A method for reducing diversity in two-sided machine learning models

Abstract

An apparatus is provided, comprising means for performing the following: transmitting information regarding at least one two-sided machine learning model supported by the device to a network node; or transmitting a machine learning profile identity of the device to a network node, wherein the machine learning profile identity is associated with at least one two-sided machine learning model supported by the device, and the at least one two-sided machine learning model is configured to enable joint inference by the device and the network node; and receiving from the network node an indication of at least a first selected two-sided machine learning model supported by the device.

Description

[Technical Field]

[0001] Various example embodiments relate to methods for reducing diversity in two-sided machine learning models inferred by user equipment and network nodes. [Background technology]

[0002] In communication networks, machine learning (ML) techniques may be used for a variety of tasks. A two-sided ML model is one in which joint inference is performed across user equipment and network nodes. Summary of the Invention

[0003] According to some aspects, the subject matter of the independent claims is provided. Some example embodiments are defined in the dependent claims. The scope of protection claimed for the various example embodiments is defined by the independent claims. Example embodiments and features described herein that do not fall within the scope of the independent claims, if any, should be interpreted as examples that are helpful for understanding the various example embodiments.

[0004] According to a fourth aspect, there is provided a method including: transmitting, by a device, to a network node information regarding at least one two-sided machine learning model supported by the device; or transmitting a machine learning profile identity of the device to the network node, wherein the machine learning profile identity is associated with at least one two-sided machine learning model supported by the device, and the at least one two-sided machine learning model is configured to enable joint inference by the device and the network node; and receiving, by the device, from the network node, an indication of at least a first selected two-sided machine learning model supported by the device.

[0005] According to an embodiment, the method includes initiating joint inference of the first selected two-sided machine learning model with the network node.

[0006] According to an embodiment, the information about the at least one double-sided machine learning model includes an identity of the at least one double-sided machine learning model supported by the device and information about the capabilities of the at least one double-sided machine learning model supported by the device.

[0007] According to an embodiment, the method includes receiving, from a network node, an instruction to switch to a second two-sided machine learning model supported by the device, the second two-sided machine learning model also supported by another device serviced by the network node.

[0008] According to an embodiment, the method includes switching to a second two-sided machine learning model and initiating joint inference of the second two-sided machine learning model with the network node.

[0009] According to an embodiment, the method includes determining that the second two-sided machine learning model is no longer supported by the device, and based on this determination, sending a negative acknowledgement to the network node, and continuing joint inference of the first selected two-sided machine learning model.

[0010] According to an embodiment, the method includes: - establishing a secondary connection with another network node; - transmitting information about at least one two-sided machine learning model supported by the device to the other network node, or transmitting a machine learning profile identity of the device to the other network node, where the machine learning profile identity is associated with the at least one two-sided machine learning model supported by the device; - initiating joint inference of the first selected two-sided machine learning model with the other network node.

[0011] According to an embodiment, the method includes receiving, from a network node, an instruction to switch to a second two-sided machine learning model supported by the device, where the second two-sided machine learning model is also supported by another network node; switching to the second two-sided machine learning model; and initiating joint inference of the second two-sided machine learning model with the network node and the other network node.

[0012] According to an embodiment, the method includes receiving, from another network, an indication of a second selected two-sided machine learning model while performing joint inference of a first selected two-sided machine learning model with the network node, and transmitting to the other network a negative acknowledgement and an indication that the device is performing joint inference of the first selected two-sided machine learning model with the network node.

[0013] According to an embodiment, the method includes receiving, from another network node, an instruction to initiate joint inference using the first selected two-sided machine learning model or an instruction to reject joint inference using the first selected two-sided machine learning model.

[0014] According to a fifth aspect, there is provided a method comprising: - receiving, by a network node, from a user equipment, information regarding at least one double-sided machine learning model supported by the user equipment; or - receiving, by the network node, from the user equipment, a machine learning profile identity of the user equipment, wherein the machine learning profile identity is associated with at least one double-sided machine learning model supported by the user equipment; and - obtaining, by the network node, information regarding the at least one double-sided machine learning model supported by the user equipment based on the machine learning profile identity of the user equipment, wherein the at least one double-sided machine learning model is configured to enable joint inference by the device and the user equipment; selecting, by the network node, a first double-sided machine learning model supported by the user equipment; and sending, by the network node, an instruction to the user equipment of selecting the at least first double-sided machine learning model supported by the user equipment.

[0015] According to an embodiment, the method includes initiating joint inference of a first two-sided machine learning model with a user device.

[0016] According to an embodiment, selecting the first two-sided machine learning model is performed based on other two-sided machine learning models currently jointly inferred by the device and other user equipment or supported by other user equipment.

[0017] According to an embodiment, the method includes: - receiving, from another user equipment serviced by the device, information regarding at least one double-sided machine learning model supported by the other user equipment; or - receiving, from another user equipment serviced by the device, a machine learning profile identity of the other user equipment, wherein the machine learning profile identity is associated with at least one double-sided machine learning model supported by the other user equipment; and obtaining information regarding the at least one double-sided machine learning model supported by the other user equipment based on the machine learning profile identity of the other user equipment.

[0018] According to an embodiment, the method includes determining a second two-sided machine learning model supported by both user equipment based on information about at least one two-sided machine learning model supported by the user equipment and by another user equipment, and sending an instruction to the user equipment to switch to the second two-sided machine learning model.

[0019] According to an embodiment, the method includes sending an instruction to another user device to select a second two-sided machine learning model, and initiating joint inference of the second two-sided machine learning model with the user device and the other user device.

[0020] According to an embodiment, the method includes receiving a negative acknowledgment from the user equipment indicating that the second two-sided machine learning model is no longer supported by the user equipment, continuing to infer the first two-sided machine learning model with the user equipment, and initiating joint inference of the second two-sided machine learning model with another user equipment.

[0021] According to an embodiment, the method includes receiving a request from another network node serving the user equipment for information regarding a two-sided machine learning model to be used for the user equipment, determining that a first two-sided machine learning model matches the capabilities of the other network node, and sending an indication to the other network node to select the first two-sided machine learning model.

[0022] According to an embodiment, the method includes receiving a request from another network node serving the user equipment for information regarding a two-sided machine learning model used for the user equipment, determining that a first two-sided machine learning model is incompatible with the capabilities of the other network node, selecting a second two-sided machine learning model that is compatible with the capabilities of the user equipment and the capabilities of the other network node based on the determination, and sending an instruction to the user equipment to switch to the second two-sided machine learning model.

[0023] According to an embodiment, the method includes sending an indication to another network node of selecting a second two-sided machine learning model, and initiating inference of the second two-sided machine learning model together with the user equipment and the other network node.

[0024] According to a sixth aspect, there is provided a method comprising: establishing a connection with a user equipment, whereby an apparatus is configured to function as a secondary network node; receiving information from a master network node configured to provide service to the user equipment regarding a two-sided machine learning model that matches capabilities of the user equipment and capabilities of the secondary network node; and initiating joint inference of the two-sided machine learning model together with the user equipment and the master network node.

[0025] According to one aspect, there is provided an apparatus comprising means for carrying out the method of the fourth aspect and any of its embodiments.

[0026] According to one aspect, there is provided an apparatus comprising means for carrying out the method of the fifth aspect and any of its embodiments.

[0027] According to one aspect, there is provided an apparatus comprising means for carrying out the method of the sixth aspect and any of its embodiments.

[0028] According to an embodiment, the means (in any of the previous aspects) comprises at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to run.

[0029] According to one aspect, there is provided a non-transitory computer-readable medium containing instructions that, when executed by an apparatus, cause the apparatus to perform at least the method of the fourth aspect and any of its embodiments.

[0030] According to one aspect, there is provided a non-transitory computer readable medium containing instructions that, when executed by an apparatus, cause the apparatus to perform at least the method of the fifth aspect and any of its embodiments.

[0031] According to one aspect, there is provided a non-transitory computer-readable medium containing instructions that, when executed by an apparatus, cause the apparatus to perform at least the method of the sixth aspect and any of its embodiments.

[0032] According to one aspect, there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform the method of the fourth aspect and any of its embodiments.

[0033] According to one aspect, there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform the method of the fifth aspect and any of its embodiments.

[0034] According to one aspect, there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform the method of the sixth aspect and any of its embodiments.

[0035] Some example embodiments will now be described with reference to the accompanying drawings. [Brief explanation of the drawings]

[0036] [Figure 1] 1 illustrates, by way of example, a network architecture of a communication system; [Figure 2a] FIG. 1 illustrates, by way of example, a network node serving multiple user equipments. [Figure 2b] FIG. 1 illustrates, by way of example, a user equipment in multiple connection mode. [Figure 3] FIG. 1 shows a flowchart of a method, by way of example. [Figure 4] FIG. 1 shows a flowchart of a method, by way of example. [Figure 5] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 6] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 7a] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 7b] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 8] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 9] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 10] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 11] FIG. 1 illustrates, by way of example, signaling between entities. [Figure 12] FIG. 1 shows, by way of example, a block diagram of an apparatus. DETAILED DESCRIPTION OF THE INVENTION

[0037] 1 shows, by way of example, a network architecture of a communication system. In the following, various exemplary embodiments are described using a radio access architecture based on New Radio (NR), also known as Long Term Evolution Advanced (LTE-A) or fifth generation (5G), as an example of an access architecture to which the embodiments may be applied, without limiting the embodiments to such an architecture. It will be clear to those skilled in the art that the embodiments may also be applied to other types of communication networks including suitable means by appropriately adjusting parameters and procedures.Some examples of other options for suitable systems are universal mobile telecommunications system (UMTS) radio access networks (UMTS radio access network (UTRAN) or E-UTRAN), long term evolution (LTE, same as E-UTRA), wireless local area networks (wireless local area networks (WLAN) or WiFi), worldwide interoperability for microwave access (WiMAX), Bluetooth®, personal communications services (PCS), ZigBee®, wideband code division multiple access (WCDMA), systems using ultra-wideband (UWB) technology, sensor networks, mobile ad-hoc networks (MANET), and Internet Protocol multimedia subsystems (IMS), or any combination thereof.

[0038] The example of Figure 1 illustrates a portion of an exemplary radio access network. Figure 1 shows user devices or user equipments (UE) 100 and 102 configured to connect wirelessly over one or more communication channels within a cell, which is served by an access node 104, such as a gNB, i.e., next generation Node B, or eNB, i.e., evolved Node B (eNodeB). The physical link from the user device to the network node is referred to as the uplink (UL) or reverse link, and the physical link from the network node to the user device is referred to as the downlink (DL) or forward link. It should be understood that network nodes or their functions may be implemented using any node, host, server, or access point, or other entity suitable for such use. A communication system typically comprises two or more network nodes, in which case the network nodes may also be configured to communicate with each other via wired or wireless links designed for this purpose. These links may also be used for signaling purposes. A network node is a computing device configured to control the radio resources of the communication system to which the network node is coupled. A network node may also be referred to as a base station (BS), an access point, or any other type of interfacing device capable of operating in a wireless environment, including a relay station. The network node includes or is coupled to a transceiver. A connection from the network node's transceiver is provided to an antenna unit to establish a bidirectional wireless link with a user device. The antenna unit may include multiple antennas or antenna elements. The network node is further connected to a core network 110 (core network (CN) or next generation core (NGC)).Depending on the system, the corresponding part on the CN side can be a serving gateway (S-GW (serving gateway) that routes and forwards user data packets), a packet data network gateway (P-GW) for providing a connection of a user device (UE) to an external packet data network, or a mobile management entity (MME), etc. An example of a network node configured to operate as a relay station is an integrated access and backhaul node (IAB). The distributed unit (DU) part of the IAB node performs the BS function of the IAB node, while the backhaul connection is performed by the mobile termination (MT) part of the IAB node. The UE function may be performed by the IAB MT, and the BS function may be performed by the IAB DU. The network architecture may include a parent node, i.e., an IAB donor, which can have a wired connection with the CN and a wireless connection with the IAB MT.

[0039] User devices or user equipment (UE) typically refer to portable computing devices, including wireless mobile communication devices that operate with or without a subscriber identification module (SIM), including, but not limited to, the following types of devices: mobile stations (cell phones), smartphones, personal digital assistants (PDAs), handsets, devices that use wireless modems (such as alarms or measuring devices), laptop and / or touchscreen computers, tablets, game consoles, notebooks, and multimedia devices. It should be understood that user devices can also be almost exclusively uplink-only devices, an example of which is a camera or camcorder that loads images or video clips onto the network. User devices can also be devices capable of operating within an Internet of Things (IoT) network, where objects have the ability to transfer data over a network without the need for human-to-human or human-to-computer interaction.

[0040] Furthermore, although the devices are shown as single entities, various units, processors and / or memory units (not all of which are shown in FIG. 1) may be implemented within these devices to enable their functionality.

[0041] 5G will incorporate multiple input - multiple output (MIMO) technology on both the UE and gNB side, allowing for the use of many more base stations or nodes than in LTE (the so-called small cell concept), working in cooperation with smaller base stations, and employing different radio technologies depending on the service need, use case, and / or available spectrum, including macro sites. 5G mobile communications will also support video streaming, augmented reality, various methods of data sharing, and various forms of machine-type applications (vehicle safety, various sensors, and real-time control) including (massive) machine-type communications (mMTC). 5G will support a wide range of use cases and related applications, including sub-7 GHz, centimeter wave, and millimeter wave. 5G is expected to include multiple air interfaces, namely, sub-7 GHz, centimeter wave, and millimeter wave, and will also be able to integrate with existing conventional radio access technologies such as LTE. The sub-7 GHz frequency range is sometimes referred to as FR1, and the frequency range above 24 GHz (or more precisely, 24-52.6 GHz) is sometimes referred to as FR2. Integration with LTE, at least initially, may be implemented as a system in which macro coverage is provided by LTE and 5G air interface access results from small cells through aggregation to LTE. In other words, 5G is planned to support both inter-RAT (e.g., LTE-5G) and inter-RI (e.g., inter-air interface operation between sub-7 GHz and centimeter wave, sub-7 GHz-centimeter wave-mm wave, etc.). One concept envisioned for use in 5G networks is network slicing, in which multiple independent and dedicated virtual subnetworks (network instances) may be created within the same infrastructure to run services with different requirements in terms of latency, reliability, throughput, and mobility.

[0042] The communication system may also communicate with or use services provided by other networks, such as the public switched telephone network or the Internet 112. The communication network may also support the use of cloud services, e.g., at least some of the operations of the core network may be performed as cloud services (this is illustrated by "cloud" 114 in FIG. 1). The communication system may also include a central control entity that provides functions for networks of different operators, e.g., to cooperate in spectrum sharing.

[0043] By utilizing network function virtualization (NVF) and software defined networking (SDN), an edge cloud may be brought to the radio access network (RAN). Using an edge cloud may mean that the operation of an access node is at least partially performed in a server, host, or node operatively coupled to a remote radio head or base station containing a radio part. It is also possible that the operation of a node is distributed among multiple servers, nodes, or hosts. The application of a Cloud RAN architecture allows the real-time functions of the RAN to be performed on the RAN side (in the distributed unit DU104) and the non-real-time functions to be performed in a centralized way (in the centralized unit CU108).

[0044] 5G may also utilize satellite communications to enhance or complement 5G service coverage, for example, by providing backhauling. Potential use cases include providing service continuity to machine-to-machine (M2M) or Internet of Things (IoT) devices, or to vehicle passengers, or ensuring service availability for critical communications and future rail, maritime, and aviation communications. Satellite communications may utilize geostationary earth orbit (GEO) satellite systems, but also low earth orbit (LEO) satellite systems, especially in megaconstellations (systems with hundreds of (small) satellites). Each satellite 106 in the constellation may serve multiple satellite-enabled network entities that create ground cells. Ground cells may be created by terrestrial relay nodes 104 or by gNBs located on the ground or on the satellite.

[0045] Radio access network (RAN) optimization algorithms may include algorithms for optimizing and / or improving RAN operation, performance, and / or one or more functions. RAN optimization may include, for example, increasing or decreasing the priority of a service. RAN optimization aimed at improving end-user perception includes, for example, capacity and coverage optimization, load distribution, load balancing, random access channel (RACH) optimization, and energy conservation. These functions may be optimized by self-organizing network (SON) algorithms. SON refers to the network's ability to function in a self-organizing manner. Modern RAN functions are expected to have, for example, self-planning, self-configuring, and self-optimizing capabilities.

[0046] The radio access network optimization algorithm may be implemented using, for example, machine learning techniques, the use of ML also being applicable in SON solutions.

[0047] Machine learning (ML) refers to algorithms and statistical models used by computer systems to perform specific tasks without explicit instructions, relying instead on patterns and inference. Machine learning (ML) is considered a subset of artificial intelligence (AI). Machine learning algorithms build mathematical models based on sample data, known as "training data," to make predictions or decisions without being explicitly programmed to perform a task. ML algorithms may be categorized, for example, as supervised learning, unsupervised learning, and reinforcement learning. ML algorithms are composed of one or more ML components that form a so-called ML pipeline, each of which may be located and executed in a different RAN network function and / or in the UE itself.

[0048] A two-sided ML model refers to a paired AI / ML model in which joint inference may be performed across the UE and the network. For example, first, a first part of the inference may be performed by the UE, and then the remaining part is performed by a network node, e.g., a gNB, or vice versa.

[0049] An example of a two-sided ML model is channel state information (CSI) feedback compression using an autoencoder. The encoder may be implemented by the UE and the decoder may be implemented by a network node, e.g., a gNB. The encoder receives the CSI H as input. The total number of feedback parameters in H is N t ×N r ×N c and N t is the number of transmit antennas at the gNB, and N r is the number of receiver antennas at the UE, and Nc is the number of subcarriers on which an orthogonal frequency-division multiplexing (OFDM) system operates. The neural network in the encoder portion compresses H into a smaller size codeword, S. In one example, S can be 64 bits. The neural network in the decoder portion receives S as input and recovers the H matrix.

[0050] The autoencoder may be trained as a whole, or the encoder and decoder portions may be trained separately. In the case of separate training, the encoder portion may be trained, for example, using an unsupervised algorithm. After the training of the encoder is accomplished, the generated labeled data may be used for supervised training of the decoder.

[0051] CSI feedback compression using an autoencoder that deploys a two-sided ML model has been shown to improve compression speed.

[0052] FIG. 2a illustrates, by way of example, a network node serving multiple user equipment (UEs), e.g., UE1 220 and UE X 225. The network node 210, e.g., a gNB, may perform full or joint inference for various UE functions using ML models hosted at the network node 210 or split between the UEs 220, 225 and the network node 210. With a large number of UEs and a large number of functions, the number of ML models running at the network node 210 can become significant. Each ML model 1....X consumes resources 230 at the network node 210, including at least random access memory (RAM), and ML accelerator capacity. The large resource consumption by ML models can cause significant and unpredictable demands on hardware resources, and in some cases, their scarcity.

[0053] 2b shows a user equipment in multi-connection mode, for example. The UE 240 may be in multi-connection mode with multiple access points (APs), e.g., AP1 250 and APX 255, or network nodes 250, 255. The UE may perform full inference or joint inference for various UE functions using ML models hosted on the UE 240 or split between the UE 240 and the access points 250, 255. Running different ML models for the same function for various connections consumes many resources 260 on the UE 240.

[0054] A method is provided for reducing the diversity of ML models running at a UE and / or at a network node at one time without impacting the performance of the feature.

[0055] FIG. 3 illustrates a flowchart of method 300, by way of example. The steps of the illustrated method 300 may be performed by a UE or by a control device configured to control the functions of the UE when installed therein. The UE may be, for example, device 510 of FIG. 5 or UE 240 of FIG. 2b, and is configured to at least perform method 300. The method includes transmitting, by the device, information regarding at least one two-sided machine learning model supported by the device to a network node (310a). Alternatively, the method may include transmitting, to the network node, a machine learning profile identity (or, equivalently, a machine learning profile identifier) ​​of the device, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the device (310b). Generally, the machine learning profile identity (or identifier) ​​is an identity or identifier that can be used to obtain information regarding the machine learning capabilities of the device (i.e., regarding the machine learning profile of the device), including a list of one or more two-sided machine learning models supported by the device. In other words, the machine learning profile identity (or identifier) ​​serves to identify the machine learning profile of the device (i.e., the machine learning capabilities of the device). The at least one two-sided machine learning model is configured to enable joint inference by the device and the network node. The method 300 includes receiving, by the device, from the network node, an indication of at least a first selected two-sided machine learning model supported by the device (320).

[0056] FIG. 4 illustrates a flowchart of method 400, by way of example. The steps of the illustrated method 400 may be performed by a network node or, when installed, by a control device configured to control the functions of a network node. The UE may be, for example, device 520 of FIG. 5 or network node 210 of FIG. 2b, e.g., a gNB, and is configured to perform at least method 400. Method 400 includes receiving, by the network node, information about at least one two-sided machine learning model supported by the user equipment from the user equipment (410a). Alternatively, method 400 includes receiving, by the network node, from the user equipment, a machine learning profile identity of the user equipment, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the user equipment, and obtaining, by the network node, information about the at least one two-sided machine learning model supported by the user equipment based on the user equipment machine learning profile identity (410b). The at least one two-sided machine learning model is configured to enable joint inference by the device and the user equipment. The method includes selecting, by the network node, a first two-sided machine learning model supported by the user equipment (420). The method includes transmitting, by the network node, to the user equipment an indication of selecting at least the first two-sided machine learning model supported by the user equipment (430).

[0057] The methods disclosed herein allow for reducing the diversity of two-sided ML models.

[0058] In the following, the methods of Figures 3 and 4 and their embodiments are described in the context of signaling diagrams.

[0059] FIG. 5 illustrates signaling between entities by way of example. The UE 510 transmits 530 information about at least one two-sided ML model supported by the UE 510 to the network node 520. In other words, the UE 510 provides information about its ML capabilities to the network node 520. The UE may have one or more ML-enabled features, e.g., best beam prediction in the spatial domain and CSI compression. Each of these features, e.g., features, may be associated with one or more ML models. The selection of an ML model for a feature may depend on requirements or input conditions. Requirements may include, for example, the size of the channel feedback after compression by the encoder, the computational complexity of the ML model, and the size of the ML model. The complexity of the ML model may be measured in terms of the number of floating-point operations (FLOPs). The size of the ML model may be given in megabytes.

[0060] For example, information about ML models may be included in a list. The list may include functions for which various ML models may be used. For example, the list may include function identities (IDs) and IDs of the supported ML models for the functions. For example, the list may include pairs of [function ID; supported ML model ID].

[0061] As an alternative to providing information about supported ML models, the UE may provide its ML profile identity, as described in the context of FIG.

[0062] The network node 520 stores in a memory information about the ML models supported by the UE. The received information may be paired with the UE ID of the UE 510. The UE ID includes, for example, information about the manufacturer or vendor of the UE, the model of the UE, and network identification information.

[0063] The network node 520 may receive information about ML models from multiple UEs connected to the network node 520. The network node 520 may maintain a database of ML models supported by different UEs for different functions.

[0064] The network node 520 selects (532) an ML model to be used for inference with the UE 510. The selected ML model may be a first two-sided ML model. This selection may be based on models specific to other UEs currently inferring with the network node 520. For example, the network node 520 may select an ML model currently being used by other UEs for the same function. This may reduce the diversity of two-sided ML models that are simultaneously inferred for the UE. Selecting an ML model to be used for the UE 510 that is already being used or supported by other UEs served by the network node 520 may reduce the diversity of ML models simultaneously running at the network node.

[0065] The network node 520 sends 534 to the UE 510 an indication of at least the selected ML model or at least a first two-sided ML model supported by the UE 510.

[0066] The UE 510 receives an indication of the selected ML model 534. The UE 510 initiates inference 536 of the first two-sided ML model with the network node 520.

[0067] 6 illustrates, by way of example, signaling between entities. UE1 610 transmits 630 its ML profile ID to network node 620. The ML profile ID is associated with at least one two-sided ML model supported by UE1 610. The ML profile ID may be, for example, in the form of a UE identifier, such as an international mobile equipment identity (IMEI). As another example, the ML profile ID may be a dedicated ID to indicate the ML model supported by the UE.

[0068] After receiving 630 the ML profile ID of UE1 610, network node 620 may be enabled to retrieve or obtain information regarding at least one two-sided ML model supported by UE1 610. Network node 620 may obtain 631 information regarding the ML models supported by UE1 610 based on the ML profile ID. This information may be included in UE1's ML profile.

[0069] The network node 620 may obtain the ML profile from a server, such as an operations, administration and maintenance (O&M) server 622. The O&M server 622 may receive a request from the network node 620 for an ML profile that corresponds to the ML profile ID of UE1 610. The O&M server 622 may obtain the requested information about UE1's ML capabilities from the vendor's server 622.

[0070] The network node 620 receives information about the ML models supported by UE1 from a server, for example, from an O&M server 622.

[0071] The network node 620 stores information about the ML models supported by UE1 610 in a memory.

[0072] The network node 620 selects (632) an ML model to be used for inference with UE1 610. The selected ML model may be a first two-sided ML model. This selection may be based on models specific to other UEs currently inferring with the network node 620. For example, the network node 620 may select an ML model currently being used by other UEs for the same function. This may reduce the diversity of two-sided ML models that are simultaneously inferred for the UE. Selecting an ML model to use for UE1 610 that is already being used or supported by other UEs served by the network node 620 may reduce the diversity of ML models simultaneously running at the network node.

[0073] The network node 620 sends (634) to the UE 610 an indication of at least the selected ML model or at least a first two-sided ML model supported by the UE 610.

[0074] UE1 610 receives an indication of the selected ML model 634. UE1 610 begins inferring 636 the first two-sided ML model with network node 620.

[0075] Figure 7a illustrates, by way of example, signaling between entities. In the example of Figure 7a, multiple UEs connect to a network node 720. For example, UE1 710 and UE2 722 join the same network node 720. In the example of Figure 7a, it is assumed that UE1 and UE2 support at least one ML model that is supported by both UE1 and UE2. This type of ML model that is supported by both UE1 and UE2 may be referred to as a common ML model.

[0076] The network node 720 receives 730 information about the ML models supported by UE1 710. For example, ML models 1 and 2 are supported by UE1 710.

[0077] The network node 720 stores the received information and performs an ML model selection 732. In the example of Figure 7a, ML model 1 is selected.

[0078] The network node 720 sends 734 an indication of the selected ML model to UE1 710.

[0079] UE1 710 and network node 720 initiate inference 736 of ML model 1.

[0080] The network node 720 receives 738 information about the ML models supported by another user equipment, UE2 722. For example, ML models 2 and 3 are supported by UE2 722.

[0081] Alternatively, the network node 720 may receive an ML profile ID for UE2 722, based on which the network node 720 may retrieve or obtain information about supported ML models from a server, as described in the context of FIG. 6.

[0082] The network node 720 performs the ML model selection based on information about the ML models supported by UE1 710 and UE2 722. In the example of Figure 7a, UE1 710 and the network node 720 implement two-sided ML model 1 (first two-sided ML model). However, UE2 722 does not support ML model 1.

[0083] The network node 720 determines a second two-sided ML model supported by both user equipment, i.e., UE1 and UE2, based on information about the two-sided ML models supported by UE1 710 and UE2 722. The network node 720 determines a common ML model that can be inferred for both UEs. In the example of Figure 7a, the common model is ML model 2 (the second two-sided ML model).

[0084] Network node 720 sends (742) an instruction to UE1 710 to switch to ML model 2. UE1 710 receives (742) the instruction to switch to another ML model supported by UE1, which ML model is also supported by another UE served by network node 720. UE1 710 switches to ML model 2 (the second two-sided ML model).

[0085] The network node 720 sends 744 an indication of the selected ML model, i.e., ML model 2, to UE2 722.

[0086] The network node 720 begins inferring ML model 2 with UE1 710 and UE2 722 (746). The network node 720 selects a common ML model for both UEs or multiple UEs to reduce resource utilization for inferring the ML models. The consistency of the ML models inferred for different UEs allows the network node 720 to conserve resources.

[0087] The network node 720 is assumed to maintain up-to-date information about the ML models supported by the UEs connected to the network node 720. The UE may notify the network node of any changes in the information about the ML models supported by the UE. An example where the network node 720 has an out-of-date list of the ML models for the UE is shown in Figure 7b.

[0088] Figure 7b shows, by way of example, signaling between entities. The signaling in Figure 7b continues from the signaling in Figure 7a. In the example of Figure 7b, UE1 receives an instruction to switch to ML model 2 (742) but determines that ML model 2 is no longer supported by UE1 710. Based on determining that ML model 2 is not supported by UE1 710, UE1 sends a negative acknowledgement NACK to network node 720.

[0089] UE1 710 continues inference of the first selected two-sided ML model with network node 720 (749).

[0090] Network node 720 receives 743 a NACK from UE1 710. This NACK indicates to network node 720 that ML model 2 is no longer supported by UE1 710. Network node 720 may update its database accordingly.

[0091] The network node 720 may decide to run different ML models for UE1 and UE2. The network node 720 may continue inferring ML model 1 with UE1 (749) and may begin inferring ML model 2 with UE2 722 (747), as shown at 744. To this end, the network 720 may verify that it has sufficient resources to run different ML models for different UEs. As an alternative to running different ML models for different UEs, the network 720 may decide to switch one or more UEs to a non-ML algorithm.

[0092] A UE may support two or more simultaneous connections with the network. A UE in a multiple connection mode is shown in the example of Figure 2b. For example, a radio access network (RAN) may support NR-NR dual connectivity (NR-DC), in which the UE may be connected to a network node, e.g., a gNB functioning as a master node (MN), and another network node, e.g., a gNB functioning as a secondary node (SN). The master node may be a first network node, and the secondary node may be a second network node.

[0093] As another example, NR-DC may be used when a UE is connected to a single network node that functions as both an MN and an SN and that constitutes both a master cell group (MCG) and a secondary cell group (SCG).

[0094] 8 illustrates, by way of example, signaling between entities: UE 810 is performing ML Model 1 inference 830 with Network Node 1 820, which may be a first network node or master node (MN).

[0095] The UE 810 establishes 832 a secondary connection with network node 2 822, which may be a second network node or secondary node (SN) or another network node. The UE 810 sends information about at least one two-sided ML model supported by the UE 810 to network node 2 822. Network node 2 822 may maintain a list of ML models supported by UEs connected to network node 2 822.

[0096] Network node 2 822 requests 834 from network node 1 information regarding the ML model used for the UE 810. For example, network node 2 822 may request the ML model ID used for the UE 810, e.g., the ML model ID currently being used for the UE 810.

[0097] If inference with network node 1 820 began before the establishment of the secondary connection with network node 2 822, the UE may indicate the currently inferred ML model to network node 2 822 when indicating the UE's ML capabilities in step 832.

[0098] Network node 1 820 maintains a list of ML models supported by UEs connected to network node 1 820 and a list of ML models supported by neighboring network nodes that can serve the UE in DC mode.

[0099] Network node 1 820 checks whether ML model 1 currently inferred with UE 810 is compatible with network node 2 822. Network node 1 820 may determine or check (836) that ML model 1 is compatible with network node 2 822. Network node 1 820 indicates (838) ML model 1 to network node 2 822.

[0100] The UE 810 initiates inference of ML model 1 for both network node 1 and network node 2 (840).

[0101] Thus, network node 1 820 ensures that the UE does not need to run different ML models for the same function for connections with different network nodes, thereby enabling UE resources to be saved.

[0102] 9 illustrates, by way of example, signaling between entities: UE 910 is performing ML Model 1 inference 930 with Network Node 1 920, which may be a first network node or master node (MN).

[0103] The UE 910 establishes 932 a secondary connection with network node 2 922, which may be a second network node or secondary node (SN). The UE 910 sends information about at least one two-sided ML model supported by the UE 910 to network node 2 922. Network node 2 922 may maintain a list of ML models supported by UEs connected to network node 2 922.

[0104] Network node 2 922 requests (934) from network node 1 information regarding the ML model used for UE 910. For example, network node 2 922 may request the ML model ID used for UE 910, e.g., the ML model ID currently being used for UE 910.

[0105] If inference with network node 1 920 began before the establishment of the secondary connection with network node 2 922, UE 910 may indicate the currently inferred ML model to network node 2 922 when indicating the ML capabilities of UE 910 in step 932.

[0106] Network node 1 920 maintains a list of ML models supported by UEs connected to network node 1 920 and a list of ML models supported by neighboring network nodes that can serve the UE in DC mode.

[0107] Network node 1 920 checks whether ML model 1 currently inferred with UE 910 is compatible with network node 2 922. Network node 1 920 may determine (936) that the first two-sided machine learning model is not compatible with network node 2 922. Network node 1 920 selects (938) ML model 2 that is compatible with the user equipment and network node 2 922.

[0108] Network node 1 920 sends 940 an instruction to UE 910 to switch to ML model 2. Network node 1 920 sends the ML model 2 indication to network node 2.

[0109] UE 910, together with network node 1 and network node 2, initiates inference of a common model, i.e., ML model 2.

[0110] In this way, network node 1 920 enables model matching of the UE in DC mode.

[0111] If the network node does not support ML-related coordination, the UE itself may ensure model matching in DC mode. Figure 10 shows, as an example, signaling between entities. UE 1010 performs 1030 inference of ML model 1 with network node 1 1020. UE 1010 establishes 1032 a secondary connection with network node 2 1022. The UE indicates its ML capabilities to network node 2 1022. Network node 2 1022 may maintain a list of ML models supported by UEs connected to network node 2 1022.

[0112] The UE 1010 receives 1034 an indication of the selected ML model 2 from network node 2 1022. This indication may be received while the UE 1010 is performing inference of ML model 1 with network node 2 1020.

[0113] The UE 1010 sends a NACK and an indication to network node 2 that the UE will perform inference of another ML model, for example, ML model 1, together with network node 1. Network node 1 is an MN and network node 2 is an SN.

[0114] If inference with network node 1 1020 has begun prior to the establishment of the secondary connection with network node 2 1022, UE 1010 may indicate the currently inferred ML model to network node 2 1022 when indicating the ML capabilities of UE 1010 in step 1032.

[0115] Network node 2 1022 receives an indication of the ML model currently running at the UE. Network node 2 may accept 1038 the inference of ML model 1 and send 1040 an acknowledgment ACK. Network node 2 may reject 1038 the inference of ML model 1 and send 1040 a NACK. This acceptance or rejection may depend on, for example, the ML capabilities of network node 2 or resources at network node 2, such as available central processing unit (CPU) resources and / or occupancy of hardware accelerators for ML.

[0116] In this way, the UE in the DC may indicate to the secondary cell group (SCG) the ML model used for the master cell group (MCG), allowing the secondary node to determine whether to implement the same ML model as already inferred by the UE.

[0117] Referring back to the O&M server, the O&M server is aware of the network topology and can verify that neighboring network nodes have at least one compatible ML model for each ML-enabled function. Compatible ML models mean that the ML models at network nodes are compatible when the UE is in DC mode and utilizes the same portion of a two-sided ML model with both network nodes. For example, for CSI compression using an autoencoder, a UE in DC may run the same encoder model for both network nodes, while the network nodes may have different models for the decoder. The different models provide the same or sufficiently similar output for the same input. The procedure for configuring compatible ML models at neighboring network nodes may be driven by the O&M server.

[0118] 11 shows, by way of example, signaling between entities: A network node 1110 joins 1130 a network managed by a management server, e.g., an O&M server 1120. When a new network node joins the network, a process may be triggered to obtain and configure a matching ML model.

[0119] The O&M server 1120 identifies (1132) neighboring network nodes that can participate in DC for the UE served by the network node 1110.

[0120] The O&M server 1120 determines ML models compatible with neighboring network nodes (1134). The server 1120 may maintain a list of compatible ML models. This list may include pairs of [neighbor ID; ID of supported ML model]. The IDs of neighboring network nodes may include a new radio cell global identifier (NCGI) for neighbor 1, an NCGI for neighbor 2, etc. The IDs of the supported ML models may be, for example, ID1, ID2, and ID3 for neighbor 1, and ID3, ID5, and ID6 for neighbor 2.

[0121] Optionally, the O&M server 1120 may perform training of one or more models adapted to neighboring network nodes (1134).

[0122] The O&M server 1120 may configure the network node 1110 with an ML model that is compatible with neighboring network nodes. For example, the O&M server 1120 may provide the ML model to the network node 1110. Thus, the O&M server may be configured to provide a common ML model or information about the common ML model to the MCG and SCG that are serving the UE in the DC.

[0123] FIG. 12 illustrates, by way of example, a block diagram of an apparatus capable of at least performing the methods disclosed herein. A device 1200 is shown, which may include, for example, a mobile communication device such as the UE 510 of FIG. 5 or the network node 520 of FIG. 5. The device 1200 includes a processor 1210, which may include, for example, a single-core processor or a multi-core processor, where a single-core processor has one processing core and a multi-core processor has two or more processing cores. The processor 1210 may generally include a control device. The processor 1210 may include two or more processors. The processor 1210 may be the control device. The processing core may include, for example, a Cortex-A8 processing core manufactured by ARM Holdings or a Steamroller processing core designed by Advanced Micro Devices Corporation. The processor 1210 may include at least one Qualcomm Snapdragon and / or Intel Atom processor. The processor 1210 may include at least one application-specific integrated circuit (ASIC). The processor 1210 may include at least one field-programmable gate array (FPGA). The processor 1210 may be a means for performing steps of a method in the device 1200. The processor 1210 may be configured to perform actions at least in part by computer instructions.

[0124] A processor may comprise circuitry or may be configured as one or more circuits, the one or more circuits configured to perform method steps in accordance with the example embodiments described herein. As used in this application, the term "circuitry" may refer to one or more or all of: (a) a hardware-only circuit implementation, such as an implementation in analog and / or digital circuitry only; (b) where applicable, (i) a combination of analog and / or digital hardware circuitry with software / firmware, and (ii) any portion of a hardware processor, including software (including a digital signal processor), software, and memory, that function together to cause a device, such as user equipment or a network node, to perform various functions; and (c) a hardware circuit and / or processor, such as a microprocessor or portion of a microprocessor, that requires software (e.g., firmware) to operate but may not be present if software is not required for operation.

[0125] This definition of circuit applies to all uses of the term in this application, including in any claims. As a further example, when used in this application, the term circuit also covers simply a hardware circuit or processor (or processors), or a portion of a hardware circuit or processor, as well as its (or their) accompanying software and / or firmware implementation. The term circuit also covers, for example, a baseband or processor integrated circuit for a mobile device, or a similar integrated circuit in a server, cellular network device, or other computing or network device, if applicable to certain claim elements.

[0126] The device 1200 may include a memory 1220. The memory 1220 may include random access memory and / or fixed memory. The memory 1220 may include at least one RAM chip. The memory 1220 may include, for example, semiconductor memory, magnetic memory, optical memory, and / or holographic memory. The memory 1220 may be at least partially accessible to the processor 1210. The memory 1220 may be at least partially included in the processor 1210. The memory 1220 may be a means for storing information. The memory 1220 may include instructions, such as computer instructions or computer program code, and the processor 1210 is configured to execute these instructions. If instructions configured to cause the processor 1210 to perform a particular operation are stored in the memory 1220 and the entire device 1200 is configured to use the instructions from the memory 1220 and execute as directed by the processor 1210, the processor 1210 and / or at least one processing core thereof may be considered to be configured to perform the aforementioned particular operation. The memory 1220 may be at least partially external to the device 1200 but accessible to the device 1200 .

[0127] Device 1200 may include a transmitter 1230. Device 1200 may include a receiver 1240. Transmitter 1230 and receiver 1240 may be configured to transmit and receive information, respectively, according to at least one cellular or non-cellular standard. Transmitter 1230 may include two or more transmitters. Receiver 1240 may include two or more receivers. Transmitter 1230 and / or receiver 1240 may be configured to operate according to, for example, Global System for Mobile Communications (GSM), Wideband Code Division Multiple Access (WCDMA), 5G, Long Term Evolution (LTE), IS-95, Wireless Local Area Network (WLAN), Ethernet, and / or Worldwide Interoperability for Microwave Access (WiMAX) standards.

[0128] Device 1200 may include a user interface (UI) 1260. UI 1260 may include at least one of a display, a keyboard, a touchscreen, a vibrator positioned to signal the user by vibrating device 1200, a speaker, and a microphone. A user may be able to operate device 1200 via UI 1260, for example, to accept incoming calls, make phone or video calls, browse the Internet, manage digital files stored in memory 1220 or stored in the cloud accessible via transmitter 1230 and receiver 1240 or via NFC transceiver 1250, and / or play games.

[0129] Processor 1210 may include a transmitter arranged to output information from processor 1210 to other devices included in device 1200 via conductors internal to device 1200. Such a transmitter may include, for example, a serial bus transmitter arranged to output information via at least one conductor to memory 1220 for storage in memory 1220. Instead of a serial bus, the transmitter may include a parallel bus transmitter. Similarly, processor 1210 may include a receiver arranged to receive information at processor 1210 from other devices included in device 1200 via conductors internal to device 1200. Such a receiver may include, for example, a serial bus receiver arranged to receive information from receiver 1240 via at least one conductor for processing in processor 1210. Instead of a serial bus, the receiver may include a parallel bus receiver.

[0130] The term "non-transient" as used herein is a limitation of the medium itself (i.e., tangible rather than signal) as opposed to a limitation on the persistence of data storage (e.g., RAM vs. ROM).

[0131] As used herein, "at least one of: " and "at least one of " and similar phrases, when a list of two or more elements is joined by "and" or "or", mean at least any one of the elements, or at least any two or more of the elements, or at least all of the elements.

Claims

1. transmitting information about at least one two-sided machine learning model supported by the device to a network node; or transmitting a machine learning profile identity of the device to a network node, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the device; transmitting the at least one two-sided machine learning model configured to enable joint inference by the device and the network node; receiving, from the network node, an indication of at least a first selected two-sided machine learning model supported by the device.

2. The means comprises: The apparatus of claim 1 , further configured to: initiate joint inference of the first selected two-sided machine learning model with the network node.

3. The information regarding at least one two-sided machine learning model may include: the identity of the at least one two-sided machine learning model supported by the device; and Information about the capabilities of the at least one two-sided machine learning model supported by the device.

3. The apparatus of claim 1 or 2, comprising:

4. The means comprises: receiving, from the network node, an instruction to switch to a second two-sided machine learning model supported by the device, the second two-sided machine learning model also supported by another device serviced by the network node; The apparatus of any of claims 1 to 3, further configured to perform:

5. The means comprises: switching to the second two-sided machine learning model; and initiating joint inference of the second two-sided machine learning model with the network node; and The apparatus of claim 4 , further configured to perform:

6. The means comprises: determining that the second two-sided machine learning model is no longer supported by the device; transmitting a negative acknowledgement to the network node based on said determination; Continuing the joint inference of the first selected two-sided machine learning model. The apparatus of claim 4 , further configured to perform:

7. The means comprises: - Establishing a secondary connection with another network node; transmitting information about at least one two-sided machine learning model supported by said device to said other network node; or transmitting a machine learning profile identity of the device to the other network node, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the device; initiating joint inference of the first selected two-sided machine learning model with the other network node; The apparatus of any of claims 1 to 6, further configured to perform:

8. The means comprises: receiving, from the network node, an instruction to switch to a second two-sided machine learning model supported by the device, the second two-sided machine learning model also supported by the other network node; switching to the second two-sided machine learning model; initiating joint inference of the second two-sided machine learning model with the network node and the other network node. The apparatus of claim 7 , further configured to perform:

9. The means comprises: While performing joint inference of the first selected two-sided machine learning model with the network nodes, receiving an indication of a second selected two-sided machine learning model from the other network; sending a negative acknowledgement and an indication to the other network node that the device will perform joint inference of the first selected two-sided machine learning model with the network node; The apparatus of claim 7 , further configured to perform:

10. The means comprises: receiving, from the other network node, an instruction to initiate joint inference using the first selected two-sided machine learning model or an instruction to reject joint inference using the first selected two-sided machine learning model; The apparatus of claim 9 , further configured to perform:

11. (a) receiving, from a user equipment, information regarding at least one two-sided machine learning model supported by the user equipment; or (b) receiving, from a user equipment, a machine learning profile identity for the user equipment, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the user equipment; and Obtaining information regarding at least one two-sided machine learning model supported by the user equipment based on the machine learning profile identity of the user equipment, obtaining the at least one two-sided machine learning model configured to enable joint inference by the device and the user equipment; selecting a first two-sided machine learning model supported by the user equipment; sending an indication to the user device to select at least the first two-sided machine learning model supported by the user device; An apparatus comprising: means for performing

12. The means comprises: Initiating joint inference of the first two-sided machine learning model with the user device. The apparatus of claim 11 , further configured to perform:

13. 13. The device of claim 11 or 12, wherein selecting the first two-sided machine learning model is performed based on other two-sided machine learning models currently being jointly inferred by the device and other user equipment or supported by the other user equipment.

14. The means comprises: (a) receiving, from another user equipment serviced by the device, information regarding at least one two-sided machine learning model supported by the other user equipment; or (b) receiving, from another user equipment serviced by the device, a machine learning profile identity of the other user equipment, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the other user equipment; and and obtaining information about at least one two-sided machine learning model supported by the other user device based on the machine learning profile identity of the other user device. The apparatus of any of claims 11 to 13, further configured to perform:

15. The means comprises: determining a second two-sided machine learning model supported by both user equipment based on the information about at least one two-sided machine learning model supported by the user equipment and by the other user equipment; sending an instruction to the user device to switch to the second two-sided machine learning model; The apparatus of claim 14 , further configured to perform:

16. The means comprises: sending an indication to the other user device of selecting the second two-sided machine learning model; initiating joint inference of the second two-sided machine learning model with the user device and the other user device; and The apparatus of claim 15 , further configured to perform:

17. The means comprises: receiving a negative acknowledgment from the user equipment indicating that the second two-sided machine learning model is no longer supported by the user equipment; continuing to interact with the first two-sided machine learning model with the user device; and initiating joint inference of the second two-sided machine learning model with the other user device; and 17. The apparatus of claim 15 or 16, further configured to perform:

18. The means comprises: receiving a request from another network node serving the user equipment for information about a two-sided machine learning model used by the user equipment; determining that the first two-sided machine learning model matches the capabilities of the other network node; sending an indication to the other network node of selecting the first two-sided machine learning model; The apparatus of any of claims 11 to 17, further configured to perform:

19. The means comprises: receiving a request from another network node serving the user equipment for information about a two-sided machine learning model used by the user equipment; determining that the first two-sided machine learning model does not match the capabilities of the other network node; selecting a second two-sided machine learning model that matches the capabilities of the user equipment and the capabilities of the other network node based on the determination; sending an instruction to the user device to switch to the second two-sided machine learning model; The apparatus of any of claims 11 to 17, further configured to perform:

20. The means comprises: sending an indication to the other network node of selecting the second two-sided machine learning model; initiating inference of the second two-sided machine learning model together with the user equipment and the other network node; and 20. The apparatus of claim 19, further configured to perform:

21. Establishing a connection with a user equipment, whereby said device is configured to act as a secondary network node; receiving, from a master network node configured to provide service to the user equipment, information regarding a two-sided machine learning model that matches the capabilities of the user equipment and the capabilities of the secondary network node; initiating joint inference of the two-sided machine learning model with the user equipment and the master network node. An apparatus comprising: means for performing

22. 22. The apparatus of any preceding claim, wherein said means comprises at least one processor and at least one memory storing instructions that, when executed by said at least one processor, cause said execution of said apparatus.

23. transmitting, by a device, to a network node, information about at least one two-sided machine learning model supported by said device; or transmitting, by the device, a machine learning profile identity of the device to a network node, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the device; transmitting the at least one two-sided machine learning model configured to enable joint inference by the device and the network node; receiving, by the device, from the network node, an indication of at least a first selected two-sided machine learning model supported by the device; A method comprising:

24. (a) receiving, by a network node, from a user equipment, information regarding at least one two-sided machine learning model supported by the user equipment; or (b) receiving, by a network node, from a user equipment, a machine learning profile identity of the user equipment, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the user equipment; and Obtaining information regarding at least one two-sided machine learning model supported by the user equipment based on the machine learning profile identity of the user equipment, obtaining the at least one two-sided machine learning model configured to enable joint inference by the device and the user equipment; selecting, by the network node, a first two-sided machine learning model supported by the user equipment; sending, by the network node, an indication to the user equipment of selecting at least the first two-sided machine learning model supported by the user equipment; A method comprising:

25. establishing, by a device, a connection with a user equipment, whereby said device is configured to function as a secondary network node; receiving, by an apparatus, from a master network node configured to provide service to the user equipment, information regarding a two-sided machine learning model that matches capabilities of the user equipment and capabilities of the secondary network node; initiating, by the device, joint inference of the two-sided machine learning model with the user equipment and the master network node. A method comprising:

26. A computer program comprising instructions that, when executed by an apparatus, cause the apparatus to: transmitting information to a network node about at least one two-sided machine learning model supported by said device; or transmitting a machine learning profile identity of the device to a network node, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the device; transmitting the at least one two-sided machine learning model configured to enable joint inference by the device and the network node; receiving from the network node an indication of at least a first selected two-sided machine learning model supported by the device; A computer program that executes

27. A computer program comprising instructions that, when executed by an apparatus, cause the apparatus to: (a) receiving, from a user equipment, information regarding at least one two-sided machine learning model supported by the user equipment; or (b) receiving, from a user equipment, a machine learning profile identity for the user equipment, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the user equipment; and Obtaining information regarding at least one two-sided machine learning model supported by the user equipment based on the machine learning profile identity of the user equipment, obtaining the at least one two-sided machine learning model configured to enable joint inference by the device and the user equipment; selecting a first two-sided machine learning model supported by the user equipment; 12. A computer program product comprising: a computer program causing the computer to execute: sending an instruction to the user device to select at least the first two-sided machine learning model supported by the user device.

28. A computer program comprising instructions that, when executed by an apparatus, cause the apparatus to: Establishing a connection with a user equipment, whereby said device is configured to act as a secondary network node; receiving, from a master network node configured to provide service to the user equipment, information regarding a two-sided machine learning model that matches the capabilities of the user equipment and the capabilities of the secondary network node; initiating joint inference of the two-sided machine learning model with the user equipment and the master network node. A computer program that executes

29. A non-transitory computer-readable medium containing instructions that, when executed by a device, cause the device to: transmitting information to a network node about at least one two-sided machine learning model supported by said device; or transmitting a machine learning profile identity of the device to a network node, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the device; transmitting the at least one two-sided machine learning model configured to enable joint inference by the device and the network node; receiving from the network node an indication of at least a first selected two-sided machine learning model supported by the device; A non-transitory computer-readable medium that executes at least the above.

30. A non-transitory computer-readable medium containing instructions that, when executed by a device, cause the device to: (a) receiving, from a user equipment, information regarding at least one two-sided machine learning model supported by the user equipment; or (b) receiving, from a user equipment, a machine learning profile identity for the user equipment, the machine learning profile identity being associated with at least one two-sided machine learning model supported by the user equipment; and Obtaining information regarding at least one two-sided machine learning model supported by the user equipment based on the machine learning profile identity of the user equipment, obtaining the at least one two-sided machine learning model configured to enable joint inference by the device and the user equipment; selecting a first two-sided machine learning model supported by the user equipment; sending an indication to the user device to select at least the first two-sided machine learning model supported by the user device; A non-transitory computer-readable medium that executes at least the above.

31. A non-transitory computer-readable medium containing instructions that, when executed by a device, cause the device to: Establishing a connection with a user equipment, whereby said device is configured to act as a secondary network node; receiving, from a master network node configured to provide service to the user equipment, information regarding a two-sided machine learning model that matches the capabilities of the user equipment and the capabilities of the secondary network node; initiating joint inference of the two-sided machine learning model with the user equipment and the master network node. A non-transitory computer-readable medium that causes at least one of the following to be executed.

Images