Low complexity ML model training over multiple gns

Abstract

Embodiments of the present disclosure relate to low complexity ML model training over multiple GBTs. An apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: receive, from a plurality of network nodes, information relating to at least one parameter of a model of the plurality of network nodes; determining at least one cluster of the plurality of network nodes based on at least one similarity criterion and information relating to at least one parameter of a model of the plurality of network nodes; and determining at least one global model for the at least one cluster using the local models of the network nodes belonging to the at least one cluster.

Description

Technical Field

[0001] The exemplary and non-limiting example embodiments relate generally to communications and, more particularly, to low-complexity ML model training across multiple gNBs. Background Art

[0002] It is known that a communication device obtains access to the network via a cell covered by a transmission reception point in the communication network. Summary of the invention

[0003] According to one aspect, a device includes at least one processor; and at least one memory for storing instructions that, when executed by the at least one processor, cause the device to at least: receive information related to at least one parameter of a model of the multiple network nodes from a plurality of network nodes; determine at least one cluster of the multiple network nodes based on at least one similarity criterion and the information related to at least one parameter of the model of the multiple network nodes; and determine at least one global model for the at least one cluster using local models of the network nodes belonging to the at least one cluster.

[0004] According to one aspect, a device includes at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the device to at least: send information related to at least one parameter of a model of the device to a network entity; in response to the device being located within a cluster of network nodes similar to the device based on at least one similarity criterion and the information related to at least one parameter of the model of the device, receive an indication from the network entity to perform federated learning with the network entity; and in response to receiving an indication from the network entity to perform federated learning with the network entity, perform federated learning with the network entity.

[0005] According to one aspect, a device includes at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the device to at least: receive or access a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; receive or access a local model of a network node; and perform inference using at least one of: a global model based on federated learning or a local model of a network node. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The foregoing aspects and other features are explained in the following description taken in conjunction with the accompanying drawings.

[0007] Figure 1 is a block diagram of one possible and non-limiting system in which example embodiments may be practiced.

[0008] Figure 2 is an example block diagram of a federated learning model in a wireless network.

[0009] Figure 3 is a flow chart of the method described herein.

[0010] Figure 4 Cell grouping embedding (simplified to 2d) and grouping using T-SNE and PCA algorithms are shown.

[0011] Figure 5 The CA Scell ​​selection NN model is shown.

[0012] Figure 6 A signaling diagram with an example gNB for FL procedure and a gNB for transfer learning is shown.

[0013] Figure 7 is an example apparatus configured to implement the examples described herein.

[0014] Figure 8 Representations of examples of non-volatile storage media for storing instructions implementing the examples described herein are shown.

[0015] Fig. 9 is an example method based on the examples described herein.

[0016] Fig.10 is an example method based on the examples described herein.

[0017] Fig.11 is an example method based on the examples described herein. DETAILED DESCRIPTION

[0018] Go to Figure 1 , which shows a block diagram of one possible and non-limiting example in which the example may be practiced. A user equipment (UE) 110, a radio access network (RAN) node 170, and a network element 190 are shown. Figure 1In the example of , a user equipment (UE) 110 wirelessly communicates with a wireless network 100. A UE is a wireless device that can access the wireless network 100. The UE 110 includes one or more processors 120, one or more memories 125, and one or more transceivers 130 interconnected by one or more buses 127. Each of the one or more transceivers 130 includes a receiver Rx 132 and a transmitter Tx 133. The one or more buses 127 may be address, data, or control buses, and may include any interconnection mechanism, such as a series of lines on a motherboard or integrated circuit, an optical fiber or other optical communication device, etc. The one or more transceivers 130 are connected to one or more antennas 128. The one or more memories 125 include computer program code 123. The UE 110 includes a module 140, which includes one or both of the parts 140-1 and / or 140-2, which may be implemented in a variety of ways. The module 140 may be implemented in hardware as the module 140-1, such as as part of the one or more processors 120. Module 140-1 may also be implemented as an integrated circuit or by other hardware (such as a programmable gate array). In another example, module 140 may be implemented as module 140-2, which is implemented as computer program code 123 and executed by one or more processors 120. For example, one or more memories 125 and computer program code 123 may be configured to perform one or more operations described herein with one or more processors 120 using user equipment 110. UE 110 communicates with RAN node 170 via wireless link 111.

[0019] The RAN node 170 in this example is a base station that provides access to the wireless network 100 for wireless devices such as UE 110. The RAN node 170 may be, for example, a base station for 5G, also known as New Radio (NR). In 5G, the RAN node 170 may be a NG-RAN node, which is defined as a gNB or ng-eNB. A gNB is a node that provides NR user plane and control plane protocol termination to the UE and is connected to a 5GC such as, for example, network element (s) 190 via an NG interface such as connection 131. An ng-eNB is a node that provides E-UTRA user plane and control plane protocol termination to the UE and is connected to a 5GC via an NG interface such as connection 131. The NG-RAN node may include multiple gNBs, which may also include a central unit (CU) (gNB-CU) 196 and (multiple) distributed units (DU) (gNB-DU), of which DU 195 is shown. Note that the DU 195 may include or be coupled to and control a radio unit (RU). The gNB-CU 196 is a logical node that carries the radio resource control (RRC), SDAP and PDCP protocols of the gNB or the RRC and PDCP protocols of the en-gNB, and controls the operation of one or more gNB-DUs. The gNB-CU 196 terminates the F1 interface connected to the gNB-DU 195. The F1 interface is shown as reference 198, although reference 198 also shows a link between a remote element of the RAN node 170 and a centralized element of the RAN node 170, such as a link between the gNB-CU 196 and the gNB-DU 195. The gNB-DU 195 is a logical node that carries the RLC, MAC and PHY layers of the gNB or en-gNB, and its operation is partially controlled by the gNB-CU 196. One gNB-CU 196 supports one or more cells. One cell can be supported by one gNB-DU 195, or one cell can be supported / shared by multiple DUs under RAN sharing. The gNB-DU 195 terminates the F1 interface 198 connected to the gNB-CU 196. Note that DU 195 is considered to include transceiver 160, e.g., as part of a RU, but some examples of this may include transceiver 160 as part of a separate RU, e.g., under the control of and connected to DU 195. RAN node 170 may also be an eNB (evolved NodeB) base station for LTE (Long Term Evolution), or any other suitable base station or node.

[0020] The RAN node 170 includes one or more processors 152, one or more memories 155, one or more network interfaces (N / WI / F) 161, and one or more transceivers 160 interconnected by one or more buses 157. Each of the one or more transceivers 160 includes a receiver Rx 162 and a transmitter Tx 163. The one or more transceivers 160 are connected to one or more antennas 158. The one or more memories 155 include computer program code 153. The CU 196 may include the processor 152, the one or more memories 155, and the network interface 161. Note that the DU 195 may also contain its own one or more memories and (multiple) processors and / or other hardware, but these are not shown.

[0021] The RAN node 170 includes a module 150, which includes one or both of the parts 150-1 and / or 150-2, which can be implemented in a variety of ways. The module 150 can be implemented in hardware as a module 150-1, such as as part of one or more processors 152. The module 150-1 can also be implemented as an integrated circuit or by other hardware (such as a programmable gate array). In another example, the module 150 can be implemented as a module 150-2, which is implemented as a computer program code 153 and executed by one or more processors 152. For example, one or more memories 155 and computer program code 153 are configured to, together with one or more processors 152, cause the RAN node 170 to perform one or more operations described herein. Note that the functionality of the module 150 can be distributed, such as distributed between the DU 195 and the CU 196, or implemented only in the DU 195.

[0022] One or more network interfaces 161 communicate over a network, such as via links 176 and 131. Two or more gNBs 170 may communicate using, for example, link 176. Link 176 may be wired or wireless or both, and may implement, for example, an Xn interface for 5G, an X2 interface for LTE, or other suitable interfaces for other standards.

[0023] The one or more buses 157 may be address, data, or control buses, and may include any interconnection mechanism, such as a series of wires on a motherboard or integrated circuit, optical fiber or other optical communication device, wireless channel, etc. For example, the one or more transceivers 160 may be implemented as a remote radio head (RRH) 195 for LTE or a distributed unit (DU) 195 for a gNB implementation for 5G, where other elements of the RAN node 170 may be physically located at a different location from the RRH / DU 195, and the one or more buses 157 may be partially implemented as, for example, a fiber optic cable or other suitable network connection to connect the other elements of the RAN node 170 (e.g., central unit (CU), gNB-CU 196) to the RRH / DU 195. Reference 198 also indicates those suitable network link(s).

[0024] The RAN node / gNB may include one or more TRPs, and the methods described in this document may be applied to these TRPs. Figure 1 RAN node 170 is shown to include TRP 51 and TRP 52 in addition to the TRP represented by transceiver 160. Similar to transceiver 160, TRP 51 and TRP 52 may each include a transmitter and a receiver. RAN node 170 may carry or include Figure 1 Other TRPs not shown.

[0025] Relay nodes in NR are called integrated access and backhaul nodes. The mobile termination part of the IAB node facilitates the backhaul (parent link) connection. In other words, the mobile termination part includes the functionality of carrying UE functions. The distributed unit part of the IAB node facilitates the so-called access link (sub-link) connection (i.e., for the access link UE, and in the case of multi-hop IAB, for the backhaul for other IAB nodes). In other words, the distributed unit part is responsible for certain base station functions. The IAB scenario may follow a so-called split architecture, where the central unit hosts the higher layer protocols to the UE and terminates the control plane and user plane interfaces to the 5G core network.

[0026] It is worth noting that the description herein indicates that a "cell" performs functions, but it should be clear that the devices that form the cell can perform these functions. The cell constitutes part of the base station. That is, each base station can have multiple cells. For example, for a single carrier frequency and associated bandwidth, there can be three cells, each covering one-third of a 360-degree area, so that the coverage area of ​​a single base station covers an approximate ellipse or circle. In addition, each cell can correspond to a single carrier, and the base station can use multiple carriers. Therefore, if there are three 120-degree cells per carrier and there are two carriers, the base station has a total of 6 cells.

[0027] The wireless network 100 may include one or more network elements 190, which may include core network functions, and which provide connectivity to another network (which is to a telephone network and / or a data communication network (e.g., the Internet)) via one or more links 181. Such core network functions for 5G may include a location management function (LMF) and / or an access and mobility management function (AMF) and / or a user plane function (UPF) and / or a session management function (SMF). Such core network functions for LTE may include MME (mobility management entity) / SGW (serving gateway) functions. Such core network functions may include SON (self-organizing / optimizing network) functions. These are merely example functions that may be supported by (multiple) network elements 190, and it is noted that both 5G and LTE functions may be supported. The RAN node 170 is coupled to the network element 190 via a link 131. The link 131 may be implemented, for example, as an NG interface for 5G, an S1 interface for LTE, or other suitable interfaces for other standards. The network element 190 includes one or more processors 175, one or more memories 171, and one or more network interfaces (N / WI / F) 180, which are interconnected by one or more buses 185. The one or more memories 171 include computer program code 173. The computer program code 173 may include SON and / or MRO functionality 172.

[0028] The wireless network 100 may implement network virtualization, which is the process of combining hardware and software network resources and network functions into a single software-based management entity or virtual network. Network virtualization involves platform virtualization, often combined with resource virtualization. Network virtualization is categorized into external virtualization, which combines many networks or parts of networks into virtual units, and internal virtualization, which provides network-like functions to software containers on a single system. Note that the virtualized entities produced by network virtualization are still implemented to some extent using hardware, such as processors 152 or 175 and memories 155 and 171, and such virtualized entities also produce technical effects.

[0029] Computer readable memories 125, 155, and 171 may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, non-transitory memory, transient memory, fixed memory, and removable memory. Computer readable memories 125, 155, and 171 may be components for performing storage functions. Processors 120, 152, and 175 may be of any type suitable for the local technical environment and may include, as non-limiting examples, one or more of a general purpose computer, a special purpose computer, a microprocessor, a digital signal processor (DSP), and a processor based on a multi-core processor architecture. Processors 120, 152, and 175 may be components for performing functions, such as controlling UE 110, RAN node 170, (multiple) network elements 190, and other functions described herein.

[0030] In general, various example embodiments of user equipment 110 may include, but are not limited to, cellular phones, such as smartphones, tablet computers, personal digital assistants (PDAs) with wireless communication capabilities, portable computers with wireless communication capabilities, image capture devices (such as digital cameras with wireless communication capabilities), gaming devices with wireless communication capabilities, music storage and playback devices with wireless communication capabilities, Internet devices (including those that allow wireless Internet access and browsing), tablet computers with wireless communication capabilities, head-mounted displays (such as those that implement virtual / augmented / mixed reality), and portable units or terminals that combine these functions. UE 110 may also be a vehicle (such as a car) or a UE installed in a vehicle, a UAV (such as, for example, a drone) or a UE installed in a UAV. User equipment 110 may be a terminal device, such as a mobile phone, a mobile device, a sensor device, etc., which is a device used by a user or a device not used by a user.

[0031] UE 110, RAN node 170 and / or network element(s) 190 (and associated memory, computer program code and modules) may be configured to (e.g., partially) implement the methods described herein. Thus, computer program code 123 of UE 110, module 140-1, module 140-2 and Figure 1 Other elements / features shown in the figure may implement relevant aspects of the example user equipment described herein. Similarly, the computer program code 153, module 150-1, module 150-2 and Figure 1 Other elements / features shown in the figure may implement relevant aspects of the example gNB / TRP described herein. Computer program code 173 of network element(s) 190 and Figure 1 Other elements / features shown in the figure may be configured to implement relevant aspects of the example network elements described herein.

[0032] Having thus introduced a suitable but non-limiting technical context for practicing example embodiments, example embodiments are now described in more detail.

[0033] Many applications in mobile networks require large amounts of data from multiple distributed sources, such as UEs or distributed gNBs, to be used to train a single common model. In order to minimize the data exchange between the distributed units where the data is generated and the centralized unit where the common model needs to be created, the concept of federated learning (FL) can be applied. FL is a form of machine learning where instead of model training at a single node, different versions of the model are trained locally on different distributed hosts and then aggregated at a central entity. This is different from distributed machine learning, where a single ML model is trained at distributed nodes to use the computing power of different nodes. In other words, FL differs from distributed learning in that for FL: 1) each distributed node in the FL scenario has its own local training data, which may not be from the same distribution as the data of other nodes; 2) each node calculates the parameters for its local ML model and 3) the central host does not calculate versions or parts of the model, but combines the parameters of all distributed models to generate a global model. The global model is then shared with the distributed nodes. The purpose of this approach is to keep the training dataset where it is generated and perform model training locally at each individual learner in the federation.

[0034] Figure 2 is an example block diagram for a federated learning model in a wireless network. Note that both the partial model (sent at 204-1, 204-2, 204-3) and the aggregated model (sent at 208-1, 208-2, 208-3) are sent over a conventional communication link. Figure 2UEs (110-1, 110-2, 110-3) acting as local learners and gNB 170 acting as aggregator nodes are shown. After training the local models, each individual learner (110-1, 110-2, 110-3) transmits (at 204-1, 204-2, 204-3) its local model parameters (202-1, 202-2, 203-3) (instead of the original training data set) to the aggregation unit 170. The aggregation unit 170 utilizes the local model parameters (202-1, 202-2, 203-3) to update the global model 206, which can eventually be fed back (at 208-1, 208-2, 208-3) to the local learners (110-1, 110-2, 110-3) for further iterations until the global model 206 converges. Therefore, each local learner (110-1, 110-2, 110-3) benefits from the datasets of other local learners (110-1, 110-2, 110-3) only through the global model 206 shared by the aggregator 170, without explicitly accessing the large amount of privacy-sensitive data available at each of the other local learners. Figure 2 is shown in .

[0035] In summary, the FL training process includes the following main steps (1-4):

[0036] 1. Initialization: A machine learning model (e.g., linear regression, neural network) is selected to be trained on the local node and initialized.

[0037] 2. Client selection: A small number of local nodes are selected to start training on local data. The selected nodes obtain the current statistical model, while other nodes wait for the next round of federation.

[0038] 3. Reporting and aggregation: Each selected node sends its local model to the server for aggregation. The central server aggregates the received models and sends model updates back to the nodes.

[0039] 4. Termination: Once the predefined termination criteria are met (e.g., the maximum number of iterations is reached), the central server aggregates the updates and finalizes the global model.

[0040] This paper describes the use of FL concepts to train a single model for a group of gNBs with similar training data.

[0041] One approach is to use an iterative method to initialize a Q-learning model that is applied to the training data of multiple gNBs. However, this approach does not differentiate gNBs based on the data, and data from all gNBs are processed to train one ML model. The approach described in this paper addresses the problem when the training data from different gNBs is different enough and various gNB groups with similar data can be identified.

[0042] Training supervised learning models with target accuracy and generalization can be costly and time-consuming, especially when trained separately in each gNB. In many cases, ML functions are run on several cells.

[0043] Federated learning can be adopted to ensure the generalization of the model over several cells while making a trade-off in terms of model accuracy. However, training in federated learning can be very time-consuming and resource-intensive if the data from various nodes belong to completely different distributions.

[0044] Therefore, the problem addressed by the example described in this article is as follows: How to improve the training process for ML models that need to be trained using data from multiple cells with sufficiently different data distributions by jointly using federated learning and transfer learning?

[0045] The solution described herein includes procedures and related signaling enhancements that allow optimizing the learning strategy of a group of cells by jointly employing a combination of federated learning and transfer learning.

[0046] The concept of embedding is used with the examples described in this article. The concept of embedding is defined as: embedding is a plurality of vectors / position coordinates on a real-valued line (continuous line). Embedding is learned as part of model training. If embedding is regarded as a position coordinate, similar models will be closer to each other. The Euclidean distance between vectors can be used to find the most similar models. Embedding is a vector or digital array that represents the meaning and context of the tags processed and generated by the model. Embedding is derived from the parameters or weights of the model and is used to encode and decode the input and output text. The embedding of model parameters may cause the embedding to have a smaller dimension than the model or the parameters of the model, so the embedding process can be a data dimensionality reduction process. The decoding or reconstruction of embedding can also be used by the examples described in this article.

[0047] Key Features:

[0048] Consider a set of N cells, each with a unique training dataset. Instead of training a separate ML model for each cell (which may not be feasible for fast-moving UEs that frequently switch between different cells), the following ML model training scheme combining FL and TL methods is described in this paper:

[0049] Step 1 : Grouping is first performed on the cells using a similarity criterion. Example #1: The similarity criterion here is specifically determined based on the embeddings emitted by each local node together with the local model. Example #2: The similarity criterion test can also be based on the local training data sample. Note that only similar cells are grouped together. Therefore, a set of n1 cell groups can be obtained, and the remaining individual cells do not belong to any group.

[0050] Step 2 : Federated learning is then applied to each cell group to obtain n1 global FL models.

[0051] Step 3 : Step 3 involves the process for each individual cell. Embodiment #1: Search for the closest group of cells with similar training data and train the model on them via transfer learning to obtain a new model. Embodiment #2: Perform local ML training on cells with data that is not similar to any cell's data.

[0052] The solution described herein is described in further detail below, assuming that the same ML framework or function is running on multiple cells or nodes, which was previously used to train the ML model separately in each cell.

[0053] Find nodes with similar local models. As part of NN training in local nodes, embeddings (large vectors) can be learned for local models. Embeddings can be viewed as local model signatures; similar models are close in the embedding vector space. Each node sends the embedding to the OAM along with the trained local model. The OAM determines the group based on the embedding and creates a global FL model for each group using the local models of the cells belonging to the group, or using the corresponding local models of the corresponding cells belonging to the group (e.g., the corresponding subset of the local models of the corresponding subset of cells belonging to the group). Transfer learning can be used for nodes that do not belong to any group (non-similar nodes).

[0054] The detailed steps of the proposed solution are summarized in Figure 3 In the flowchart shown in Figure 3 , gNB 170 performs operations 302, 304, 306, 308, and 318, and OAM 190 performs operations 310, 312, 314, and 316. At 302, the process starts. At 304, gNB 170 performs NN training. At 306, embedding is learned. At 306, the cell features used to learn the embedding may be DL or UL PRB usage time series, throughput time series, etc. At 308, gNB 170 sends the learned embedding and local model to OAM 190.

[0055] At 310, OAM 190 identifies similarity criteria and clusters using ML algorithms (such as embedding) - using PCA and / or cluster K-means methods. At 312, OAM 190 determines whether there are similar nodes that belong to the cluster. If at 312 OAM 190 determines that there are similar nodes that belong to the cluster, the method transitions to 314. If at 312 OAM 190 determines that there are nodes that do not belong to the cluster, or there are no similar nodes that belong to the cluster, the method transitions to 316. At 314, OAM 190 creates a global model for each cluster using the local models that belong to the cluster. At 316, OAM 190 obtains the nearest set of cells by calculating the Euclidean distance.

[0056] At 318, gNB 170 performs transfer learning on the nearest set of cells. Specifically, gNB 170 performs transfer learning using the trained model of the cluster whose embedding appears to be closest to the embedding of gNB 170. In other words, gNB 170 is similar to a particular cluster, but not similar enough to be part of that cluster. The cluster's trained model is then used as a starting point to train the gNB's local model, which is called transfer learning. Transfer learning is used to quickly train a local model. Once the gNB 170's local model is trained, gNB 170 can use the trained model locally.

[0057] Embed (306). Some cell characteristics to be used for grouping cells include, but are not limited to, DL or ULPRB usage time series, throughput time series, and number of RRC connected users or active users.

[0058] Identify similarity criteria, reduce dimensionality and group (310). Using unsupervised machine learning, various ML algorithms can be used to first propose embeddings for each cell (say up to 10 dimensions), and based on the embeddings groups of cells can be obtained, keeping in mind the number and size of these groups. The ML algorithms or tools to be used may include (these are examples, other methods can be used) 1) for embedding, PCA or t-SNE, 2) for clustering, -K-means, GMM or DBSCAN, and 3) for anomaly detection, density estimation or thresholding.

[0059] Figure 4Field data results for embedding 406 (reduced to 2d) and grouping 408 of cells using T-SNE (402) and PCA (404) algorithms are shown. The data used to generate the results includes 500 cells and 1 week of report data for each cell. The cell characteristics used to derive the embedding include DL and UL PRB usage time series, the entire time series, and the number of RRC connected users and active users. Using the ML algorithm, the dimensionality of the data is reduced from 10d to 2d for each cell, and further grouping of cells using the ML algorithm is performed.

[0060] Figure 5 The problems related to the CA secondary cell selection feature and how the solution described herein achieves secondary cell selection optimization are shown.

[0061] In case of CA Scell ​​selection, a local model that predicts the Scell ​​spectral efficiency (SE) for one cell may not be applicable to another cell, although the input used to predict the SE remains the same. The reason is that the environment has an impact on the prediction. The model learns the cell surroundings, so the model becomes very cell-specific. And with federated learning, the global model tends to lose this cell-specific information when averaging and / or aggregating, which causes performance degradation. Therefore, it becomes imperative to find similar cells and form a group.

[0062] In particular, Figure 5 A NN model 502 is shown that is learned based on the examples described herein. The NN model 502 takes as input the primary cell spectral efficiency (SE) 504, the primary path loss 506, the primary carrier load 508, the secondary carrier load 510, and the arrival or departure angle 512. The arrival or departure angle can be derived from the SRS, CRI, PMI, RI, or LI. Based on the inputs, the NN model 502 generates a prediction or estimate of the secondary cell spectral efficiency.

[0063] Main implementation: Signaling enhancement

[0064] The ML embedded at the gNB sends local model information to OAM. The ML embedded at OAM further performs group identification and global model generation.

[0065] Figure 6 The signalling exchange between the NG-RAN node and the OAM is highlighted (but it may be similar for other entities in the network that handle the establishment of these rules). Figure 6 The signaling of the gNB with an example gNB for FL process and gNB for transfer learning is shown.

[0066] At 601, OAM 190 performs federated learning initialization for the ML function ID. At 602-1, OAM 190 requests a local model and a specific metric or embedding from gNB1 170-1. At 602-2, OAM 190 requests a local model and a specific metric or embedding from gNB2 170-2. At 603-1, gNB1 170-1 computes the embedding. At 603-2, gNB2 170-2 computes the embedding. At 604-1, gNB1 170-1 sends the computed embedding to OAM 190. At 604-2, gNB2 170-2 sends the computed embedding to OAM 190. At 605, OAM 190 identifies a cluster based on the embeddings received from gNB1 170-1 and gNB2 170-2, and identifies the gNBs on which to perform federated learning or transfer learning.

[0067] At 606, OAM 190 sends an indication to gNB1 170-1 to perform transfer learning (e.g., OAM 190 at 605 identifies that gNB1 170-1 performs transfer learning). At 607, OAM 190 sends an indication to gNB2 170-2 to perform federated learning (e.g., OAM 190 at 605 identifies that gNB2 170-2 performs federated learning). At 608, gNB1 170-1, gNB2 170-2, and OAM 190 perform FL training iterations. At 609, OAM 190 identifies the closest local model (e.g., gNB2 170-2) for gNB1 170-1 to perform transfer learning by determining the nearest cell using Euclidean distance. At 610, OAM 190 sends an indication to gNB1 170-1 that the model for gNB2 170-2 is closest to perform transfer learning. At 611, gNB1 170-1 requests a local model from gNB2 170-2. At 612, once the FL is terminated, gNB2 170-2 sends the local model to gNB1 170-1. At 613, gNB1 170-1 performs transfer learning using the local model received from gNB2 170-2. Transfer learning begins and accelerates the process of training the gNB1 170-1 model. At 614, gNB1 170-1 checks whether the local model has been trained. At 615, gNB1 170-1 sends an indication of training completion to OAM 190.

[0068] Alternative Embodiment: In the methods described herein, transfer learning has been proposed to speed up the training process for cells that do not have similar training data as any other cell (not part of any group). However, this is an optional feature and these cells that do not have similar training data as any other cell (not part of any group) may not use TL and perform only local model learning. In this case, no communication is required between gNBs to receive data for starting transfer learning and the gNBs may perform learning based only on their local data.

[0069] Applicability: The solution described in this article can be implemented within AI / ML based 5G CA Scell ​​select machines, functions or features, where the ML training is done at the base station (using an embedded ML framework).

[0070] Figure 7 700 is an example apparatus 700 that can be implemented in hardware and configured to implement the examples described herein. The apparatus 700 includes at least one processor 702 (e.g., FPGA and / or CPU), one or more memories 704, including computer program code 705, the computer program code 705 having instructions for performing the methods described herein, wherein the at least one memory 704 and the computer program code 705 are configured to, together with the at least one processor 702, cause the apparatus 700 to implement a circuit system, process, component, module, or function (implemented using a control module 706) to implement the examples described herein. The memory 704 can be a non-transitory memory, a transient memory, a volatile memory (e.g., RAM), or a non-volatile memory (e.g., ROM).

[0071] The optionally included cluster 730 may enable identification based on embedded clusters, such as in Figure 6 The optional included identifier 740 may implement the identification of the gNB to perform federated learning or transfer learning, such as in Figure 6 The identification 740 may be implemented to identify the closest local model of the gNB to perform transfer learning, for example, Figure 6 Item 609 of the present invention is performed using OAM 190. Optionally included transfer learning 750 can implement transfer learning as described herein, such as in Figure 6 Item 613 of is performed using gNB1 170-1. Optionally included federated learning 760 may implement federated learning as described herein, such as in Figure 6 Item 608 is performed using gNB1170-1, gNB2 170-2 and OAM 190.

[0072] The device 700 includes a display and / or I / O interface 708, which includes user interface (UI) circuit systems and components, which can be used to display aspects or states of the methods described herein (e.g., while one of the methods is being performed or at a later time), or receive input from a user, such as using a keyboard, camera, touch screen, touch area, microphone, biometrics, one or more sensors, etc. The device 700 includes one or more communications, such as network (N / W) interface (I / F) 710. The communication I / F(s) 710 can be wired and / or wireless, and can communicate over the Internet / (other) networks via any communication technology, including via one or more links 724. The link(s) 724 can be Figure 1 (multiple) links 131 and / or 176 in. Figure 1 The link(s) 131 and / or 176 in the communication I / F 710 may also be implemented using the transceiver(s) 716 and the corresponding wireless link(s) 726. The communication I / F(s) 710 may include one or more transmitters or one or more receivers.

[0073] The transceiver 716 includes one or more transmitters 718 and one or more receivers 720. The transceiver 716 and / or the communication I / F(s) 710 may include standard well-known components such as amplifiers, filters, frequency converters, (de)modulators and encoder / decoder circuitry and one or more antennas, such as the antenna 714 used for communicating over the wireless link 726.

[0074] The control module 706 of the device 700 includes one or both of the parts 706-1 and / or 706-2, which can be implemented in a variety of ways. The control module 706 can be implemented in hardware as the control module 706-1, such as being implemented as part of one or more processors 702. The control module 706-1 can also be implemented as an integrated circuit or by other hardware (such as a programmable gate array). In another example, the control module 706 can be implemented as a control module 706-2, which is implemented as a computer program code (with corresponding instructions) 705 and executed by one or more processors 702. For example, one or more memories 704 store instructions that, when executed by one or more processors 702, cause the device 700 to perform one or more operations described herein. In addition, one or more processors 702, one or more memories 704, and example algorithms (e.g., as flow charts and / or signaling diagrams) are encoded as instructions, programs, or codes, and are components for causing the execution of the operations described herein.

[0075] The apparatus 700 implementing the functionality of the control 706 may be a UE 110, a RAN node 170 (e.g., a gNB), or (multiple) network elements 190 (e.g., a LMF 190). Therefore, the processor 702 can correspond to (multiple) processors 120, (multiple) processors 152 and / or (multiple) processors 175, the memory 704 can correspond to one or more memories 125, one or more memories 155 and / or one or more memories 171, the computer program code 705 can correspond to computer program code 123, computer program code 153 and / or computer program code 173, the control module 706 can correspond to module 140-1, module 140-2, module 150-1 and / or module 150-2 and (multiple) communication I / Fs 710 and / or transceiver 716 can correspond to transceiver 130, (multiple) antennas 128, transceiver 160, (multiple) antennas 158, (multiple) N / WI / Fs 161 and / or (multiple) N / WI / Fs 180. Alternatively, apparatus 700 and its elements may not correspond to UE 110, RAN node 170, or network element 190 and their respective elements, as apparatus 700 may be part of a Self-Organizing / Optimizing Network (SON) node or other node, such as a node in a cloud.

[0076] Apparatus 700 may also correspond to UE1 110-1, UE2 110-2, UE3 110-3 (wherein UE1 110-1, UE2 110-2, UE3 110-3 are configured similarly to UE 110), gNB1 170-1, gNB2 170-2 (wherein gNB1 170-1 and gNB2 170-2 are configured similarly to RAN node 170), or OAM 190 (for example, when OAM 190 is configured similarly to one or more network elements 190).

[0077] The apparatus 700 may also be distributed throughout a network (eg, 100 ), including within and between the apparatus 700 and any network elements, such as a network control element (NCE) 190 and / or a RAN node 170 and / or a UE 110 .

[0078] Interface 712 enables data communication and signaling between various items of apparatus 700, such as Figure 7As shown. For example, interface 712 can be one or more buses, such as an address, data, or control bus, and can include any interconnection mechanism, such as a series of lines on a motherboard or integrated circuit, optical fiber or other optical communication device, etc. Computer program code (e.g., instructions) 705 (including control 706) can include object-oriented software that is configured to pass data or messages between objects within computer program code 705. Device 700 need not include each of the features mentioned, or may also include other features. The various components of device 700 can be at least partially located in a common housing 728, or subsets of the various components of device 700 can be at least partially located in different housings, which can include housing 728.

[0079] Figure 8 Schematic representations of non-volatile storage media 800a (e.g., computer / compact disk (CD) or digital versatile disk (DVD)) and 800b (e.g., universal serial bus (USB) memory stick) and 800c (e.g., cloud storage for downloading instructions and / or parameters 802 or receiving instructions and / or parameters 802 sent via email) storing instructions and / or parameters 802 that, when executed by a processor, allow the processor to perform one or more steps of the methods described herein. Instructions and / or parameters 802 may represent non-transitory computer-readable media.

[0080] Fig. 9 The present invention is an example method 900 based on example embodiments described herein. At 910, the method includes receiving information related to at least one parameter of a model of the plurality of network nodes from a plurality of network nodes. At 920, the method includes determining at least one cluster of the plurality of network nodes based on at least one similarity criterion and the information related to at least one parameter of the model of the plurality of network nodes. At 930, the method includes determining at least one global model for the at least one cluster using local models of network nodes belonging to the at least one cluster. The method 900 may be performed using one or more network elements 190 (e.g., OAM 190) or apparatus 700.

[0081] Fig.10An example method 1000 based on example embodiments described herein. At 1010, the method includes sending information related to at least one parameter of a model of a device to a network entity. At 1020, the method includes receiving an indication from the network entity to perform federated learning with the network entity in response to the device being located within a cluster of network nodes that are similar to the device based on at least one similarity criterion and the information related to at least one parameter of the model of the device. At 1030, the method includes performing federated learning with the network entity in response to receiving an indication from the network entity to perform federated learning with the network entity. Method 1000 may be performed using RAN node 170 (e.g., gNB 170) or device 700.

[0082] Fig.11 The method 1100 is based on the example embodiments described herein. At 1110, the method includes receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes. At 1120, the method includes receiving or accessing a local model of the network node. At 1130, the method includes performing inference using at least one of: a global model based on federated learning, or a local model of the network node. The method 1100 may be performed using the UE 110 or the apparatus 700.

[0083] The following examples are provided and described herein.

[0084] Example 1. A device comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the device to at least: receive information related to at least one parameter of a model of the multiple network nodes from a plurality of network nodes; determine at least one cluster of the multiple network nodes based on at least one similarity criterion and the information related to at least one parameter of the model of the multiple network nodes; and determine at least one global model for at least one cluster using local models of network nodes belonging to at least one cluster.

[0085] Example 2. The apparatus of Example 1, wherein determining at least one global model for at least one cluster is performed using federated learning with network nodes belonging to the at least one cluster.

[0086] Example 3. An apparatus of any one of Examples 1 to 2, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: send an instruction to perform local model training for federated learning to a network node within at least one cluster; wherein the instruction to perform federated learning is sent to the network node within at least one cluster in response to the network node belonging to at least one cluster.

[0087] Example 4. An apparatus of any one of Examples 1 to 3, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: determine that the network node does not belong to any of at least one cluster; and in response to determining that the network node does not belong to any of the at least one cluster, send an instruction to perform local model training to the network node that does not belong to any of the at least one cluster.

[0088] Example 5. The apparatus of Example 4, wherein local model training is performed using transfer learning.

[0089] Example 6. An apparatus of any one of Examples 4 to 5, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: determine a model of a second network node, the model of the second network node being configured to be used together with the first network node to perform transfer learning; wherein the first network node includes: a network node that does not belong to any of the at least one cluster; and send an instruction to the first network node to perform transfer learning using the determined model of the second network node.

[0090] Example 7. The apparatus of Example 6, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: determine a model of the second network node based on a Euclidean distance between the model of the first network node and the model of the second network node, the model of the second network node being configured to be used with the first network node to perform transfer learning; wherein the Euclidean distance between the model of the first network node and the model of the second network node is less than a Euclidean distance between the model of the first network node and models of other network nodes in the plurality of network nodes.

[0091] Example 8. The apparatus of any one of Examples 4 to 7, wherein determining that the network node does not belong to any of the at least one cluster is based on at least one or more of a density estimate or a threshold.

[0092] Example 9. An apparatus of any one of Examples 4 to 8, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: determine a trained global model of a cluster, the trained global model being configured to be used with network nodes that do not belong to any of the at least one cluster; determine, based on a similarity criterion, that a similarity measure of a network node that does not belong to any of the at least one cluster is more similar to a similarity measure of a network node that belongs to a cluster than to a similarity measure of a network node that does not belong to a cluster; and send an instruction to a network node that does not belong to any of the at least one cluster to perform transfer learning using the trained global model of the cluster.

[0093] Example 10. An apparatus of any one of Examples 1 to 9, wherein information related to at least one parameter of a model of multiple network nodes used to determine at least one cluster of multiple network nodes includes at least one of the following: an embedding of at least one parameter of the model of multiple network nodes, or a local training data sample used to generate the model of multiple network nodes.

[0094] Example 11. An apparatus of Example 10, wherein the embedding is based on at least one of: a downlink physical resource block usage time series, or an uplink physical resource block usage time series, or a throughput time series, or the number of radio resource control connected users, or the number of radio resource control active users, or principal component analysis, or t-distributed random adjacent embedding.

[0095] Example 12. An apparatus of any one of Examples 1 to 11, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: send a request to a plurality of network nodes for information related to at least one parameter of a model of the plurality of network nodes; wherein the information related to at least one parameter of the model of the plurality of network nodes is received in response to sending the request for information.

[0096] Example 13. An apparatus of any one of Examples 1 to 12, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive trained local models of multiple network nodes from the multiple network nodes; and determine at least one global model using at least some of the trained local models received from the multiple network nodes; wherein the information received from the multiple network nodes related to at least one parameter of the models of the multiple network nodes includes the trained local models.

[0097] Example 14. The apparatus of any one of Examples 1 to 13, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: send at least one global model to a network node within at least one cluster.

[0098] Example 15. The apparatus of any one of Examples 1 to 14, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: provide access to at least one global model to network nodes within at least one cluster.

[0099] Example 16. The apparatus of any one of Examples 1 to 15, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: determine a global model for each at least one cluster using information related to at least one parameter of the models of the plurality of network nodes.

[0100] Example 17. The apparatus of any one of Examples 1 to 16, wherein the plurality of network nodes comprises a radio access network node.

[0101] Example 18. The apparatus of any one of Examples 1 to 17, wherein at least one global model is a trained global model for the corresponding at least one cluster.

[0102] Example 19. A device comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the device to at least: send information related to at least one parameter of a model of the device to a network entity; in response to the device being located within a cluster of network nodes similar to the device based on at least one similarity criterion and the information related to at least one parameter of the model of the device, receive an indication from the network entity to perform federated learning with the network entity; and in response to receiving an indication from the network entity to perform federated learning with the network entity, perform federated learning with the network entity.

[0103] Example 20. The apparatus of Example 19, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive an instruction from a network entity to perform local model training for federated learning.

[0104] Example 21. The apparatus of any one of Examples 19 to 20, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive an instruction from a network entity to perform local model training in response to the apparatus not belonging to a cluster of any network nodes.

[0105] Example 22. The apparatus of Example 21, wherein local model training is performed using transfer learning.

[0106] Example 23. The device of any of Examples 21 to 22, wherein the device not belonging to any cluster is based on at least one or more of a density estimate or a threshold.

[0107] Example 24. An apparatus of any one of Examples 19 to 23, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive an instruction from a network entity to perform transfer learning using a network node in response to the apparatus not belonging to any cluster of network nodes; and perform transfer learning using the network node that received the instruction to perform transfer learning received from the network entity.

[0108] Example 25. The apparatus of Example 24, wherein the network node belongs to a cluster of network nodes having similar local models.

[0109] Example 26. The apparatus of any one of Examples 24 to 25, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive, from a network entity, an indication of a model of a network node with which to perform transfer learning.

[0110] Example 27. The apparatus of Example 26, wherein the Euclidean distance between the model of the apparatus and the model of the network node with which transfer learning is performed is smaller than the Euclidean distance between the model of the apparatus and the models of other network nodes.

[0111] Example 28. An apparatus of any one of Examples 19 to 27, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive an instruction from a network entity to perform transfer learning using a trained global model of a cluster in response to the apparatus not belonging to any cluster of network nodes; wherein based on a similarity criterion, a similarity measure of the apparatus that does not belong to any cluster of network nodes is more similar to a similarity measure of a network node that belongs to a cluster having a trained global model than to a similarity measure of a network node that does not belong to a cluster having a trained global model.

[0112] Example 29. The device of any of Examples 19 to 28, wherein the information related to at least one parameter of the device's model includes at least one of: an embedding of at least one parameter of the device's model, or a local training data sample used to generate the device's model.

[0113] Example 30. An apparatus of Example 29, wherein the embedding is based on at least one of: a downlink physical resource block usage time series, or an uplink physical resource block usage time series, or a throughput time series, or the number of radio resource control connected users, or the number of radio resource control active users, or principal component analysis, or t-distributed random adjacent embedding.

[0114] Example 31. An apparatus of any of Examples 19 to 30, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive a request for information related to at least one parameter of a model of the apparatus from a network entity; wherein the information related to at least one parameter of the model of the apparatus is sent in response to receiving the request for information.

[0115] Example 32. An apparatus of any one of Examples 19 to 31, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: send a trained local model of the apparatus to a network entity, the trained local model being configured to be used to learn a global model for a cluster of network nodes; wherein the information related to at least one parameter of the model of the apparatus that is sent to the network entity includes: the trained local model of the apparatus.

[0116] Example 33. The apparatus of any one of Examples 19 to 32, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: receive the global model from a network entity; and perform inference using the global model.

[0117] Example 34. The apparatus of any one of Examples 19 to 33, wherein the instructions, when executed by at least one processor, cause the apparatus to at least: obtain access to a global model; and perform inference using the global model.

[0118] Example 35. The apparatus of any one of Examples 19 to 34, wherein the apparatus comprises a radio access network node.

[0119] Example 36. A device comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the device to at least: receive or access a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; receive or access a local model of a network node; and perform inference using at least one of: a global model based on federated learning, or a local model of a network node.

[0120] Example 37. The apparatus of Example 36, wherein the local model is based on transfer learning with network nodes that are not part of the cluster.

[0121] Example 38. The apparatus of any of Examples 36 to 37, wherein the apparatus comprises a user device.

[0122] Example 39. A method comprising: receiving information related to at least one parameter of a model of the multiple network nodes from the multiple network nodes; determining at least one cluster of the multiple network nodes based on at least one similarity criterion and the information related to at least one parameter of the model of the multiple network nodes; and determining at least one global model for the at least one cluster using local models of the network nodes belonging to the at least one cluster.

[0123] Example 40. A method comprising: sending information related to at least one parameter of a model of a device to a network entity; in response to the device being located within a cluster of network nodes similar to the device based on at least one similarity criterion and the information related to at least one parameter of the model of the device, receiving an indication from the network entity to perform federated learning with the network entity; and in response to receiving an indication from the network entity to perform federated learning with the network entity, performing federated learning with the network entity.

[0124] Example 41. A method comprising: receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; receiving or accessing a local model of a network node; and performing inference using at least one of: a global model based on federated learning, or a local model of a network node.

[0125] Example 42. An apparatus comprising: a component for receiving information related to at least one parameter of a model of the multiple network nodes from a plurality of network nodes; a component for determining at least one cluster of the multiple network nodes based on at least one similarity criterion and the information related to at least one parameter of the model of the multiple network nodes; and a component for determining at least one global model for at least one cluster using local models of network nodes belonging to at least one cluster.

[0126] Example 43. A device comprising: a component for sending information related to at least one parameter of a model of the device to a network entity; a component for receiving an indication from the network entity to perform federated learning with the network entity in response to the device being located within a cluster of network nodes similar to the device based on at least one similarity criterion and the information related to at least one parameter of the model of the device; and a component for performing federated learning with the network entity in response to receiving an indication from the network entity to perform federated learning with the network entity.

[0127] Example 44. An apparatus comprising: a component for receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; a component for receiving or accessing a local model of a network node; and a component for performing inference using at least one of: a global model based on federated learning, or a local model of a network node.

[0128] Example 45. A computer-readable medium comprising instructions stored thereon for performing at least the following: receiving information related to at least one parameter of a model of the multiple network nodes from the multiple network nodes; determining at least one cluster of the multiple network nodes based on at least one similarity criterion and the information related to at least one parameter of the model of the multiple network nodes; and determining at least one global model for the at least one cluster using local models of the network nodes belonging to the at least one cluster.

[0129] Example 46. A computer-readable medium comprising instructions stored thereon for performing at least the following: sending information related to at least one parameter of a model of a device to a network entity; in response to the device being located within a cluster of network nodes similar to the device based on at least one similarity criterion and the information related to at least one parameter of the model of the device, receiving an indication from the network entity to perform federated learning with the network entity; and in response to receiving an indication from the network entity to perform federated learning with the network entity, performing federated learning with the network entity.

[0130] Example 47. A computer-readable medium comprising instructions stored thereon for performing at least the following: receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; receiving or accessing a local model of a network node; and performing inference using at least one of: a global model based on federated learning, or a local model of a network node.

[0131] It should be understood that references to "computers," "processors," and the like include not only computers having different architectures (such as single / multi-processor architectures and sequential or parallel architectures), but also special-purpose circuits (such as field programmable gate arrays (FPGAs), application-specific circuits (ASICs), signal processing devices, and other processing circuit systems). It should be understood that references to computer programs, instructions, codes, and the like include software or firmware for programmable processors, such as programmable content of hardware devices, whether instructions for a processor, or configuration settings for a fixed-function device, gate array, or programmable logic device, and the like.

[0132] The memory described herein can be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, non-transitory memory, transient memory, fixed memory, and removable memory. The memory can include a database for storing data.

[0133] As used herein, the term "circuitry" may refer to the following: (a) hardware circuit implementations, such as implementations in analog and / or digital circuitry, and (b) combinations of circuitry and software (and / or firmware), such as, as applicable: (i) a combination of (multiple) processors or (ii) portions of (multiple) processors / software, including (multiple) digital signal processors, software, and memory, which work together to enable the device to perform various functions, and (c) circuitry, which is associated with (multiple) microprocessors or portions of (multiple) microprocessors that require software or firmware to operate, even if the software or firmware is not physically present. As a further example, as used herein, the term "circuitry" would also cover implementations of only a processor (or multiple processors) or a portion of a processor and its (their) accompanying software and / or firmware. For example, if applicable to the particular element, the term "circuitry" would also cover a baseband integrated circuit or application processor integrated circuit for a mobile phone or a similar integrated circuit in a server, cellular network device, or other network device.

[0134] It should be understood that the above description is illustrative only. Various alternatives and modifications can be designed by those skilled in the art. For example, the features described in each dependent claim can be combined with each other in any suitable (multiple) combination. In addition, the features from the above different example embodiments can be selectively combined into new example embodiments. Therefore, this description is intended to cover all such alternatives, modifications and variations that belong to the scope of the appended claims.

[0135] The following acronyms and abbreviations that may be found in the specification and / or drawings are as follows (abbreviations and acronyms may be appended / combined with each other or with other characters such as dashes, hyphens, slashes, letters or numbers, and may not be case sensitive):

[0136] 4G Fourth Generation

[0137] 5G Fifth Generation

[0138] 5GC 5G Core Network

[0139] AI

[0140] AMF Access and Mobility Management Function

[0141] ASIC Application-Specific Integrated Circuit

[0142] CA Carrier Aggregation

[0143] CD / Computer CD

[0144] CPU Central Processing Unit

[0145] CRI CSI reference signal resource indicator

[0146] CSI Channel State Information

[0147] CU Central Unit or Centralized Unit

[0148] d dimension, dimension (e.g. 2d, 10d)

[0149] DBSCAN Density-based spatial clustering of applications with noise

[0150] dim dimension

[0151] DL Downlink

[0152] DSP Digital Signal Processor

[0153] DU Distributed Unit

[0154] DVD Digital Versatile Disc

[0155] eNB Evolved Node B (e.g. LTE base station)

[0156] EN-DC E-UTRAN New Radio – Dual Connectivity

[0157] en-gNB provides the NR user plane and control plane protocol termination node to the UE and acts as a secondary node in EN-DC

[0158] E-UTRA Evolved UMTS Terrestrial Radio Access, also known as LTE radio access technology

[0159] E-UTRAN E-UTRA Network

[0160] F1 Interface between CU and DU

[0161] FL Federated Learning

[0162] FPGA Field Programmable Gate Array

[0163] GMM Gaussian Mixture Modeling

[0164] gNB is a base station for 5G / NR, which is a node that provides NR user plane and control plane protocol termination to UE and is connected to 5GC via NG interface.

[0165] K is the number of clusters (e.g. k-means)

[0166] IAB Integrated Access and Backhaul

[0167] ID Identifier

[0168] I / F Interface

[0169] I / O Input / Output

[0170] LI layer indicator

[0171] LMF Location Management Function

[0172] LTE Long Term Evolution (4G)

[0173] MAC Media Access Control

[0174] ML Machine Learning

[0175] MME Mobility Management Entity

[0176] MN Mobile Network

[0177] MRO Mobility Robustness Optimization

[0178] NCE Network Control Element

[0179] ng or NG new generation

[0180] ng-eNB Next Generation eNB

[0181] NG-RAN Next Generation Radio Access Network

[0182] NN Neural Network

[0183] NR New Radio

[0184] N / W Network

[0185] OAM Operations, Administration and Maintenance

[0186] PCA Principal Component Analysis

[0187] PDA Personal Digital Assistant

[0188] PDCP Packet Data Convergence Protocol

[0189] PHY Physical Layer

[0190] PMI Precoding matrix indicator

[0191] PRB Physical Resource Block

[0192] Q one or more expected rewards for one or more actions taken in a given state (e.g. Q-learning)

[0193] RAM Random Access Memory

[0194] RAN Radio Access Network

[0195] RI Rating Indicator

[0196] RLC Radio Link Control

[0197] ROM Read Only Memory

[0198] RRC Radio Resource Control

[0199] RU Radio Unit

[0200] Rx Receive, or Receiver or Reception

[0201] Scell

[0202] SDAP Service Data Adaptation Protocol

[0203] SE Spectral Efficiency

[0204] SGW Service Gateway

[0205] SMF session management functions

[0206] SON self-organizing / optimizing network

[0207] SRS Sounding Reference Signal

[0208] TL Transfer Learning

[0209] TRP Transmission and Reception Point

[0210] t-SNE t-Distributed Stochastic Neighbor Embedding

[0211] Tx Send, or transmitter or transmission

[0212] UAV

[0213] UE User Equipment (e.g., wireless, usually mobile)

[0214] UI User Interface

[0215] UL Uplink

[0216] UMTS Universal Mobile Telecommunications System

[0217] UPF User Plane Function

[0218] USB Universal Serial Bus

[0219] UTRAN UMTS Terrestrial Radio Access Network

[0220] X2 Network interface between RAN nodes and between RAN and core network

[0221] Xn Network interface between NG-RAN nodes

Claims

1. A device for communication, comprising: at least one processor; as well as at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: receiving, from a plurality of network nodes, information related to at least one parameter of a model of the plurality of network nodes; determining at least one cluster of the plurality of network nodes based on at least one similarity criterion and the information related to the at least one parameter of the model of the plurality of network nodes; as well as At least one global model for the at least one cluster is determined using local models of network nodes belonging to the at least one cluster. 2 . The apparatus of claim 1 , wherein determining the at least one global model for the at least one cluster is performed using federated learning with the network nodes belonging to the at least one cluster.

3. The apparatus according to any one of claims 1 to 2, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: Sending an instruction to the network nodes within the at least one cluster to perform local model training for federated learning; Wherein, in response to the network node belonging to the at least one cluster, the instruction to perform federated learning is sent to the network node within the at least one cluster.

4. The apparatus according to any one of claims 1 to 3, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: determining that the network node does not belong to any of the at least one cluster; and In response to determining that the network node does not belong to any of the at least one cluster, an instruction to perform local model training is sent to the network node that does not belong to any of the at least one cluster.

5. The apparatus of claim 4, wherein the local model training is performed using transfer learning.

6. The apparatus according to any one of claims 4 to 5, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: determining a model of a second network node, the second network node configured to be used with the first network node to perform transfer learning; The first network node comprises: said network node not belonging to any of said at least one cluster; as well as An instruction is sent to the first network node to perform transfer learning using the determined model of the second network node.

7. The apparatus of claim 6, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: determining the model of the second network node based on a Euclidean distance between the model of the first network node and the model of the second network node, the second network node being configured to be used with the first network node to perform transfer learning; The Euclidean distance between the model of the first network node and the model of the second network node is smaller than the Euclidean distance between the model of the first network node and models of other network nodes in the plurality of network nodes.

8. The apparatus according to any one of claims 4 to 7, wherein determining that the network node does not belong to any of the at least one cluster is based on at least one or more of density estimation or thresholding.

9. The apparatus of any one of claims 4 to 8, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: determining a trained global model of clusters, the trained global model configured to be used with the network nodes that do not belong to any of the at least one cluster; determining, based on a similarity criterion, that a similarity measure of the network nodes that do not belong to any of the at least one cluster is more similar to a similarity measure of a network node that belongs to the cluster than to a similarity measure of a network node that does not belong to the cluster; An instruction to perform transfer learning using the trained global model of the cluster is sent to the network node that does not belong to any of the at least one cluster.

10. The apparatus according to any one of claims 1 to 9, wherein the information related to the at least one parameter of the model of the plurality of network nodes used to determine the at least one cluster of the plurality of network nodes comprises at least one of the following: embedding of the at least one parameter of the model of the plurality of network nodes, or Local training data samples are used to generate the models of the plurality of network nodes.

11. The apparatus of claim 10, wherein the embedding is based on at least one of: Downlink physical resource blocks use time sequence, or Uplink physical resource blocks use time sequence, or Throughput time series, or The number of connected users of the radio resource control, or The number of active RRC users, or Principal components analysis, or t-Distributed Random Neighbor Embedding.

12. The apparatus of any one of claims 1 to 11, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: sending, to the plurality of network nodes, a request for the information related to at least one parameter of a model of the plurality of network nodes; wherein the information relating to the at least one parameter of the model of the plurality of network nodes is received in response to sending the request for the information.

13. The apparatus of any one of claims 1 to 12, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: receiving, from the plurality of network nodes, trained local models of the plurality of network nodes; and determining the at least one global model using at least some of the trained local models received from the plurality of network nodes; The information received from the plurality of network nodes and related to at least one parameter of the model of the plurality of network nodes comprises: The local model is trained.

14. The apparatus of any one of claims 1 to 13, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: The at least one global model is sent to the network nodes within the at least one cluster.

15. The apparatus of any one of claims 1 to 14, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: Access to the at least one global model is provided to the network nodes within the at least one cluster.

16. The apparatus of any one of claims 1 to 15, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: A global model for each of the at least one cluster is determined using the information related to the at least one parameter of the models of the plurality of network nodes.

17. The apparatus according to any one of claims 1 to 16, wherein the plurality of network nodes comprises radio access network nodes.

18. The apparatus according to any one of claims 1 to 17, wherein the at least one global model is a trained global model for the corresponding at least one cluster.

19. An apparatus for communication, comprising: at least one processor; as well as at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: sending information related to at least one parameter of a model of the device to a network entity; receiving, in response to the device being located within a cluster of network nodes that are similar to the device based on at least one similarity criterion and the information related to the at least one parameter of the model of the device, an indication from the network entity to perform federated learning with the network entity; as well as In response to receiving the indication from the network entity to perform federated learning with the network entity, performing federated learning with the network entity.

20. The apparatus of claim 19, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: An instruction to perform local model training for the federated learning is received from the network entity.

21. The apparatus of any one of claims 19 to 20, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: In response to the device not belonging to any cluster of network nodes, an indication is received from the network entity to perform local model training.

22. The apparatus of claim 21, wherein the local model training is performed using transfer learning.

23. The device of any one of claims 21 to 22, wherein the device not belonging to any cluster is based on at least one or more of density estimation or thresholding.

24. The apparatus of any one of claims 19 to 23, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: In response to the apparatus not belonging to any cluster of network nodes, receiving an indication from the network entity to perform transfer learning with the network nodes; and Transfer learning is performed using the network node received with the indication to perform transfer learning received from the network entity.

25. The apparatus of claim 24, wherein the network node belongs to a cluster of network nodes having similar local models.

26. The apparatus of any one of claims 24 to 25, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: An indication of a model of the network node for performing the transfer learning is received from the network entity.

27. The device of claim 26, wherein the Euclidean distance between the model of the device and the model of the network node used to perform transfer learning is smaller than the Euclidean distance between the model of the device and models of other network nodes.

28. The apparatus of any one of claims 19 to 27, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: In response to the device not belonging to any cluster of network nodes, receiving an instruction from the network entity to perform transfer learning using a trained global model of a cluster; Wherein based on the similarity criterion, a similarity measure of the device not belonging to any cluster of network nodes is more similar to a similarity measure of network nodes belonging to the cluster with the trained global model than to a similarity measure of network nodes not belonging to the cluster with the trained global model.

29. The device according to any one of claims 19 to 28, wherein the information related to the at least one parameter of the model of the device comprises at least one of the following: embedding of said at least one parameter of said model of said device, or A local training data sample is used to generate the model of the device.

30. The apparatus of claim 29, wherein the embedding is based on at least one of: Downlink physical resource blocks use time sequence, or Uplink physical resource blocks use time sequence, or Throughput time series, or The number of connected users of the radio resource control, or The number of active RRC users, or Principal components analysis, or t-Distributed Random Neighbor Embedding.

31. The apparatus of any one of claims 19 to 30, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: receiving, from the network entity, a request for the information related to the at least one parameter of the model of the device; Wherein the information related to the at least one parameter of the model of the device is sent in response to receiving the request for the information.

32. An apparatus according to any one of claims 19 to 31, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: sending a trained local model of the apparatus to the network entity, the trained local model configured for learning a global model for the cluster of network nodes; The information related to the at least one parameter of the model of the device sent to the network entity comprises: The trained local model of the device.

33. An apparatus according to any one of claims 19 to 32, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: receiving the global model from the network entity; and Inference is performed using the global model.

34. An apparatus according to any one of claims 19 to 33, wherein the instructions, when executed by the at least one processor, cause the apparatus to at least: obtaining access to the global model; and Inference is performed using the global model.

35. An apparatus according to any one of claims 19 to 34, wherein the apparatus comprises a radio access network node.

36. An apparatus for communication, comprising: at least one processor; as well as at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; a local model of the receiving or accessing network node; and Inference is performed using at least one of: the global model based on federated learning, or the local model of the network node.

37. The apparatus of claim 36, wherein the local model is based on transfer learning utilizing network nodes that are not part of the cluster.

38. An apparatus according to any one of claims 36 to 37, wherein the apparatus comprises user equipment.

39. A method for communication, comprising: receiving, from a plurality of network nodes, information related to at least one parameter of a model of the plurality of network nodes; determining at least one cluster of the plurality of network nodes based on at least one similarity criterion and the information related to the at least one parameter of the model of the plurality of network nodes; as well as At least one global model for the at least one cluster is determined using local models of network nodes belonging to the at least one cluster.

40. A method for communication, comprising: sending information related to at least one parameter of a model of the device to a network entity; receiving, in response to the device being located within a cluster of network nodes that are similar to the device based on at least one similarity criterion and the information related to the at least one parameter of the model of the device, an indication from the network entity to perform federated learning with the network entity; as well as In response to receiving the indication from the network entity to perform federated learning with the network entity, performing federated learning with the network entity.

41. A method for communication, comprising: receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; A local model of a receiving or accessing network node; as well as Inference is performed using at least one of: the global model based on federated learning, or the local model of the network node.

42. An apparatus for communication, comprising: means for receiving, from a plurality of network nodes, information relating to at least one parameter of a model of the plurality of network nodes; means for determining at least one cluster of said plurality of network nodes based on at least one similarity criterion and said information relating to said at least one parameter of a model of said plurality of network nodes; as well as Means for determining at least one global model for the at least one cluster using local models of network nodes belonging to the at least one cluster.

43. An apparatus for communication, comprising: means for sending information related to at least one parameter of a model of the device to a network entity; means for receiving, from the network entity, an indication to perform federated learning with the network entity in response to the apparatus being located within a cluster of network nodes that are similar to the apparatus based on at least one similarity criterion and the information related to the at least one parameter of the model of the apparatus; as well as Means for performing federated learning with the network entity in response to receiving the indication from the network entity to perform federated learning with the network entity.

44. An apparatus for communication, comprising: means for receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; means for receiving or accessing a local model of a network node; as well as Means for performing inference using at least one of: the global model based on federated learning, or the local model of the network node.

45. A computer readable medium comprising instructions stored thereon, the instructions for performing at least the following: receiving, from a plurality of network nodes, information related to at least one parameter of a model of the plurality of network nodes; determining at least one cluster of the plurality of network nodes based on at least one similarity criterion and the information related to the at least one parameter of the model of the plurality of network nodes; as well as At least one global model for the at least one cluster is determined using local models of network nodes belonging to the at least one cluster.

46. ​​A computer readable medium comprising instructions stored thereon for performing at least the following: sending information related to at least one parameter of a model of the device to a network entity; receiving, in response to the device being located within a cluster of network nodes that are similar to the device based on at least one similarity criterion and the information related to the at least one parameter of the model of the device, an indication from the network entity to perform federated learning with the network entity; and In response to receiving the indication from the network entity to perform federated learning with the network entity, performing federated learning with the network entity.

47. A computer readable medium comprising instructions stored thereon, the instructions for performing at least the following: receiving or accessing a global model, wherein the global model is based on federated learning with a cluster of similar network nodes; a local model of the receiving or accessing network node; and Inference is performed using at least one of: the global model based on federated learning, or the local model of the network node.

Images