Enhancing quality of analog network data using generative adversarial networks

Abstract

Embodiments of the present disclosure relate to enhancing quality of analog network data using generative adversarial networks. A device may receive real network data associated with a network, and may receive random potential vectors and random process samples. The device may generate synthetic network data using random potential vectors with a generative adversarial network (GAN) model, and train the GAN model with the real network data and the synthetic network data to generate a trained GAN model. The device may utilize the random process samples with the random process to generate simulated network data, and may apply weights to the real network data, the composite network data, and the simulated network data to generate weighted real network data, weighted composite network data, and weighted simulated network data. The device may combine the weighted real network data, the weighted synthetic network data, and the weighted analog network data to generate interpolated network data, and may perform an action based on the interpolated network data.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computers, and more particularly, to enhancing the quality of data. Background Art

[0002] Machine learning models have become an integral part of computer network and cloud management systems. For example, machine learning models can be used to monitor and / or control computer networks and / or cloud management systems. Summary of the invention

[0003] Some embodiments described herein relate to a method. The method may include receiving real network data associated with a network, and receiving a random latent vector and a random process sample. The method may include generating synthetic network data using the random latent vector and a generative adversarial network (GAN) model, and training the GAN model with the real network data and the synthetic network data to generate a trained GAN model. The method may include generating simulated network data using the random process sample and the random process, and applying weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data. The method may include combining the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data, and performing one or more actions based on the interpolated network data.

[0004] Some embodiments described herein relate to a device. The device may include one or more memories and one or more processors. One or more processors may be configured to receive real network data associated with a network, wherein the real network data includes a multivariate data set, and receive a random potential vector and a random process sample. One or more processors may be configured to generate synthetic network data using a random potential vector and a GAN model, and train the GAN model with real network data and synthetic network data to generate a trained GAN model. One or more processors may be configured to generate simulated network data using random process samples and a random process, and weights are applied to real network data, synthetic network data, and simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data. One or more processors may be configured to combine weighted real network data, weighted synthetic network data, and weighted simulated network data to generate interpolated network data, and perform one or more actions based on the interpolated network data.

[0005] Some embodiments described herein relate to a non-transitory computer-readable medium storing a set of instructions. When the set of instructions is executed by one or more processors of a device, the device may receive real network data associated with a network, and receive a random potential vector and a random process sample. When the set of instructions is executed by one or more processors of the device, the device may generate synthetic network data using a random potential vector and a GAN model, wherein the GAN model is a Wasserstein recursive GAN model, and the GAN model is trained with real network data and synthetic network data to generate a trained GAN model. When the set of instructions is executed by one or more processors of the device, the device may generate simulated network data using random process samples and a random process, and weights are applied to real network data, synthetic network data, and simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data. When the set of instructions is executed by one or more processors of the device, the device may combine the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data, and perform one or more actions based on the interpolated network data.

[0006] One aspect of the present disclosure provides a method, including: receiving, by a device, real network data associated with a network; receiving, by the device, a random latent vector and a random process sample; generating, by the device, synthetic network data using the random latent vector and a generative adversarial network (GAN) model; training, by the device, the GAN model using the real network data and the synthetic network data to generate a trained GAN model; generating, by the device, simulated network data using the random process sample and a random process; applying, by the device, weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data; combining, by the device, the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data; and performing, by the device, one or more actions based on the interpolated network data.

[0007] According to one or more embodiments of the present disclosure, the real network data includes a multivariate data set.

[0008] According to one or more embodiments of the present disclosure, the GAN model is a Wasserstein recursive GAN model.

[0009] According to one or more embodiments of the present disclosure, wherein training the GAN model using the real network data and the synthetic network data to generate the trained GAN model includes: training the GAN model using the real network data and the synthetic network data to generate the weights.

[0010] According to one or more embodiments of the present disclosure, applying the weights to the real network data, the synthetic network data and the simulated network data to generate the weighted real network data, the weighted synthetic network data and the weighted simulated network data includes: applying a first weight to the real network data to generate the weighted real network data; applying a second weight to the synthetic network data to generate the weighted synthetic network data; and applying a third weight to the simulated network data to generate the weighted simulated network data, wherein the sum of the first weight, the second weight and the third weight is equal to 1.

[0011] According to one or more embodiments of the present disclosure, the value of the first weight determines the amount of the masked real network data.

[0012] According to one or more embodiments of the present disclosure, using the random process sample and the random process to generate the simulated network data includes: using the random process sample and the real network data to generate two Poisson distributions; and superimposing the two Poisson distributions to generate the simulated network data.

[0013] Another aspect of the present disclosure provides a device, comprising: one or more memories; and one or more processors, configured to: receive real network data associated with a network,

[0014] The real network data includes a multivariate data set; receiving a random latent vector and a random process sample; generating synthetic network data using the random latent vector and a generative adversarial network (GAN) model; training the GAN model using the real network data and the synthetic network data to generate a trained GAN model; generating simulated network data using the random process samples and a random process; applying weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data; combining the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data; and performing one or more actions based on the interpolated network data.

[0015] According to one or more embodiments of the present disclosure, the one or more processors that utilize the random process samples and the random process to generate the simulated network data are used to: utilize the random process samples and the real network data to generate two Poisson distributions; and superimpose the two Poisson distributions to generate the simulated network data.

[0016] According to one or more embodiments of the present disclosure, the GAN model includes a generator component and a discriminator component.

[0017] According to one or more embodiments of the present disclosure, the one or more processors for performing the one or more actions are used for one or more of the following: providing the interpolated network data for display; or retraining the GAN model based on the interpolated network data.

[0018] According to one or more embodiments of the present disclosure, the one or more processors for executing the one or more actions are used for one or more of the following: training a network anomaly detection model using the interpolated network data; or training a network prediction model using the interpolated network data.

[0019] According to one or more embodiments of the present disclosure, the one or more processors for executing the one or more actions are used to: perform initial training of a network anomaly detection model using the interpolated network data; and perform fine-tuning training of the network anomaly detection model using the real network data.

[0020] According to one or more embodiments of the present disclosure, the one or more processors for executing the one or more actions are used to: deploy in the network a network prediction model or a network anomaly detection model trained using the interpolated network data.

[0021] Another aspect of the present disclosure provides a non-transitory computer-readable medium storing a set of instructions, the set of instructions including: one or more instructions, which, when executed by one or more processors of a device, cause the device to: receive real network data associated with a network; receive a random latent vector and a random process sample; generate synthetic network data using the random latent vector and a generative adversarial network (GAN) model, wherein the GAN model is a Wasserstein recursive GAN model; train the GAN model using the real network data and the synthetic network data to generate a trained GAN model; generate simulated network data using the random process sample and a random process; apply weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data; combine the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data; and perform one or more actions based on the interpolated network data.

[0022] According to one or more embodiments of the present disclosure, the one or more instructions that cause the device to train the GAN model using the real network data and the synthetic network data to generate the trained GAN model cause the device to: train the GAN model using the real network data and the synthetic network data to generate the weights.

[0023] According to one or more embodiments of the present disclosure, the one or more instructions that cause the device to apply the weights to the real network data, the synthetic network data, and the simulated network data to generate the weighted real network data, the weighted synthetic network data, and the weighted simulated network data cause the device to: apply a first weight to the real network data to generate the weighted real network data; apply a second weight to the synthetic network data to generate the weighted synthetic network data; and apply a third weight to the simulated network data to generate the weighted simulated network data, wherein the sum of the first weight, the second weight, and the third weight is equal to 1.

[0024] According to one or more embodiments of the present disclosure, the value of the first weight determines the amount of the masked real network data.

[0025] According to one or more embodiments of the present disclosure, the one or more instructions that cause the device to use the random process sample and the random process to generate the simulated network data cause the device to: use the random process sample and the real network data to generate two Poisson distributions; and superimpose the two Poisson distributions to generate the simulated network data.

[0026] According to one or more embodiments of the present disclosure, the GAN model includes a generator component and a discriminator component. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1A-Figure 1I is a schematic diagram of an example associated with using a generative adversarial network to enhance the quality of simulated network data.

[0028] Figure 2 is a diagram illustrating an example of training and using a machine learning model.

[0029] Figure 3 is a schematic diagram of an example environment in which the systems and / or methods described herein may be implemented.

[0030] Figure 4 yes Figure 3 A schematic diagram of example components of one or more devices.

[0031] Figure 5 is a flowchart of an example process for using generative adversarial networks to enhance the quality of simulated network data. DETAILED DESCRIPTION

[0032] The following detailed description of example embodiments refers to the accompanying drawings.The same reference numbers in different drawings may identify the same or similar elements.

[0033] Traditional rule-based or statistical models lack the ability to model the multivariate time behavior of network service throughput. In such cases, machine learning models can be used to capture complex patterns and assist in monitoring and control functionality. However, the usefulness of machine learning models is reduced due to the lack of availability of training data and the lack of high-quality training data. Collecting data from physical networks is very expensive and sometimes even impossible. This problem is particularly serious in network and cloud systems due to factors such as dynamic configuration, tail events, frequent introduction of new technologies, and / or similar items. Obtaining high-quality training data for machine learning models may be more important than developing machine learning models that meet specific indicators. Therefore, current technologies for training machine learning models consume computing resources (e.g., processing resources, memory resources, communication resources, and / or similar items), networking resources, and / or other resources associated with failure to generate sufficient or high-quality training data for machine learning models, failure to train machine learning models with insufficient training data, generating wrong machine learning models based on insufficient training data, generating wrong outputs with wrong machine learning models, and / or similar items.

[0034] Some embodiments described herein relate to a generation system that uses a generative adversarial network to enhance the quality of simulated network data. For example, the generation system may receive real network data associated with a network, and may receive random potential vectors and random process (e.g., Poisson) samples. The generation system may generate synthetic network data using random potential vectors and a GAN model, and may train the GAN model with real network data and synthetic network data to generate a trained GAN model and weights. The generation system may generate simulated network data using random process samples and a random process, and may apply weights to real network data, synthetic network data, and simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data. The generation system may combine weighted real network data, weighted synthetic network data, and weighted simulated network data to generate interpolated network data, and may perform one or more actions based on the interpolated network data.

[0035] In this way, the generation system enhances the quality of simulated network data using generative adversarial networks. For example, the generation system can use GANs to augment real telemetry data collected from the network. The generation system can use a stochastic process (e.g., a Poisson process for network data) to create a simulated rule-based time series dataset that matches the expected data distribution. The generation system can design and train Wasserstein recursive GANs for multivariate datasets (e.g., network data), capturing the underlying pattern with weights learned from pre-training on each feature that simulates the temporal distribution of each feature on limited real network data and prevents mode collapse. The generation system can design a training method for continuous, multivariate, streaming data generation that trains faster and converges better than existing models. The generation system can generate interpolated datasets based on real network data, simulated network data that captures general trends of real network data, and synthetic network data that provides high fidelity. Thus, the generation system can save computing resources, networking resources, and / or other resources that would otherwise be consumed due to failing to generate sufficient or high-quality training data for a machine learning model, using insufficient training data to train a machine learning model, generating an erroneous machine learning model based on insufficient training data, generating erroneous outputs using an erroneous machine learning model, and / or the like.

[0036] Although the embodiments described herein relate to GAN models, the embodiments may be utilized with any type of gated neural network model, such as a long short-term memory (LSTM) model, a gated recurrent unit (GRU) model, and / or the like. For example, the weight transfer process described herein from a univariate GAM model to a multivariate GAN model may be utilized with any type of gated neural network model.

[0037] Figure 1A-1I is a schematic diagram of example 100 associated with using a generative adversarial network to enhance the quality of simulated network data. Figure 1A-1I As shown, example 100 includes a generation system associated with a network and a user device. The generation system may include a system that uses a generative adversarial network to enhance the quality of simulated network data. The network may include a cellular network, such as a fifth generation (5G) network, a fourth generation (4G) network, a long term evolution (LTE) network, a third generation (3G) network, a code division multiple access (CDMA) network, and / or the like. The user device may include a wireless communication device, a mobile phone, a user device, a laptop computer, a tablet computer, a desktop computer, and / or the like. Additional details of the generation system, network, and user device are provided elsewhere herein.

[0038] like Figure 1AAs shown, and by reference numeral 105, the generation system may receive real network data associated with the network. For example, the network may include one or more network devices (e.g., base stations, switches, routers, gateways, multi-access edge computing (MEC) devices, and / or the like) that generate network data associated with network operations. The network data may include time series data with multivariate time behavior, such as business throughput, bandwidth, business demand, and / or the like. In some embodiments, the generation system may continuously receive real network data from the network in real time or near real time, may periodically receive real network data from the network, may receive real network data from the network based on requesting real network data from the network, and / or the like. In some embodiments, the generation system may store the real network data in a data structure (e.g., a database, a table, a list, and / or the like) associated with the generation system.

[0039] like Figure 1A As further shown, and by reference numeral 110, the generation system may receive a random latent vector and a random process (e.g., Poisson) sample. For example, the generation system may determine the random latent vector and the random process sample. Alternatively, a user of a user device may cause the user device to provide the random latent vector and the random process sample to the generation system. The generation system may receive the random latent vector and the random process sample from the user device. In some embodiments, the random latent vector may include a random noise vector that the generation system utilizes to generate synthetic network data, as described below. The random process sample may include a random process variable that the generation system uses to generate simulated network data, as described below.

[0040] like Figure 1A As further shown, and by reference numeral 115, the generation system may utilize the random latent vector and the GAN model to generate synthetic network data. For example, the generation system may include the GAN model. The GAN model may include a generator component and a discriminator component (e.g., Figure 1B). Conventional GAN ​​models are prone to problems such as vanishing gradients and mode collapse caused by the stability of the discriminator component during the training process. To address this issue, the Wasserstein GAN model can replace the standard binary cross entropy objective function with the earth mover's distance. Another version of the WassersteinGAN model adds a gradient penalty term as regularization to improve the stability of the discriminator component. Equation 1 below represents the loss of the generator component, which is the expected value of the prediction of the discriminator component on synthetic samples. Equation 2 represents the Wasserstein distance between the loss of the discriminator component and the expected value of the discriminator component from synthetic samples and real samples, as well as the gradient penalty term. Equation 3 represents the calculation of the gradient penalty term to maintain 1-L continuity by constraining the maximum norm of the gradient to 1. By adding a regularization term according to Equation 4, the loss function of the discriminator component can be modified to ensure that the difference between the expected scores does not increase dramatically.

[0041]

[0042]

[0043]

[0044]

[0045] The gradient penalty can be calculated on the interpolated sequence, which is the real sequence (x) and the synthetic sequence A weighted combination of .

[0046] In some embodiments, the GAN model may include a simple recursive GAN (RGAN) with Wasserstein distance and gradient penalty for time series generation (e.g., WRGAN-GP). By utilizing Wasserstein loss with gradient penalty as the loss function, the GAN model can prevent problems such as mode collapse while still generating data with sufficient fidelity, diversity, and usefulness.

[0047] In some embodiments, a generator component (G) and a discriminator component (D) (e.g., see Figure 1B ) can be a long short-term memory (LSTM) based on a process sequence. The generator component can return a window of synthetic sequence (f) (e.g., synthetic network data) based on a random latent vector (z) and hidden state (h,c). At time (t), based on a sliding window of size (w) of data (d), a series of real network data x(t)∈R can be defined on the historical time series data used for training. w×dThe LSTM unit of the generator element can use the random latent vector (z) as a sequence, together with the hidden state (h,c) to predict the sequence window (f t ), as shown in Equations 5 and 6.

[0048]

[0049]

[0050] The size of the random latent vector (z) can be a hyperparameter. As the value of the random latent vector (z) increases, a more deterministic mapping can be provided. The discriminator component can utilize real network data (x) or synthetic network data. to determine the difference.

[0051] like Figure 1B As shown, and by reference numeral 120, the generation system can train a GAN model with real network data and synthetic network data to generate a trained GAN model. For example, the generation system can train a GAN model for each feature of the real network data and the synthetic network data. In some embodiments, the generation system can train n different univariate models for n-feature input data provided by the gates of the LSTM unit (e.g., input gate (i), output gate (o), and forget gate (f)). Internally, the LSTM unit can cascade the weights associated with different gates into a matrix while training the gates. The input data can be divided into block matrices to show the transfer of weights. The input gate can be associated with two weight matrices or four layers. The first column of the block matrix can be the weight matrix of the LSTM unit associated with the input vector, and the second column of the block matrix can represent the weight matrix associated with the hidden vector. The univariate model has learned the temporal dynamics of the feature on which the univariate model is trained. However, the n-generators may not be synchronized, and their phases may not be aligned to combine the cross-correlations between multiple features. To enhance the training of multivariate models having generator and discriminator components with associated weights, the learned weights and cross-correlation weights of the individual models may be randomly assigned to be trained in a later epoch.

[0052] In the first layer (e.g., the layer with weights associated with the forget gates), the weight matrices of the forget gates from the LSTM units in the n different univariate models may be placed diagonally, and the remaining entries may be set to zero and may be associated with cross-correlations (e.g., learned later in training). The remaining layers may be generated similarly. The input vector may be formed by the concatenation of the input vectors provided to the n-univariate models. Input vector m H 入 =n*H 入 and m H 单元 =n*H 单元This is true for a successful weight transfer, where n is the number of features in the dataset. This type of weight transfer can facilitate the training process, make the GAN model robust, and prevent mode collapse.

[0053] In some embodiments, the generation system may train the GAN model with real network data and synthetic network data to generate a trained GAN model and generate weights for generating interpolated network data, as described below. In some embodiments, the generation system may train the GAN model for a certain number of steps (e.g., one thousand, two thousand, three thousand, and / or the like) using mini-batch gradient descent. In each epoch, the discriminator component of the GAN model may be trained five times, and the generator component of the GAN model may be trained once. The generator component may be updated based on feedback from the discriminator component. The loss (L) of the generator component G ) (e.g., according to Equation 1) may be the expected value of synthetic network data equivalent to real network data. The loss (L) of the discriminator component D ) can be real network data (X) and synthetic network data The Wasserstein distance between has a gradient penalty term according to Equation 2. Figure 1G and Figure 1H Provides additional details for training GAN models.

[0054] like Figure 1C As shown, and by reference numeral 125, the generation system can utilize random process samples and random processes to generate simulated network data. For example, the simulated network data can be application-based, can be generated from a Poisson mathematical process, and can include network traffic tracking data at a (samples / second) frequency. The network traffic throughput can include seasonal characteristics related to demand changes at different times during a day. The generation system can generate the simulated network data from a random (e.g., Poisson) process that defines packets arriving per second, where the arrival rate of packets during the day (e.g., 8:00 a.m. to 10:00 p.m.) can be determined by the random process sample (e.g., λ). 日 = 10) is defined. In order to reduce network traffic at night, the Poisson process variable (λ) can be multiplied by the deactivation rate factor (ω D ). As a result, the generating system can be based on the random process samples (such as λ 夜 =2) to generate simulated network data during the night (e.g., 10:00 p.m. to 8:00 a.m.). The resulting simulated network data (X p ) can be generated as a superposition of two Poisson distributions. Two of the Poisson process variables can be trained separately using a grid search on each attribute of the real network data to obtain the lowest mean square error (MSE).

[0055] like Figure 1D As shown, and by reference numeral 130, the generation system may apply weights to real network data, synthetic network data, and simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data. For example, the generation system may apply weights generated based on training a GAN model with real network data and synthetic network data. Alternatively, the generation system may determine different weights for each of the real network data, synthetic network data, and simulated network data. In some embodiments, when weights are applied to real network data, synthetic network data, and simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data, the generation system may apply a first weight (w1) to the real network data to generate weighted real network data. The generation system may apply a second weight (e.g., w2 different from the first weight) to the synthetic network data to generate weighted synthetic network data, and may apply a third weight (e.g., w3 different from the first weight and the second weight) to the simulated network data to generate weighted simulated network data. In some embodiments, the sum of the first weight, the second weight, and the third weight may be equal to 1. In some implementations, the value of the first weight may determine the amount of real network data that is masked.

[0056] like Figure 1E As shown and indicated by reference numeral 135, the generation system may combine the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data. For example, the generation system may utilize the synthetic network data Simulated network data (X p ) and a small contribution from real network data (X) to generate interpolated network data (I(λ,z)) that approximates the network traffic of the network. The interpolated network data may be a weighted sum of real network data, synthetic network data, and simulated network data (e.g., the sum of weighted real network data, weighted synthetic network data, and weighted simulated network data). The generation system may use a random process to model user activities (e.g., by changing the variable λ). The generator component of the GAN model may model a time series (e.g., network data) by learning the underlying pattern using a random latent vector. The second and third weights (e.g., w2 and w3) may control the weights of the synthetic network data and the simulated network data, respectively. Interpolated network data similar to real network data may include a high value of the third weight. The contribution of real network data (e.g., the first weight w1) may be diluted to ensure adequate masking of sensitive information in the interpolated network data. Equation 7 below may indicate a weighted combination of real network data, synthetic network data, and simulated network data (e.g., interpolated network data).

[0057]

[0058] W1+w2+w3=1(8)

[0059] Combined with the following Fig. 1I Additional details on generating interpolated network data are provided.

[0060] like Figure 1F As shown, and by reference numeral 140, the generation system may perform one or more actions based on the interpolated network data. In some embodiments, performing the one or more actions includes the generation system retraining the GAN model based on the interpolated network data. For example, the generation system may utilize the interpolated network data as additional training data for retraining the GAN model, thereby increasing the amount of training data available for training the GAN model. Thus, relative to other systems for identifying, obtaining, and / or generating historical data for training machine learning models, the generation system may save computing resources associated with identifying, obtaining, and / or generating historical data for training the GAN model.

[0061] In some embodiments, performing one or more actions includes the generation system training the network anomaly detection model with interpolated network data. For example, since the interpolated network data is a weighted combination of real network data, simulated network data, and synthetic network data, the generation system can use the interpolated network data to train the network anomaly detection model according to changes in user behavior by controlling the parameter (λ). The generation system can apply meta-learning by performing initial training of the network anomaly detection model with interpolated network data and fine-tuning the network anomaly detection model with real network data. If an attribute is missing some values ​​or a feature is sparse, the generation system may use interpolated network data to estimate the attribute value. In this way, the generation system can save computing resources, networking resources, and / or other resources that would otherwise be consumed due to failure to generate sufficient or high-quality training data for the machine learning model.

[0062] In some embodiments, performing one or more actions includes the generation system training the network prediction model with the interpolated network data. For example, the generation system may train the network prediction model with the interpolated network data to generate a trained network prediction model that can be used for self-correcting network devices in a networking application. In this manner, the generation system can save computing resources, networking resources, and / or other resources that would otherwise be consumed due to insufficient training data to train the machine learning model.

[0063] In some embodiments, performing one or more actions includes the generation system performing initial training of the network anomaly detection model with interpolated network data, and fine-tuning the network anomaly detection model with real network data. For example, the generation system may perform initial training of the network anomaly detection model with interpolated network data, and fine-tune the network anomaly detection model with real network data. If an attribute is missing some values ​​or a feature is sparse, the generation system may use interpolated network data to estimate the attribute value. In this way, the generation system can save computing resources, networking resources, and / or other resources that would otherwise be consumed due to generating an erroneous machine learning model based on insufficient training data.

[0064] In some embodiments, performing one or more actions includes the generation system deploying a network prediction model or a network anomaly detection model trained with interpolated network data in the network. For example, the generation system may train the network prediction model or the network anomaly detection model with interpolated network data to generate a trained network prediction model or a trained network anomaly detection model. The trained model may be used for self-correcting network devices in networked applications. In this way, the generation system may save computing resources, networking resources, and / or other resources that would otherwise be consumed by generating incorrect outputs with incorrect machine learning models.

[0065] Figure 1G It is a schematic diagram depicting the LSTM unit structure of the GAN model trained on various features of the dataset. 入 is 1, and H 单元 represents the hidden unit dimension. Figure 1G In Figure 1, the four layers represent the gates of the LSTM unit, an input gate (i), an output gate (o), and a forget gate (f). Internally, the LSTM unit cascades the weights associated with different gates into a matrix and trains them simultaneously. The weights are divided into block matrices to show the transfer of weights. The input gate is associated with two weight matrices (e.g., four layers). The first column of block matrices is the weight matrix of the LSTM unit associated with the input vector, and the second column of block matrices represents the weight matrix associated with the hidden vector. For simplicity, the bias vector associated with the gate is not shown. These univariate models have learned the temporal dynamics of the features they are trained on. However, the n-generators are not synchronized, and their phases are not aligned to incorporate the cross-correlations between multiple features.

[0066] Figure 1H is a schematic diagram depicting the LSTM unit structure of a GAN model trained on a multi-feature dataset. m H 入 is n, where n is the number of features in the dataset, m H 单元represents the hidden unit dimensions of the LSTM unit (e.g., m H 单元 =n*H 单元 ). Figure 1H Depicts training a multivariate GAN model with a similar generator and evaluator architecture, whose weights are passed as the learned weights of the individual models, where the cross-correlation weights are randomly assigned to be trained in a later epoch. In the first layer (e.g., the layer with weights associated with the forget gate), there are weight matrices from the forget gates of the LSTM units in n different univariate models placed diagonally, and the remaining entries are the entries associated with the cross-correlation, which will be learned later in training. The remaining layers follow the same logic. The input vector is formed by the concatenation of the input vectors to the n univariate models. m H 入 =n*H 入 and m H 单元 =n*H 单元 For successful weight transfer, where n is the number of features in the dataset. The transfer of weights facilitates the training process, makes the GAN model robust, and prevents mode collapse.

[0067] Fig. 1I is depicted using synthetic network data Simulated network data (X p ) and the real network data (X) to generate interpolated network data (I(λ,z)).

[0068] In this way, the generation system uses a generative adversarial network to enhance the quality of simulated network data. For example, the generation system can use GANs to augment real telemetry data collected from the network. The generation system can use a stochastic process to create a simulated rule-based time series dataset that matches the expected data distribution. The generation system can design and train Wasserstein recursive GANs for multivariate datasets (e.g., network data), capturing the underlying pattern with weights learned from pre-training on each feature that simulates the temporal distribution of each feature on limited real network data and prevents mode collapse. The generation system can design a training method for continuous, multivariate, streaming data generation that trains faster and converges better than existing models. The generation system can generate interpolated datasets based on real network data, simulated network data that captures general trends of real network data, and synthetic network data that provides high fidelity. Thus, the generation system can save computing resources, networking resources, and / or other resources that would otherwise be consumed due to failing to generate sufficient or high-quality training data for a machine learning model, using insufficient training data to train a machine learning model, generating an erroneous machine learning model based on insufficient training data, generating erroneous outputs using an erroneous machine learning model, and / or the like.

[0069] As mentioned above, Figure 1A-Figure 1I Provided as an example. Other examples may vary from Figure 1A-1I as described. Figure 1A-1I The number and arrangement of devices shown are provided as examples. Figure 1A-1I There may be additional devices, fewer devices, different devices, or differently arranged devices than those shown. Figure 1A-1I Two or more of the devices shown may be implemented in a single device, or Figure 1A-1I The single device shown may be implemented as multiple distributed devices. Additionally or alternatively, Figure 1A-Figure 1I The illustrated set of devices (e.g., one or more devices) may perform the operations described as being performed by Figure 1A-Figure 1I Another group of devices is shown performing one or more functions.

[0070] Figure 2 2 is a schematic diagram illustrating an example 200 of training and using a machine learning model. The machine learning model training and use described herein may be performed using a machine learning system. The machine learning system may include or may be included in a computing device, a server, a cloud computing environment, or the like, such as a production system.

[0071] As indicated by reference numeral 205, a set of observations may be used to train a machine learning model. The set of observations may be obtained from training data (e.g., historical data), such as data collected during one or more processes described herein. In some embodiments, the machine learning system may receive the set of observations (e.g., as input) from a generation system, as described elsewhere herein.

[0072] As shown by reference numeral 210, the set of observations may include a feature set. A feature set may include a set of variables, and the variables may be referred to as features. A particular observation may include a set of variable values ​​(or feature values) corresponding to the set of variables. In some embodiments, the machine learning system may determine the variables of a set of observations and / or the variable values ​​of a particular observation based on input received from a generating system. For example, the machine learning system may identify a feature set (e.g., one or more features and / or feature values) by extracting the feature set from structured data, by performing natural language processing to extract the feature set from unstructured data, and / or by receiving input from an operator.

[0073] As an example, a feature set for a set of observations may include a first feature of real network data, a second feature of synthetic network data, a third feature of simulated network data, etc. As shown, for the first observation, the first feature may have a value of real network data 1, the second feature may have a value of synthetic network data 1, and the third feature may have a value of simulated network data 1, etc. These features and feature values ​​are provided as examples and may be different in other examples.

[0074] As indicated by reference numeral 215, the set of observations may be associated with a target variable. A target variable may represent a variable having a numerical value, may represent a variable having a numerical value that falls within a range of values ​​or has some discrete possible values, may represent a variable that may be selected from a plurality of options (e.g., one of a plurality of categories, classifications, or labels), and / or may represent a variable having a Boolean value. A target variable may be associated with a target variable value, and the target variable value may be specific to an observation. In example 200, the target variable is interpolated data, which has a value of interpolated data 1 for the first observation. The above feature sets and target variables are provided as examples, and other examples may differ from the above.

[0075] The target variable may represent the value that the machine learning model is being trained to predict, and the feature set may represent the variables that are input to the trained machine learning model to predict the value of the target variable. The set of observations may include the target variable value, so that the machine learning model may be trained to identify patterns in the feature set that lead to the target variable value. A machine learning model trained to predict the value of the target variable may be referred to as a supervised learning model.

[0076] In some embodiments, a machine learning model may be trained on a set of observations that do not include a target variable. This may be referred to as an unsupervised learning model. In this case, the machine learning model may learn patterns from the set of observations without labeling or supervision, and may provide outputs indicative of such patterns, such as by identifying groups of related items within the set of observations using clustering and / or association.

[0077] As shown by reference numeral 220, the machine learning system can train a machine learning model using a set of observations and one or more machine learning algorithms, such as a regression algorithm, a decision tree algorithm, a neural network algorithm, a k-nearest neighbor algorithm, a support vector machine algorithm, or the like. After training, the machine learning system can store the machine learning model as a trained machine learning model 225 to be used to analyze new observations.

[0078] As indicated by reference numeral 230, the machine learning system may apply the trained machine learning model 225 to the new observation, such as by receiving the new observation and inputting the new observation into the trained machine learning model 225. As shown, as an example, the new observation may include a first feature of real network data X, a second feature of synthetic network data Y, a third feature of simulated network data Z, and so on. The machine learning system may apply the trained machine learning model 225 to the new observation to generate an output (e.g., a result). The type of output may depend on the type of machine learning model and / or the type of machine learning task being performed. For example, the output may include a predicted value of a target variable, such as when supervised learning is employed. Additionally or alternatively, the output may include information identifying the cluster to which the new observation belongs and / or information indicating the similarity between the new observation and one or more other observations, such as when unsupervised learning is employed.

[0079] As one example, the trained machine learning model 225 may predict a value of the interpolated data A for the target variable of the interpolated data for the new observation, as indicated by reference numeral 235. Based on the prediction, the machine learning system may provide a first recommended value, may provide an output for determining the first recommended value, may perform a first automated action, and / or may cause performance of the first automated action (e.g., by instructing another device to perform the automated action), among other examples.

[0080] In some implementations, the trained machine learning model 225 may classify (e.g., cluster) new observations in clusters, as indicated by reference numeral 240. The observations within a cluster may have a threshold similarity. As an example, if the machine learning system classifies the new observation into a first cluster (e.g., a real network data cluster), the machine learning system may provide a first recommendation. Additionally or alternatively, the machine learning system may perform a first automated action based on classifying the new observation in the first cluster and / or may cause the first automated action to be performed (e.g., by instructing another device to perform the automated action).

[0081] As another example, if the machine learning system is to classify the new observation into a second cluster (e.g., a synthetic network data cluster), the machine learning system may provide a second (e.g., different) recommendation and / or may perform or cause a second (e.g., different) automated action.

[0082] In some embodiments, the recommended value and / or automated action associated with a new observation may be based on a target variable value having a particular label (e.g., a classification or categorization), may be based on whether the target variable value satisfies one or more thresholds (e.g., whether the target variable value is greater than a threshold, less than a threshold, equal to a threshold, falls within a threshold range, or the like), and / or may be based on a clustering that classifies the new observation.

[0083] In some embodiments, the trained machine learning model 225 can be retrained using feedback information. For example, feedback can be provided to the machine learning model. The feedback can be associated with actions performed based on the recommended values ​​provided by the trained machine learning model 225 and / or automated actions performed or caused by the trained machine learning model 225. In other words, the recommended values ​​and / or actions output by the trained machine learning model 225 can be used as input to retrain the machine learning model (e.g., a feedback loop can be used to train and / or update the machine learning model).

[0084] In this way, a machine learning system can apply a rigorous and automated process to determine interpolated network data. The machine learning system can identify and / or label dozens, hundreds, thousands, or millions of features and / or feature values ​​for dozens, hundreds, thousands, or millions of observations, thereby improving accuracy and consistency and reducing latency associated with determining interpolated network data relative to the need to allocate computing resources to dozens, hundreds, or thousands of operators to manually determine interpolated network data using features or feature values.

[0085] As mentioned above, Figure 2 Provided as an example. Other examples may differ from the combination Figure 2 as described.

[0086] Figure 3 is a schematic diagram of an example environment 300 in which the systems and / or methods described herein may be implemented. Figure 3 As shown, environment 300 may include a generation system 301, which may include one or more elements of a cloud computing system 302 and / or may be implemented within the cloud computing system 302. The cloud computing system 302 may include one or more elements 303-313, as described in more detail below. Figure 3 As further shown, environment 300 may include network 320 and / or user device 330. Devices and / or elements of environment 300 may be interconnected via wired connections and / or wireless connections.

[0087] The cloud computing system 302 includes computing hardware 303, a resource management component 304, a host operating system (OS) 305, and / or one or more virtual computing systems 306. The cloud computing system 302 may be implemented, for example, on an Amazon Web Services platform, a Microsoft Azure platform, or a Snowflake platform. The resource management component 304 may perform virtualization (e.g., abstraction) of the computing hardware 303 to create one or more virtual computing systems 306. Using virtualization, the resource management component 304 enables a single computing device (e.g., a computer or server) to operate like multiple computing devices, such as by creating multiple isolated virtual computing systems 306 from the computing hardware 303 of a single computing device. In this way, the computing hardware 303 may operate more efficiently than using separate computing devices, with lower power consumption, higher reliability, higher availability, higher utilization, greater flexibility, and lower cost.

[0088] Computing hardware 303 includes hardware and corresponding resources from one or more computing devices. For example, computing hardware 303 may include hardware from a single computing device (e.g., a single server) or from multiple computing devices (e.g., multiple servers), such as multiple computing devices in one or more data centers. As shown, computing hardware 303 may include one or more processors 307, one or more memories 308, one or more storage components 309, and / or one or more networking components 310. Examples of processors, memories, storage components, and networking components (e.g., communication components) are described elsewhere herein.

[0089] The resource management component 304 includes a virtualization application (e.g., implemented on hardware, such as the computing hardware 303) that can virtualize the computing hardware 303 to start, stop, and / or manage one or more virtual computing systems 306. For example, such as when the virtual computing system 306 is a virtual machine 311, the resource management component 304 may include a hypervisor (e.g., a bare metal or type 1 hypervisor, a hosted or type 2 hypervisor, or another type of hypervisor) or a virtual machine monitor. Additionally or alternatively, the resource management component 304 may include a container manager, such as when the virtual computing system 306 is a container 312. In some embodiments, the resource management component 304 is implemented within and / or in coordination with the host operating system 305.

[0090] The virtual computing system 306 includes a cloud-based virtual environment that implements the operations and / or processes described herein using computing hardware 303. As shown, the virtual computing system 306 may include a virtual machine 311, a container 312, or a hybrid environment 313 including virtual machines and containers, among other examples. The virtual computing system 306 may implement one or more applications using a file system that includes binary files, software libraries, and / or other resources required to implement applications on a client operating system (e.g., within the virtual computing system 306) or a host operating system 305.

[0091] Although generation system 301 may include one or more elements 303-313 of cloud computing system 302, may be implemented within cloud computing system 302, and / or may be hosted within cloud computing system 302, in some embodiments, generation system 301 may not be cloud-based (e.g., may be implemented outside of a cloud computing system) or may be partially cloud-based. For example, generation system 301 may include one or more devices that are not part of cloud computing system 302, such as Figure 4 The device 400 may include a stand-alone server or another type of computing device. The generation system 301 may perform one or more operations and / or processes described in more detail elsewhere herein.

[0092] Network 320 includes one or more wired and / or wireless networks. For example, network 320 may include a cellular network, a public land mobile network (PLMN), a local area network (LAN), a wide area network (WAN), a private network, the Internet, and / or a combination of these or other types of networks. Network 320 enables communication between devices in environment 300.

[0093] User device 330 may include one or more devices capable of receiving, generating, storing, processing, and / or providing information, as described elsewhere herein. User device 330 may include a communication device and / or a computing device. For example, user device 330 may include a wireless communication device, a mobile phone, a user device, a laptop computer, a tablet computer, a desktop computer, a game console, a set-top box, a wearable communication device (e.g., a smart watch, a pair of smart glasses, a head-mounted display, or a virtual reality headset), or a similar type of device.

[0094] Figure 3 The number and arrangement of devices and networks shown are provided as examples. Figure 3 There may be additional devices and / or networks, fewer devices and / or networks, different devices and / or networks, or differently arranged devices and / or networks than those shown. Figure 3 Two or more of the devices shown may be implemented in a single device, or Figure 3 The single device shown may be implemented as multiple distributed devices. Additionally or alternatively, one set of devices (eg, one or more devices) of environment 300 may perform one or more functions described as being performed by another set of devices of environment 300.

[0095] Figure 4 4 is a schematic diagram of example components of device 400, which may correspond to generation system 301 and / or user device 330. In some embodiments, generation system 301 and / or user device 330 may include one or more devices 400 and / or one or more components of device 400. Figure 4 As shown, device 400 may include a bus 410 , a processor 420 , a memory 430 , an input component 440 , an output component 450 , and a communication component 460 .

[0096] Bus 410 includes one or more components that enable wired and / or wireless communication between components of device 400. Bus 410 may Figure 4 Two or more components of a computer program product are coupled together, such as via an operational coupling, a communication coupling, an electronic coupling, and / or an electrical coupling. Processor 420 includes a central processing unit, a graphics processing unit, a microprocessor, a controller, a microcontroller, a digital signal processor, a field programmable gate array, an application specific integrated circuit, and / or another type of processing component. Processor 420 is implemented in hardware, firmware, or a combination of hardware and software. In some embodiments, processor 420 includes one or more processors that can be programmed to perform one or more operations or processes described elsewhere herein.

[0097] Memory 430 includes volatile and / or non-volatile memory. For example, memory 430 may include random access memory (RAM), read-only memory (ROM), a hard drive, and / or another type of memory (e.g., flash memory, magnetic memory, and / or optical memory). Memory 430 may include internal memory (e.g., RAM, ROM, or hard drive) and / or removable memory (e.g., removable via a universal serial bus connection). Memory 430 may be a non-temporary computer-readable medium. Memory 430 stores information, instructions, and / or software (e.g., one or more software applications) related to the operation of device 400. In some embodiments, memory 430 includes one or more memories such as coupled to one or more processors (e.g., processor 420) via bus 410.

[0098] Input assembly 440 enables device 400 to receive input, such as user input and / or sensor input. For example, input assembly 440 may include a touch screen, a keyboard, a keypad, a mouse, a button, a microphone, a switch, a sensor, a global positioning system sensor, an accelerometer, a gyroscope and / or an actuator. Output assembly 450 enables device 400 to provide output, such as via a display, a speaker and / or a light emitting diode. Communication assembly 460 enables device 400 to communicate with other devices via a wired connection and / or a wireless connection. For example, communication assembly 460 may include a receiver, a transmitter, a transceiver, a modem, a network interface card and / or an antenna.

[0099] Device 400 may perform one or more operations or processes described herein. For example, a non-transitory computer-readable medium (e.g., memory 430) may store a set of instructions (e.g., one or more instructions or codes) implemented by processor 420. Processor 420 may implement the set of instructions to perform one or more operations or processes described herein. In some embodiments, the implementation of the set of instructions by one or more processors 420 prompts one or more processors 420 and / or device 400 to perform one or more operations or processes described herein. In some embodiments, hard-wired circuitry may be used instead of instructions or in combination with instructions to perform one or more operations or processes described herein. Additionally or alternatively, processor 420 may be configured to perform one or more operations or processes described herein. Therefore, the embodiments described herein are not limited to any particular combination of hardware circuitry and software.

[0100] Figure 4 The number and arrangement of components shown are provided as examples. Figure 4The device 400 may include additional components, fewer components, different components, or differently arranged components than shown. Additionally or alternatively, a set of components (e.g., one or more components) of the device 400 may perform one or more functions described as being performed by another set of components of the device 400.

[0101] Figure 5 A flow chart depicts an example process 500 for enhancing the quality of simulated network data using a generative adversarial network. In some embodiments, Figure 5 One or more process blocks of may be performed by a device (e.g., generating system 301). In some embodiments, Figure 5 One or more process blocks of may be performed by another device or a group of devices separate from or including the device. Additionally or alternatively, Figure 5 One or more process blocks of may be performed by one or more components of device 400 , such as processor 420 , memory 430 , input component 440 , output component 450 , and / or communication component 460 .

[0102] like Figure 5 As shown, process 500 may include receiving real network data associated with a network (block 510). For example, as described above, a device may receive real network data associated with a network. In some embodiments, the real network data includes a multivariate data set.

[0103] like Figure 5 As further shown, process 500 may include receiving a random latent vector and a random process sample (block 520). For example, as described above, a device may receive a random latent vector and a random process sample.

[0104] like Figure 5 As further shown, process 500 may include generating synthetic network data using a random latent vector and a GAN model (block 530). For example, as described above, the device may generate synthetic network data using a random latent vector and a GAN model. In some embodiments, the GAN model is a Wasserstein recursive GAN model. In some embodiments, the GAN model includes a generator component and a discriminator component.

[0105] like Figure 5 As further shown, process 500 may include training the GAN model with the real network data and the synthetic network data to generate a trained GAN model (block 540). For example, as described above, the device may train the GAN model with the real network data and the synthetic network data to generate a trained GAN model. In some embodiments, training the GAN model with the real network data and the synthetic network data to generate a trained GAN model includes training the GAN model with the real network data and the synthetic network data to generate weights.

[0106] like Figure 5 As further shown, process 500 may include generating simulated network data using the random process sample and the random process (block 550). For example, as described above, the device may generate simulated network data using the random process sample and the random process. In some embodiments, generating simulated network data using the random process sample and the random process includes generating two Poisson distributions using the random process sample and the real network data, and superimposing the two Poisson distributions to generate simulated network data.

[0107] like Figure 5 As further shown, process 500 may include applying weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data (block 560). For example, as described above, the device may apply weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data. In some embodiments, applying weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data includes applying a first weight to the real network data to generate weighted real network data, applying a second weight to the synthetic network data to generate weighted synthetic network data, and applying a third weight to the simulated network data to generate weighted simulated network data, wherein the sum of the first weight, the second weight, and the third weight is equal to 1. In some embodiments, the value of the first weight determines the amount of real network data that is masked.

[0108] like Figure 5 As further shown, process 500 may include combining the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data (block 570). For example, as described above, the device may combine the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data.

[0109] like Figure 5 As further shown, process 500 may include performing one or more actions based on the interpolated network data (block 580). For example, as described above, the device may perform one or more actions based on the interpolated network data. In some embodiments, performing the one or more actions includes providing the interpolated network data for display, or retraining one or more of the GAN models based on the interpolated network data. In some embodiments, performing the one or more actions includes training one or more of a network anomaly detection model with the interpolated network data or training a network prediction model with the interpolated network data.

[0110] In some embodiments, performing one or more actions includes performing initial training of a network anomaly detection model using interpolated network data, and performing fine-tuning training of the network anomaly detection model using real network data. In some embodiments, performing one or more actions includes deploying a network prediction model or a network anomaly detection model trained using interpolated network data in the network.

[0111] although Figure 5 An example block diagram of process 500 is shown, but in some implementations, Figure 5 Process 500 may include additional blocks, fewer blocks, different blocks, or differently arranged blocks than shown. Additionally or alternatively, two or more blocks of process 500 may be performed in parallel.

[0112] As used herein, the term "component" is intended to be broadly interpreted as hardware, firmware, or a combination of hardware and software. It is apparent that the systems and / or methods described herein may be implemented with different forms of hardware, firmware, and / or a combination of hardware and software. The actual dedicated control hardware or software code used to implement these systems and / or methods does not limit these implementations. Therefore, the operation and behavior of the systems and / or methods are described herein without reference to specific software code - it should be understood that, based on the description herein, software and hardware may be used to implement the systems and / or methods.

[0113] As used herein, satisfying a threshold may refer to a value that is greater than a threshold, greater than or equal to a threshold, less than or equal to a threshold, equal to a threshold, not equal to a threshold, or the like, depending on the context.

[0114] To the extent that the above-described embodiments collect, store, or use an individual's personal information, it should be understood that such information should be used in accordance with all applicable laws related to the protection of personal information. In addition, the collection, storage, and use of such information may require the individual's consent to such activities, for example, through well-known "opt-in" or "opt-out" processes appropriate to the circumstances and type of information. The storage and use of personal information may be carried out in an appropriately secure manner reflecting the type of information, for example, through various encryption and anonymization techniques for particularly sensitive information.

[0115] Even if specific combinations of features are listed in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of the various embodiments. In fact, many of these features can be combined in ways not specifically listed in the claims and / or disclosed in the specification. Although each dependent claim listed below may directly depend on only one claim, the disclosure of the various embodiments includes the combination of each dependent claim with each other claim in the claim set. As used herein, a phrase referring to "at least one of" a series of items refers to any combination of those items, including a single member. As an example, "at least one of: a, b, or c" is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination with multiple identical items.

[0116] Unless so clearly described, any element, action or instruction used in this article should not be interpreted as key or necessary. In addition, as used herein, the article "a" and "an" are intended to include one or more projects, and can be used interchangeably with "one or more (one or more)". In addition, as used herein, the article "the" is intended to include one or more projects related to the article "the", and can be used interchangeably with "the one or more (the one or more)". In addition, as used herein, the term "set" is intended to include one or more projects (for example, related projects, unrelated projects or a combination of related and unrelated projects), and can be used interchangeably with "one or more (one or more)". In the case of only being intended to a project, the phrase "only one (onlyone)" or similar language is used. In addition, as used herein, the term "has", "have", "having (having)" or similar items are intended to be open terms. In addition, the phrase "based on" is intended to mean "based, at least in part, on," unless expressly stated otherwise. Furthermore, as used herein, the term "or" is inclusive when used in a series and is used interchangeably with "and / or," unless expressly stated otherwise (e.g., if used in conjunction with "either" or "only one of").

[0117] In the foregoing description, various example embodiments have been described with reference to the accompanying drawings. However, it will be apparent that various modifications and changes may be made thereto, and that additional embodiments may be implemented, without departing from the broader scope of the invention as set forth in the appended claims. Accordingly, the description and drawings are to be regarded as illustrative rather than restrictive.

Claims

1. A method comprising: receiving, by the device, actual network data associated with the network; receiving, by the device, a random latent vector and a random process sample; The device generates synthetic network data using the random latent vector and a generative adversarial network (GAN) model; Training the GAN model by the device using the real network data and the synthetic network data to generate a trained GAN model; The device generates simulated network data using the random process sample and the random process; applying weights by the device to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data; combining, by the device, the weighted real network data, the weighted synthetic network data, and the weighted simulated network data to generate interpolated network data; as well as One or more actions are performed by the device based on the interpolated network data.

2. The method of claim 1, wherein the real network data comprises a multivariate dataset.

3. The method of claim 1, wherein the GAN model is a Wasserstein recursive GAN model.

4. The method according to claim 1, wherein training the GAN model using the real network data and the synthetic network data to generate the trained GAN model comprises: The GAN model is trained using the real network data and the synthetic network data to generate the weights.

5. The method of claim 1 , wherein applying the weights to the real network data, the synthetic network data, and the simulated network data to generate the weighted real network data, the weighted synthetic network data, and the weighted simulated network data comprises: applying a first weight to the real network data to generate the weighted real network data; applying a second weight to the synthetic network data to generate the weighted synthetic network data; as well as applying a third weight to the simulated network data to generate the weighted simulated network data, The sum of the first weight, the second weight and the third weight is equal to 1. The method of claim 5 , wherein the value of the first weight determines the amount of the real network data that is masked.

7. The method according to claim 1, wherein using the random process sample and the random process to generate the simulated network data comprises: generating two Poisson distributions using the random process sample and the real network data; as well as The two Poisson distributions are superimposed to generate the simulated network data.

8. A device comprising: one or more memories; as well as One or more processors for: Receive real network data associated with the network, wherein the real network data includes a multivariate dataset; receiving a random latent vector and a random process sample; Generate synthetic network data using the random latent vector and a generative adversarial network (GAN) model; Training the GAN model using the real network data and the synthetic network data to generate a trained GAN model; Generate simulated network data using the random process sample and the random process; applying weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data; combining the weighted real network data, the weighted synthetic network data and the weighted simulated network data to generate interpolated network data; as well as One or more actions are performed based on the interpolated network data.

9. The apparatus of claim 8, wherein the one or more processors that utilize the random process samples and the random process to generate the simulated network data are configured to: generating two Poisson distributions using the random process sample and the real network data; and The two Poisson distributions are superimposed to generate the simulated network data.

10. The apparatus of claim 8, wherein the GAN model comprises a generator component and a discriminator component.

11. The apparatus of claim 8, wherein the one or more processors for performing the one or more actions are configured to perform one or more of: providing the interpolated network data for display; or The GAN model is retrained based on the interpolated network data.

12. The apparatus of claim 8, wherein the one or more processors for performing the one or more actions are configured to: Training a network anomaly detection model using the interpolated network data; or The interpolated network data is used to train a network prediction model.

13. The apparatus of claim 8, wherein the one or more processors for performing the one or more actions are configured to: performing initial training of a network anomaly detection model using the interpolated network data; and The real network data is used to perform fine-tuning training of the network anomaly detection model.

14. The apparatus of claim 8, wherein the one or more processors for performing the one or more actions are configured to: A network prediction model or a network anomaly detection model trained using the interpolated network data is deployed in the network.

15. A non-transitory computer readable medium storing a set of instructions, the set of instructions comprising: One or more instructions that, when executed by one or more processors of a device, cause the device to: receiving actual network data associated with the network; receiving a random latent vector and a random process sample; Using the random latent vector and the generative adversarial network (GAN) model to generate synthetic network data, The GAN model is a Wasserstein recursive GAN model; Training the GAN model using the real network data and the synthetic network data to generate a trained GAN model; Generate simulated network data using the random process sample and the random process; applying weights to the real network data, the synthetic network data, and the simulated network data to generate weighted real network data, weighted synthetic network data, and weighted simulated network data; combining the weighted real network data, the weighted synthetic network data and the weighted simulated network data to generate interpolated network data; as well as One or more actions are performed based on the interpolated network data.

16. The non-transitory computer-readable medium of claim 15, wherein the one or more instructions that cause the device to train the GAN model using the real network data and the synthetic network data to generate the trained GAN model cause the device to: The GAN model is trained using the real network data and the synthetic network data to generate the weights.

17. The non-transitory computer-readable medium of claim 15, wherein the one or more instructions causing the device to apply the weights to the real network data, the synthetic network data, and the simulated network data to generate the weighted real network data, the weighted synthetic network data, and the weighted simulated network data cause the device to: applying a first weight to the real network data to generate the weighted real network data; applying a second weight to the synthetic network data to generate the weighted synthetic network data; as well as applying a third weight to the simulated network data to generate the weighted simulated network data, The sum of the first weight, the second weight and the third weight is equal to 1.

18. The non-transitory computer-readable medium of claim 17, wherein a value of the first weight determines an amount of the real network data that is masked.

19. The non-transitory computer-readable medium of claim 15, wherein the one or more instructions causing the device to generate the simulated network data using the random process sample and the random process causes the device to: generating two Poisson distributions using the random process sample and the real network data; and The two Poisson distributions are superimposed to generate the simulated network data.

20. The non-transitory computer-readable medium of claim 15, wherein the GAN model comprises a generator component and a discriminator component.

Images