Communication management system and communication management method

Abstract

The aim is to further improve the efficiency of communication resource utilization. [Solution] The communication management system includes a generator 131 that generates pseudo-optimal communication operations statistically similar to true appropriate communication operations, with each communication resource usage for each time slot corresponding to an appropriate communication operation being considered a true appropriate communication operation, and a discriminator 132 that distinguishes between the pseudo-optimal communication operations generated by the generator 131 and true appropriate communication operations. The system also includes a third learning unit 13 configured to learn multiple generative models, each having an objective function set with a probability distribution of appropriate and inappropriate communication operations determined from the ratio estimated by the first learning unit 11, and a second parameter estimated by the second learning unit 12.

Description

[Technical Field]

[0001] The present invention relates to a communication management system and a communication management method. [Background technology]

[0002] Generally, communication traffic tends to be concentrated during certain time periods. For example, communication traffic tends to increase during the daytime (e.g., 11am-1pm) and in the evening (e.g., 5pm-8pm). Traditionally, services have been provided for IoT devices and the like that direct communication to nighttime hours when traffic is low (see Non-Patent Document 1).

[0003] However, nighttime is not necessarily the only time of day with low traffic; there may be available traffic even outside of nighttime hours. Therefore, Patent Document 1 discloses a system that uses a learning model generated by employing reinforcement learning such as Deep Q-Network (DQN) to create control information indicating whether or not a communication terminal should perform communication for each time slot.

[0004] However, when using DQN to train the system to determine whether or not to perform communication based on rewards, designing rewards for actions is extremely difficult. Even when employing GAIL (Generative Adversarial Imitation Learning), which uses inverse reinforcement learning for rewards, learning may not converge, and there were cases where the desired reward was not obtained. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Patent No. 7478300 [Non-patent literature]

[0006] [Non-Patent Document 1] "M2M Access", [online], Internet Initiative Japan Inc., [searched December 7, 2025], Internet<https: / / mineo.jp / business / m2m-iot / m2m / > [Overview of the Initiative] [Problems that the invention aims to solve]

[0007] With conventional technology, it was difficult to improve the efficiency of communication resource utilization.

[0008] This invention was made to solve the above-mentioned problems and aims to further improve the efficiency of communication resource utilization. [Means for solving the problem]

[0009] To solve the above-mentioned problems, the communication management system according to the present invention includes a first learning unit configured to learn a first parameter representing the ratio between a probability distribution of appropriate communication operations, which shows a series of communication resource usage for each time slot among a plurality of time slots corresponding to the communication operations of the communication terminals whose communication resource usage requests correspond to the communication operations of the communication terminals whose communication resource usage requests are lower than a second standard (lower than the first standard), and to estimate the ratio based on the learned first parameter; and each of the series of transmitted data amounts for each time slot is treated as an observed value, and each observed value is treated as a conditionally independent discrete value, and based on the frequency of occurrence of each observed value, the system determines the class of appropriate communication operations and the class of inappropriate communication operations respectively. The system comprises: a second learning unit configured to estimate a second parameter of a corresponding class-specific probability distribution model; a generator that generates pseudo-appropriate communication operations statistically similar to the true appropriate communication operations, with each of the communication resource usage amounts for each time slot corresponding to the appropriate communication operations being treated as a true appropriate communication operation; a discriminator that distinguishes between the pseudo-appropriate communication operations generated by the generator and the true appropriate communication operations; a third learning unit configured to learn a plurality of generative models, each having an objective function set with the probability distribution of the appropriate communication operations and the probability distribution of the inappropriate communication operations determined from the ratio estimated by the first learning unit, and the second parameter estimated by the second learning unit; and a generation unit configured to generate the pseudo-appropriate communication operations using each of the plurality of trained generators obtained through learning by the third learning unit.

[0010] Furthermore, the communication management system according to the present invention may also include a fourth learning unit configured to learn a communication operation prediction model that takes the pseudo-appropriate communication operation generated by each of the trained generators as training data, takes information indicating one of the multiple time slots as input, and outputs the amount of communication resources used in that time slot.

[0011] Furthermore, the communication management device according to the present invention may further include: an acquisition unit configured to acquire information on time slots assigned to the communication terminal; a calculation unit configured to provide the time slot information assigned to the communication terminal acquired by the acquisition unit as an unknown input to a trained communication operation prediction model obtained by the fourth learning unit, perform calculations on the trained communication operation prediction model, and output the amount of communication resources used in the time slot assigned to the communication terminal; and an instruction unit configured to send an instruction to perform communication according to the amount of communication resources used in the time slot assigned to the communication terminal output by the calculation unit.

[0012] Furthermore, in the communication management system according to the present invention, the communication management system comprises a communication management device and a communication terminal that is communicatively connected to the communication management device, the communication management device comprises a first learning unit, a second learning unit, a generation unit, and a fourth learning unit, and the communication terminal may comprise an acquisition unit, a calculation unit, and an instruction unit.

[0013] To solve the above-mentioned problems, the communication management method according to the present invention includes a first learning step of learning a first parameter that represents the ratio between a probability distribution of appropriate communication operations, which shows a series of communication resource usage for each time slot among a plurality of time slots, corresponding to the communication operations of the communication terminals whose communication resource usage requests correspond to the communication operations of the communication terminals which are lower than a second standard, and a first learning step of estimating the ratio based on the learned first parameter, and treating each of the series of transmitted data amounts for each time slot as an observed value, treating each observed value as a conditionally independent discrete value, and determining the class of appropriate communication operations and the class of inappropriate communication operations based on the frequency of occurrence of each observed value. The system comprises: a second learning step of estimating a second parameter of a corresponding class-specific probability distribution model; a generator that generates pseudo-appropriate communication operations statistically similar to the true appropriate communication operations, with each of the communication resource usage amounts for each time slot corresponding to the appropriate communication operations being considered a true appropriate communication operation; a discriminator that distinguishes between the pseudo-appropriate communication operations generated by the generator and the true appropriate communication operations; a third learning step of learning a plurality of generative models, each having an objective function set with the probability distribution of the appropriate communication operations and the probability distribution of the inappropriate communication operations determined from the ratio estimated in the first learning step, and the second parameter estimated in the second learning step; and a generation step of generating the pseudo-appropriate communication operations using each of the plurality of trained generators obtained through learning in the third learning step.

[0014] Furthermore, the communication management method according to the present invention may further include a fourth learning step in which a communication operation prediction model is trained, which takes the pseudo-appropriate communication operation generated by each of the trained generators as training data, takes information indicating one of the multiple time slots as input, and outputs the amount of communication resources used in that time slot.

[0015] Furthermore, the communication management method according to the present invention may further include: an acquisition step of acquiring information on a time slot assigned to the communication terminal; a calculation step of providing the time slot information assigned to the communication terminal acquired in the acquisition step as an unknown input to a trained communication operation prediction model obtained in the fourth learning step, performing calculations on the trained communication operation prediction model, and outputting the amount of communication resources used in the time slot assigned to the communication terminal; and an instruction step of sending an instruction to perform communication according to the amount of communication resources used in the time slot assigned to the communication terminal output in the calculation step. [Effects of the Invention]

[0016] According to the present invention, the system includes a generator that generates pseudo-optimal communication operations statistically similar to true optimal communication operations, with each communication resource usage amount for each time slot corresponding to an optimal communication operation being considered a true optimal communication operation, and a discriminator that distinguishes between the pseudo-optimal communication operations generated by the generator and true optimal communication operations. The system performs learning of multiple generative models, each having an objective function set with a probability distribution of optimal communication operations and a probability distribution of non-optimal communication operations determined from a ratio estimated by a first learning unit, and a second parameter estimated by a second learning unit. Therefore, the efficiency of communication resource utilization can be further improved. [Brief explanation of the drawing]

[0017] [Figure 1] Figure 1 is a block diagram showing the configuration of a communication management system according to an embodiment of the present invention. [Figure 2] Figure 2 is a diagram illustrating the overview of the communication management system according to this embodiment. [Figure 3] Figure 3 is a diagram illustrating the third learning unit included in the communication management device according to this embodiment. [Figure 4] Figure 4 is a diagram illustrating the third learning unit included in the communication management device according to this embodiment. [Figure 5]Figure 5 is a diagram illustrating the third learning unit included in the communication management device according to this embodiment. [Figure 6] Figure 6 is a diagram illustrating the fourth learning unit included in the communication management device according to this embodiment. [Figure 7] Figure 7 is a block diagram showing an example of the hardware configuration of the communication management device according to this embodiment. [Figure 8] Figure 8 is a sequence diagram showing the operation of the communication management system according to this embodiment. [Figure 9] Figure 9 is a flowchart showing the operation of the communication management device according to this embodiment. [Figure 10] Figure 10 is a flowchart showing the operation of the communication management device according to this embodiment. [Figure 11] Figure 11 is a flowchart showing the operation of the communication management device according to this embodiment. [Modes for carrying out the invention]

[0018] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to Figures 1 to 11.

[0019] [Configuration of the communication management system] First, with reference to Figure 1, an overview of a communication management system comprising a communication management device 1, a communication terminal 2, a gateway device 3, and a data network 4 according to an embodiment of the present invention will be described. The communication management system manages communication by the communication terminal 2 by generating communication operations that represent the amount of communication resources used in each of multiple time slots, reflecting the communication operations of the communication terminal 2, which has high throughput characteristics.

[0020] The communication terminal 2 is configured as an information processing device including a processor, a memory, a communication interface, etc., and can realize a communication function by a program operating on the information processing device. The communication terminal 2 includes a communication circuit for realizing the communication function. The communication circuit includes a communication control circuit (such as a MAC layer control circuit) for controlling the transmission timing and the transmission amount, and a wireless transmission circuit (RF circuit) for modulating and transmitting the transmission data as a wireless signal.

[0021] The communication terminal 2 can be realized as a mobile communication terminal such as a smartphone, a tablet computer, a laptop computer, a wearable device, an industrial robot, etc., and can be configured as an IoT device to which a terminal IP address for uniquely identifying the terminal is assigned. In this embodiment, there is one or more communication terminals 2. The communication terminal 2 can also include a SIM (Subscriber Identity Module).

[0022] The communication terminal 2 can communicate with other devices (not shown) connected to another data network 4 (such as the Internet) via the gateway device 3. The communication terminal 2 has a communication function of transmitting communication data in a time slot permitted to be used among a plurality of time slots assigned by the communication network NW. Here, the time slot is a time unit obtained by dividing the communication period into M, and the time slots are assigned identification numbers of time slots 1 to time slots M. Each time slot typically has the same time width, but may have an unequal time width. The exemplary time width of the time slot may be an arbitrary appropriate length, such as several seconds, dozens of seconds, several minutes, dozens of minutes, etc. The number of time slots used by each communication terminal 2 is common. For example, N = 100 (N < M) time slots out of M = 1000 total time slots are used. The functional blocks of the communication terminal 2 will be described later.

[0023] The gateway device 3 receives communication data transmitted from the communication terminal 2 and has a relay function that relays the data based on the transfer path within the communication network NW. Furthermore, when the gateway device 3 receives communication data transmitted by the communication terminal 2, it observes the amount of transmitted data in each time slot (time slot 1 to time slot M) during the communication period and records the amount of transmitted data. The history of transmitted data recorded by the gateway device 3 represents the amount of transmitted data for each time slot of the communication terminal 2, which refers to the amount of data actually transmitted by the communication terminal 2 in each time slot. In addition, the gateway device 3 transmits the recorded history of transmitted data for each time slot to the communication management device 1.

[0024] The data network 4 is an external network connected to the communication terminal 2 via the gateway device 3, and includes, for example, the internet, an internal corporate network, or a wide-area network such as a cloud service network. Various communication devices, such as server devices, storage devices, and application provisioning devices (not shown), which are the targets of communication by the communication terminal 2, are connected to the data network 4.

[0025] Figures 2(a) and 2(b) show an example of the communication operation of communication terminal 2, and indicate the observed value of communication resource usage in each time slot when the communication period is divided into M time slots. Communication resource usage includes, for example, the amount of data transmitted, the number of allocated RBs, and the amount of power transmitted in each time slot, but in the following explanation, we will use the case where communication resource usage is the amount of data transmitted for each time slot as an example.

[0026] In Figures 2(a) and (b), the horizontal axis represents time slots [ms], and each value indicates a time slot identification number. The vertical axis represents the amount of data transmitted by one communication terminal 2 in each time slot [Kbit]. Figure 2(a) shows an example corresponding to proper communication operation, and (b) shows an example corresponding to improper communication operation.

[0027] In the example shown in Figure 2(a), a relatively uniform amount of transmitted data is observed across multiple time slots that are permitted to be used out of the M time slots, indicating that the communication terminal 2 is performing high-throughput communication that continuously meets a predetermined level of communication resource usage (for example, a data rate request of 50 kbit or more per time slot, or 10 Mbps or more). Such communication operation is defined as proper communication operation as it satisfies the communication resource usage requirement of the first criterion or higher.

[0028] On the other hand, in the example shown in Figure 2(b), communication operations are concentrated in a limited number of time slots, and the amount of data transmitted in each time slot is also small (for example, a few kilobits per time slot, or a data rate request of less than 1 Mbps). Such communication operations fall under the second criterion, which is a lower criterion than the first criterion, and are defined as inappropriate communication operations as low-throughput communication.

[0029] The specific values ​​for the first and second criteria can be set as appropriate depending on network conditions and communication methods. As mentioned above, thresholds may be set so that high-throughput communication operation is classified when the amount of transmitted data in a time slot is above a certain amount, and low-throughput communication operation is classified when it is below a certain amount. In this case, the first criterion may be set when the amount of transmitted data per time slot is 50 kbit or more, and the second criterion may be set when it is less than 5 kbit. Alternatively, the first criterion may be set when the communication rate is 10 Mbps or more, and the second criterion may be set when it is less than 1 Mbps. Furthermore, when based on communication resource utilization, an example is to set the first criterion when 20% or more of all resources are utilized, and the second criterion when less than 5% is utilized. The first and second criteria are thresholds for distinguishing the communication operation of communication terminal 2, and can be set not only based on the amount of transmitted data, but also on other indicators such as communication resource utilization, communication quality indicators, or service request bandwidth.

[0030] [Functional blocks of the communication management device] Next, the functional blocks of the communication management device 1 according to this embodiment will be described with reference to the block diagram in Figure 1. As shown in Figure 1, the communication management device 1 includes a first acquisition unit 10, a first learning unit 11, a second learning unit 12, a third learning unit 13, a generation unit 14, a fourth learning unit 15, a transmission unit 16, and a first storage unit 17.

[0031] The first acquisition unit 10 acquires the past amount of data transmitted by the communication terminal 2 for each of the multiple time slots from the gateway device 3. More specifically, the first acquisition unit 10 acquires a series of data transmission amounts by the communication terminal 2 for each time slot that corresponds to a communication operation where the communication resource usage request is equal to or greater than the first standard, from the history of past data transmission amounts by the communication terminal 2 for each time slot, as appropriate communication operations. The first acquisition unit 10 also acquires a series of data transmission amounts by the communication terminal 2 for each time slot that corresponds to a communication operation where the communication resource usage request is lower than the second standard, which is lower than the first standard, from the history of past data transmission amounts by the communication terminal 2 for each time slot, as inappropriate communication operations. The first acquisition unit 10 may also acquire pre-prepared data of inappropriate communication operations from an external source.

[0032] The first learning unit 11 learns a parameter (first parameter) that represents the ratio between the probability distribution of appropriate communication operations, which shows the sequence of data transmitted by the communication terminal 2 for each time slot among multiple time slots, corresponding to communication operations where the communication resource usage requirement is equal to or greater than the first criterion, and the probability distribution of inappropriate communication operations, which shows the sequence of data transmitted by the communication terminal 2 for each time slot, corresponding to communication operations where the communication resource usage requirement is lower than the second criterion, and estimates the ratio based on the learned parameter.

[0033] The first learning unit 11 learns a parameter representing the density ratio (ratio) of the probability density functions, which are the probability distributions of proper communication operations by the high-throughput characteristic communication terminal 2 satisfying the first criterion and improper communication operations by the low-throughput characteristic communication terminal 2 not satisfying the second criterion, using data of proper communication operations and improper communication operations. Also, the first learning unit 11 estimates the density ratio from the learned parameter.

[0034] Here, let the set of training data related to proper communication operations be D = {x (1) , x (2) , …, x (N)}, and the set of training data related to improper communication operations be D’ = {x’ (1) , x’ (2) , …, x’ (N’)}. x indicates the amount of transmitted data in each time slot. Each observed data x (n) is M-dimensional, and x (n) = (x1 (n) , x2 (n) , …, x M (n) ). When there are 1000 (= M) time slots in the communication period, it is 1000-dimensional. Each component x i (n) indicates the amount of transmitted data by the communication terminal 2 in a certain time slot i.

[0035] Hereinafter, let D be proper communication operations and D’ be improper communication operations, and let the probability density function of proper communication operations D be p(x) and the probability density function of improper communication operations D’ be p’(x). The probability density function p(x) of proper communication operations D indicates the distribution of the probability that the amount of transmitted data x in each time slot corresponding to proper communication operations is observed. Also, the probability density function of improper communication operations D’ indicates the distribution of the probability that the amount of transmitted data x in each time slot constituting improper communication operations is observed. The density ratio r(x) between the probability density function p(x) of proper communication operations D and the probability density function p’(x) of improper communication operations D’ is represented by the following formula (1).

Equation

[0036]

number

[0037] The basis function ψ(x) is defined using the RBF (Radial Basis Function) kernel and expressed by the following equation (3).

number

[0038] Here, based on equation (1) above, the specific form of equation (3) where the number of basis vectors b is the number of training data N (b=N) is given by the following equation (4).

number

[0039] Equation (4) above expresses the density ratio as a linear sum of RBFs centered on all learning points. Here, we introduce the generalized Kullback-Leibler divergence, which measures the information-theoretic distance between the non-negative functions f and g shown in equation (5).

number

[0040] In density ratio estimation, f=p(x) and g=r θ Substitute p'(x) into equation (5) above, and the objective function is given by equation (6).

number

[0041] In equation (6) above, each x n , x' n’ The optimization objective function obtained by approximating the integral with an empirical distribution that sets all values ​​except θ to 0, ignoring terms that do not depend on the parameter θ, and removing constants is expressed by the following equation (7).

number

[0042] By minimizing J(θ) in equation (7) above, the density ratio r θ This is estimated. Since J(θ) is a convex function, the first learning unit 11 updates the parameter θ from the initial value to convergence using the parameter θ update formula by gradient descent shown in equation (8) below.

number

[0043] The result of specifically calculating the gradient in equation (7) above is expressed by the following equation (9).

number

[0044] In equation (9) above, the first term represents the contribution from improper communication operations, and the second term represents the contribution from proper communication operations. Since the second term is dominant in equation (9), stable estimation is possible when the number of proper communication operation data N is large. In other words, even when the number of improper communication operation data N' is small, it is possible to stably determine the parameter θ.

[0045] Furthermore, before using equations (7) to (9) above to find the optimal solution for parameter θ, the first learning unit 11 finds an appropriate value for the bandwidth h in equation (4) above using cross-validation or an information criterion (KL divergence minimization criterion). Based on the optimal solution for parameter θ obtained by the KL density ratio estimation method, the first learning unit 11 uses equation (2) above to determine the density ratio r for any input x. θ An estimated value can be obtained. The density ratio estimated by the first learning unit 11 is passed to the third learning unit 13, which will be described later.

[0046] The second learning unit 12 takes the amount of data transmitted by the communication terminal 2 for each time slot in multiple time slots as an observed value, treats each observed value as a conditionally independent discrete value, and estimates the parameters (second parameters) of the class-specific probability distribution model corresponding to the class of proper communication operation and the class of improper communication operation, respectively, based on the frequency of occurrence of each observed value. The second learning unit 12 sets up a class-specific probability distribution model using a multinomial distribution for the amount of data transmitted in each time slot related to proper communication operation and the amount of data transmitted in each time slot related to improper communication operation, estimates its parameters, and calculates the log-likelihood of the series of observed values ​​based on the class-specific probability distribution model. The second learning unit 12 first rounds the amount of data transmitted by the communication terminal 2 for each time slot in multiple time slots to integer discrete values, and then constructs the class-specific probability distribution model and estimates its parameters.

[0047] The second learning unit 12 focuses on the fact that the simple Bayes method treats each observation as a conditionally independent discrete variable, and extends the simple Bayes method to a multinomial distribution model to set up a class-based probability distribution model. Here, as shown in equation (10) below, the observed value X is a vector representing the amount of transmitted data in each time slot as the number of occurrences of M categories. X = x = (x1, x2, ..., x M ) ···(10)

[0048] Furthermore, the class label to which the observed value X belongs, i.e., the event Y, is defined as a binary variable by the following equation (11).

number

[0049] If the conditional probabilities P(X|Y) for the occurrence of observed value X under the given conditions of event Y are independent, then the observed variable x under the given conditions of event Y is... i The variables are conditionally independent of each other, and the likelihood ρ(x|y) is given by the simple Bayesian estimation method in equation (12).

[0050]

number

[0051] In this embodiment, since we are dealing with a sequence of transmitted data amounts for each time slot of integer values, we are dealing with a vector of occurrence counts for each category x=(x1,x2,···,x M We consider ) as one sample and extend it to a multinomial distribution probability model while maintaining the naive assumption that the amount of transmitted data in each time slot is conditionally independent. The occurrence probabilities θ1,···,θ of each marginal probability P(X) in which the observed value X is observed are given. M Under the constraints that the variables are independent and the sum of their probabilities of occurrence is 1, the probability model of the multinomial distribution can be expressed by the following equation (13).

[0052]

number

[0053] As shown in equation (13) above, the observed value x is the count value of the amount of data transmitted in each time slot. i The sum of (x1 + x2 + ... + x M Once the value of x is determined, the distribution becomes the product of the following equation (14), and the observed value of the amount of transmitted data in the i-th time slot in the series is x. i Each with an independent probability θ i ^x i It is possible to find this.

number

[0054] Therefore, it can be seen that equation (14) above has a similar relationship to that of the simple Bayes method. Here, the prior probability P(Y) is a binary problem between the class of proper communication operation (Y=0) and the class of improper communication operation (Y=1), and the unknown parameters of each are θ. 0 ,θ 1 Let's define D as the marginal probability (result), and the observed value for an inappropriate value (Y=0) be D. 0 The observed value for the appropriate state (Y=1) is D 1 Let's assume that equation (14) above can be expressed as equation (15) by transforming the product form into a sum under the Naive Bayes independence assumption and decomposing the log-likelihood for the data set, the set of observed values ​​D, into class-specific forms.

[0055]

number

[0056] Here, the constraint is expressed by equation (16).

number

[0057] Furthermore, applying the Lagrangian multiplier method, the parameter θ related to the class of improper communication behavior is obtained. 0 The maximum value of the log-likelihood for the i-th component is given by equation (17).

number

[0058] Parameter θ related to the class of proper communication operation 1 Similarly, when we find the maximum value of the log-likelihood for this as well, under the constraint of equation (16) above, the parameter θ corresponds to the class of proper communication operation and the class of improper communication operation, respectively. 1 θ 0 This can be expressed by the following equation (18).

number

[0059] Note that x i Since x takes an integer value, i To prevent the multinomial distribution from diverging when ^(n) is 0, smoothing can be performed by specifying +α (e.g., α=1) smoothing. In this way, the second learning unit 12 considers the sequence of transmitted data amounts in each of the multiple time slots as a multinomial distribution, estimates the parameters from the count of occurrences, and measures the degree of appropriateness and inappropriateness using likelihood. The parameters θ estimated by the second learning unit 12 1 , θ 0 It will be handed over to the 3rd Learning Department, 13.

[0060] The third learning unit 13 includes a generator 131 that generates pseudo-proper communication operations statistically similar to true proper communication operations, with each transmission data amount for each time slot corresponding to proper communication operation being considered a true proper communication operation, and a discriminator 132 that distinguishes between the pseudo-proper communication operations generated by the generator 131 and true proper communication operations. It learns multiple generative models, each having an objective function set with the probability distribution of proper communication operations and the probability distribution of non-proper communication operations determined from the density ratio estimated by the first learning unit 11, and the parameters (second parameters) estimated by the second learning unit 12.

[0061] As shown in Figure 3, the third learning unit 13 performs adversarial learning on a GAN (Generative Adversarial Network) having a generator 131 and a discriminator 132. As shown in Figure 3, the third learning unit 13 performs adversarial learning on each of the multiple GANs provided for each observation value i (i=1,...,M), i.e., for each time slot i. Through the learning of the third learning unit 13, a total of M trained generators 131' are constructed, each corresponding to the amount of transmitted data 1 to M for each time slot.

[0062] Figures 4 and 5 schematically represent the neural network configuration of the GAN generator 131 and discriminator 132 used by the first learning unit 11. As shown in Figure 4, the generator 131 consists of a neural network having an input layer, a hidden layer, and an output layer. The generator 131 is a model that generates pseudo-appropriate communication behavior from random noise. For example, m randomly sampled Gaussian noise vectors are input to the input nodes of the generator 131 (z1~z m ).

[0063] The generator 131 outputs an output G(z) after performing a sum-of-products operation on the input and weight parameters, and thresholding using an activation function. The output G(z) from the generator 131 is a pseudo-optimal communication operation that is statistically similar to the amount of data transmitted in the i-th time slot corresponding to proper communication operation. A CNN or ResNet can be used as the neural network that constitutes the generator 131.

[0064] The classifier 132 shown in Figure 5 consists of a neural network having an input layer, a hidden layer, and an output layer. In the example in Figure 5, the amount of data x transmitted in the i-th time slot corresponding to proper communication operation, acquired by the first acquisition unit 10, is given as the input for training data.

[0065] The classifier 132 outputs a probability value in the range of 1 to 0 after performing a sum-of-products operation on the input and weight parameters and thresholding using an activation function. When the classifier 132 correctly identifies the input training data relating to true proper communication behavior as true proper communication behavior, it outputs a probability value close to y=1. On the other hand, when it correctly identifies the input training data relating to pseudo-proper communication behavior as pseudo-proper communication behavior, it outputs a probability value close to y=0. In this way, the classifier 132 is a model that distinguishes the model distribution generated by the generator 131 from the data distribution of the training data, which is the true distribution. A CNN can be used as the neural network that constitutes the classifier 132.

[0066] Figure 3 is a block diagram illustrating the adversarial learning of the GAN by the third learning unit 13. The GAN generator 131 adopted by the first learning unit 11 is denoted as function G, and the discriminator 132 as function D. Furthermore, the true proper communication behavior is denoted as x, the predicted value output by the discriminator 132 is denoted as y, and the correct label is denoted as t. The correct label t is set to 1 for the true proper communication behavior and 0 for the pseudo-proper communication behavior generated by the generator 131. In this case, the discriminator 132 is a binary classification problem with the cross-entropy E given by equation (19) below. CE It can be expressed as follows.

[0067]

number

[0068] The first term inside the brace in equation (19) above represents t n lny n In this case, the predicted value y of the classifier 132 n However, the correct label for true proper communication operation t n It is desirable to approach the value of =1. On the other hand, the second term inside the brace represents (1-t n )ln(1-y n In this case, the predicted value y of the classifier 132 n However, the value of the correct label (1-t) that identifies the pseudo-proper communication operation is... n It is desirable for cross-entropy E to approach 0. CE This value is maximized when the predicted value matches the correct label value.

[0069] Here, the generator 131 that constitutes the GAN has parameter w G ,θ G It has a function G(w G ,θ G ) is expressed as. Also, the classifier 132 has parameter w D ,θ D It has a function D(w D ,θ D This is expressed as ). The cross-entropy E in equation (19) above CE The objective function E of a GAN comprising a generator 131 and a discriminator 132 based on the above can be expressed by the following equation (20).

number

[0070] The first term of equation (20) above represents E D(x)=1 lnD(w D ,θ D ) is the expected value that the discriminator 132 identifies as a true proper communication operation. The second term of equation (20) above represents E D(x)=0 ln(1-D(G(w G ,θ G ),w D ,θ D )) is the expected value that the classifier 132 identifies the pseudo-proper communication operation generated by the generator 131 as a pseudo-proper communication operation. Here, the expected value of equation (20) above can be expressed as equation (21) using a probability distribution.

[0071]

number

[0072] Here, for the probability density function p(x) for proper communication operation and the probability density function p'(x) for improper communication operation in equation (1) above, we set p(x)≡ρ(x|y=1) and p'(x)≡ρ(x|y=0) in order to make them into a probabilistic labeled classification problem for min-max learning of the GAN. The density ratio r estimated by the first learning unit 11 θ (x) is defined by the following equation (22).

number

[0073] The probability density function ρ(x|y=1), which is the conditional probability distribution of proper communication operation in equation (22) above, can be calculated using maximum likelihood estimation or the like from M-dimensional data on the amount of transmitted data for each time slot of proper communication operation, which can be obtained from the communication history. Using the calculated probability density function ρ(x|y=1) of proper communication operation, the probability density function ρ(x|y=0), which is the conditional probability distribution of improper communication operation, can be expressed by the following equation (23).

number

[0074] Substitute the probability density function ρ(x|y=1) for proper communication operation and the probability density function ρ(x|y=0) for improper communication operation in equation (23) into the objective function E in equation (21) above, and set the prior probability ρ(y=1) for the class of proper communication operation (y=1) and the prior probability ρ(y=0) for the class of improper communication operation (y=0) to a parameter θ corresponding to the amount of data transmitted in the i-th time slot corresponding to the class of proper communication operation and the class of improper communication operation estimated by the second learning unit 12. 1 θ 0 We use the following. Furthermore, the posterior probability ρ(y=1|x) which is unique to the amount of data x transmitted in the i-th time slot is given by D(w) in the optimal solution of equation (22) above. D ,θ D ) corresponds to the posterior probability ρ(y=0|x) of improper communication operation for the amount of data transmitted x in the i-th time slot, which is 1-D(G(w) in the optimal solution of equation (21) above. G ,θ G ),w D ,θ D This corresponds to the density ratio r. θ And it can be determined from the prior probability.

[0075] In GAN training, the generator 131 and the discriminator 132 are trained adversarially by min-max optimization of the objective function E. Therefore, the generator 131 is trained to generate pseudo-appropriate communication behavior that deceives the discriminator 132, and the discriminator 132 is trained to recognize the pseudo-appropriate communication behavior generated by the generator 131 as pseudo-appropriate communication behavior.

[0076] During the training of classifier 132, when true proper communication behavior is given, classifier 132 is trained to maximize the first term of the objective function E in equation (21) above by producing an output close to y=1. On the other hand, when pseudo-proper communication behavior is given, classifier 132 is trained to maximize the second term of the objective function E by producing an output close to y=0.

[0077] In the learning of generator 131, D(G(w) in equation (21) above G ,θ G ),w D ,θ D )(D(G(z)) in Figure 3) is close to 1 G(w G ,θ G The objective function E is minimized by outputting (G(z) in Figure 3). The third learning unit 13 uses a learning procedure that alternately updates the parameters of the generator 131 and the discriminator 132. Details of the learning procedure for the generator 131 and discriminator 132 by the third learning unit 13 will be described later.

[0078] When the objective function E of the GAN is optimized, the third learning unit 13 passes the trained generator 131', which has been constructed for each observation in time slot i, to the generation unit 14. Thus, ultimately M trained generators 131' are constructed.

[0079] The generation unit 14 generates pseudo-appropriate communication operations using each of the learned generators 131' obtained through learning by the third learning unit 13. Therefore, the generation unit 14 generates M sets of pseudo-appropriate communication operations.

[0080] The fourth learning unit 15 uses the pseudo-appropriate communication operation generated by each of the trained generators 131' as training data, takes information indicating one of the multiple time slots as input, and trains a communication operation prediction model that outputs the amount of data transmitted in that time slot.

[0081] Figure 6 is a schematic diagram showing the structure of the neural network used as the path prediction model in this embodiment. The neural network can use a multilayer structure consisting of an input layer X, a hidden layer H, and an output layer Y. The input nodes of the input layer X are given time slot identification numbers.

[0082] The neural network shown in Figure 6 applies an activation function to the weighted sum of the inputs for the time slot identification number given to the input layer X, and passes the output determined by thresholding to the output layer Y. The output node of the output layer Y outputs the amount of data transmitted by the communication terminal 2 in the time slot related to the input.

[0083] The fourth learning unit 15 adjusts parameters such as the weight w and bias θ of the connections between nodes, which are neurons, so that the output when a time slot identification number is given as input becomes the value of the pseudo-optimal communication operation of the correct label, that is, the value of the amount of transmitted data indicated by the pseudo-optimal communication operation. The fourth learning unit 15 compares the obtained output value with the given input value, for example using backpropagation, examines the error of each weight, propagates it in the reverse direction, and determines the optimized parameters. Through this supervised learning process, the fourth learning unit 15 constructs a trained communication behavior prediction model.

[0084] The transmitting unit 16 transmits the trained communication behavior prediction model, constructed by the learning process of the fourth learning unit 15, to the communication terminal 2 via the communication network NW. The transmitting unit 16 can transmit the trained communication behavior prediction model to multiple communication terminals 2.

[0085] The first memory unit 17 stores the parameters and density ratios estimated by the first learning unit 11, as well as the parameters estimated by the second learning unit 12. The first memory unit 17 stores M learned generators 131'. The first memory unit 17 also stores the learned communication behavior prediction model learned by the fourth learning unit 15.

[0086] [Functional Blocks of Communication Terminal] As shown in FIG. 1, the communication terminal 2 includes a second acquisition unit (acquisition unit) 20, a second memory unit 21, an arithmetic unit 22, an instruction unit 23, and a communication control unit 24.

[0087] The second acquisition unit 20 acquires information on the time slots assigned to the communication terminal 2. The second acquisition unit 20 acquires the information on the time slots assigned to the communication terminal 2 based on the scheduling control information transmitted from a base station (or scheduler) not shown in the communication network NW. The time slot information acquired by the second acquisition unit 20 may be sequential assignment time slot information according to the communication situation, in addition to the case where N (N<M) time slots out of M time slots are assigned in a batch for one communication request.

[0088] The second memory unit 21 stores the learned communication operation prediction model transmitted from the communication management device 1.

[0089] The arithmetic unit 22 provides the information on the time slots assigned to the communication terminal 2 acquired by the second acquisition unit 20 as an unknown input to the learned communication behavior prediction model obtained by the fourth learning unit 15, performs the calculation of the learned communication behavior prediction model, and outputs the amount of transmission data by the communication terminal 2 in the time slot. The arithmetic unit 22 outputs the amount of transmission data in each of a plurality of time slots assigned to the communication terminal 2 through a product-sum calculation and a threshold process based on parameters such as learned weights.

[0090] The instruction unit 23 sends an instruction to communicate according to the amount of data to be transmitted in the time slot allocated to the communication terminal 2, which is output by the calculation unit 22. The instruction unit 23 then sends an instruction to the communication control unit 24.

[0091] The communication control unit 24 reads data from the transmission buffer according to the amount of data to be transmitted, in response to instructions from the instruction unit 23, and transmits it from the wireless transmission circuit.

[0092] [Hardware configuration of the communication management device] Next, an example of a hardware configuration for realizing the communication management device 1 having the functions described above will be explained using Figure 7.

[0093] As shown in Figure 7, the communication management device 1 can be implemented, for example, by a computer equipped with a processor 102, main memory 103, communication interface 104, auxiliary storage 105, and input / output I / O 106 connected via a bus 101, and a program that controls these hardware resources. Furthermore, the communication management device 1 includes a display device 107.

[0094] The processor 102 is a circuit or device that performs arithmetic processing, and can be implemented by, for example, a general-purpose central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), etc. Alternatively, some or all of these may be combined to form the processor.

[0095] The main memory 103 is composed of, for example, volatile random access memory (RAM), and pre-stores programs for the processor 102 to perform various controls and calculations. The processor 102 and the main memory 103 work together to realize the various functions of the communication management device 1, such as the first acquisition unit 10, the first learning unit 11, the second learning unit 12, the third learning unit 13, the generation unit 14, and the fourth learning unit 15 shown in Figure 1.

[0096] The communication interface 104 is an interface circuit for networking the communication management device 1 with various external electronic devices.

[0097] The auxiliary storage device 105 consists of a read / write storage medium and a drive device for reading and writing various information such as programs and data to the storage medium. The auxiliary storage device 105 can use non-volatile storage such as a hard disk or flash memory as the storage medium.

[0098] The auxiliary storage device 105 has a program storage area for storing the communication management program. The auxiliary storage device 105 also has a program storage area for storing parameters representing the density ratio of the probability density functions of proper and improper communication operations executed by the communication management device 1, and a first learning program for estimating the density ratio. The auxiliary storage device 105 also has a program storage area for storing a second learning program executed by the second learning unit 12 for estimating parameters using polynomial Bayes. The auxiliary storage device 105 also has a program storage area for storing a third learning program executed by the third learning unit 13 for adversarial learning of a generative model. The auxiliary storage device 105 also has a program storage area for storing a fourth learning program executed by the fourth learning unit 15 for supervised learning of a communication operation prediction model.

[0099] The auxiliary storage device 105 enables the first storage unit 17 described in Figure 1. Furthermore, it may also have, for example, a backup area for backing up the aforementioned data and programs.

[0100] The input / output I / O 106 is an input / output device that inputs signals from external devices or outputs signals to external devices.

[0101] The display device 107 is composed of an organic EL display, a liquid crystal display, or the like. The display device 107 can display time-series data of the transmission data amount on the screen.

[0102] [Operation of the Communication Management System] Next, the operation of the communication management system having the above-described configuration will be described with reference to the sequence diagram of FIG. 8 and the flowcharts of FIGS. 9 to 11.

[0103] As shown in FIG. 8, first, the gateway device 3 transmits, to the communication management device 1, the history of the transmission data amount of each communication terminal 2 in each of the M time slots by the plurality of communication terminals 2 (step S1). Next, the first acquisition unit 10 of the communication management device 1 obtains, from the history of the transmission data amount in each time slot received from the gateway device 3 in step S1, an appropriate communication operation indicating a series of transmission data amounts in each time slot corresponding to the communication operation of the communication terminal 2 whose communication resource usage requirement is greater than or equal to the first criterion, and a non-appropriate communication operation indicating a series of transmission data amounts for each time slot corresponding to the communication operation less than the second criterion lower than the first criterion (step S2).

[0104] Next, the first learning unit 11 performs the first learning process (step S2). FIG. 9 is a flowchart for explaining the first learning process in step S2. As shown in FIG. 9, the first learning unit 11 learns a parameter θ representing the density ratio r θ between the probability density function of the appropriate communication operation and the probability density function of the non-appropriate communication operation (step S20). The first learning unit 11 updates the parameter θ by using the above equations (7) to (9) and the gradient descent method or the like, and obtains the optimal solution of the parameter θ.

[0105] Next, the first learning unit 11 estimates the density ratio r from the learned parameters θ obtained in step S20 based on the above equation (2). θ (Step S21).

[0106] Returning to FIG. 8, the second learning unit 12 executes the second learning process (step S3). FIG. 10 is a flowchart for explaining the second learning process of step S3. As shown in FIG. 10, the second learning unit 12 sets parameters θ 1 , θ 0 for the amount of transmission data in each time slot related to the class of proper communication operations and the class of improper communication operations (step S31). Subsequently, the second learning unit 12 defines the log-likelihood according to the above equation (13) (step S32). In step S32, the second learning unit 12 changes the objective function for optimization from the product form (Equation (14)) to the sum form. Next, the second learning unit 12 uses the Lagrange multiplier method of the above equation (17) to estimate the parameters θ 1 corresponding to the class of proper communication operations that maximizes the log-likelihood (Equation (15)) under the constraint (Equation (16)), and the parameters θ 0 corresponding to the class of improper communication operations (Equation (18)) (step S33). The estimated parameters θ 1 , θ 0 are stored in the first storage unit 17, and the process proceeds to step S6 in FIG. 8.

[0107] Next, the third learning unit 13 executes the third learning process (step S6). In step S6, the third learning unit 13 uses a generator 131 that generates a pseudo-proper communication operation that is statistically similar to the true proper communication operation, with the amount of transmission data in each time slot corresponding to the proper communication operation as the true proper communication operation, and a discriminator 132 that discriminates between the pseudo-proper communication operation generated by the generator 131 and the true proper communication operation, and learns a plurality of generation models each having an objective function set with the probability density functions of the proper communication operation and the improper communication operation determined from the density ratio estimated by the first learning unit 11 and the parameters estimated by the second learning unit 12.

[0108] Figure 11 is a flowchart illustrating the third learning process in step S6. As shown in Figure 11, first, the third learning unit processes the density ratio r estimated in the first learning process in step S2 of Figure 8. θ The probability density function ρ(x|y=1) for proper communication operation and the probability density function ρ(x|y=0) for improper communication operation (equation (21) above), determined from the above, are set as the objective function E of the GAN (equation (21)) (step S50). More specifically, the third learning unit 13 calculates the probability density function ρ(x|y=1), which is the conditional probability distribution of proper communication operation, from the data of proper communication operation acquired by the first acquisition unit 10, using the maximum likelihood estimation method or the like. The third learning unit 13 also determines the probability density function ρ(x|y=0), which is the conditional probability distribution of improper communication operation expressed in equation (23) above, from the calculated probability density function ρ(x|y=1) for proper communication operation.

[0109] Next, the third learning unit 13 calculates the parameters θ corresponding to the classes of proper and improper communication operations, respectively, which were estimated in the second learning process of step S3 in Figure 8. 1 ,θ 0 The prior probability ρ(y=1) for a favorable outcome and the prior probability ρ(y=0) for an unfavorable outcome are set as the objective function E of the GAN (Equation (22)) (Step S51). Furthermore, in the optimal solution of the objective function E in Equation (21) above, D(w D ,θ D ) is the posterior probability ρ(y=1|x) that is appropriate for the amount of transmitted data x observed in the time slot, and 1-D(G(w G ,θ G ),w D ,θ D ) is the posterior probability ρ(y=0|x) that is inappropriate for the amount of transmitted data x observed in the time slot. These posterior probabilities ρ(y=1|x) and ρ(y=0|x) are given by the estimated density ratio r θ It can also be calculated from the prior probabilities ρ(y=1) and ρ(y=0).

[0110] Next, the third learning unit 13 acquires each of the transmission data amounts for each time slot corresponding to the proper communication operation acquired in step S2 of Figure 8 as the true proper communication operation (step S52). In step S52, the transmission data amount for the time slot with i=1, which corresponds to one of the M transmission data amounts for each time slot corresponding to the proper communication operation (i=1), is defined as the true proper communication operation.

[0111] Next, the third learning unit 13 inputs the true proper communication operation, which is the amount of transmitted data in the time slot i=1 related to the proper communication operation acquired in step S52, as training data 134 to the classifier 132, and sets the parameter w of the classifier 132 so that it can distinguish the true proper communication operation from the true proper communication operation (y=1). D ,θ D Learn and update (step S53). As shown in the block diagram of the third learning unit 13 in Figure 3, the true proper communication operation is used as training data 134 input when learning the discriminator 132.

[0112] In step S53, the third learning unit 13 can, for example, use the backpropagation method to teach the classifier 132 true proper communication operation. Step S53 pre-constructs a classifier 132 that can distinguish true proper communication operation from true proper communication operation.

[0113] Next, the third learning unit 13 generates Gaussian noise and provides a random vector of the generated Gaussian noise as input to the generator 131 (step S54). Subsequently, the generator 131 uses the input z and weight parameter w based on the given Gaussian noise. G ,θ G A sum-of-products operation and threshold processing using an activation function are performed to generate a pseudo-optimal communication operation G(z) (step S55).

[0114] Next, the third learning unit 13 trains the classifier 132. The training of the classifier 132 involves the parameters w of the generator 131. D ,θ DThis is done with fixed parameters. First, the third learning unit 13 provides the true proper communication operation acquired in step S52 as training data 134 to the classifier 132 as input. Then, the third learning unit 13 uses backpropagation or the like to maximize the objective function E in equation (22) above, and the parameters w D ,θ D Update (step S56). Note that the label for training data 134 is set to 1 (true proper communication operation).

[0115] Next, the third learning unit 13 provides the pseudo-appropriate communication operation generated by the generator 131 in step S55 as input to the discriminator 132, and parameters w are set using backpropagation or the like so that the objective function E in equation (21) above is maximized. D ,θ D Update (step S57). That is, in steps S56 and S57, in order to maximize the objective function E in equation (21) above, the first term is D(w D ,θ D )=1 is output, and the second term is D(G(w G ,θ G ),w D ,θ D The optimization is performed so that ) = 0. Note that the training data 134 has the label 0 (pseudo-proper communication operation) set.

[0116] The learning of the classifier 132 in steps S56 and S57 corresponds to the dashed arrows in the block diagram of the third learning unit 13 shown in Figure 3, which indicate that the classifier error is calculated in block 135 of the objective function E based on the output 133 from the classifier 132, and then the error is backpropagated to the classifier 132.

[0117] Next, the third learning unit 13 trains the generator 131. During the training of the generator 131, the parameters of the discriminator 132 are fixed. The third learning unit 13 trains the generator 131 so that when random Gaussian noise is applied to the generator 131, pseudo-optimal communication operation is generated. Specifically, the first learning unit 11 minimizes the objective function E in equation (21) above by using methods such as backpropagation to control the parameters w G ,θG Update (Step S58).

[0118] The learning in step S58 corresponds to the dashed arrow flow in the block diagram of the third learning unit 13 in Figure 3, which indicates that the error is backpropagated to the generator 131. In other words, step S58 corresponds to the dashed arrow flow in which the pseudo-appropriate communication operation generated by the generator 131 in Figure 4 is input to the discriminator 132, the generator error is calculated from its output 133 in block 135 of the objective function E, and then the error is backpropagated to the generator 131.

[0119] Subsequently, the learning of the discriminator 132 and generator 131 from steps S55 to S58 is repeated until the value of the objective function E reaches a Nash equilibrium and converges (step S59: NO). On the other hand, if the value of the objective function E converges (step S59: YES), the processing from steps S53 to S59 is repeated until the learning of the generator 131 and discriminator 132 is performed using all true proper communication operations (step S60: NO). Note that the processing from steps S52 to S58 and from steps S59 to S60 can be performed in batches.

[0120] Subsequently, if the generator 131 and discriminator 132 have been trained using all true proper communication operations (step S60: YES), the third learning unit 13 stores the trained generator 131' corresponding to the time slot i=1 in the first storage unit 17 (step S61). Further, as shown in connector a, adversarial learning is performed on each of the generative models constructed for each of the time slots corresponding to the observed values ​​i=2, ..., M (step S61: NO), and after storing M trained generators 131' in the first storage unit 17 (step S61: YES), the process moves to step S7 in Figure 8.

[0121] Next, the generation unit 14 generates pseudo-appropriate communication operations corresponding to each time slot using each of the trained generators 131' constructed by the third learning unit 13 (step S6). After that, the fourth learning unit 15 executes the fourth learning process (step S7). In step S7, the second learning unit 12 uses the pseudo-appropriate communication operations generated by each of the trained generators 131' as training data, takes information from one of the multiple time slots as input, and learns a communication behavior prediction model that outputs the amount of data transmitted in that time slot. In step S7, the fourth learning unit 15 performs supervised learning of the communication behavior prediction model having the neural network configuration shown in Figure 6.

[0122] Subsequently, the transmission unit 16 transmits the trained communication behavior prediction model constructed in step S7 to the communication terminal 2 via the communication network NW (step S8). Next, the communication terminal 2 stores the trained path prediction model received in step S9 in the second storage unit 21 (step S9). Then, the second acquisition unit 20 acquires the time slot information assigned to the communication terminal 2 (step S10).

[0123] Next, the calculation unit 22 provides the time slot information assigned to the communication terminal 2, acquired in step S10, as an unknown input to the trained communication behavior prediction model, performs calculations on the trained communication behavior prediction model, and outputs the amount of data to be transmitted in that time slot (step S11). Next, the instruction unit 23 instructs the communication control unit 24 to perform communication control according to the amount of data to be transmitted in the time slot output in step S11 (step S12).

[0124] Subsequently, the communication control unit 24 reads data corresponding to the amount of data to be transmitted from the transmission buffer according to the instructions in step S13, and transmits it from the wireless transmission circuit (step S13). The same process from step S9 to step S13 is performed for the other communication terminals 2.

[0125] As described above, according to the communication management system of this embodiment, the probability density function and parameters obtained by the first learning unit 11 and the second learning unit 12 are set as the objective function of the GAN, and adversarial learning of the GAN is performed with each of the transmitted data amounts for each time slot corresponding to proper communication operation as the true proper communication operation, and pseudo-proper communication operations are generated using each of the generated trained generators 131'. Furthermore, the pseudo-proper communication operations that are similar to the true proper communication operations generated by each of the trained generators 131' are used as training data to train the communication behavior prediction model on the relationship between each time slot and the transmitted data amount. As a result, the efficiency of communication resource utilization can be further improved.

[0126] According to the communication management system of this embodiment, a generation model is constructed for each of the M time slots, and the objective function for each generation model corresponding to the amount of data transmitted in each time slot is set with parameters for appropriate and inappropriate communication operations corresponding to the amount of data transmitted in that time slot estimated by the second learning unit 12. Therefore, the learning accuracy of the generation model can be improved.

[0127] In addition, the communication management system described in the embodiment described above is one in which the communication management device 1 has the functions of the first learning unit 11 to the fourth learning unit 15, but it is also possible to configure the system so that these learning functions are installed in the communication terminal 2.

[0128] Although embodiments of the communication management system and communication management method of the present invention have been described above, the present invention is not limited to the embodiments described above, and various modifications that a person skilled in the art can envision are possible within the scope of the invention described in the claims. [Explanation of symbols]

[0129] 1...Communication management device, 2...Communication terminal, 10...First acquisition unit, 11...First learning unit, 12...Second learning unit, 13...Third learning unit, 14...Generation unit, 15...Fourth learning unit, 16...Transmission unit, 17...First storage unit, 20...Second acquisition unit, 21...Second storage unit, 22...Calculation unit, 23...Instruction unit, 24...Communication control unit, 101...Bus, 102...Processor, 103...Main memory, 104...Communication interface, 105...Auxiliary storage, 106...Input / output I / O, 107...Display device, 131...Generator, 132...Discriminator, 133...Output, 134...Training data, 135...Block of objective function E, NW...Communication network.

Claims

1. A first learning unit is configured to learn a first parameter that represents the ratio between a probability distribution of appropriate communication operation, which shows a series of communication resource usage for each time slot among a plurality of time slots corresponding to the communication operation of a communication terminal whose communication resource usage request is equal to or greater than a first criterion, and a probability distribution of improper communication operation, which shows a series of communication resource usage for each time slot corresponding to the communication operation of the communication terminal whose communication resource usage request is lower than a second criterion (lower than the first criterion), and to estimate the ratio based on the learned first parameter. A second learning unit is configured to take each of the sequences of transmitted data amounts for each time slot as an observed value, treat each observed value as a conditionally independent discrete value, and estimate a second parameter of a class-specific probability distribution model corresponding to the class of proper communication operation and the class of improper communication operation, respectively, based on the frequency of occurrence of each observed value. A third learning unit is configured to learn multiple generative models, each having an objective function set by the probability distribution of the appropriate communication operation and the probability distribution of the inappropriate communication operation determined from the ratio estimated by the first learning unit, and the second parameter estimated by the second learning unit. The third learning unit includes a generator that generates pseudo-appropriate communication operations that are statistically similar to the true appropriate communication operation, with each of the communication resource usage amounts for each time slot corresponding to the appropriate communication operation being considered as a true appropriate communication operation, and a discriminator that distinguishes between the pseudo-appropriate communication operation and the true appropriate communication operation generated by the generator, and the third learning unit having an objective function set by the probability distribution of the appropriate communication operation and the probability distribution of the inappropriate communication operation determined from the ratio estimated by the first learning unit, and the second parameter estimated by the second learning unit. A generation unit configured to generate the pseudo-appropriate communication operation using each of the multiple trained generators obtained through learning by the third learning unit, A communication management system equipped with the following features.

2. In the communication management system described in claim 1, Furthermore, the system includes a fourth learning unit configured to learn a communication operation prediction model that uses the pseudo-optimal communication operation generated by each of the trained generators as training data, takes information indicating one of the multiple time slots as input, and outputs the amount of communication resources used in that time slot. A communication management system characterized by the following features.

3. In the communication management system described in claim 2, Furthermore, an acquisition unit configured to acquire information on the time slot assigned to the communication terminal, A calculation unit is configured to provide the time slot information assigned to the communication terminal, acquired by the acquisition unit, as an unknown input to a trained communication operation prediction model obtained by the fourth learning unit, perform calculations on the trained communication operation prediction model, and output the amount of communication resource usage in the time slot assigned to the communication terminal. An instruction unit configured to send instructions to communicate according to the amount of communication resources used in the time slot allocated to the communication terminal, which is output by the calculation unit. A communication management system characterized by having the following features.

4. In the communication management system described in claim 3, The communication management system comprises a communication management device and a communication terminal that is connected to the communication management device in a manner that enables communication with the communication management device. The aforementioned communication management device is The system comprises the first learning unit, the second learning unit, the generation unit, and the fourth learning unit, The aforementioned communication terminal is The system comprises the acquisition unit, the calculation unit, and the instruction unit. A route management system characterized by the following:

5. A first learning step involves learning a first parameter that represents the ratio between a probability distribution of appropriate communication operation, which shows a series of communication resource usage for each time slot among multiple time slots corresponding to the communication operation of a communication terminal whose communication resource usage request is equal to or greater than a first criterion, and a probability distribution of improper communication operation, which shows a series of communication resource usage for each time slot corresponding to the communication operation of the communication terminal whose communication resource usage request is lower than a second criterion (lower than the first criterion), and estimating the ratio based on the learned first parameter. A second learning step involves treating each of the sequences of transmitted data amounts for each time slot as an observed value, treating each observed value as a conditionally independent discrete value, and estimating the second parameter of a class-specific probability distribution model corresponding to the class of proper communication operation and the class of improper communication operation, respectively, based on the frequency of occurrence of each observed value. A third learning step involves learning a plurality of generative models, each having an objective function set by the probability distribution of the appropriate communication operation and the probability distribution of the inappropriate communication operation determined from the ratio estimated in the first learning step, and the second parameter estimated in the second learning step. A generation step is performed to generate the pseudo-appropriate communication operation using each of the multiple trained generators obtained through the learning in the third learning step. A communication management method comprising the following features.

6. In the communication management method described in claim 5, Furthermore, the system includes a fourth learning step in which a communication operation prediction model is trained, which takes the pseudo-optimal communication operation generated by each of the trained generators as training data, takes information indicating one of the multiple time slots as input, and outputs the amount of communication resources used in that time slot. A communication management method characterized by the following features.

7. In the communication management method described in claim 6, Furthermore, the acquisition step involves obtaining information about the time slot assigned to the communication terminal, A calculation step which involves providing the time slot information assigned to the communication terminal obtained in the acquisition step as an unknown input to the trained communication operation prediction model obtained in the fourth learning step, performing calculations on the trained communication operation prediction model, and outputting the amount of communication resource usage in the time slot assigned to the communication terminal; An instruction step that sends an instruction to communicate according to the amount of communication resources used in the time slot allocated to the communication terminal, which was output in the calculation step; A communication management method characterized by comprising:

Images