request to the UE 712 for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data.

[0115] exist Figure 7 In step 10, the UE 712 may provide (or send to the network node 710) SSM-specific training data and any UE-specific constraints (e.g., limitations on the size of the ML model supported by the UE, or certain features or capabilities supported or not supported by the UE 712 (e.g., which may be related to the ML model or the RRM task for which the ML model may be used), or other constraints).

[0116] exist Figure 7 In step 11, the network node 710 (eg, which may include a GMTU and / or an MMTU) performs training of the SSM, for example based on a generic model, received subtask-specific training data and / or any UE-specific constraints provided by the UE 712.

[0117] exist Figure 7 In step 12 , the network node 710 may deploy (provide or send) the trained SSM model to (multiple) UEs, including UE 712 .

[0118] exist Figure 7 In step 13, the UE (eg, including UE 712) uses or applies the new SSM model to solve the problem or perform the subtask.

[0119] As mentioned earlier, Figure 8 is a diagram of a network in which general model (GM) training and subtask specific model (SSM) training are performed at a UE or user equipment. Figure 8 In , the network node 810 can communicate with the UE 812. Figure 8In the example, the training of the general model and the sub-task specific models may be performed by the UE 812. The SSM may be trained or modified to perform the same task as the general model is used for or a different task (e.g., a sub-task). The training of the GM and / or SSM may be performed by the UE 812 (e.g., a GMTU and / or MMTU provided within the UE 812), and the training may be controlled and / or configured by the network node 810. In addition, in Figure 8 In a network with a plurality of SSMs, the network node 810 may configure and / or test ML-enabled functions implemented in the UE 812 or performed by the SSM at the UE 812. The UE may perform ML model training, for example, without the network node 810 necessarily knowing (and the network node may or may not know) implementation details of the SSM model in the UE 812. In addition, according to an example embodiment, in Figure 8 For example, the GMTU may be set in the network node 810 and the MMTU may be set within the UE 812.

[0120] exist Figure 8 In step 1 of the embodiment, the network node 812 may send a request to the UE to train a model for the selected ML-enabled function (e.g., a request to the UE 812 to train a generic ML model for a generic task), which includes one or more configuration parameters. The network node does not need to indicate to the UE that this is a generic meta-model, thereby avoiding the possibility of the UE performing some 'cheating' when generating the ML model.

[0121] exist Figure 8 In step 2, UE 812 trains its ML model using input signals, data, configuration parameters, etc. provided by network node 810, for example, to generate a general model to perform a general task. Details of the input signals / data and / or configuration parameters required for the general model may be provided by network node 810. UE 812 does not need to know that the model is considered a general model by network node 810.

[0122] exist Figure 8 In step 3, the UE 812 sends an indication to the network node that the ML model is trained and ready to be used to perform a general task or solve a problem.

[0123] exist Figure 8 In step 4, the network node 810 triggers or causes the UE 812 to use or apply the general model, for example, the network node 810 issues or sends a request to use or apply the general model to the UE 812. The required input signal / data can also be provided to the UE 812 by the network node 810.

[0124] exist Figure 8In step 5, the UE 812 uses or applies the trained general ML model and performs the general ML-enabled functions using the provided input signals / data.

[0125] exist Figure 8 In step 6, the UE 812 provides the configured feedback to the network node 810 when using the ML-enabled function (based on the GM). This step can be performed periodically or in a one-time manner (once), depending on the ML-enabled function (generic task) under test. As part of this step, the UE 812 can provide or send the output based on the generic model (such as, for example, the output of the trained generic model) to the network node 810, so that the network node 810 can verify or confirm that the output is correct and that the generic model is operating well or as expected.

[0126] exist Figure 8 In step 7, the network node 810 verifies the UE output by comparing the provided feedback (e.g., output based on the generic model provided by the UE) with the expected feedback generated using its own generic model (trained in the NW using the same data and configuration, etc.). If this step is declared "passed" (the expected output matches the actual UE output / feedback), the NW proceeds to step 8 for new subtask specific tests. Otherwise (the step is not declared passed) the network node can proceed back to step 1 to perform the same or new task (potentially using different input conditions, parameters, etc.).

[0127] exist Figure 8 In step 8, the network node 810 may request the UE 812 to modify the ML model for the subtask specific purpose (e.g., generating a subtask specific model based on retraining of the general model). The network node 810 may provide the UE 812 with configuration parameters, subtask description, etc.).

[0128] exist Figure 8 In step 9, UE 812 verifies a request from the network node to modify a generic ML model for a particular subtask (e.g., the UE verifies a request to generate a subtask-specific model based on the trained generic model and other information provided by the network node 810). The verification can be performed based on the provided subtask description, an existing database of possible subtasks, and / or available computing power for such training tasks. Thus, for example, the verification can include verifying at least one of the following of the subtask-specific model: the requested subtask-specific model is on a list of allowed subtask-specific models; a threshold amount of training data and / or input signals are available for training the subtask-specific model; and / or a threshold amount of processor resources and / or memory resources are available for training and / or using the subtask-specific model.

[0129] exist Figure 8 In step 10, the UE 812 trains the SSM (sub-task specific model) with input signals / data from the network according to the received configuration (eg, based on the trained general model).

[0130] exist Figure 8 In step 11, UE 812 indicates to the network node (GMTU) that the ML model (SSM) has been trained and is ready for use.

[0131] exist Figure 8 In step 12, the network node 810 sends a message or signal to the UE 812 to trigger or cause the UE 812 to use or apply the SSM to perform the subtask.

[0132] exist Figure 8 In step 12, the UE 812 uses the subtask specific ML model (SSM) and performs the ML-enabled function (to perform the subtask) using the provided input signal / data.

[0133] exist Figure 8 In step 13, the UE 812 provides the configured feedback when using the ML-enabled function (based on SSM), for example, the UE 812 may provide an SSM-based output (or an SSM output) to the network node 810. This step 13 may occur periodically or once (or in a one-time manner), depending on the ML-enabled function (e.g., subtask) under test.

[0134] exist Figure 8 In step 14, the network node 810 verifies the UE output by comparing the provided feedback (e.g., the output based on the SSM) with the expected feedback generated with its own SSM (trained in the NW with the same data and configuration, etc.). If this step is declared passed (e.g., the provided feedback is sufficiently matched within the threshold of the output based on the SSM generated by the network node), the network node 810 proceeds to step 8 for a new subtask specific test, or proceeds to step 1 for a new task. Otherwise (the step fails, e.g., the feedback does not match the expected feedback), the network node 810 continues to perform step 8 for the same subtask (potentially using a different set of input conditions).

[0135] Example use case when the NR element is a UE:

[0136] ML-based models (part of one or more UE ML-enabled functions) can be built to perform common tasks in the UE. For example, a common task is to use a DNN (Deep Neural Network) based model to output a class / label and its probability, which is then used to trigger an action for UE beam selection.

[0137] For the same task: the same task can be performed in two different (groups of) UEs using different numbers of beams: assuming that the ML-based model is initially trained based on only one UE type, for example X=3 beams; or, alternatively, when only X of the Y beams are enabled for each UE during training (the X beams can be selected by the UE or network node). Based on appropriate feedback / measurements from the selected UE, general model training (including testing / validation, etc.) can occur or be performed in the network node or GMTU. The network node or GMTU deploys (or configures) the general model to the UE (for example, issues a general model configuration, constraints, or other parameters of the general model trained by the UE). After receiving and using the general model, some UEs in the NW indicate the ability or need to use a subtask specific model (SSM) to control the Z beam instead of X (Z<>X). The network node or GMTU initiates subtask-specific (UE-specific) training or adjustment of the general model for the corresponding UE (other UE-specific information may be used in addition if available at the NW). The subtask-specific model can be generated in the MMTU or network node. The network node deploys (or configures) a subtask specific model (SSM) to the UE requesting the subtask specific model.

[0138] For a similar (e.g., but slightly different) task: A similar (e.g., but slightly different) task for an ML-based model may be, for example, selecting antenna panels instead of antenna beams in (multiple) the same or different UEs: Assume that the ML-based model is initially trained based on only one UE type, e.g., X=3 beams. Or, alternatively, when X of Y beams are enabled for each UE during training (the X beams may be selected by the UE or the NW). Model training occurs in the NW / GMTU based on appropriate feedback / measurements from the selected UEs. The network node or GMTU may deploy (or configure or send model and configuration information) a generic model to the UE. After receiving the generic model, certain UEs in the NW indicate the need to use only ML-enabled antenna panel selection instead of beam selection. The NW generates sub-task-specific (UE-specific) training or adjustments of the generic model for the corresponding UE (other UE-specific information may be used in addition if available at the NW). The NW deploys (or configures) specific models to UEs that request sub-task-specific models.

[0139] As can be seen from the above illustrative examples, the availability of the proposed mechanism for deploying the same ML model for the same task on different UEs (different numbers of beams) can be directly used to enable consistency testing of the beam selection function of the UE ML. For example, the choice and advantage of using meta-learning can be in the network aspect. Figure 8 The figure depicts some example signaling for this use case.

[0140] Conformance testing using the proposed method can also be extended to the case where the ML model is not fully trained in the network node, but the training is at least partially configurable by the network node (e.g., this may be a 3GPP prerequisite for UE-gNB collaboration). The UE ML model may be trained in the UE using exposed configuration parameter settings received from the network node and test signals provided by the network node. For example, a generic task may be to use only X=2 beams (out of Y>X supported by the UE). After training and functional verification for X=2, the network node may reconfigure (e.g., modify or retrain) the functionality enabling the UE ML (e.g., modify or retrain the generic model to become or generate an SSM) to request the use of Y>=Z>2 beams. In response to this request, the UE may indicate that more training and test signals from the NW are required. In this case, the choice and advantage of using meta-learning is entirely up to the UE. However, its use may be detected by measuring the time required to adjust the generic / initial model for X beams to the new requirement for Z beams. If an additional 'retune time' limit is imposed on the UE, then meta-learning will more or less be forced to be implemented in the UE in order to comply with that limit. Figure 8 The figure depicts examples of signaling that can be used for these types of UE conformance test cases.

[0141] Example when the NR element is a gNB or a network node:

[0142] An example of a situation when the NR element running the ML-enabled functionality is a gNB may be one in which UL (uplink) power control parameters are optimized: determining OLPC (open loop power control) P0 and / or alpha, and / or determining CLPC (closed loop power control) adjustment steps. Figure 7 , the following implementations may be used or are possible: GMTU / MMTU is located at the NW side. The NR element is the gNB. The SSM collector is the selected cell edge / cell center UE.

[0143] OLPC may be an example use case. Typically, OLPC (Open Loop Power Control, as a specific usage function or subtask of the SSM model) parameters (P0 and alpha) are configured at the cell level, e.g., the same values ​​may be used for the UEs served in a cell. However, the specification supports per-UE (UE-specific) signaling of these parameters, so a more optimized approach may also be used or implemented. One such approach is to use clusters of UEs based on their radio proximity (DL (Downlink) RSRP (Reference Signal Received Power)) to the serving cell or neighboring cells. In this case, there may be at least two UE clusters (“cell edge” and “cell center”) with different sets of OLPC parameters, and controlled separately by ML-enabled functions. Using the meta-learning approach described in this article, a general model may be trained to control both OLPC P0 and / or alpha using available radio measurements from one or more cells (in a selected geographical area) as input. The general model may then be deployed to each gNB or network node, potentially also including network nodes or gNBs for which no training data was used. During operation, some network nodes or gNBs identify that a third UE cluster may be beneficial to control separately from the already included "cell edge" and "cell center" clusters (e.g., a high-speed UE cluster). These gNBs or network nodes may then request subtask specific models (e.g., for the same task) from the GMTU, and upon confirmation or verification of the request, the SSM may be trained (by the UE) and redistributed or sent back to the corresponding network nodes or gNBs. The SSM may also be differentiated based on which OLPC (similar task) is being controlled. For example, the system may also include an identification algorithm to determine the need for subtask specific models and provide and / or collect input training data.

[0144] Some additional examples will be provided.

[0145] Example 1. A method may include: a first model training unit receiving from a second model training unit a trigger indication that triggers or causes modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and a subtask or one or more parameters of the subtask-specific model to be used to modify the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; the first model training unit receiving training data for the subtask-specific model from one or more subtask-specific model collectors; the first model training unit modifying the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model and the training data received from the one or more subtask-specific model collectors; and the first model training unit sending the modified subtask-specific model to a first wireless node.

[0146] Example 2. A method according to Example 1, wherein the modification includes at least one of the following: modifying one or more weights of the subtask-specific model; training the subtask-specific model; retraining the subtask-specific model; configuring or updating one or more weights or parameters of the subtask-specific model; and / or upgrading or downgrading the subtask-specific model.

[0147] Example 3. A method according to any one of Examples 1 to 2, wherein the receiving a trigger indication includes: receiving, by the first model training unit from the second model training unit, one or more of the following: information about the general model, including one or more of the following: the architecture of the general model, the weights of the general model, the loss function and / or activation function of the general model, and / or the output type of the general model; subtask parameterization, including constraints of the subtask-specific model or subtask-specific cost functions of the subtask-specific model; and / or identifiers of the one or more subtask-specific model collectors.

[0148] Example 4. A method according to any one of Examples 1 to 3, wherein the modification of the subtask-specific model by the first model training unit based on the general model includes performing one or more of the following based on one or more constraints of the general model and the subtask-specific model: pruning or reducing the size of the general model so that the subtask-specific model will adapt to the maximum allowed subtask-specific model constraint; deactivating one or more inputs of the general model so that the input of the subtask-specific model depth or size will adapt to the format, size or depth of the training data received from the one or more subtask-specific model collectors; replacing the general model activation function with an activation function specific to the subtask-specific model; or defining a cost function specific to the subtask-specific model.

[0149] Example 5. The method according to any one of Examples 1 to 4 further includes: sending the modified sub-task specific model by the first model training unit to the second model training unit.

[0150] Example 6. The method according to any one of Examples 1 to 5 further includes: sending the modified subtask-specific model by the first model training unit to at least one subtask-specific model collector among the one or more subtask-specific model collectors.

[0151] Example 7. A method according to any one of Examples 1 to 6, wherein any one of the following: the first model training unit and the second model training unit are set in a second wireless node; or the first model training unit is set in the second wireless node, and the second model training unit is set in a third wireless node.

[0152] Example 8. The method according to any one of Examples 1 to 7 further includes: the first model training unit sending a request for sub-task specific model training or meta-learning of the sub-task to the second model training unit.

[0153] Example 9. A method according to any one of Examples 1 to 8, wherein: the second model training unit includes: a general model training unit configured to modify the general model for a general task; and the first model training unit is a meta-learning model training unit or a specific model training unit configured to modify the specific model or subtask specific model for the subtask.

[0154] Example 10. A method according to any one of Examples 1 to 8, wherein one or more of the first wireless node, the second wireless node, or the third wireless node includes at least one of the following: user equipment, user device, base station, or gNB.

[0155] Example 11. A device comprising: at least one processor; and at least one memory comprising computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: receive, by a first model training unit, from a second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and a subtask or one or more parameters of the subtask-specific model to be used to modify the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; receive, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors; modify, by the first model training unit, the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model, and the training data received from the one or more subtask-specific model collectors; and send, by the first model training unit, the modified subtask-specific model to a first wireless node.

[0156] Example 12. A device comprising: a component for receiving, by a first model training unit, from a second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and a component for modifying a subtask of the subtask-specific model or one or more parameters of the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; a component for receiving, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors; a component for modifying, by the first model training unit, the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model, and the training data received from the one or more subtask-specific model collectors; and a component for sending, by the first model training unit, the modified subtask-specific model to a first wireless node.

[0157] Example 13. A non-transitory computer-readable storage medium, comprising instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: receive, by a first model training unit, from a second model training unit, a trigger indication that triggers or causes modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and a subtask or one or more parameters of the subtask-specific model to be used to modify the subtask-specific model based on the general model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; receive, by the first model training unit, training data for the subtask-specific model from one or more subtask-specific model collectors; modify, by the first model training unit, the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model, and the training data received from the one or more subtask-specific model collectors; and send, by the first model training unit, the modified subtask-specific model to a first wireless node.

[0158] Example 14. A method comprising: receiving a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determining a need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; sending, by a second model training unit, to a first model training unit a trigger indication that triggers or causes the first model training unit to modify the subtask-specific model, an indication of the general model trained by the second model training unit for a general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; configuring, by the second model training unit, one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and receiving, by the second model training unit, the modified subtask-specific model modified by the first model training unit.

[0159] Example 15. A method according to Example 14, wherein the sending includes: sending one or more of the following by the second model training unit to the first model training unit: information of the general model, including one or more of the following: the architecture of the general model, the weights of the general model, the loss function and / or activation function of the general model, and / or the output type of the general model; subtask parameterization, including constraints of the subtask-specific model, or subtask-specific cost functions of the subtask-specific model; and / or identifiers of the one or more subtask-specific model collectors.

[0160] Example 16. A method according to any one of Examples 14 to 15, wherein any one of the following: the first model training unit and the second model training unit are set in a second wireless node; or the first model training unit is set in the second wireless node, and the second model training unit is set in a third wireless node.

[0161] Example 17. A method according to any one of Examples 14 to 16, wherein receiving a request to modify a subtask-specific model, or otherwise determining the need for it, includes: receiving, by the second model training unit from the first model training unit, a request to modify or train the subtask-specific model for the subtask; and verifying the request to modify or train the subtask-specific model for the subtask.

[0162] Example 18. A method according to any one of Examples 14 to 17, wherein: the second model training unit includes: a general model training unit configured to modify or train the general model for a general task; and the first model training unit is a meta-learning model training unit or a specific model training unit configured to modify or train a specific model or a subtask specific model for the subtask.

[0163] Example 19. A method according to any one of Examples 14 to 18, wherein one or more of the first wireless node, the second wireless node, or the third wireless node includes at least one of the following: user equipment, user device, base station, or gNB.

[0164] Example 20. A device comprising: at least one processor; and at least one memory comprising computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: receive a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determine the need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; a second model training unit sends a trigger indication to the first model training unit that triggers or causes the first model training unit to modify the subtask-specific model, an indication of the general model trained by the second model training unit for a general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; the second model training unit configures one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and the second model training unit receives the modified subtask-specific model modified by the first model training unit from the first model training unit.

[0165] Example 21. A device comprising: a component for receiving a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determining a need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; a component for sending, by a second model training unit, to a first model training unit a trigger indication that triggers or causes the first model training unit to modify the subtask-specific model, an indication of the general model trained by the second model training unit for a general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; a component for configuring, by the second model training unit, one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and a component for receiving, by the second model training unit, from the first model training unit, the modified subtask-specific model modified by the first model training unit.

[0166] Example 22. A non-transitory computer-readable storage medium comprising instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: receive a request to modify a subtask-specific model for a first wireless node based on a general model, or otherwise determine a need for it, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; send, by a second model training unit, to a first model training unit a trigger indication that triggers or causes the first model training unit to modify the subtask-specific model, an indication of the general model trained by the second model training unit for a general task, and one or more parameters of a subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; configure, by the second model training unit, one or more subtask-specific model collectors to provide training data for modifying the subtask-specific model to the first model training unit; and receive, by the second model training unit, from the first model training unit the modified subtask-specific model modified by the first model training unit.

[0167] Example 23. A method, comprising: determining, by a user device, a trained general model to perform or assist in performing a general task that enables machine learning; determining, based on the trained general model, one or more general model-based outputs according to one or more signals or inputs; sending, by the user device, the one or more general model-based outputs to a network node; receiving, by the user device at least in part based on the one or more general model-based outputs, a request for a subtask-specific model from the network node, including configuration parameters of the subtask-specific model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; verifying the request for the subtask-specific model; modifying, by the user device, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or the subtask-specific model; and performing or completing a subtask that enables machine learning based on or using the modified subtask-specific model by the user device; and sending the subtask-specific model output to the network node by the user device based on performing or completing the subtask that enables machine learning according to or using the modified subtask-specific model.

[0168] Example 24. A method according to Example 23, wherein the modification includes at least one of the following: modifying one or more weights of the subtask specific model; training the subtask specific model; retraining the subtask specific model; configuring or updating one or more weights or parameters of the subtask specific model; and / or upgrading or downgrading the subtask specific model.

[0169] Example 25. A method according to any one of Examples 23 to 24, wherein the verifying the request for the subtask-specific model includes: verifying at least one of the following for the subtask-specific model: the requested subtask-specific model is on a list of allowed subtask-specific models; a threshold amount of training data and / or input signals are available for training the subtask-specific model; and / or a threshold amount of processor resources and / or memory resources are available for training and / or using the subtask-specific model.

[0170] Example 26. A method according to any one of Examples 23 to 25, wherein the receiving of a request for a subtask-specific model includes configuration parameters of the subtask-specific model, including receiving: a trigger indication that triggers or causes meta-learning or training of the subtask-specific model, and one or more parameters of the subtask or subtask-specific model to be used to train the subtask-specific model based on the general model, including receiving subtask parameterization, the subtask parameterization including: constraints of the subtask-specific model, or a subtask-specific cost function of the subtask-specific model.

[0171] Example 27. A method according to any one of Examples 23 to 26, wherein the configuration parameters of the subtask or the subtask-specific model include: one or more constraints of the subtask-specific model, and wherein the modification of the subtask-specific model by the user device based on the general model includes: performing one or more of the following based on the general model, and one or more constraints of the subtask-specific model: pruning or reducing the size of the general model so that the subtask-specific model will adapt to the maximum allowed subtask-specific model depth or size; deactivating one or more inputs of the general model so that the input of the subtask-specific model depth or size will adapt to the format, size or depth of the training data received from the one or more subtask-specific model collectors; replacing the general model activation function with an activation function specific to the subtask-specific model; or defining a cost function specific to the subtask-specific model.

[0172] Example 28. A method according to any one of Examples 23 to 27, wherein determining one or more general model-based outputs includes: receiving, by the user device, a request from a network node to train a general model for the general task enabling machine learning; training the general model by the user device based on a configuration or input received from the network node; and using the trained general model to perform or complete the general task enabling machine learning to obtain the one or more general model-based outputs.

[0173] Example 29. A method according to Example 28, wherein the request for the sub-task specific model is received by the user device in response to the user device sending the one or more general model-based outputs of the trained general model to the network node.

[0174] Example 30. A device comprising: at least one processor; and at least one memory comprising computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: determine, by a user device, a trained general model for performing or assisting in performing a general task enabling machine learning; determine, based on the trained general model, one or more general model-based outputs according to one or more signals or inputs; send, by the user device, the one or more general model-based outputs to a network node; receive, by the user device, a response to a subtask from the network node based at least in part on the one or more general model-based outputs A request for a subtask-specific model, including configuration parameters of the subtask-specific model, wherein the subtask-specific model will perform or assist in performing a subtask that enables machine learning; verifying the request for the subtask-specific model; modifying the subtask-specific model by the user device based on the trained general model and the configuration parameters of the subtask or the subtask-specific model; and performing or completing the subtask that enables machine learning based on or using the modified subtask-specific model by the user device; and sending a subtask-specific model output by the user device to the network node based on performing or completing the subtask that enables machine learning according to or using the trained subtask-specific model.

[0175] Example 31. An apparatus comprising: a component for determining, by a user device, a trained general model for performing or assisting in performing a general task that enables machine learning; a component for determining, based on the trained general model, one or more general model-based outputs according to one or more signals or inputs; a component for sending, by the user device, the one or more general model-based outputs to a network node; a component for receiving, by the user device, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, including configuration parameters of the subtask-specific model, wherein the subtask-specific model Perform or assist in performing a subtask that enables machine learning; a component for verifying the request for the subtask-specific model; a component for modifying the subtask-specific model by the user device based on the trained general model and the configuration parameters of the subtask or the subtask-specific model; and a component for performing or completing the subtask that enables machine learning by the user device based on or using the modified subtask-specific model; and a component for sending a subtask-specific model output to the network node by the user device based on performing or completing the subtask that enables machine learning according to or using the trained subtask-specific model.

[0176] Example 32. A non-transitory computer-readable storage medium, comprising instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: determine, by a user device, a trained general model for performing or assisting in performing a general task that enables machine learning; determine, based on the trained general model, one or more general model-based outputs according to one or more signals or inputs; send, by the user device, the one or more general model-based outputs to a network node; receive, by the user device, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, including the subtask-specific model; The method comprises the steps of: receiving, by the user device, a subtask-specific model configured to perform or assist in performing a machine learning-enabled subtask; verifying the request for the subtask-specific model; modifying, by the user device, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or the subtask-specific model; and performing or completing the machine learning-enabled subtask based on or using the modified subtask-specific model by the user device; and sending a subtask-specific model output to the network node by the user device based on performing or completing the machine learning-enabled subtask according to or using the trained subtask-specific model.

[0177] Example 33. A method, comprising: determining, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; providing, by the network node, the trained general model to a user device; receiving, by the network node, a request for a subtask-specific model from the user device; verifying the request for the subtask-specific model; sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device; receiving, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device; modifying, by the network node, the subtask-specific model based on the general model, and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and sending, by the network node, the modified subtask-specific model to the user device.

[0178] Example 34. A method according to Example 33, wherein the modification includes at least one of the following: modifying one or more weights of the subtask specific model; training the subtask specific model; retraining the subtask specific model; configuring or updating one or more weights or parameters of the subtask specific model; and / or upgrading or downgrading the subtask specific model.

[0179] Example 35. A method according to any one of Examples 33 to 34, wherein the verifying the request for the subtask-specific model includes: verifying at least one of the following for the subtask-specific model: the requested subtask-specific model is on a list of allowed subtask-specific models; a threshold amount of training data and / or input signals are available for training the subtask-specific model; and / or a threshold amount of processor resources and / or memory resources are available for training and / or using the subtask-specific model.

[0180] Example 36. A method according to any one of Examples 33 to 35, wherein the modification of the subtask-specific model by the network node based on the general model includes performing one or more of the following based on the general model and one or more constraints of the subtask-specific model: pruning or reducing the size of the general model so that the subtask-specific model will adapt to the maximum allowed subtask-specific model depth or size; deactivating one or more inputs of the general model so that the input of the subtask-specific model depth or size will adapt to the format, size or depth of the training data received from the one or more subtask-specific model collectors; replacing the general model activation function with an activation function specific to the subtask-specific model; or defining a cost function specific to the subtask-specific model.

[0181] Example 37. A device comprising: at least one processor; and at least one memory comprising computer program code; the at least one memory and the computer program code are configured to, together with the at least one processor, enable the device to at least: determine a trained general model by a network node to perform or assist in performing a general task that enables machine learning; provide the trained general model to a user device by the network node; receive a request for a subtask-specific model from the user device by the network node; verify the request for the subtask-specific model; send a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data by the network node to the user device; receive at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data by the network node from the user device; modify the subtask-specific model by the network node based on the general model and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and send the modified subtask-specific model to the user device by the network node.

[0182] Example 38. An apparatus comprising: a component for determining, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; a component for providing, by the network node, the trained general model to a user device; a component for receiving, by the network node, a request for a subtask-specific model from the user device; a component for verifying the request for the subtask-specific model; a component for sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device; a component for receiving, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device; a component for modifying, by the network node, the subtask-specific model based on the general model, and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and a component for sending the modified subtask-specific model to the user device.

[0183] Example 39. A non-transitory computer-readable storage medium, comprising instructions stored thereon, which, when executed by at least one processor, are configured to cause a computing system to: determine, by a network node, a trained general model to perform or assist in performing a general task that enables machine learning; provide, by the network node, the trained general model to a user device; receive, by the network node, a request for a subtask-specific model from the user device; verify the request for the subtask-specific model; send, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user device; receive, by the network node, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data from the user device; modify, by the network node, the subtask-specific model based on the general model, and at least one of the subtask-specific model configuration or constraint and / or subtask-specific model training data; and send, by the network node, the modified subtask-specific model to the user device.

[0184] Fig. 9 1 is a block diagram of a wireless station or node (e.g., UE, user equipment, AP, BS, eNB, gNB, RAN node, network node, TRP, or other node) 1200 according to an example embodiment. The wireless station 1200 may include, for example, one or more (e.g., Fig. 9The wireless station also includes two RF (radio frequency) or wireless transceivers 1202A, 1202B shown, each of which includes a transmitter for sending signals and a receiver for receiving signals. The wireless station also includes a processor or control unit / entity (controller) 1204 for executing instructions or software and controlling the transmission and reception of signals, and a memory 1206 for storing data and / or instructions.

[0185] The processor 1204 may also make decisions or determinations, generate frames, packets or messages for transmission, decode received frames or messages for further processing, and perform other tasks or functions described herein. For example, the processor 1204, which may be a baseband processor, may generate messages, packets, frames or other signals for transmission via the wireless transceiver 1202 (1202A or 1202B). The processor 1204 may control the transmission of signals or messages through a wireless network, and may control the reception of signals or messages, etc., via a wireless network (e.g., after being down-converted by the wireless transceiver 1202). The processor 1204 may be programmable and may be capable of executing software or other instructions stored in a memory or on other computer media to perform the various tasks and functions described above, such as one or more of the tasks or methods described above. For example, the processor 1204 may be (or may include) hardware, programmable logic, a programmable processor that executes software or firmware, and / or any combination of these. For example, using other terms, the processor 1204 and the transceiver 1202 may be considered together as a wireless transmitter / receiver system.

[0186] In addition, reference Fig. 9 , the controller (or processor) 1208 can execute software and instructions and can provide overall control for the station 1200 and can Fig. 9 Other systems not shown provide controls, such as controlling input / output devices (e.g., display, keypad), and / or software that can execute one or more applications that may be provided on wireless station 1200, such as an email program, audio / video applications, a word processor, voice over IP applications, or other applications or software.

[0187] Additionally, a storage medium may be provided that includes stored instructions that, when executed by a controller or processor, may cause the processor 1204 or other controller or processor to perform one or more of the functions or tasks described above.

[0188] According to another example embodiment, the RF or (multiple) wireless transceiver 1202A / 1202B can receive signals or data and / or send or transmit signals or data. The processor 1204 (and possibly the transceiver 1202A / 1202B) can control the RF or wireless transceiver 1202A or 1202B to receive, issue, broadcast or transmit signals or data.

[0189] Embodiments of the various techniques described herein may be implemented in a digital electronic circuit system, or in computer hardware, firmware, software, or a combination thereof. Embodiments may be implemented as computer program products, i.e., computer programs tangibly embodied in information carriers, for example, in machine-readable storage devices or in propagated signals, for data processing devices to execute or control the operation of data processing devices, for example, programmable processors, one computer or multiple computers. Embodiments may also be provided on a computer-readable medium or a computer-readable storage medium, which may be a non-transitory medium. Embodiments of various techniques may also include embodiments provided via transient signals or media, and / or downloadable programs and / or software embodiments via the Internet or (multiple) other networks (wired networks and / or wireless networks). In addition, embodiments may be provided via machine type communications (MTC) or via the Internet of Things (IOT).

[0190] A computer program may be in source code form, object code form or some intermediate form and may be stored in some carrier, distribution medium or computer readable medium, which may be any entity or device capable of carrying the program. Such carriers include, for example, recording media, computer memories, read-only memories, optoelectronic and / or electrical carrier signals, telecommunication signals and software distribution packages. Depending on the processing power required, a computer program may be executed in a single electronic digital computer or distributed among multiple computers.

[0191] In addition, embodiments of the various techniques described herein may use cyber-physical systems (CPS) (systems that enable computing elements that control physical entities to collaborate). CPS can implement and utilize a large number of interconnected ICT devices (sensors, actuators, processors, microcontrollers, etc.) embedded in different locations in physical objects. Mobile cyber-physical systems (where the physical system in question has inherent mobility) are a subcategory of cyber-physical systems. Examples of mobile physical systems include mobile robots and electronic devices transported by humans or animals. The popularity of smartphones has increased interest in the field of mobile cyber-physical systems. Therefore, various embodiments of the techniques described herein may be provided by one or more of these techniques.

[0192] Computer programs such as the above-mentioned (multiple) computer programs can be written in any form of programming language, including compiled or interpreted languages, and can be deployed in any form, including as a stand-alone program, or as a module, component, subroutine or other unit or part thereof suitable for a computing environment. A computer program can be deployed to execute on one computer, or on multiple computers at one site, or on multiple computers distributed between multiple sites and interconnected by a communication network.

[0193] The method steps may be performed by one or more programmable processors executing a computer program or a portion of a computer program to perform functions by operating on input data and generating output. The method steps may also be performed by, and the apparatus may be implemented as, a special purpose logic circuit system, such as an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).

[0194] For example, processors suitable for executing computer programs include both general and special purpose microprocessors, and any one or more processors of any kind of digital computer, chip or chipset. Typically, the processor will receive instructions and data from a read-only memory or a random access memory or both. Elements of a computer may include at least one processor for executing instructions, and one or more memory devices for storing instructions and data. Typically, a computer may also include or be operatively coupled to receive data from or transfer data to or both of one or more mass storage devices (e.g., magnetic, magneto-optical disks, or optical disks) for storing data. Information carriers suitable for embodying computer program instructions and data include all forms of non-volatile memory, including, for example, semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; magnetic disks, such as internal hard disks or removable disks; magneto-optical disks; and CD ROM and DVD-ROM disks. The processor and memory may be supplemented by or incorporated in a dedicated logic circuit system.

[0195] To provide interaction with a user, an embodiment may be implemented on a computer having a display device (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) display) for displaying information to the user, and a user interface (such as a keyboard and a pointing device, such as a mouse or trackball) through which the user can provide input to the computer. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback, such as visual feedback, auditory feedback, or tactile feedback; and the input from the user may be received in any form, including sound, voice, or tactile input.

[0196] Embodiments may be implemented in a computing system that includes: a back-end component, for example, as a data server; or includes a middleware component, for example, an application server; or includes a front-end component, for example, a client computer having a graphical user interface or a web browser through which a user can interact with the embodiments; or any combination of such back-end, middleware, or front-end components. The components may be interconnected by any form or media of digital data communication, such as a communication network. Examples of communication networks include local area networks (LANs) and wide area networks (WANs), such as the Internet.

[0197] Although certain features of the described embodiments have been illustrated as described herein, those skilled in the art will now be able to conceive of many modifications, substitutions, changes and equivalents. Therefore, it should be understood that the appended claims are intended to cover all such modifications and changes that fall within the true spirit of the various embodiments.

Claims

1. A method comprising: Receiving, by the first model training unit from the second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of the subtask-specific model to be used to modify the subtask or the subtask-specific model based on the general model, wherein the subtask-specific model is to perform or assist in performing a subtask that enables machine learning; The first model training unit receives the training data of the subtask specific model from one or more subtask specific model collectors; modifying, by the first model training unit, the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model, and the training data received from the one or more subtask-specific model collectors; as well as The modified sub-task specific model is sent by the first model training unit to the first wireless node.

2. The method of claim 1, wherein the modification comprises at least one of: modifying one or more weights of the subtask specific model; training the subtask specific model; retraining the subtask specific model; configuring or updating one or more weights or parameters of the subtask specific model; and / or Upgrade or downgrade the subtask specific model.

3. The method according to any one of claims 1 to 2, wherein the receiving a trigger indication comprises: The first model training unit receives one or more of the following from the second model training unit: The information of the general model includes one or more of the following information: the architecture of the general model, the weight of the general model, the loss function and / or activation function of the general model, and / or the output type of the general model; subtask parameterization, including constraints of the subtask-specific model, or subtask-specific cost functions of the subtask-specific model; and / or The one or more subtask-specific model collector identifiers.

4. The method according to any one of claims 1 to 3, wherein the modification of the sub-task specific model by the first model training unit based on the general model comprises: Based on the one or more constraints of the general model and the subtask-specific model, one or more of the following are performed: pruning or reducing the size of the generic model so that the subtask specific models will fit within a maximum allowed subtask specific model constraint; deactivating one or more inputs of the generic model so that the input of the subtask specific model depth or size will adapt to the format, size or depth of the training data received from the one or more subtask specific model collectors; Replace the generic model activation function with a subtask-specific model-specific activation function; or Define a cost function specific to the subtask-specific model.

5. The method according to any one of claims 1 to 4, further comprising: The modified sub-task specific model is sent by the first model training unit to the second model training unit.

6. The method according to any one of claims 1 to 5, further comprising: The modified sub-task specific model is sent by the first model training unit to at least one sub-task specific model collector among the one or more sub-task specific model collectors.

7. The method according to any one of claims 1 to 6, wherein any one of the following: The first model training unit and the second model training unit are arranged in a second wireless node; or The first model training unit is arranged in a second wireless node, and the second model training unit is arranged in a third wireless node.

8. The method according to any one of claims 1 to 7, further comprising: A request for sub-task specific model training or meta-learning of the sub-task is sent by the first model training unit to the second model training unit.

9. The method according to any one of claims 1 to 8, wherein: The second model training unit includes: a general model training unit configured to modify the general model for a general task; and The first model training unit is a meta-learning model training unit or a specific model training unit configured to modify a specific model or a subtask specific model for the subtask.

10. The method of any one of claims 1 to 8, wherein one or more of the first wireless node, the second wireless node, or the third wireless node comprises at least one of the following: User equipment, UE, base station, or gNB.

11. An apparatus comprising: at least one processor; as well as at least one memory including computer program code; The at least one memory and the computer program code are configured to, with the at least one processor, cause the apparatus to at least: Receiving, by the first model training unit from the second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of the subtask-specific model to be used to modify the subtask or the subtask-specific model based on the general model, wherein the subtask-specific model is to perform or assist in performing a subtask that enables machine learning; The first model training unit receives the training data of the subtask specific model from one or more subtask specific model collectors; modifying, by the first model training unit, the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model, and the training data received from the one or more subtask-specific model collectors; as well as The modified sub-task specific model is sent by the first model training unit to the first wireless node.

12. An apparatus comprising: means for receiving, by a first model training unit, from a second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a generic model trained by the second model training unit for a generic task, and means for modifying a subtask of the subtask-specific model or one or more parameters of the subtask-specific model based on the generic model, wherein the subtask-specific model is to perform or assist in performing a subtask that enables machine learning; means for receiving, by the first model training unit, training data of the subtask specific model from one or more subtask specific model collectors; means for modifying, by the first model training unit, the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model, and the training data received from the one or more subtask-specific model collectors; as well as Means for sending, by the first model training unit, the modified sub-task specific model to a first wireless node.

13. A non-transitory computer-readable storage medium comprising instructions stored thereon, which instructions, when executed by at least one processor, are configured to cause a computing system to: Receiving, by the first model training unit from the second model training unit, a trigger indication for triggering or causing modification of a subtask-specific model, an indication of a general model trained by the second model training unit for a general task, and one or more parameters of the subtask-specific model to be used to modify the subtask or the subtask-specific model based on the general model, wherein the subtask-specific model is to perform or assist in performing a subtask that enables machine learning; The first model training unit receives the training data of the subtask specific model from one or more subtask specific model collectors; modifying, by the first model training unit, the subtask-specific model based on the general model, the one or more parameters of the subtask or the subtask-specific model, and the training data received from the one or more subtask-specific model collectors; as well as The modified sub-task specific model is sent by the first model training unit to the first wireless node.

14. A method comprising: receiving a request to modify, or otherwise determine a need for, a subtask-specific model for a first wireless node based on the general model, wherein the subtask-specific model will perform or assist in performing a machine learning-enabled subtask; The second model training unit sends to the first model training unit a trigger indication for triggering or causing the first model training unit to modify the subtask specific model, an indication of the general model trained by the second model training unit for the general task, and one or more parameters of the subtask or subtask specific model to be used to modify the subtask specific model based on the general model; The second model training unit configures one or more sub-task specific model collectors to provide the first model training unit with training data for modifying the sub-task specific model; as well as The modified sub-task-specific model modified by the first model training unit is received by the second model training unit from the first model training unit.

15. The method of claim 14, wherein the sending comprises: The second model training unit sends one or more of the following to the first model training unit: The information of the general model includes one or more of the following information: the architecture of the general model, the weight of the general model, the loss function and / or activation function of the general model, and / or the output type of the general model; subtask parameterization, including constraints of the subtask-specific model, or subtask-specific cost functions of the subtask-specific model; and / or The one or more subtask-specific model collector identifiers.

16. The method according to any one of claims 14 to 15, wherein any one of the following: The first model training unit and the second model training unit are arranged in a second wireless node; or The first model training unit is arranged in a second wireless node, and the second model training unit is arranged in a third wireless node.

17. The method according to any one of claims 14 to 16, wherein receiving a request for modification of a subtask specific model, or otherwise determining a need for the same, comprises: receiving, by the second model training unit from the first model training unit, a request to modify or train the subtask-specific model for the subtask; as well as The request to modify or train the subtask-specific model for the subtask is validated.

18. A method according to any one of claims 14 to 17, wherein: The second model training unit includes: a general model training unit configured to modify or train the general model for a general task; and The first model training unit is a meta-learning model training unit or a specific model training unit configured to modify or train a specific model or a subtask specific model for the subtask.

19. The method of any one of claims 14 to 18, wherein one or more of the first wireless node, the second wireless node, or the third wireless node comprises at least one of: User equipment, UE, base station, or gNB.

20. An apparatus comprising: at least one processor; as well as at least one memory including computer program code; The at least one memory and the computer program code are configured to, with the at least one processor, cause the apparatus to at least: receiving a request to modify, or otherwise determine a need for, a subtask-specific model for a first wireless node based on the general model, wherein the subtask-specific model will perform or assist in performing a machine learning-enabled subtask; The second model training unit sends to the first model training unit a trigger indication for triggering or causing the first model training unit to modify the subtask specific model, an indication of the general model trained by the second model training unit for the general task, and one or more parameters of the subtask or subtask specific model to be used to modify the subtask specific model based on the general model; The second model training unit configures one or more sub-task specific model collectors to provide the first model training unit with training data for modifying the sub-task specific model; as well as The modified sub-task-specific model modified by the first model training unit is received by the second model training unit from the first model training unit.

21. An apparatus comprising: means for receiving a request to modify, or otherwise determining a need for, a subtask-specific model for a first wireless node based on a generic model, wherein the subtask-specific model is to perform or assist in performing a machine learning-enabled subtask; means for sending, by the second model training unit, to the first model training unit, a trigger indication for triggering or causing the first model training unit to modify a subtask-specific model, an indication of the general model trained by the second model training unit for a general task, and one or more parameters of the subtask or subtask-specific model to be used to modify the subtask-specific model based on the general model; means for configuring, by the second model training unit, one or more sub-task specific model collectors to provide the first model training unit with training data for modifying the sub-task specific model; as well as Means for receiving, by the second model training unit, from the first model training unit, the modified sub-task specific model modified by the first model training unit.

22. A non-transitory computer-readable storage medium comprising instructions stored thereon, which instructions, when executed by at least one processor, are configured to cause a computing system to: receiving a request to modify, or otherwise determine a need for, a subtask-specific model for a first wireless node based on the general model, wherein the subtask-specific model will perform or assist in performing a machine learning-enabled subtask; The second model training unit sends to the first model training unit a trigger indication for triggering or causing the first model training unit to modify the subtask specific model, an indication of the general model trained by the second model training unit for the general task, and one or more parameters of the subtask or subtask specific model to be used to modify the subtask specific model based on the general model; The second model training unit configures one or more sub-task specific model collectors to provide the first model training unit with training data for modifying the sub-task specific model; as well as The modified sub-task-specific model modified by the first model training unit is received by the second model training unit from the first model training unit.

23. A method comprising: Determining, by a user device, a trained generic model to perform or assist in performing a generic task that enables machine learning; Based on the trained general model, determining one or more general model-based outputs according to one or more signals or inputs; Sending, by the user equipment, the one or more general model-based outputs to a network node; receiving, by the user equipment, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, including configuration parameters of the subtask-specific model, wherein the subtask-specific model is to perform or assist in performing a machine learning-enabled subtask; validating said request for said subtask specific model; modifying, by the user device, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or the subtask-specific model; as well as performing or completing, by the user device, a machine learning-enabled subtask based on or using the modified subtask-specific model; as well as The subtask-specific model output is sent by the user equipment to the network node based on performing or completing the machine learning enabled subtask according to or using the modified subtask-specific model.

24. The method of claim 23, wherein the modification comprises at least one of: modifying one or more weights of the subtask specific model; training the subtask specific model; retraining the subtask specific model; configuring or updating one or more weights or parameters of the subtask specific model; and / or Upgrade or downgrade the subtask specific model.

25. The method of any one of claims 23 to 24, wherein said validating said request for said subtask specific model comprises: Verify at least one of the following for the subtask-specific model: The requested subtask-specific model is on the list of allowed subtask-specific models; A threshold amount of training data and / or input signals may be used to train the subtask specific model; and / or A threshold amount of processor resources and / or memory resources may be used to train and / or use the sub-task specific model.

26. The method according to any one of claims 23 to 25, wherein receiving a request for a subtask specific model, including configuration parameters of the subtask specific model, comprises receiving: A trigger indication for triggering or causing meta-learning or training of a subtask-specific model, and one or more parameters of a subtask or subtask-specific model to be used for training the subtask-specific model based on the general model, comprising receiving a subtask parameterization, the subtask parameterization comprising: constraints of the subtask specific model, or a subtask specific cost function of the subtask specific model.

27. The method according to any one of claims 23 to 26, wherein the configuration parameters of the subtask or the subtask-specific model include: The one or more constraints of the subtask-specific model, and wherein the modification of the subtask-specific model by the user device based on the general model comprises: performing one or more of the following based on the general model and the one or more constraints of the subtask-specific model: pruning or reducing the size of the generic model so that the subtask specific models will fit within a maximum allowed subtask specific model depth or size; deactivating one or more inputs of the generic model so that the input of the subtask specific model depth or size will adapt to the format, size or depth of the training data received from the one or more subtask specific model collectors; Replace the generic model activation function with a subtask-specific model-specific activation function; or Define a cost function specific to the subtask-specific model.

28. The method of any one of claims 23 to 27, wherein determining one or more generic model-based outputs comprises: Receiving, by the user equipment from a network node, a request to train a generic model for the generic task enabling machine learning; Training the general model by the user equipment based on a configuration or input received from the network node; as well as Use the trained general model to perform or complete the general task enabling machine learning to obtain the one or more general model-based outputs.

29. The method of claim 28, wherein the request for a sub-task-specific model is received by the user device in response to the user device sending the one or more general model-based outputs of the trained general model to the network node.

30. An apparatus comprising: at least one processor; as well as at least one memory including computer program code; The at least one memory and the computer program code are configured to, with the at least one processor, cause the apparatus to at least: Determining, by a user device, a trained generic model for performing or assisting in performing a generic task that enables machine learning; Based on the trained general model, determining one or more general model-based outputs according to one or more signals or inputs; Sending, by the user equipment, the one or more general model-based outputs to a network node; receiving, by the user equipment, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, including configuration parameters of the subtask-specific model, wherein the subtask-specific model is to perform or assist in performing a machine learning-enabled subtask; validating said request for said subtask specific model; modifying, by the user device, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or the subtask-specific model; as well as performing or completing, by the user device, a machine learning-enabled subtask based on or using the modified subtask-specific model; as well as The subtask-specific model output is sent by the user equipment to the network node based on performing or completing the machine learning enabled subtask according to or using the trained subtask-specific model.

31. An apparatus comprising: means for determining, by a user device, a trained generic model for performing or assisting in performing a generic task enabling machine learning; means for determining one or more general model-based outputs based on the trained general model and from one or more signals or inputs; means for sending, by the user equipment, the one or more general model-based outputs to a network node; means for receiving, by the user equipment, a request from the network node for a subtask-specific model based at least in part on the one or more generic model-based outputs, including configuration parameters of the subtask-specific model, wherein the subtask-specific model is to perform or assist in performing the machine learning-enabled subtask; means for validating said request for said subtask specific model; means for modifying, by the user device, the subtask specific model based on the trained generic model and the configuration parameters of the subtask or the subtask specific model; as well as means for performing or completing, by the user device, a machine learning-enabled subtask based on or using the modified subtask-specific model; as well as means for sending, by the user equipment, a subtask-specific model output to the network node based on performing or completing the machine learning enabled subtask in accordance with or using the trained subtask-specific model.

32. A non-transitory computer-readable storage medium comprising instructions stored thereon, which instructions, when executed by at least one processor, are configured to cause a computing system to: Determining, by a user device, a trained generic model for performing or assisting in performing a generic task that enables machine learning; Based on the trained general model, determining one or more general model-based outputs according to one or more signals or inputs; Sending, by the user equipment, the one or more general model-based outputs to a network node; receiving, by the user equipment, a request for a subtask-specific model from the network node based at least in part on the one or more general model-based outputs, including configuration parameters of the subtask-specific model, wherein the subtask-specific model is to perform or assist in performing a machine learning-enabled subtask; validating said request for said subtask specific model; modifying, by the user device, the subtask-specific model based on the trained general model and the configuration parameters of the subtask or the subtask-specific model; as well as performing or completing, by the user device, a machine learning-enabled subtask based on or using the modified subtask-specific model; as well as The subtask-specific model output is sent to the network node by the user equipment based on performing or completing the machine learning enabled subtask according to or using the trained subtask-specific model.

33. A method comprising: Determining, by a network node, a trained general purpose model to perform or assist in performing a general purpose task that enables machine learning; Providing, by the network node, the trained general model to a user equipment; Receiving, by the network node, a request for a subtask specific model from the user equipment; validating said request for said subtask specific model; Sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user equipment; Receiving, by the network node from the user equipment, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data; modifying, by the network node, the subtask specific model based on the general model and at least one of the subtask specific model configuration or constraint and / or subtask specific model training data; as well as The modified sub-task specific model is sent by the network node to the user equipment.

34. The method of claim 33, wherein the modification comprises at least one of: modifying one or more weights of the subtask specific model; training the subtask specific model; retraining the subtask specific model; configuring or updating one or more weights or parameters of the subtask specific model; and / or Upgrade or downgrade the subtask specific model.

35. The method of any one of claims 33 to 34, wherein said validating said request for said subtask-specific model comprises: Verify at least one of the following for the subtask-specific model: The requested subtask-specific model is on the list of allowed subtask-specific models; A threshold amount of training data and / or input signals may be used to train the subtask specific model; and / or A threshold amount of processor resources and / or memory resources may be used to train and / or use the sub-task specific model.

36. The method according to any one of claims 33 to 35, wherein the modification of the subtask specific model by the network node based on the general model comprises: Based on the general model and one or more constraints of the subtask specific model, one or more of the following are performed: pruning or reducing the size of the generic model so that the subtask specific models will fit within a maximum allowed subtask specific model depth or size; deactivating one or more inputs of the generic model so that the input of the subtask specific model depth or size will adapt to the format, size or depth of the training data received from the one or more subtask specific model collectors; Replace the generic model activation function with a subtask-specific model-specific activation function; or Define a cost function specific to the subtask-specific model.

37. An apparatus comprising: at least one processor; as well as at least one memory including computer program code; The at least one memory and the computer program code are configured to, with the at least one processor, cause the apparatus to at least: Determining, by a network node, a trained general purpose model to perform or assist in performing a general purpose task that enables machine learning; Providing, by the network node, the trained general model to a user equipment; Receiving, by the network node, a request for a subtask specific model from the user equipment; validating said request for said subtask specific model; Sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user equipment; Receiving, by the network node from the user equipment, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data; modifying, by the network node, the subtask specific model based on the general model and at least one of the subtask specific model configuration or constraint and / or subtask specific model training data; as well as The modified sub-task specific model is sent by the network node to the user equipment.

38. An apparatus comprising: Means for determining, by a network node, a trained generic model to perform or assist in performing a generic task that enables machine learning; means for providing, by the network node, the trained generic model to a user equipment; means for receiving, by the network node, a request for a subtask specific model from the user equipment; means for validating said request for said subtask specific model; means for sending, by the network node to the user equipment, a request for at least one of a subtask specific model configuration or constraint and / or subtask specific model training data; means for receiving, by the network node from the user equipment, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data; means for modifying, by the network node, the subtask specific model based on the general model and at least one of the subtask specific model configuration or constraints and / or subtask specific model training data; as well as Means for sending, by the network node, the modified sub-task specific model to the user equipment.

39. A non-transitory computer-readable storage medium comprising instructions stored thereon, which instructions, when executed by at least one processor, are configured to cause a computing system to: Determining, by a network node, a trained general purpose model to perform or assist in performing a general purpose task that enables machine learning; Providing, by the network node, the trained general model to a user equipment; Receiving, by the network node, a request for a subtask specific model from the user equipment; validating said request for said subtask specific model; Sending, by the network node, a request for at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data to the user equipment; Receiving, by the network node from the user equipment, at least one of a subtask-specific model configuration or constraint and / or subtask-specific model training data; modifying, by the network node, the subtask specific model based on the general model and at least one of the subtask specific model configuration or constraint and / or subtask specific model training data; as well as The modified sub-task specific model is sent by the network node to the user equipment.

Images