Method for optimizing network by using feature extracted from network, and electronic device for performing same

Abstract

This method may comprise the steps of: obtaining, from each of one or more network entities, network entity data associated with each network entity; for the one or more network entities, generating network embeddings on the basis of the network entity data by using an encoder model; transforming the network embeddings into a predefined number of parameters by using a transformation model; inputting the predefined number of parameters to an inference model; obtaining, from the inference model, an output regarding the predefined number of parameters; and on the basis of the output of the inference model, determining one or more parameters associated with a network control.

Description

A method for optimizing a network using features extracted from a network and an electronic device for performing the same

[0001] The present disclosure relates to a wireless communication method and a wireless communication electronic device, and more specifically, to a method for optimizing a network using features extracted from a network and an electronic device for performing the same.

[0002] Wireless communication technologies have been developed primarily for human-oriented services, such as voice, multimedia, and data. With the development of 5G (5th-generation) communication systems, the number of connected devices linked to communication networks is increasing. Examples of networked devices include vehicles, robots, drones, home appliances, displays, smart sensors installed in various infrastructures, construction machinery, and factory equipment. Mobile devices are expected to evolve into various form factors, such as augmented reality glasses, virtual reality headsets, and holographic devices. In the 6G (6th-generation) era, efforts are underway to develop improved 6G communication systems to connect hundreds of billions of devices and objects to provide diverse services. For this reason, 6G communication systems are referred to as "beyond 5G" systems.

[0003] In a 6G communication system, the maximum transmission speed is tera (i.e., 1,000 gigabit) bps, and the wireless latency is 100 microseconds (μsec). In other words, compared to a 5G communication system, the transmission speed in a 6G communication system is 50 times faster, and the wireless latency is reduced to one-tenth.

[0004] To achieve such high data transmission speeds and ultra-low latency, 6G communication systems are being considered for implementation in the terahertz band (e.g., the 95 GHz to 3 terahertz (3 THz) band). In the terahertz band, due to more severe path loss and atmospheric absorption compared to the millimeter wave (mmWave) band introduced in 5G, the importance of technology capable of guaranteeing signal reach, or coverage, will increase. As key technologies to ensure coverage, radio frequency (RF) devices, antennas, new waveforms that offer better coverage than orthogonal frequency division multiplexing (OFDM), beamforming, and multi-antenna transmission technologies such as massive multiple-input and multiple-output (massive MIMO), full-dimensional MIMO (FD-MIMO), array antennas, and large-scale antennas must be developed. In addition, new technologies such as metamaterial-based lenses and antennas, high-dimensional spatial multiplexing technology using orbital angular momentum (OAM), and reconfigurable intelligent surface (RIS) are being discussed to improve coverage of terahertz band signals.

[0005] In addition, to improve frequency efficiency and system network, development is underway in 6G communication systems for full duplex technology, in which uplink and downlink simultaneously utilize the same frequency resources at the same time; network technology that integrates satellites and HAPS (high-altitude platform stations); network structure innovation technology that supports mobile base stations and enables network operation optimization and automation; dynamic spectrum sharing technology through collision avoidance based on spectrum usage prediction; AI-based communication technology that utilizes AI (artificial intelligence) from the design stage and internalizes end-to-end AI support functions to realize system optimization; and next-generation distributed computing technology that realizes services of complexity exceeding the limits of terminal computing capabilities by utilizing ultra-high performance communication and computing resources (mobile edge computing (MEC), cloud, etc.). In addition, attempts are continuing to further strengthen connectivity between devices, further optimize networks, promote the softwareization of network entities, and increase the openness of wireless communication through the design of new protocols to be used in 6G communication systems, the implementation of hardware-based security environments, the development of mechanisms for the safe utilization of data, and the development of technologies regarding privacy maintenance methods.

[0006] Through the research and development of such 6G communication systems, a new dimension of hyper-connected experience will become possible via the hyper-connectivity of 6G communication systems, which encompasses not only connections between objects but also connections between people and objects. Specifically, 6G communication systems will enable the provision of services such as truly immersive extended reality (truly immersive XR), high-fidelity mobile holograms, and digital replicas. Furthermore, services such as remote surgery, industrial automation, and emergency response, which enhance security and reliability, will be provided through 6G communication systems and applied in various fields including industry, healthcare, automotive, and home appliances.

[0007] According to one aspect of the present disclosure, the method may comprise: obtaining network entity data associated with each network entity from each of one or more network entities; for each of the one or more network entities, generating a network embedding based on the network entity data associated with each network entity using an encoder model; converting the network embeddings into a predetermined number of parameters using a transformation model; inputting the predetermined number of parameters into an inference model; obtaining an output regarding the predetermined number of parameters from the inference model; and determining one or more parameters associated with the control of the network based on the output of the inference model.

[0008] According to one aspect of the present disclosure, an electronic device may include: at least one processor comprising processing circuitry; and a memory comprising one or more storage media for storing one or more instructions. When the one or more instructions are executed by the at least one processor, the electronic device may: obtain network entity data associated with each network entity from each of one or more network entities; for each of the one or more network entities, generate a network embedding based on the network entity data associated with each network entity using an encoder model and convert the network embeddings into a predetermined number of parameters using a transformation model; input the predetermined number of parameters into an inference model; obtain an output regarding the predetermined number of parameters from the inference model; and determine one or more parameters associated with the control of a network based on the output of the inference model.

[0009] According to one aspect of the present disclosure, a computer-readable recording medium may record a program for performing any combination of methods, steps, operations, or functions according to one embodiment of the present disclosure on a computer.

[0010] FIG. 1 illustrates an example of a wireless communication system according to one embodiment of the present disclosure.

[0011] FIG. 2 illustrates an example of network optimization using features extracted from a network according to one embodiment of the present disclosure.

[0012] FIGS. 3a and 3b illustrate examples of network optimization according to one embodiment of the present disclosure.

[0013] FIG. 4 illustrates an example of an encoder model according to one embodiment of the present disclosure.

[0014] FIG. 5 illustrates an example of a conversion model according to one embodiment of the present disclosure.

[0015] FIG. 6 illustrates an example of training an encoder model according to one embodiment of the present disclosure.

[0016] FIG. 7 illustrates an example of training a transformation model and an inference model according to one embodiment of the present disclosure.

[0017] FIG. 8 illustrates an exemplary example of optimizing a base station policy according to one embodiment of the present disclosure.

[0018] FIG. 9 illustrates an example of a block diagram of an electronic device according to one embodiment of the present disclosure.

[0019] FIG. 10 illustrates an example of a flowchart of a method according to one embodiment of the present disclosure.

[0020] The terms used in the embodiments of this specification have been selected to be as widely used as possible, taking into account the functions of the present disclosure; however, these terms may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the description section of the relevant embodiments. Therefore, the terms used in this specification are defined not merely by their names, but based on the meanings of specific terms and the overall content of the present disclosure.

[0021] The terms used in this disclosure are used merely to describe specific embodiments and are not intended to limit the scope of other embodiments. A singular expression may include a plural expression unless the context clearly indicates otherwise. Terms used herein, including technical or scientific terms, may have the same meaning as generally understood by those skilled in the art described in this disclosure. Terms used in this disclosure that are defined in a general dictionary may be interpreted as having the same or similar meaning as they have in the context of the relevant technology, and are not interpreted in an ideal or overly formal sense unless explicitly defined in this disclosure. In some cases, even terms defined in this disclosure are not to be interpreted to exclude the embodiments of this disclosure.

[0022] In one or more embodiments of the present disclosure described below, a hardware-based approach is described as an example. However, since one or more embodiments of the present disclosure include techniques using both hardware and software, one or more embodiments of the present disclosure do not exclude a software-based approach.

[0023] Throughout this disclosure, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "...part," "...module," etc., as used in this specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software.

[0024] As used in this disclosure, the expression “configured to” may be replaced, depending on the context, with, for example, “suitable for,” “having the capacity to,” “designed to,” “adapted to,” “made to,” or “capable of.” The term “configured to” may not necessarily mean only “specifically designed to” in hardware. Instead, in some situations, the expression “system configured to” may mean that the system is “capable of” in conjunction with other devices or components. For example, the phrase “processor configured to perform A, B, and C” may mean a dedicated processor for performing the said operations (e.g., an embedded processor) or a generic-purpose processor capable of performing said operations by executing one or more software programs stored in memory.

[0025] In addition, when a component is described in the present disclosure as being "connected" or "connected" to another component, it should be understood that the component may be directly connected to or directly connected to the other component, but unless otherwise specifically stated, it may also be connected or connected through another component in between.

[0026] Additionally, in this disclosure, expressions such as "greater than" or "less than" may be used to determine whether a specific condition is satisfied or fulfilled; however, this is merely for the purpose of expressing an example and does not exclude descriptions such as "greater than" or "less than." Conditions described as "greater than" may be replaced with "greater than," conditions described as "less than" may be replaced with "less than," and conditions described as "greater than and less than" may be replaced with "greater than and less than."

[0027] This disclosure uses terms and names defined in the 3GPP LTE (Long Term Evolution) or NR (New Radio) standards, or terms and names modified therefrom. However, this disclosure is not limited to the terms and names described above and may be equally applied to systems conforming to other standards. In this disclosure, eNB (eNode B) may be used interchangeably with gNB (gNode B) for convenience of explanation. For example, a base station referred to as eNB may be understood as a gNB. In this disclosure, the term terminal may refer to User Equipment (UE), Mobile Station (MS), mobile phones, NB-IoT (narrowband-internet of things) devices, sensors, as well as various wireless communication devices.

[0028] In one embodiment, an 'Artificial Intelligence (AI) model' may be an algorithm, system, or model designed to perform a specific task by analyzing given input data. For example, an AI model may be an algorithm, system, or model that learns patterns from input data to perform inference, such as prediction, classification, or decision-making. An AI model may include explicit rule-based algorithms, machine learning (ML) models, and deep learning models. An AI model can learn patterns from training data and improve its performance on its own.

[0029] In one embodiment, 'machine learning' may be a technique that learns from given data and generalizes to unseen data to perform tasks without explicit instructions. Machine learning may include supervised learning, unsupervised learning, semi-supervised learning, and / or reinforcement learning, but is not limited to the examples described above. Through machine learning, one or more weights and / or parameters of an AI model may be optimized. For example, an AI model may improve its performance by adjusting weights and parameters during the learning (or training) process. During learning, the AI ​​model may update weights and parameters to minimize a loss or cost value. An AI model learned or trained through machine learning may be referred to as an 'ML model' or an 'AI / ML model'.

[0030] In one embodiment, the 'deep learning model' may refer to an AI model comprising a plurality of neural network layers. Each neural network layer may include one or more neurons, and each neuron may have one or more weights that are optimized through learning. The neurons of one layer may perform operations between the operation result (or output) of the previous layer and the corresponding weights of the current layer. Through these operations, the deep learning model can learn data and extract features from input data. The 'deep learning model' may be referred to as a 'neural network model'.

[0031] In one embodiment, the terms 'AI model', 'ML model', 'AI / ML model', 'deep learning model', and 'neural network model' may be used interchangeably.

[0032] In one embodiment, the AI ​​model may include various AI / ML models such as linear regression, logistic regression, Gaussian Mixture Model (GMM), Support Vector Machine (SVM), Latent Dirichlet Allocation (LDA), decision tree, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Deep Neural Network (DNN), Recurrent Neural Network (RNN), Restricted Boltzmann Machine (RBM), Deep Belief Network, Bidirectional Recurrent Deep Neural Network (BRDNN), Transformer, Deep Q-Networks (DQN), or proximal policy optimization (PPO), but is not limited thereto.

[0033] In the present disclosure, functions related to AI are operated through a processor and memory. The processor may be composed of one or more processors. In this case, the one or more processors may be a general-purpose processor, a graphics-only processor, or an AI-only processor. The one or more processors may process input data according to predefined operation rules or AI models stored in memory. In one embodiment, where the one or more processors are AI-only processors, the AI-only processors may be designed with a hardware structure specialized for processing a specific AI model.

[0034] The predefined operation rules or AI model are characterized by being created through learning. Here, being created through learning means that a predefined operation rules or AI model configured to perform a desired characteristic (or objective) is created by a basic AI model being trained using a number of training data by a learning algorithm. Such learning may be performed on the device itself where the AI ​​according to the present disclosure is executed, or it may be performed through a separate server and / or system. In one embodiment, the learning or training of the AI ​​model may be performed based on machine learning.

[0035] Hereinafter, embodiments of the present disclosure are described in detail with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present disclosure may be embodied in various different forms and is not limited to the embodiments described herein.

[0036] FIG. 1 illustrates an example of a wireless communication system (10) according to one embodiment of the present disclosure.

[0037] Referring to FIG. 1, an electronic device (100) may collect data from one or more network entities included in the core network (12) and / or one or more network entities connected to the core network (12). A wireless communication system (10) may include various nodes utilizing a wireless channel. For example, a wireless communication system (10) may include an electronic device (100), a core network (12), base stations (14, 16), and User Equipment (UE) (18). Each of the base stations (14, 16) may communicate with one or more user terminals within a corresponding cell. A user terminal may communicate with the core network (12) of the wireless communication system (10) via the base station of the cell to which the user terminal belongs. For example, a UE (18) may communicate with the core network (12) via the base station (14). In the following, each component constituting the wireless communication system (10) may be referred to as a network entity.

[0038] In one embodiment, the wireless communication system (10) is 5G (5 thIt may be a Generation) communication system. In such a wireless communication system (10), the core network (12) may be referred to as a 5G core network (5GC). Base stations (14, 16) may be referred to as an access point (AP), eNodeB (eNB), 5th generation node, next generation nodeB (gNB), wireless point, transmission / reception point (TRP), or other terms having an equivalent technical meaning. Base stations (14, 16) may be network infrastructure or network entities that provide wireless access to terminals (104, 106). Each of the base stations (14, 16) may have a coverage (or cell) defined as a certain geographical area based on the distance at which a signal can be transmitted. For the processing of data transmission and reception of one or more user devices connected to each base station, control signal processing and data signal processing can be performed at each base station.

[0039] A UE (18) is used by a user and can communicate with a base station (14) via a wireless channel. The link from the base station (14) to the UE (18) may be referred to as a downlink (DL). The link from the UE (18) to the base station (14) may be referred to as an uplink (UL). The UE (18) can communicate with other UEs via a mutual wireless channel. In one embodiment, the device-to-device (D2D) link between UEs may be referred to as a sidelink or an LTE (Long Term Evolution) V2X (Vehicle-to-everything) PC5 interface.

[0040] In one embodiment, the UE (18) may be operated without user involvement. For example, the UE (18) may be a device that performs machine type communication (MTC) and may not be carried by a user. The UE (18) may be referred to as customer premises equipment (CPE), mobile station, subscriber station, remote terminal, wireless terminal, electronic device, user device, or other terms having an equivalent technical meaning.

[0041] The electronic device (100) can optimize one or more network entities within a wireless communication system (10) based on an AI / ML model. The electronic device (100) can collect data related to data transmission and reception in the wireless communication system (10) from one or more network entities within the wireless communication system (10) (Block 102). From the collected data, the electronic device (100) can extract network features and generate network embeddings representing the extracted network features (Block 104). Based on the network embeddings, the electronic device (100) can analyze the collected data using an AI / ML model (Block 106). Based on the AI-based analysis, the electronic device (100) can perform network optimization (Block 108).

[0042] In one embodiment, with the introduction of 5G, operational complexity and operational expenditure (OPEX) may increase due to network congestion. To operate the network efficiently, AI / ML models may be utilized to predict the state of the network (or wireless communication system) or to optimize one or more parameters used in the network.

[0043] The types and ranges of network data input to AI / ML models can be very wide or diverse, and collecting such data can also be time-consuming. Furthermore, to train AI / ML models to propose optimal parameters, data may be acquired by directly applying various parameters from commercial networks. For example, to optimize base stations, various base station parameters can be applied to collect data, which is then used to train the AI / ML model. To achieve high performance metrics, it may be necessary to collect training data through aggressive parameter recommendations for base stations. However, from a network operation perspective, a failure in a single base station can have a cascading effect on other base stations; therefore, this method of data acquisition may not be preferred as it can compromise user QoS. Consequently, to ensure stable network operation, the values ​​of parameters used for training data collection may be restricted, which could consequently limit the performance of AI / ML-based network optimization.

[0044] As the number of parameter combinations input into an AI / ML model increases, the amount of data that needs to be collected also increases; therefore, it may be necessary to establish a sound collection strategy. For example, considering the difficulty of directly inputting every combination and every value to collect data, a strategy of collecting data by quantizing one or more parameter combinations and / or values ​​can be used. For instance, data can be collected by changing the value of a single parameter in increments of 10 or 20 instead of 1. Additionally, data augmentation can be used to collect more data based on a smaller amount of data; however, since the variation of parameter values ​​is limited, the improvement of the AI / ML model may also be restricted.

[0045] Meanwhile, collecting large amounts of data using a simulator for communication systems (e.g., the NS-3 simulator) may be one method. However, due to the nature of simulators, it may be difficult to perfectly replicate commercial networks. Additionally, due to limited computational power, using a simulator to acquire data corresponding to a single transmission time interval (TTI) (e.g., 1 ms (millisecond)) may take 1 second or more. For example, acquiring one day's worth of data may take 24,000 hours (approximately 1,000 days). Considering that network operations require data corresponding to a period equal to or greater than one year, there may be limitations to collecting data using simulators.

[0046] In one embodiment, instead of directly inputting collected network data into an AI / ML model, the electronic device (100) may first extract hidden features of the network from the network data and input the extracted features into the input layer of the AI / ML model. For example, the electronic device (100) may generate network embeddings representing one or more features of the network from the network data and input the generated network embeddings into the AI / ML model. Accordingly, the convergence speed of training the AI / ML model can be improved even when using a small amount of training data, and the performance of the AI / ML model can also be improved.

[0047] In one embodiment, a generative AI model may be utilized to generate network embeddings. For example, an electronic device (100) may generate network embeddings from collected network data using at least one of various generative AI models such as an autoencoder (AE), a variational autoencoder (VAE), a diffusion model, and / or a generative adversarial network (GAN).

[0048] In one embodiment, to analyze network data and determine one or more parameters for network optimization, the electronic device (100) may use at least one of various supervised learning models, unsupervised learning models, reinforcement learning models, and / or neural network models.

[0049] FIG. 2 illustrates an example of network optimization using an inference model according to one embodiment of the present disclosure.

[0050] Referring to FIG. 2, in order to optimize a wireless communication system, data collected from one or more network entities included in a wireless communication system or network may be analyzed using an inference model (210). In one embodiment, an electronic device (100) may include an encoder model (204), a transformation model (208), and an inference model (210). The electronic device (100) may analyze network data using the encoder model (204), the transformation model (208), and the inference model (210), and optimize the network based on the analysis results. For example, in FIG. 2, UE data (202) associated with a wireless communication system may be obtained from one or more User Equipment (UE). Information associated with a wireless communication system and / or a wireless network may be obtained from each UE at a specific point in time or during a specific time interval, periodically, non-periodically, or in response to or based on one or more predefined triggers. For example, each of one or more UEs may provide UE data (202) containing information associated with a wireless communication system and / or a wireless network for each transmission time interval (TTI), for example, to the electronic device (100) of FIG. 1.

[0051] In one embodiment, the UE data (202) is data associated with performance obtained from each UE at a specific point in time or during a specific time interval (e.g., Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), Signal-to-Interference-plus-Noise Ratio (SINR), Channel Quality Indicator (CQI), and / or Timing Advance (TA)), data associated with network mobility (e.g., data associated with Primary Cell (PCell) and / or Secondary Cell (SCell), data associated with Carrier Aggregation (CA), data associated with Handover, data associated with Radio Resource Control (RRC) connection, and / or data associated with surrounding cells), and data associated with data processing and traffic (e.g., downlink and / or uplink) It may include information related to throughput, packet loss and / or latency, and / or information related to a wireless bearer, and / or information related to UE status (e.g., power consumption, data related to Discontinuous Reception (DRX), power headroom reports, transmit power information, and / or information for identifying the network connected to the UE).

[0052] In FIG. 2, UE data (202) from one or more UEs during TTI t (e.g., UE data collected from UE 1 during TTI t) 1, t , ..., UE data collected from UE i during TTI UEi, t ; i is a natural number) can be collected. UE data (202) corresponding to TTI t can be first input into an encoder model (204) before being analyzed using an inference model (210). Based on the UE data (202) corresponding to TTI t, the encoder model (204) can generate network embeddings (206) corresponding to TTI t.

[0053] The encoder model (204) can extract one or more features associated with the network from input data and generate network embeddings that represent the extracted features. In one embodiment, the 'embedding' may be an operation of extracting features from data and representing them as vectors, or the result of such an operation. Through embeddings, high-dimensional unstructured data, such as text, images, graphs, and user data, can be converted into low-dimensional structured data, for example, numeric vectors. For example, in natural language processing, embeddings may be an operation of converting natural language used by humans into vectors in a numerical form that a machine can understand, or the result of such conversion. In one embodiment, based on the use of a neural network for feature extraction, the extracted features may be represented as neural network embedding vectors. For example, the neural network model for embeddings may be trained so that similar values ​​within the input data are converted into vector values ​​of similar values.

[0054] For example, the encoder model (204) can generate a network embedding that represents the features of the corresponding UE during TTI t from each UE data. The encoder model (204) can extract the features of the corresponding UE during TTI t from each UE data and represent the extracted features as a latent vector or latent representation in a latent space. The latent vector or latent representation corresponding to TTI t for each UE may be referred to as the network embedding corresponding to TTI t of the corresponding UE (or associated with it). UE data collected from UE 1 during TTI t UE 1, t Based on this, the encoder model (204) has a network embedding Z that represents the network features of UE 1 during TTI t. 1, t It can generate network embeddings Z based on each UE data during TTI t. In a similar manner, the encoder model (204) generates network embeddings Z based on each UE data during TTI t. 2,t , Z 3, t , ..., Z i, t It can generate.

[0055] In one embodiment, the encoder model (204) may be a combination of components that generate latent vectors or latent representations of a generative AI model, or may correspond thereto. For example, the encoder model (204) may be an encoder model of an encoder-decoder based generative model such as an autoencoder (AE), a denoising autoencoder (DAE), or a variational autoencoder (VAE), an encoder model of a Generative Adversarial Networks (GAN) based model, a generator model, or an encoder model of a diffusion based model, or may correspond thereto.

[0056] Network embeddings (206) corresponding to TTI t, generated by the encoder model (204), can be input to the transformation model (208). The transformation model (208) can transform the network embeddings (206) corresponding to TTI t into a predetermined number of parameter(s). The transformation model (208) can transform the input network embeddings into one or more predetermined number of parameters, regardless of the number of input network embeddings (e.g., the size of the input dimension). The one or more transformed parameters can be input to the inference model (210).

[0057] The number of network entities from which data is acquired may vary over time. For example, in an actual wireless communication system, the number of UEs connected to a single base station may vary over time due to various causes such as handover, RRC connection establishment or release, changes in the base station's cell coverage, load balancing, and / or cell reselection. Accordingly, for example, data may be acquired from i UEs at TTI t, but data may be acquired from k UEs during TTI t+1 (k is a natural number different from i). Therefore, the number of network embeddings corresponding to each TTI generated from the encoder model (204) may also vary. The transformation model (208) can transform the input embeddings into a predefined number of parameter(s) regardless of the number of input embeddings. Accordingly, even if the number of network entities (from which data is collected) in the actual wireless communication system changes, the size of the input dimension of the data input to the inference model (210) via the encoder model (204) and the transformation model (208) can be fixed.

[0058] In one embodiment, the transformation model (208) may be implemented based on any algorithm or model that transforms an input of variable length into an output of a predefined dimension (or length). For example, the transformation model (208) may be implemented based on at least one of various models, such as a self-attention algorithm or a transformer model, an RNN model such as an LSTM or a GRU (gated recurrent unit), a pooling function such as mean pooling or max pooling, or an encoder model for compressing a variable-length input into a fixed-size vector; however, the embodiments of the present disclosure are not limited to the exemplary embodiments described above.

[0059] One or more parameters transformed by the transformation model (208) can be input into the inference model (210). The inference model (210) can perform inference based on the one or more transformed parameters. An output regarding the transformed parameters can be obtained from the inference model (210). Based on the inference result by the inference model (210), network optimization can be performed (Block 212).

[0060] In one embodiment of the present disclosure, an inference model (210) may be trained to infer one or more parameters or policies to be used in a wireless communication system based on network data. For example, the inference model (210) may be trained to infer one or more parameters and / or one or more policies regarding network entity(s) belonging to a wireless communication system based on network data collected from network entities, which is transformed through an encoder model (204) and a transformation model (208). Network optimization may be performed using one or more parameters and / or one or more policies inferred by the inference model (210). For example, one or more parameters and / or policies used in a wireless communication system may be modified based on the inference results of the inference model (210).

[0061] In one embodiment, the inference model (210) may be trained to predict one or more Key Performance Indicators (KPIs) of a wireless communication system based on network data. For example, the inference model (210) may be trained to predict one or more KPIs regarding network entity(s) belonging to a wireless communication system based on network data collected from network entities, which is transformed through an encoder model (204) and a transformation model (208). Network optimization may be performed using one or more KPIs predicted by the inference model (210). For example, one or more parameters and / or policies used in the wireless communication system may be modified based on the inference results of the inference model (210).

[0062] In one embodiment, the inference model (210) may be implemented based on at least one of various AI / ML models such as linear regression, logistic regression, random forest, RNN, long short-term memory (LSTM), transformer, deep Q-network (DQN), and / or proximal policy optimization (PPO), but is not limited thereto.

[0063] FIGS. 3a and 3b illustrate examples of network optimization according to one embodiment of the present disclosure.

[0064] Referring to FIG. 3a, the inference model (210) may be an AI / ML model trained to recommend one or more optimal base station parameters from given network data, e.g., a base station parameter inference model (302), or may correspond thereto. A wireless communication system (30) may include a base station (32) and one or more UEs (34). One or more UEs (34) may be connected to the base station (32). The base station (32) included in the wireless communication system (30) may be optimized using an encoder model (204), a transformation model (208), and a base station parameter inference model (302).

[0065] The electronic device (100) can obtain network data (data in tabular format) collected from a data collection entity of the wireless communication system (30) (e.g., OAM (Operations Administration Maintenance)) or from a simulator simulating the wireless communication system (30). The electronic device (100) can generate network embeddings from the given network data through an encoder model (204). The embedded network data can be used for AI tasks.

[0066] In one embodiment, the embedded network data may be used for configuration measurement (CM) / policy recommendation. For example, the inference model (210) may be used to change base station parameters (e.g., one or more parameters for base station control included in the CM) to optimize the operation or performance of the base station, or to change the base station policy. Accordingly, the embedded graph (e.g., network embedding) may be input to the inference model (210), and the inference model (210) may be trained to output the best CM or the best policy. The inference model (210) may be applied to various tasks such as energy saving, load balancing, C-DRX (Connected Mode Discontinuous Reception) optimization, or resource scheduling optimization. In one embodiment, the inference model (210) trained for CM or policy recommendation may be trained based on reinforcement learning.

[0067] For example, a wireless communication system (30) may provide network data obtained from at least one of a base station (32) or one or more UEs (34) to an encoder model (204) periodically, non-periodically, in response to a trigger, or based thereon. The encoder model (204) may generate network embeddings based on the network data. A transformation model (208) may transform the network embeddings into a predetermined number of parameters (or a vector or matrix composed of a predetermined number of parameters). The parameters transformed by the transformation model (208) may be input to a base station parameter inference model (302). The base station parameter inference model (302) may infer one or more base station parameters based on the given transformation parameters. One or more base station parameters output from the base station parameter inference model (302) may be used to determine one or more parameters associated with the base station (32) of the wireless communication system (30). For example, the wireless communication system (30) can change one or more parameters used in the base station (32) to one or more base station parameters inferred by the base station parameter inference model (302).

[0068] In one embodiment, various base station parameters used in the base station (32) can be adjusted based on one or more base station parameters inferred by the base station parameter inference model (302). For example, base station power status (e.g., turn-on or turn-off), radio frequency-related parameters (e.g., frequency band, bandwidth, transmit power, antenna gain, and / or cell radius), coverage and handover-related parameters (e.g., RSRP, RSRQ, SINR, and / or handover trigger value), capacity and Quality of Service (QoS)-related parameters (e.g., maximum users per cell, number of Physical Resource Blocks (PRB), QoS classification identifier, and / or delay requirements), time synchronization and delay-related parameters (e.g., timing advance, propagation delay, HARQ (Hybrid Automatic Repeat request) timer, and / or slot duration), antenna and beamforming-related parameters (e.g., beam switching rate, and / or beamwidth), and / or scheduling algorithm parameters (e.g., uplink or downlink target block error rate, At least one of various base station parameters, such as BLER or SINR offset, can be adjusted to a value inferred by the base station parameter inference model (302).

[0069] Referring to FIG. 3b, the inference model (210) may be implemented as an AI / ML model trained to predict the performance of a network and / or wireless communication system (30) from given network data, e.g., a performance prediction model (304). The wireless communication system (30) may include a base station (32) and one or more UEs (34). One or more UEs (34) may be connected to the base station (32). Network entities included in the wireless communication system (30) (e.g., base station (32) and / or one or more UEs (34)) may be optimized using an encoder model (204), a transformation model (208), and a performance prediction model (304).

[0070] In one embodiment, the embedded network data may be used for KPI, channel, or traffic estimation. For example, to prevent network condition degradation in advance, the inference model (210) may be trained to predict the state of the network based on the embedded network data. Accordingly, the embedded graph (e.g., network embedding) may be input to the inference model (210), and the inference model (210) may infer (or predict) KPI values ​​(e.g., IP throughput, UE throughput, potential throughput, call drop rate), channel estimation values ​​(e.g., CQI, SINR, BLER), and / or traffic estimation values ​​(PRB usage and / or downlink / uplink PDCP (Packet Data Convergence Protocol) data volume) based on the embedded graph. The inference model (210) may be utilized for various tasks such as root cause analysis or cell planning.

[0071] For example, a wireless communication system (30) may provide network data obtained from at least one of a base station (32) or one or more UEs (34) to an encoder model (204) periodically, non-periodically, in response to a trigger, or based thereon. The encoder model (204) may generate network embeddings based on the network data. A transformation model (208) may transform the network embeddings into a predefined number of parameters (or a vector or matrix composed of a predefined number of parameters). The parameters transformed by the transformation model (208) may be input into a performance prediction model (304). The performance prediction model (304) may predict one or more KPIs associated with the wireless communication system (30) based on the given transformation parameters. One or more predicted KPIs output from the base station parameter inference model (302) may be used to adjust at least one parameter of the network entities of the wireless communication system (30). For example, the wireless communication system (30) can adjust one or more parameters used in the base station (32) and / or one or more UEs (34) based on predicted KPIs.

[0072] In one embodiment, various parameters used in base stations (32) and / or UEs (34) may be adjusted based on one or more predicted KPIs inferred by a performance prediction model (304). A performance prediction model (304) has at least one of various KPIs, such as, for example, coverage and signal quality indicators (e.g., RSRP, RSRQ, SINR, CQI, Modulation and Coding Scheme (MCS), and / or coverage hole rate), capacity and traffic indicators (e.g., cell throughput, user throughput, PRB utilization, traffic load, and / or peak time traffic), handover indicators (e.g., cell setup success rate, drop call rate, handover success rate, RRC setup success rate), latency and QoS indicators (e.g., end-to-end latency, packet loss rate, jitter, BLER, and / or QoS satisfaction), and / or energy and operational efficiency indicators (e.g., power consumption, DRX cycle length, RRC disable time, spectral efficiency, and / or network availability), given network It can be trained to predict from data. In one embodiment, the inference model (210) trained for KPI prediction can be trained based on a regression model such as linear regression or logistic regression.

[0073] FIG. 4 illustrates an example of an encoder model (400) according to one embodiment of the present disclosure.

[0074] Referring to FIG. 4, the encoder model (204) of the electronic device (100) may be a VAE-based encoder model (400) or may correspond thereto. For example, the encoder model (400) may be trained to infer a probability distribution representing each UE state variation from given UE data. Based on the probability distribution, a latent vector encapsulating network state structures may be generated.

[0075] The encoder model (400) can receive UE data (202) collected from one or more UEs during TTI t. The encoder model (400) can infer a probability distribution (402) for each UE. For example, the encoder model (400) can, as in Equation 1, the state or data of UE i corresponding to TTI t. i, t latent representation (or embedding) z i, t It can be mapped to.

[0076]

[0077] Referring to mathematical formula 1, It can correspond to an encoder model (400). is network status UE i, t It can correspond to the mean of the latent distribution. is a latent expression It can respond to the dispersion of. It can adjust the degree of uncertainty. In one embodiment, The value of can be preset during training of the encoder model (400) or dynamically adjusted during inference. High The value introduces high noise into the latent representation, which can capture high uncertainty within the UE's state. Accordingly, a flexible latent representation capable of adapting to diverse and complex network conditions can be allowed. Low The value lowers the noise level, which results in a stable and accurate latent representation, and such latent representation can be beneficial for relatively stable network conditions. In one embodiment, can be associated with parameters output from the inference model (210). For example, It can be dynamically adjusted to correspond to the target value to be output using an inference model (210).

[0078] Network embeddings (404) can be generated for each UE through reparameterization based on probability distributions (402) output from the encoder model (400). During reparameterization, network embeddings can be sampled based on the probability distribution and Gaussian noise for each UE. For example, network embeddings for UE i corresponding to TTI t It can be derived based on mathematical formula 2.

[0079]

[0080] Referring to mathematical formula 2, and It can correspond to the mean and variance output from the encoder model (400). It can be standard Gaussian noise. The higher the value, the more, More noise is introduced, which captures greater uncertainty within the state of the UE, causing the encoder model (400) to It is possible to adaptively adjust latent expressions based on values.

[0081] In one embodiment, multiple encoder models may be used to process data collected from multiple network entities simultaneously or in parallel. For example, an electronic device (100) may include multiple encoder models (400). Network data obtained from different UEs may be input to each of the multiple encoder models. For example, data from a first UE may be input to a first encoder model among the multiple encoder models, and simultaneously, data from a second UE may be input to a second encoder model among the multiple encoder models. The first encoder model and the second encoder model may process their respective input data simultaneously. For example, while a network embedding for the first UE is being generated, a network embedding for the second UE may be generated. Accordingly, the processing speed of data received from multiple network entities (e.g., embedding speed) may be improved.

[0082] By embedding the correlation between network states and control parameters, the encoder model (400) can help the inference model (210) interpret network conditions more easily. The structured representation generated by the encoder model (400) can reduce the exploration space, which can allow the inference model (210) to infer optimized policies and / or parameters with fewer attempts and errors. Additionally, the latent vectors of the encoder model (400) can capture common patterns between different states, which can facilitate stable learning and rapid generalization under uncertain conditions.

[0083] FIG. 5 illustrates an example of a conversion model (500) according to one embodiment of the present disclosure.

[0084] Referring to FIG. 5, the transformation model (208) of the electronic device (100) may be a self-attention-based transformation model (500) or may correspond thereto. For example, the transformation model (500) may transform network embeddings (206) generated by the encoder model (204) into a predefined number of parameters based on a self-attention mechanism. In response to or based on the number of activated UEs varying for each TTI, the self-attention-based transformation model (500) may dynamically handle variable-length embeddings.

[0085] The network embeddings corresponding to TTI t generated by the encoder model (204) are a single vector It can be combined as. is a latent representation (or network embedding) generated by the encoder model (204) for each UE i at time t (or TTI t) It may be a set of. In one embodiment, can be a row vector or a column vector. Combined embedding It can be input into the transformation model (500). In the transformation model (500), About (502), (504), Self-attention processing using (506) can be performed. (502) can be referred to as the query weight matrix used in query transformation. (504) can be referred to as the key weight matrix used for key transformation. (506) can be referred to as a value weight matrix used for value transformation.

[0086] Combined embeddings silver (502), (504), (506) can be multiplied by each. (502) and The product of is the query vector It can be referred to as (508). (504) and The product of is the key vector It can be referred to as. (506) and The product of is the value vector It can be referred to as (512). Query vector (508) and key vector (510) can be taken as a dot product. For example, key vector is transposed as a query vector Can be multiplied by (508). Query vector (508) and key vector The softmax function can be applied to the inner product of (510). Finally, the value vector for the inner product with the softmax applied It can be multiplied with (512). Accordingly, the final output of the transformation model (500) can be calculated. The processing of the transformation model (500) can be understood as Equation 3.

[0087]

[0088] Referring to mathematical formula 3, It can correspond to a softmax function that generates a probability distribution for attention weights. In one embodiment, the self-attention-based transformation model (500) can output a predefined number of parameters even if the number of input network embeddings varies. For example, Even if the number of UEs N, which is the length of , is variable, the size of the output matrix by the self-attention operation can be fixed. The self-attention mechanism can capture complex inter-UE relationships, which can improve the capabilities of the inference model (210) even under dynamically changing network conditions. Additionally, the transformation model (500) can leverage inter-UE dependencies through weight matrices to handle variable input dimensions and optimize network performance.

[0089] FIG. 6 illustrates an example of training an encoder model (600) according to one embodiment of the present disclosure.

[0090] Referring to FIG. 6, the encoder model (400) can be obtained by training the encoder model (600) and the decoder model (602). The encoder model (600) and the decoder model (602) may correspond to the encoder model and the decoder model of the VAE, respectively. The encoder model (600) can be trained using training network data (604) obtained from a single UE (e.g., UE i). The training network data (604) may include network data obtained from UE i for each TTI. For example, the training network data (604) may include one or more indicators associated with the traffic or channel state of UE i, such as CQI, MCS, SINR, or BLER.

[0091] Network data acquired from UE i during TTI UE i, tThe data can be input to an encoder model (600) for training. The encoder model (600) can infer a probability distribution (606) for UE i from the training network data (604) in a manner similar to the method described above with reference to Equation 1. The probability distribution (606) may include a mean μ and a variance σ. To generate an input to a decoder model (602), noise (610) may be sampled from Gaussian noise (608). Based on the probability distribution (606) and the noise (610), a latent representation z (612) may be computed in a manner similar to the method described above with reference to Equation 2.

[0092] The decoder model (602) can recover information of UE i from a latent representation (612) based on a probability distribution (606) and noise (610). The decoder model (602) can be trained to recover by removing noise. Recovering UE information by removing noise can correspond to removing base station parameter information from given UE data. In one embodiment, the noise (610) can correspond to base station parameters. The higher the base station parameter value, the more noise can be reflected in the latent representation (612), and thus the recovery of the decoder model (602) can be difficult. The decoder model (602) can generate recovery data from the latent representation (612) based on Equation 4.

[0093]

[0094] Referring to mathematical formula 4, is the output of the decoder model (602) It can correspond to. The decoder model (602) is a probability distribution that restores the original data from the latent vector z. It can model. This probabilistic decoding process is that the reconstructed UE state It can be guaranteed to reflect network instability caused by integrating.

[0095] Based on the output of the decoder model (602), a loss function (616) can be calculated. In one embodiment, to achieve optimal encoding, a VAE model comprising an encoder model (600) and a decoder model (602) can be trained to maximize an evidence lower bound (ELBO). Accordingly, the VAE model can accurately reconstruct the input data and ensure a structured latent space. ELBO objective It can be expressed as in mathematical formula 5.

[0096]

[0097] In mathematical formula 5, It can correspond to Kullback-Leibler Divergence (KL divergence). KL divergence is a latent representation The distribution that precedes it By matching to It can be regularized. It may be referred to as a training error or loss function. Based on the loss function (616), the weight matrices of the encoder model (600) and the decoder model (602) may be updated.

[0098] The training algorithms of the encoder model (600) and decoder model (602) of FIG. 6 can be expressed as shown in Table 1. In Table 1, It can correspond to an encoder model (600), and It can correspond to the decoder model (602). It can correspond to the loss function of VAE.

[0099]

[0100] In one embodiment, one or more hyperparameters may be predefined for training the encoder model (600) and the decoder model (602). For example, various hyperparameters such as latent size, encoder hidden layers, decoder hidden layers, type of activation function, training rate, batch size, or number of training epochs may be predefined.

[0101] In one embodiment, the encoder model (600) may be trained for at least one of other types of network entities as well as a UE. For example, the encoder model (600) may be trained based on data collected from at least one of network entities included in a base station (e.g., base station (14) of FIG. 1) (e.g., RU, distributed unit (DU), and / or centralized unit (CU)) or network entities included in a core network (e.g., core network (12) of FIG. 1) (e.g., Access and Mobility Management Function (AMF), Session Management Function (SMF), User Plane Function (UPF), Policy Control Function (PCF), Authentication Server Function (AUSF), Unified Data Management (UDM), Network Slice Selection Function (NSSF), and / or Network Exposure Function (NEF)). The aforementioned network entities are merely examples, and the present disclosure is not limited to the examples described above.

[0102] In one embodiment, the electronic device (100) may include a plurality of encoder models trained for different network entities. For example, the electronic device (100) may include not only an encoder model (400) trained based on UE data, but also an encoder model trained based on data for other network entities, e.g., a base station or an encoder model trained based on data for a core network. The electronic device (100) may identify the source of the acquired network data and select an encoder model corresponding to the source among the plurality of encoder models. For example, the electronic device (100) may identify that the source of the network data corresponding to TTI t is a UE based on the acquired network data. The electronic device (100) may select an encoder model trained based on UE data among the plurality of encoder models. The electronic device (100) may generate a network embedding from the network data corresponding to TTI t using the selected encoder model. Accordingly, the performance of the network embedding may be improved.

[0103] FIG. 7 illustrates an example of training a transformation model (700) and an inference model (702) according to one embodiment of the present disclosure.

[0104] Referring to FIG. 7, the transformation model (700) and the inference model (702) can be trained using an already trained encoder model (204). The encoder model (204) can generate training network embeddings based on training network data. The training network embeddings can be input to the transformation model (700). The transformed parameters can be input to the inference model (702). The inference model (702) can perform inference based on the transformed parameters. A loss function (706) can be calculated based on the inference of the inference model (702). Based on the loss function (706), the weight matrices of the transformation model (700) and the inference model (702) can be updated. The training network embeddings can be used as part of the input to the inference model (702), and accordingly, the decision of the inference model (702) can reflect the uncertainty of the network state.

[0105] In one embodiment, the inference model (702) may be based on deep reinforcement learning. For example, the inference model (702) may be implemented using a deep Q network (DQN) algorithm. For a DQN-based inference model (702), the loss function (706) may be calculated as shown in Equation 6.

[0106]

[0107] Referring to mathematical formula 6, It can correspond to the loss function. It can represent a mini-batch sampled from replay memory. It can represent the current reward. It can express the discount factor. It can correspond to DQN's policy network. is the next state We can express actions that maximize the Q-value at.

[0108] In one embodiment, the inference model (702) may be implemented using a Proximal Policy Optimization (PPO) algorithm. For the PPO-based inference model (702), the loss function (706) may be calculated as shown in Equation 7.

[0109]

[0110] Referring to mathematical formula 7, It can correspond to the loss function. is a new policy old policy The ratio for can be expressed. For clarity, term This can be used. Clipping action silver range It can be restricted, thereby preventing updates from deviating excessively from the previous policy. Constraining the clipping behavior can help stabilize training by preventing large policy updates that could destabilize the model's training process. The reward Value function for and criticism The advantage at time t, calculated based on, can be expressed.

[0111] The training algorithm of the inference model (702) of FIG. 7 can be expressed as shown in Table 2.

[0112]

[0113] In one embodiment of the present disclosure, one or more hyperparameters for training an inference model (702) may be predefined or adaptively adjusted. For example, for training a DQN-based inference model (702), one or more hyperparameters such as an action dimension, hidden layers, training rate, discount rate, target update frequency, replay buffer size, batch size, epsilon (search rate), optimization algorithm, or mini-batch may be predefined. For training a PPO-based inference model (702), one or more hyperparameters such as an action dimension, hidden layers, training rate, discount rate, clipping rate, entropy coefficient, value coefficient, maximum gradient norm, optimization algorithm, or mini-batch size may be predefined.

[0114] In one embodiment of the present disclosure, prior to training the transformation model (208) and the inference model (210), an encoder model (204) may be trained first. Using the encoder model (204) trained for network embeddings, an inference model (210) may be trained for network embeddings. For example, the inference model (210) may be based on DQN or PPO. The DQN-based or PPO-based model may be trained based on the algorithms described in Table 2. For example, the inference model (210) may be based on a regression model such as linear regression or logistic regression. The regression-based model may be trained using a loss function such as mean square error (MSE).

[0115] In one embodiment, as the inference model (210) is trained, weights or parameters on one of the transformation models (208) may be trained together. For example, the transformation model (208) may be a self-attention-based model or a corresponding model, and as the inference model (210) is updated, the weight matrices of the self-attention-based model (e.g., query weight matrix, key weight matrix, and / or value weight matrix) may be updated together.

[0116] FIG. 8 illustrates an exemplary example of optimizing a base station policy according to one embodiment of the present disclosure.

[0117] Referring to FIG. 8, the base station (810) may include one or more network entities (802), an OAM (804), and a SON (Self-Organization Network) agent (806), but is not limited thereto. One or more network entities (802) may include one or more wireless communication network devices, such as a radio unit (RU), a scheduler, and / or a modem, but are not limited thereto. For example, the base station (810) may include other elements or may not include some of the elements shown in FIG. 8. At least some of the one or more network entities (802), OAM (804), and SON (Self-Organization Network) agent (806) included in the base station (810) may be elements that are logically, functionally, softwarely, or hardware-separable from the other elements.

[0118] The Element Management System (EMS) (820) may be connected to the base station (810) via wired or wireless connection. The EMS (820) may collect information such as the status, performance, and errors of the base station (810). The EMS (820) may perform functions such as managing the settings or configuration of the base station (810) or resolving problems on the network. The EMS (820) may include, but is not limited to, a Manage Plane (822), an electronic device (100), and a SON manager (824). For example, the EMS (820) may include additional elements or may not include some elements shown in FIG. 8. For example, the electronic device (100) may be an external module, external server, or external device of the EMS (820) and may be connected to or communicate with the EMS (820) via wired or wireless connection. At least some of the management plane (822), electronic device (100), or SON management module (824) included in the EMS (820) may be elements that are logically, functionally, software-wise, or hardware-wise distinct from other elements. In one embodiment, the electronic device (100) may be referred to as an AI server or an AI server device.

[0119] The OAM (804) of the base station (810) can collect, acquire, or store statistical data regarding the wireless network of the base station (810). For example, the OAM (804) can collect or acquire statistical information of the communication network from wireless communication network devices (802), such as a RU, a scheduler, or a modem (① data collection). The OAM (804) can transmit the collected or acquired statistical data to the management plane (822) of the EMS (820) (② data transmission).

[0120] The management plane (822) can manage the configuration of network equipment and systems, monitor performance, and / or maintain network entities. For example, the management plane (822) can configure the initial settings of one or more network entities directly or indirectly connected to the EMS (820) and adjust the settings. The management plane (822) can measure or estimate performance indicators of one or more network entities directly or indirectly connected to the EMS (820). The management plane (822) can detect and recover from network faults. The management plane (822) can manage network security policies and access controls. The management plane (822) can manage the firmware and / or software of one or more network entities directly or indirectly connected to the EMS (820).

[0121] The management plane (822) can determine a target application among a plurality of applications. For example, the plurality of applications may include an application for energy saving, an application for load balancing, an application for a scheduler, and / or an application for anomaly detection. For example, the management plane (822) can determine, among the plurality of applications, an application for a function or effect to be optimized through AI-based network analysis as the target application. The management plane (822) can transmit information about the determined target application to the AI ​​server (240) (③ Target Application Transmission).

[0122] The SON manager (824) can perform automatic configuration, automatic optimization, and / or automatic healing functions of the network. For example, the SON manager (824) can automatically register a new base station to the network, automatically set up a list of neighboring cells, and / or automatically adjust the initial configuration of each base station. The SON manager (824) can optimize the cell coverage of the base station (820), manage interference, and / or optimize handover. The SON manager (824) can automatically recover from a failure of the base station (820) and / or detect and prevent a failure of the base station (820) in advance.

[0123] The SON manager (824) can determine parameters and / or policies to be inferred through the electronic device (100) based on information regarding the target application received from the management plane (822). For example, the SON manager (824) can select a policy to perform AI-based analysis from among a plurality of policies supported by the electronic device (100) based on information regarding the target application. The selected policy to perform AI-based analysis may be a policy corresponding to the target application. For example, based on the fact that the target application is an application for energy saving, the SON manager (824) can determine (or select or switch) an energy saving policy as the policy to perform AI-based analysis. The SON manager (824) can provide the switched policy to the electronic device (100) (④ Transmission of switched policy).

[0124] The electronic device (100) can infer optimal policies and / or parameters based on policies determined by the SON manager (824). The electronic device (100) can acquire wireless network data and generate network embeddings using an encoder model (204). The electronic device (100) can infer optimal parameters or policies based on network embeddings using a transformation model (208) and an inference model corresponding to the switched policies. The electronic device (100) can provide the inferred parameters or policies to the management plane (822) (⑤ Transmit inferred policy).

[0125] In one embodiment, the electronic device (100) may include a plurality of inference models trained for different goals. For example, the electronic device (100) may include an inference model trained to recommend optimal parameters or policies related to energy saving, an inference model trained to recommend optimal parameters or policies related to load balancing, an inference model trained to output parameters or policies related to scheduling, or an inference model trained to recommend optimal parameters or policies related to anomaly detection. The electronic device (100) may select an inference model among the plurality of inference models that corresponds to a policy determined by the SON manager (824), and infer optimal parameters or policies using the selected inference model. For example, based on the fact that the policy determined by the SON manager (824) is energy saving, the electronic device (100) may select an inference model among the plurality of inference models that is trained to recommend optimal parameters or policies related to energy saving.

[0126] In one embodiment, AI analytics may be performed based on network embeddings. An electronic device (100) may predict the network state of a base station (820) based on network embeddings. Based on the predicted network state, the electronic device (100) may decide whether to input the network embeddings into an inference model (210). For example, the electronic device (100) may determine in real time whether optimization of the parameters and / or policies of the base station (820) is required based on the predicted network state. Based on the determination that optimization of the parameters and / or policies of the base station (820) is required, the electronic device (100) may input the network embeddings into the inference model (210) to infer new policies and / or parameters.

[0127] In one embodiment, the electronic device (100) may store at least one of network embeddings or the result of AI interpretation in a database within the electronic device (100). The stored data may be used for training or inference on an inference model.

[0128] The management plane (822) can transmit the policy inferred by the electronic device (100) to the OAM (804) of the base station (820) (⑥ Policy transmission). The OAM (804) can transmit the inferred policy to the SON agent (806) (⑦ Policy transmission). The SON agent (806) can perform optimization of the base station (820) based on requests or commands from the SON manager (824). For example, the SON agent (806) can apply or reflect the policy inferred by the electronic device (100) to an application (e.g., at least one of one or more network entities (802)) (⑧ Policy application). For example, the SON agent (806) can determine parameters associated with network control, e.g., one or more parameters to be used in the target application, based on the policy inferred by the electronic device (100).

[0129] According to one embodiment of the present disclosure, an electronic device (100) can generate network embeddings from wireless network data and input the network embeddings into an inference model for a wireless network. The electronic device (100) can generate network embeddings by extracting features of a wireless network based on a generative AI model. A generative AI model may be an algorithm capable of freely generating data by changing specific values ​​of input data. To this end, the generative AI model may be pre-trained to extract features of input data and to freely vary the output in response to changes in specific values. Meanwhile, there are cases where the causal relationship between indicators or parameters of a network has not been identified. Additionally, network data collected through a base station may be affected by base station parameters. The electronic device (100) can extract network information by removing base station parameter information from the collected network data using generative AI techniques. Accordingly, the accuracy of the inference model may be improved.

[0130] According to one embodiment of the present disclosure, an electronic device (100) can input network embeddings into a conversion model (208) that converts network embeddings into a predetermined number of parameters. Network data obtained from a wireless network may include information on multiple network entities, for example, multiple UEs. Meanwhile, the number of network entities belonging to the wireless network may change in real time. For example, the number of UEs connected to a base station (810) may change incessantly. Even if the number of network embeddings changes over time, the electronic device (100) can fix the size of the input vector input to the inference model (210) using the conversion model (208). Accordingly, the inference model (210) may be able to operate even if the number of network embeddings at a specific point in time changes.

[0131] FIG. 9 illustrates an example of a block diagram of an electronic device (900) according to one embodiment of the present disclosure.

[0132] The electronic device (900) illustrated in FIG. 9 may be a computing device or server device that recommends parameters or policies for a network entity and / or predicts network performance indicators. For example, the electronic device (900) may include a device that infers one or more parameters or policies to be used for controlling a network using an AI model. For example, the electronic device (900) may be a communication device constituting a wireless network and may be a server device included in a network entity such as a base station or an EMS. The electronic device (900) may be a separate server device outside the wireless network that infers one or more parameters or policies to be used in the wireless network or predicts network performance indicators.

[0133] In one embodiment, an electronic device for training or updating at least one of an AI model for extracting network features, an AI model for converting network embeddings into a predetermined number of parameters, or an AI model for inferring network entity parameters, network entity policies, or network performance indicators may be the same as or different from an electronic device (900) that performs prediction or inference using the AI ​​models. For example, if the electronic device for training or updating the AI ​​model and the electronic device (900) (i.e., the electronic device performing the prediction or inference operation) are different, the electronic device (900) may receive the trained or updated AI model from the electronic device for training or updating the AI ​​model.

[0134] In one embodiment, the AI ​​model may be dynamically updated as a prediction or inference operation is performed. For example, at least some of the weights of the encoder model (204), the transformation model (208), or the inference model (210) may be dynamically updated as a prediction or inference operation is performed. For example, if the electronic device updating the AI ​​model and the electronic device (900) are the same, the electronic device (900) may perform a prediction or inference using the AI ​​model and simultaneously update the AI ​​model. For example, if the electronic device updating the AI ​​model and the electronic device (900) are different, the electronic device updating the AI ​​model may receive data identified, generated, or produced by the electronic device (900) while performing a prediction or inference using the AI ​​model, and update the AI ​​model based on the received data.

[0135] In one embodiment, the electronic device (900) may include at least one processor (902), memory (904), and transceiver (906), but is not limited thereto.

[0136] A processor (902) is electrically connected to components included in an electronic device (900) and can perform operations or data processing regarding the control and / or communication of components included in the electronic device (900). In one embodiment, the processor (902) can load a request, command, or data received from at least one of the other components into memory for processing and store the processing result data in memory. In one embodiment, the processor (902) can process input data or control other components to process it according to data, operation rules, algorithms, methods, or models stored in memory (904). For example, the processor (902) can perform operations of predefined operation rules, algorithms, methods, modules, or AI models (e.g., neural network models) stored in memory (904) using input data. At least one processor can execute program instructions individually or in combination to achieve or perform various functions according to the present disclosure.

[0137] According to one or more embodiments of the present disclosure, the processor (902) may include at least one of a general-purpose processor such as a CPU (central processing unit), AP (application processor), DSP (Digital Signal Processor), a graphics-dedicated processor such as a GPU (graphic processing unit) or VPU (Vision Processing Unit), or an AI-dedicated processor such as an NPU (neural processing unit). For example, if the processor (902) is an AI-dedicated processor, the AI-dedicated processor may be designed with a hardware structure specialized for processing a specific AI model.

[0138] The processor (902) may include various processing circuits and / or multiple processors. For example, the term "processor" as used in the present disclosure, including in the claims, may include various processing circuits including at least one processor. One or more of the at least one processor may be configured to perform one or more functions of the present disclosure individually and / or collectively in a distributed manner. Where the "processor," "at least one processor," or "one or more processors" are described in the present disclosure as being configured to perform multiple functions, this may include a situation where one processor performs some of the functions and other processor(s) perform other parts of the functions, and a situation where a single processor performs all functions. Additionally, the at least one processor may include a combination of processors performing various functions in a distributed manner. The at least one processor may execute program instructions to achieve or perform various functions.

[0139] The memory (904) is electrically connected to the processor (902) and can store one or more modules, algorithms, operating rules, models (e.g., machine learning models, AI models), programs, instructions, or data related to the operation of components included in the electronic device (900). For example, the memory (904) may include any non-transient computer-readable recording medium. For example, the memory (904) can store one or more modules, algorithms, operating rules, models, programs, instructions, or data for processing and control of the processor (902). The memory (904) may include at least one type of storage medium among flash memory type, hard disk type, multimedia card micro type, card type memory (e.g., SD or XD memory, etc.), RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, magnetic disk, and optical disk, but is not limited thereto.

[0140] In one embodiment, the memory (904) may store data and / or information identified, acquired, generated, or determined by the electronic device (900). For example, the memory (904) may store network data or weights of each model. The memory (904) may store network embeddings generated by the encoder model (204) or the output of the inference model (210). The memory (904) may store data and / or information identified, acquired, generated, or determined by the electronic device (900) in a compressed form. In one embodiment, the memory (904) may store predefined or determined information.

[0141] In one embodiment, the electronic device (900) may include a module that performs (or is used to perform) at least one operation. Some of the modules that perform at least one operation of the electronic device (900) may be composed of a plurality of sub-modules or may constitute a single module. The modules that perform at least one operation of the electronic device (900) may be a hardware module, a software module, and / or a combination of a hardware module and a software module, or may correspond thereto.

[0142] The memory (904) may include software modules that perform at least some of the operations of the electronic device (900) described above. In one embodiment, the module included in the memory (904) may perform operations by being executed by the processor (902). For example, the module included in the memory (904) (i.e., the software module) may include a program, model, operation rule, or algorithm configured to perform operations that derive output data for input data, which are executed according to the control or command of the processor (902).

[0143] In one embodiment, the memory (904) may include a program, instructions, a neural network model, an AI model, an ML model, a statistical model, a rule of operation, or an algorithm for processing network data. For example, the memory (904) may store an encoder model (204), a transformation model (208), and an inference model (210), or weight matrices for each model. The encoder model (204), the transformation model (208), and the inference model (210) stored in the memory (904) may be executed by at least one processor (902).

[0144] The model contained in the memory (904) can be created through training. Here, being created through training means that a foundation AI model is trained using multiple training data by a training algorithm to create an AI model configured to perform a desired characteristic (or objective). This training may be performed on the device itself where the AI ​​according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of training algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.

[0145] The transceiver (906) can perform functions for transmitting and receiving signals in a wired communication environment. The transceiver (906) may include a wired interface for controlling a direct connection between devices through a transmission medium (e.g., copper wire or optical fiber). For example, the transceiver (906) can transmit an electrical signal to another device through a copper wire or perform conversion between an electrical signal and an optical signal. The transceiver (906) can communicate with other components within a wireless communication system or wireless network.

[0146] The transceiver (906) may include an antenna section. The transceiver (906) may include at least one antenna array composed of a plurality of antenna elements. In terms of hardware, the transceiver (906) may be composed of digital circuits and / or analog circuits (e.g., a radio frequency integrated circuit (RFIC)). Here, the digital circuits and / or analog circuits may be implemented in a single package. Additionally, the transceiver (906) may include a plurality of RF chains. The transceiver (906) may perform beamforming. The transceiver (906) may apply beamforming weights to the signal to give directionality according to the processor (902) settings to the signal to be transmitted or received. According to one embodiment, the transceiver (906) may include a radio frequency (RF) block (or RF section). The transceiver (906) can transmit a synchronization signal, a reference signal, system information, a message, a control message, a stream, control information, or data.

[0147] An electronic device according to one embodiment of the present disclosure may include at least one processor. The electronic device may include a memory comprising one or more storage media for storing one or more instructions. When the one or more instructions are executed by the at least one processor, the electronic device may obtain network entity data associated with each network entity from each of the one or more network entities. When the one or more instructions are executed by the at least one processor, the electronic device may perform network embeddings for each of the one or more network entities based on the network entity data associated with each network entity using an encoder model. When the one or more instructions are executed by the at least one processor, the electronic device may convert the network embeddings into a predetermined number of parameters using a transformation model. When the one or more instructions are executed by the at least one processor, the electronic device may input the predetermined number of parameters into an inference model. When the above one or more instructions are executed by the at least one processor, the electronic device may obtain an output regarding a predetermined number of parameters from the inference model. When the above one or more instructions are executed by the at least one processor, the electronic device may determine one or more parameters related to the control of the network based on the output of the inference model.

[0148] Additionally or alternatively, when executed by the at least one processor, the above one or more instructions may cause the electronic device to input network entity data obtained from a first network entity among the one or more network entities into the encoder model. When executed by the at least one processor, the above one or more instructions may cause the electronic device to obtain a probability distribution for the first network entity from the encoder model. When executed by the at least one processor, the above one or more instructions may cause the electronic device to generate a network embedding for the first network entity by performing sampling based on the probability distribution for the first network entity.

[0149] Additionally or alternatively, the one or more instructions may cause the electronic device to extract a sample from a probability distribution for the first network entity when executed by the at least one processor. The one or more instructions may cause the electronic device to generate a network embedding for the first network entity based on the sample and Gaussian noise when executed by the at least one processor.

[0150] Additionally or alternatively, the transformation model (208) may be a self-attention-based model. The one or more instructions may cause the electronic device to combine network embeddings when executed by the at least one processor. The one or more instructions may cause the electronic device to input the combined network embeddings into the transformation model (208) when executed by the at least one processor. The one or more instructions may cause the electronic device to obtain the predefined number of parameters from the transformation model when executed by the at least one processor.

[0151] Additionally or alternatively, the network entity data may be acquired periodically. The network embedding for each network entity may be generated at each period in which the network entity data is acquired.

[0152] Additionally or alternatively, the one or more instructions may cause the electronic device to combine network embeddings generated at each period in which the network entity data is acquired when executed by the at least one processor. The one or more instructions may cause the electronic device to input the combined network embeddings into the transformation model at each period in which the network entity data is acquired when executed by the at least one processor. The one or more instructions may cause the electronic device to acquire the predefined number of parameters at each period in which the network entity data is acquired from the transformation model when executed by the at least one processor.

[0153] Additionally or alternatively, the inference model may output one or more values ​​to be used to determine one or more parameters or policies for network entities included in the network based on the aforementioned predefined number of parameters.

[0154] Additionally or alternatively, the inference model may output one or more values ​​for predicting the performance indicators of the network or the state of the network based on the aforementioned predefined number of parameters.

[0155] Additionally or alternatively, the encoder model may be trained by inputting training network entity data for a single network entity into the encoder model. The encoder model may be trained by obtaining a probability distribution for the single network entity from the encoder model. The encoder model may be trained by obtaining samples for the single network entity based on the probability distribution for the single network entity and Gaussian noise. The encoder model may be trained by using a decoder model to reconstruct network entity data for the single network entity based on the samples. The encoder model may be trained by calculating a loss function based on the network entity data reconstructed by the decoder model. The encoder model may be trained by updating one or more weights of the encoder model and one or more weights of the decoder model based on the loss function.

[0156] Additionally or alternatively, the inference model and the transformation model may be trained by using the encoder model to generate training network embeddings for each network entity from training network entity data for at least one network entity. The inference model and the transformation model may be trained by using the transformation model to transform the training network embeddings into a predetermined number of parameters. The inference model and the transformation model may be trained by inputting the parameters transformed from the training network embeddings into the inference model. The inference model and the transformation model may be trained by calculating a loss function based on the output of the inference model based on the parameters transformed from the training network embeddings. The inference model and the transformation model may be trained by updating the transformation model and the inference model based on the loss function.

[0157] In the present disclosure, descriptions that overlap in FIGS. 1 to 9 may be omitted, and one or more embodiments described in at least one of FIGS. 1 to 9 may be applied or implemented in combination with each other. FIG. 10 illustrates an example of a flowchart of a method (1000) according to one embodiment of the present disclosure.

[0158] Referring to FIG. 10, the method (1000) may include operations (1002, 1004, 1006, 1008, 1010). In one embodiment, the method (1000) may be performed by an electronic device (100). However, the present disclosure is not limited to the exemplary embodiments described above. For example, the operations (1002, 1004, 1006, 1008, 1010) may be performed individually or in combination by any electronic device. The method according to one embodiment of the present disclosure is not limited to that shown in FIG. 10. In the embodiment of FIG. 10, any one of the steps shown in FIG. 10 may be omitted, and other steps may be further included. In one embodiment, the order of at least some of the operations (1002, 1004, 1006, 1008, 1010) may be changed.

[0159] In operation (1002), the electronic device (100) can obtain network entity data associated with each network entity from each of one or more network entities. For example, the electronic device (100) can obtain UE data associated with each UE from one or more UEs. In one embodiment, in operation (1002), the electronic device (100) can obtain network entity data associated with at least one network entity. For example, the electronic device (100) can obtain UE data associated with at least one UE.

[0160] In operation (1004), the electronic device (100) can generate network embeddings based on network entity data associated with each network entity by using an encoder model (204) for each of one or more network entities. For example, the electronic device (100) can input UE data associated with the first UE to the encoder model (204) for a first UE among one or more UEs. In response to or based on the UE data associated with the first UE, the encoder model (204) can output network embeddings that represent the characteristics of the UE data associated with the first UE. In one embodiment, in operation (1004), the electronic device (100) can generate network embeddings based on network entity data associated with at least one network entity by using an encoder model (204). For example, the electronic device (100) can input UE data associated with the first UE (among one or more UEs) to the encoder model (204). The encoder model (204) can output a network embedding that represents the features of the UE data associated with the first UE in response to, or based on, the UE data associated with the first UE.

[0161] In operation (1006), the electronic device (100) can use a transformation model (208) to transform network embeddings into a predetermined number of parameters. For example, the electronic device (100) can input network embeddings generated through operation (1004) into the transformation model (208). The transformation model (208) can transform network embeddings into a predetermined number of parameters (or a predetermined fixed-size output matrix).

[0162] In operation (1008), the electronic device (100) can input a predefined number of parameters converted in operation (1006) into the inference model (210). The electronic device (100) can obtain an output regarding the predefined number of parameters from the inference model (210). In operation (1010), based on the output of the inference model (210), the electronic device (100) can determine one or more parameters associated with the control of the network.

[0163] A method according to one embodiment of the present disclosure may include: obtaining network entity data associated with each network entity from each of one or more network entities. For each of the one or more network entities, the method may include generating network embeddings based on the network entity data associated with each network entity using an encoder model. The method may include converting the network embeddings into a predetermined number of parameters using a transformation model. The method may include inputting the predetermined number of parameters into an inference model. The method may include determining one or more parameters associated with the control of the network based on the output of the inference model.

[0164] Additionally or alternatively, the step of generating a network embedding using the encoder model may include the step of inputting network entity data obtained from a first network entity among the one or more network entities into the encoder model. The step of generating a network embedding using the encoder model may include the step of obtaining a probability distribution for the first network entity from the encoder model. The step of generating a network embedding using the encoder model may include the step of generating a network embedding for the first network entity by performing sampling based on the probability distribution for the first network entity.

[0165] Additionally or alternatively, the step of generating a network embedding for the first network entity may include the step of extracting a sample from a probability distribution for the first network entity. The step of generating a network embedding for the first network entity may include the step of generating a network embedding for the first network entity based on the sample and Gaussian noise.

[0166] Additionally or alternatively, the transformation model may be based on self-attention. The step of transforming the network embeddings into a predetermined number of parameters may include a step of combining the network embeddings. The step of transforming the network embeddings into a predetermined number of parameters may include a step of inputting the combined network embeddings into the transformation model. The step of transforming the network embeddings into a predetermined number of parameters may include a step of obtaining the predetermined number of parameters from the transformation model.

[0167] Additionally or alternatively, the step of acquiring network entity data may include the step of periodically acquiring network entity data. The step of generating network embeddings for one or more network entities may include the step of generating the network embeddings for each network entity at the interval in which the network entity data is acquired.

[0168] Additionally or alternatively, the step of converting the network embeddings into a predetermined number of parameters may include the step of combining the generated network embeddings at each period in which the network entity data is acquired. The step of converting the network embeddings into a predetermined number of parameters may include the step of inputting the combined network embeddings into the conversion model at each period in which the network entity data is acquired. The step of converting the network embeddings into a predetermined number of parameters may include the step of acquiring the predetermined number of parameters at each period in which the network entity data is acquired from the conversion model.

[0169] Additionally or alternatively, the inference model may output one or more values ​​to be used to determine one or more parameters or policies for network entities included in the network based on the aforementioned predefined number of parameters.

[0170] Additionally or alternatively, the inference model may output one or more values ​​for predicting the performance indicators of the network or the state of the network based on the aforementioned predefined number of parameters.

[0171] Additionally or alternatively, the encoder model may be trained by inputting training network entity data for a single network entity into the encoder model. The encoder model may be trained by obtaining a probability distribution for the single network entity from the encoder model. The encoder model may be trained by obtaining samples for the single network entity based on the probability distribution for the single network entity and Gaussian noise. The encoder model may be trained by using a decoder model to reconstruct network entity data for the single network entity based on the samples. The encoder model may be trained by calculating a loss function based on the network entity data reconstructed by the decoder model. The encoder model may be trained by updating one or more weights of the encoder model and one or more weights of the decoder model based on the loss function.

[0172] Additionally or alternatively, the inference model and the transformation model may be trained by using the encoder model to generate training network embeddings for each network entity from training network entity data for at least one network entity. The inference model and the transformation model may be trained by using the transformation model to transform the training network embeddings into a predetermined number of parameters. The inference model and the transformation model may be trained by inputting the parameters transformed from the training network embeddings into the inference model. The inference model and the transformation model may be trained by obtaining an output regarding the transformed parameters from the inference model. The inference model and the transformation model may be trained by calculating a loss function based on the output of the inference model based on the parameters transformed from the training network embeddings. The inference model and the transformation model may be trained by updating the transformation model and the inference model based on the loss function.

[0173] According to one embodiment of the present disclosure, a computer-readable recording medium may record a program for performing any combination of methods, steps, and operations according to one embodiment of the present disclosure on a computer.

[0174] A device-readable storage medium may be provided in the form of a non-transitory storage medium. Here, 'non-transitory storage medium' simply means that it is a tangible device and does not contain a signal (e.g., electromagnetic waves), and the term does not distinguish between cases where data is stored semi-permanently and cases where it is stored temporarily. For example, a 'non-transitory storage medium' may include a buffer in which data is stored temporarily.

[0175] According to one embodiment of the present disclosure, the method may be provided by being included in a computer program product. The computer program product may be traded between a seller and a buyer as a product. The computer program product may be distributed in the form of a device-readable storage medium (e.g., compact disc read-only memory (CD-ROM)), or distributed online (e.g., download or upload) through an application store or directly between two user devices (e.g., smartphones). In the case of online distribution, at least a portion of the computer program product (e.g., downloadable app) may be temporarily stored or temporarily created on a device-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or a relay server.

[0176] Although the embodiments have been described above with reference to limited examples and drawings, those skilled in the art can make various changes and modifications from the description above. For example, appropriate results can be achieved even if the described techniques are performed in a different order than described, and / or components such as the described computer system or module are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

Claims

1. A step of obtaining network entity data associated with each network entity from each of one or more network entities; For the above one or more network entities, a step of generating network embeddings based on the network entity data using an encoder model (204); A step of converting the network embeddings into a predefined number of parameters using a conversion model (208); A step of inputting the above-mentioned number of predefined parameters into an inference model (210); A step of obtaining an output regarding the number of parameters in the predefined number from the inference model (210); and A method comprising the step of determining one or more parameters associated with the control of a network based on the output of the above inference model (210).

2. In Paragraph 1, The step of generating a network embedding using the encoder model (204) above is: A step of inputting network entity data obtained from a first network entity among the one or more network entities into the encoder model (204); A step of obtaining a probability distribution for the first network entity from the encoder model (204); and A method comprising the step of generating a network embedding for the first network entity by performing sampling based on a probability distribution for the first network entity.

3. In Paragraph 2, The step of generating a network embedding for the first network entity is, A step of extracting a sample from a probability distribution for the first network entity; and A method comprising the step of generating a network embedding for the first network entity based on the above sample and Gaussian noise.

4. In any one of paragraphs 1 through 3, The above transformation model (208) is based on self-attention, and The step of converting the above network embeddings into a predefined number of parameters is: A step of combining the above network embeddings; The step of inputting the combined network embeddings into the transformation model (208); and A method comprising the step of obtaining a predefined number of parameters from the above transformation model (208).

5. In any one of paragraphs 1 through 4, The step of acquiring the above network entity data includes the step of periodically acquiring the above network entity data, and The step of generating network embeddings for one or more network entities is a method in which the network embedding for each network entity is generated at each period in which the network entity data is acquired.

6. In Paragraph 5, The step of converting the above network embeddings into a predefined number of parameters is: A step of combining network embeddings generated at each period in which the above network entity data is acquired; A step of inputting combined network embeddings into the transformation model (208) at each period in which the above network entity data is acquired; and A method comprising the step of obtaining a predetermined number of parameters at each period in which the network entity data is obtained from the above transformation model (208).

7. In any one of paragraphs 1 through 6, The above inference model (210) outputs one or more values ​​for determining one or more parameters or policies for a network entity of the network based on the above-defined number of parameters.

8. In any one of paragraphs 1 through 6, The above inference model (210) outputs one or more values ​​for predicting the performance indicators of the network or the state of the network based on the above-defined number of parameters.

9. In any one of paragraphs 1 through 8, The above encoder model (204) is: Training network entity data for a single network entity is input into the encoder model (204) above; Obtaining a probability distribution for the single network entity from the encoder model (204); Based on the probability distribution and Gaussian noise for the single network entity, a sample for the single network entity is obtained; Using a decoder model (602), network entity data for the single network entity is restored based on the sample; Calculate a loss function based on network entity data restored by the above decoder model (602); and A method trained by updating one or more weights of the encoder model (204) and one or more weights of the decoder model (602) based on the loss function (616).

10. In any one of paragraphs 1 through 9, The above inference model (210) and the above transformation model (208) are, Using the encoder model (204) above, a training network embedding for each network entity is generated from training network entity data for at least one network entity; Using the above transformation model (208), the training network embeddings are transformed into the above-predefined number of parameters; The parameters converted from the above training network embeddings are input into the inference model (210); From the above inference model (210), an output regarding the transformed parameters is obtained; Calculate a loss function (706) based on the output of the inference model (210) based on parameters converted from the above training network embeddings; and A method trained by updating the transformation model (208) and the inference model (210) based on the loss function (706).

11. A computer-readable recording medium having a program recorded thereon for performing the method of any one of paragraphs 1 through 10 on a computer.

12. In the electronic device (900), At least one processor (902); and It includes a memory (904) that stores one or more instructions, When the above one or more instructions are executed by the above at least one processor alone or in cooperation, the electronic device: From each of one or more network entities, obtain network entity data associated with each network entity; For the above one or more network entities, a network embedding is generated based on the network entity data using an encoder model (204). Using a transformation model (208), the network embeddings are transformed into a predefined number of parameters; Input the above-mentioned number of predefined parameters into the inference model (210); Obtaining an output regarding the number of parameters in the predefined number from the above inference model (210); and An electronic device that determines one or more parameters associated with the control of a network based on the output of the above inference model (210).

13. In Paragraph 12, When the above one or more instructions are executed by the at least one processor (902) alone or in cooperation, the electronic device (900) additionally: Network entity data obtained from the first network entity among the above one or more network entities is input into the encoder model (204), and Using the encoder model (204) above, obtain a probability distribution for the first network entity; and An electronic device that generates a network embedding for the first network entity by performing sampling based on a probability distribution for the first network entity.

14. In Paragraph 13, When the above one or more instructions are executed by the at least one processor (902) alone or in cooperation, the electronic device (900) additionally: Extracting a sample from the probability distribution for the first network entity; and An electronic device that generates a network embedding for the first network entity based on the above sample and Gaussian noise.

15. In any one of paragraphs 12 through 14, The above transformation model (208) is a model based on self-attention, and When the above one or more instructions are executed by the at least one processor (902) alone or in cooperation, the electronic device (900) additionally: Combining the above network embeddings; Input the combined network embeddings above into the transformation model (208); and An electronic device that obtains the predefined number of parameters from the above conversion model (208).

Images