Network optimization device and network optimization method

Abstract

This network optimization device includes: a generation unit that executes generation processing in which a prompt including known information of a network and a problem setting of a network optimization is input to a large language model to generate a candidate solution for the network optimization; and an analysis unit that executes analysis processing in which a performance value of the network is analyzed by using one or more candidate solutions and either the network or a network model corresponding to the network. The network optimization device updates the prompt on the basis of an analysis result from the analysis unit, and executes the generation processing and the analysis processing again.

Description

Network optimization device and network optimization method

[0001] The present invention relates to a network optimization device and a network optimization method.

[0002] Network optimization is an optimization problem aimed at improving network performance or reliability. Network optimization is often formulated as an optimization problem on a graph, and in recent years, the application of large language models (LLMs) to network optimization has also been proposed (see, for example, Non-Patent Document 1).

[0003] Wang, Heng, et al. "Can language models solve graph problems in natural language?" 37th Conference on Neural Information Processing Systems (NeurIPS 2023).

[0004] However, conventional techniques can only handle formulated problems such as those based on graph theory, and have the drawback of being unable to handle problems whose problem setting is difficult to formulate, such as delay requirements or network reliability.

[0005] The embodiments of the present invention have been made in consideration of the above-mentioned problems, and enable a network optimization device that optimizes a network using a large-scale language model to deal with problems that are difficult to formulate.

[0006] In order to solve the above problems, a network optimization device according to an embodiment of the present invention includes a generation unit that executes a generation process by inputting a prompt including known information about a network and a problem setting for network optimization into a large-scale language model to generate solution candidates for the network optimization, and an analysis unit that executes an analysis process by analyzing a performance value of the network using one or more of the solution candidates and the network or a network model corresponding to the network, and updates the prompt based on the analysis result of the analysis unit, and executes the generation process and the analysis process again.

[0007] According to an embodiment of the present invention, a network optimization device that optimizes a network using a large-scale language model can handle problems that are difficult to formulate.

[0008] FIG. 1 is a diagram illustrating an example of the configuration of a network optimization system according to an embodiment; FIG. 2 is a flowchart illustrating an example of a network optimization process according to an embodiment; FIG. 3 is a diagram illustrating an example of a prompt according to an embodiment; FIG. 4 is a diagram illustrating an example of an output of an LLM according to an embodiment; and FIG. 5 is a diagram illustrating an example of the hardware configuration of a computer.

[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiment to which the present invention is applied is not limited to the following embodiment.

[0010] <About network optimization> Network optimization is an optimization problem aimed at improving network performance or reliability. Network optimization is often formulated as an optimization problem on a graph, and many classical algorithms are known. These algorithms have been used in various situations in the operation of actual communication networks.

[0011] In recent years, large language models (LLMs) have demonstrated excellent performance in various fields, and are expected to be applied to the aforementioned field of network optimization. For example, Non-Patent Document 1 attempts to solve classical problems related to graph theory or network optimization using large language models (hereinafter referred to as LLMs). In this example, mathematical problems such as connectivity determination and topological sorting are described in natural language and input into the LLM. Experiments have shown that, for small to medium-sized graphs, the LLM can grasp the graph structure and solve the graph problem. These results are thought to suggest the applicability of LLMs to network optimization.

[0012] However, there are the following differences between the graph theory problem discussed in Non-Patent Document 1 and more realistic network optimization.

[0013] First, practical network optimization often involves complex problem settings and structures, making it difficult to formulate. Generally, network optimization for communication networks requires determining appropriate network topologies and configuration values ​​based on a variety of information, such as the service delivery format, number of users, and their attributes. Furthermore, to ensure safe and secure service availability for users without interruption, networks must be designed and optimized with highly abstract characteristics, such as high security and fault tolerance, in mind. Constraints vary in their strength, ranging from fuzzy constraints that should be satisfied as much as possible to mandatory constraints that must be satisfied under contract. Thus, practical network optimization requires considering a large amount of information, which cannot necessarily be formulated uniformly.

[0014] Although Non-Patent Document 1 allows the inclusion of algorithmic advice in the input prompts of the LLM, in such realistic network optimization, such efficient algorithms are often non-trivial. Therefore, in realistic network optimization, it is difficult to directly apply the method disclosed in Non-Patent Document 1 to find a solution.

[0015] Second, in practical network optimization, complete information about the problem is not always available. For example, while the problem discussed in Non-Patent Document 1 was given all the information necessary to solve, in practical network optimization, this information is not always available. For example, information such as resource information on other companies' networks, uncertainty in traffic demand, and deviations from network device specifications is not always available, and even probabilistic modeling of some information is difficult. Even if only this known information were input directly into the LLM in natural language, it would result in an ill-posed problem, in which only incomplete information is given, and a high-quality solution would not be obtained. Therefore, it is difficult to solve practical network optimization using the method disclosed in Non-Patent Document 1.

[0016] In view of the above-mentioned problems, the present invention provides a framework for obtaining appropriate solutions for realistic network optimization problems by combining a real network or a network model simulating its behavior with an LLM and performing an iterative search. In particular, network optimization described in natural language is not only provided to the LLM to obtain candidate solutions, but the candidate solutions are also input into the real network or network model, and the LLM generates superior candidate solutions based on the network performance and analysis results at that time. Furthermore, by repeating this search process, solutions that overcome information deficiencies or ambiguity are efficiently discovered.

[0017] <Problem Setting> First, the premise for applying this embodiment to network optimization will be described.

[0018] Network optimization is a problem of determining parameters (e.g., topology, setting values, etc.) that determine network behavior so as to maximize or minimize an objective function related to network performance. Information related to network optimization that we want to solve can be divided into known information and unknown information.

[0019] Known information is information that the operator (the entity wanting to solve the problem) already has, such as the target network topology, the company's own network resource information, or the goal of network optimization (problem definition). Of course, if there is unknown information among them, it is treated as unknown information. Unknown information is information other than known information, such as information that is not actually available, such as other companies' network resource information, or information that is available but the operator has not obtained. Both known information and unknown information may be in any format.

[0020] There are countless network optimization problems that fall within the scope of this embodiment, but two examples are given below.

[0021] (1) Delay Minimization Problem Given a communications network, suppose we want to minimize the average delay between all nodes. Now, suppose we can add one link, and between which nodes can we create a new link that will minimize the average delay? The objective function in this problem is the average delay, and the space is the set of links (node ​​pairs) that can be added. Examples of known information include the network topology. On the other hand, the amount of traffic flowing can be considered unknown information, or if measurement data is available, it can be treated as known information (depending on the problem setting).

[0022] (2) Network Reliability Maximization Problem Given a communication network, each link may be disconnected with a certain probability due to failures, etc. We want to maximize the probability of communication between any two nodes (hereafter referred to as network reliability). Now, assuming that one link can be protected (the disconnection probability can be reduced to 0), which link should be protected to maximize network reliability? The objective function in this problem is network reliability, and the solution space is the set of links that can be protected. Known information can be the network topology, and unknown information can include the disconnection probability of each link.

[0023] In this embodiment, it is assumed that an operator can use an LLM and a real network or a network model. Here, the LLM is a large-scale language model constructed using a large amount of text data and deep learning technology. In this embodiment, the details of the LLM are not important as long as it is capable of question and answering based on natural language.

[0024] A network model estimates network performance values ​​(e.g., delay, etc.) according to predetermined input data (e.g., network topology information, traffic information, routing information, etc., which define network behavior). The details of the network model do not matter as long as it simulates the behavior of a real network.

[0025] A specific example of a network model is reference 1 (Rusek, Krzysztof, et al. "RouteNet: Leveraging graph neural networks for network modeling and optimization in SDN." IEEE Journal on Selected Areas in Communications 38.10 (2020): 2260-2270.). This is a neural network-based model that simulates the behavior of a network simulator. By inputting topology, device specifications, and a traffic model, it is possible to estimate network performance values ​​such as delay, jitter, and packet loss rate. This method can estimate network performance values ​​more quickly than conventional network simulators. In this embodiment, a real network may be used instead of the network model. However, in the following description, the real network and the network model will be collectively referred to simply as a "network model."

[0026] The candidate solutions for network optimization are considered to be (part of) the inputs of the network model, and the values ​​of the objective function corresponding to the candidate solutions are considered to be the outputs of the network model or calculated from the outputs. For example, in the delay minimization problem mentioned above, the added link is a candidate solution, and since network topology information is required to operate the network model, it is also considered to be part of the inputs of the network model. Furthermore, the average delay, which is the objective function, can be considered to be one of the network performance values ​​that are the outputs of the network model (or calculated from them). Consider an operator solving a network optimization problem under these settings.

[0027] <System Configuration> Fig. 1 is a diagram showing an example of the configuration of a network optimization system according to this embodiment. The network optimization system 1 is a system that optimizes a network using a large-scale language model. In the example of Fig. 1, the network optimization system 1 is configured by a network optimization device 100.

[0028] The network optimization device 100 is an information processing device having a computer configuration, or a system including multiple computers. The network optimization device 100 realizes each functional configuration shown in Fig. 1 by, for example, executing a predetermined program on a computer included in the network optimization device 100. In the example of Fig. 1, the network optimization device 100 includes an input unit 101, a generation unit 102, an analysis unit 103, an output unit 104, a known information DB (Database) 105, an LLM (solution candidate generation) 106, an experiment result DB 107, a network model 108, and an LLM (analysis statement creation) 109.

[0029] 1 is an example. For example, the known information DB 105, the LLM (solution candidate generation) 106, the experimental result DB 107, the network model 108, and the LLM (analysis statement creation) 109 may be provided outside the network optimization device 100. Furthermore, the functional components of the network optimization device 100 may be distributed across multiple computers. Furthermore, the LLM (solution candidate generation) 106 and the LLM (analysis statement creation) 109 may be the same LLM (large-scale language model), or may be an LLM external to the network optimization system 1.

[0030] The input unit 101 is an interface that accepts input of a network optimization problem setting and known information about the network. The problem setting is provided by, for example, an operator using, for example, natural language or mathematical expressions. The known information includes, for example, information about the network topology and network resources to be optimized, which is stored in advance in the known information DB 105 or the like.

[0031] The generator 102 inputs a prompt including known information about the network and a problem setting for network optimization into a large-scale language model to perform a generation process for generating solution candidates for network optimization. For example, the generator 102 performs appropriate processing or specifies an output format on the given information to generate a prompt suitable for input to the LLM (solution candidate generator) 106, and inputs the generated prompt to the LLM (solution candidate generator) 106.

[0032] Upon receiving the prompt, the LLM (solution candidate generation) 106 generates and outputs a solution candidate. If additional information is required to output the solution candidate, the generation unit 102 may query the user. If there is additional information, the generation unit 102 generates a new prompt and inputs it to the LLM (solution candidate generation) 106.

[0033] The analysis unit 103 executes an analysis process to analyze the performance value of the network using one or more solution candidates generated by the generation unit 102 using the LLM (solution candidate generation) 106 and a network or a network model. For example, the analysis unit 103 inputs the generated solution candidates into the network model 108, and stores the combination of the estimated network performance value and the solution candidate output by the network model 108 in the experiment result DB 107. Note that the experiment result DB 107 stores not only the results of a single experiment (pairs of solution candidates and estimated values ​​of network performance), but also all past experiment results. The analysis unit 103 also performs an analysis process on the past experiment results, processes the analysis results, and passes the processed experiment results to the generation unit 102.

[0034] The generation unit 102 updates the prompt requesting the LLM (solution candidate generation) 106 to generate a solution based on the analysis results of the analysis unit 103. For example, the generation unit 102 generates a prompt (prompt statement) requesting the LLM (analysis statement creation) 109 to generate an analysis statement, taking into account the problem setting information and known information described above. This prompt is intended to generate an analysis statement that takes into account the problem setting and known information, etc., the nature of the experimental results, whether the direction of the optimization search is correct, whether the search is heading toward the minimum or maximum value of the objective function to be aimed for, whether the search is violating constraints, etc.

[0035] The LLM (analysis sentence generator) 109 generates an analysis sentence based on such a prompt sentence and returns it to the generator 102 .

[0036] The generation unit 102 updates the prompt requesting the LLM (solution candidate generation) 106 to generate a solution candidate so that the prompt includes the analysis statement generated by the LLM (analysis statement creation) 109 and requests the generation of a new solution candidate based on the analysis statement, and inputs the updated prompt to the LLM (solution candidate generation) 106.

[0037] The network optimization device 100 repeatedly executes the above-described prompt update, generation process, and analysis process until a predetermined termination condition is met.

[0038] After completing the above process, the output unit 104 outputs, for example, the solution candidate that provided the best network performance as the network optimization solution. Note that if the output unit 104 identifies the best solution candidate at an early stage during the loop or obtains a sufficiently good solution that satisfies the conditions of the problem setting, it may present the solution to the user at that point.

[0039] <Processing Flow> Next, the processing flow of the network optimization method according to this embodiment will be described.

[0040] 2 is a flowchart showing an example of a network optimization process according to this embodiment, which is executed by the network optimization device 100 described with reference to FIG.

[0041] In step S201, the operator inputs the problem setting for network optimization and known information into the input unit 101.

[0042] In step S202, the generator 102 converts the input network optimization problem statement and known information into natural language or an appropriate format based on prompt engineering knowledge, and generates a prompt requesting the generation of solution candidates. Examples of prompt engineering include converting the format of input data (e.g., converting a network topology into an adjacency matrix or adjacency list) and devising a task request for the LLM (e.g., adding a role-play context). Examples of prompt engineering also include specifying an output format for subsequent steps (e.g., requesting a format suitable for input to the network model 108).

[0043] The generation unit 102 may, for example, input the network optimization problem setting and known information into a prepared prompt template, and generate a prompt that requests the LLM (solution candidate generation) 106 to generate solution candidates. Alternatively, the generation unit 102 may use the prepared prompt and LLM to generate a prompt that requests the LLM (solution candidate generation) 106 to generate solution candidates from the network optimization problem setting and known information.

[0044] In step S203, the generation unit 102 inputs the generated prompt to the LLM (solution candidate generation) 106 to generate solution candidates. The LLM (solution candidate generation) 106 generates solution candidates that are considered desirable for network optimization to the extent possible based on the given known information. However, if there is insufficient information when operating the network model 108, the generation unit 102 prompts the operator to request it. If additional information is provided by the operator, the LLM (solution candidate generation) 106 generates solution candidates taking the additional information into consideration. On the other hand, if additional information is not available, the LLM (solution candidate generation) 106 may generate solution candidates by arbitrarily assuming the additional information.

[0045] In step S204, the analysis unit 103 inputs the solution candidates generated by the generation unit 102 using the LLM (solution candidate generation) 106 into the network model 108 to obtain an estimated value of network performance. The analysis unit 103 also stores the obtained combination of the estimated value of network performance and the solution candidates in the experiment result DB 107. Note that the estimated value of network performance is an example of a network performance value. For example, the analysis unit 103 may input the generated solution candidates into a network to be optimized to obtain a network performance value.

[0046] In step S205, the analysis unit 103 analyzes one or more candidate solutions and estimated network performance values ​​(network performance values) stored in the experimental result DB 107. For example, the analysis unit 103 analyzes, as necessary, pairs of newly obtained candidate solutions and estimated network performance values, as well as all pairs of candidate solutions and estimated network performance values ​​obtained in previous loops. For example, the analysis unit 103 sorts the experimental results based on the estimated network performance values. This allows the analysis unit 103 to convert the shape of the objective function, gradient information, etc., into an LLM that is easily distinguishable. Alternatively, the analysis unit 103 may combine multiple estimated network performance values ​​to calculate a new index. For example, if the estimated network performance values ​​are obtained as a probability distribution, the analysis unit 103 may aggregate the information into the mean, standard deviation, higher-order moments, etc. Furthermore, the analysis unit 103 may perform processing such as excluding experimental results that are suspected to be due to defects in the network model 108 or outliers.

[0047] In step S306, the analysis unit 103 determines whether a termination condition has been met. The termination condition may be whether a predetermined cost (time, number of times, etc.) has been reached, or whether a target network performance value has been reached, etc. If the termination condition has not been met, the analysis unit 103 shifts the process to step S207. On the other hand, if the termination condition has been met, the analysis unit 103 shifts the process to step S209.

[0048] In step S207, the generation unit 102 generates an analytical statement indicating the analysis results. For example, the generation unit 102 converts past experimental results, previous analysis results, etc. into an appropriate format based on knowledge of prompt engineering, and generates a prompt to be input to the LLM (analysis statement generator) 109. The generation unit 102 also inputs the generated prompt to the LLM (analysis statement generator) 109 to create an analytical statement that considers the past analysis results and the analysis results.

[0049] The previous solution candidate was generated based on known information and considered desirable for network optimization using LLM (solution candidate generation) 106. This analysis statement is intended to verbalize whether the hypothesis was correct, what kind of solution candidate is considered effective for further improvement, etc.

[0050] In step S208, the generation unit 102 uses the created analysis statement to update the prompt requesting the generation of solution candidates from the LLM (solution candidate generation) 106. For example, the generation unit 102 uses known information, past experimental results, analysis results, and the analysis statement generated in step S207 to generate a prompt that urges the LLM (solution candidate generation) 106 to generate new solution candidates, and then proceeds to step S203.

[0051] On the other hand, when the process proceeds from step S206 to step S209, the output unit 104 outputs the network optimization solution. For example, the output unit 104 outputs the final estimated value of network performance or the candidate solution with the best network performance value as the network optimization solution.

[0052] According to this embodiment, the process of FIG. 2 enables the network optimization device 100, which optimizes a network using a large-scale language model, to handle problems that are difficult to formulate.

[0053] (Variations) Note that this embodiment does not necessarily follow the above steps strictly. For example, the network optimization device 100 may create a prompt that instructs the LLM to generate solution candidates in steps S202 and S203 of FIG. 2 and create an analysis statement in step S207 in a single command. An example of the prompt in this case is shown in FIG. 3, and an example of the output of the LLM is shown in FIG. 4.

[0054] 3 is a diagram showing an example of a prompt according to this embodiment. This prompt 300 shows an image of a prompt that requests the LLM (solution candidate generator) 106 or the like to generate solution candidates and write an analysis statement.

[0055] In the example of Fig. 3, the prompt 300 includes three paragraphs 301, 302, and 303. The first paragraph 301 describes the network optimization problem setting, known network problems, etc. In the example of Fig. 3, paragraph 301 describes, for example, the network structure (edge ​​set), problem instructions, initial average delay, number of trials, etc.

[0056] The second paragraph 302 describes past experimental results. Note that the example in Fig. 3 does not include past analysis statements. The third paragraph 303 includes a statement requesting the LLM (solution candidate generator) 106 or the like to generate solution candidates (Suggestion) and create analysis statements (Analysis).

[0057] 4 is a diagram showing an example of the output of the LLM according to this embodiment. This output 400 shows an example of an output statement output by the LLM (solution candidate generation) 106 or the like when a single prompt requests the generation of solution candidates and the creation of an analysis statement. This output 400 includes an analysis statement (Analysis) 401 that analyzes the experimental results and a solution candidate suggestion (Suggestion) 402 written in natural language.

[0058] Prompt 300 in Figure 3 and output 400 in Figure 4 both assume the aforementioned minimization delay problem. However, prompt 300 in Figure 3 and output 400 in Figure 4 do not necessarily correspond. This minimization delay problem also differs slightly from the aforementioned problem setup in that it presents the option of not only adding links but also increasing the capacity of existing links.

[0059] <Hardware Configuration> The network optimization device 100 according to this embodiment has, for example, the hardware configuration of a computer 500 as shown in Fig. 5. Alternatively, the network optimization device 100 is realized by a plurality of computers 500. Note that the computers are not limited to physical machines, and may be, for example, virtual machines on a cloud.

[0060] Fig. 5 is a diagram showing an example of the hardware configuration of a computer. In the example of Fig. 5, a computer 500 includes a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, and an output device 1008, all of which are interconnected by a bus B. The computer 500 may further include a GPU (Graphics Processing Unit) or the like.

[0061] A program for implementing processing on the computer 500 is provided by a recording medium 1001, such as a CD-ROM or a memory card. When the recording medium 1001 storing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001, but may be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program as well as necessary files, data, etc.

[0062] The memory device 1003 reads and stores the program from the auxiliary storage device 1002 when an instruction to start the program is received. The CPU 1004 realizes functions related to the network optimization device 100 in accordance with the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to a communication network, etc. The display device 1006 displays a GUI (Graphical User Interface) or the like according to the program. The input device 1007 is composed of a keyboard, mouse, buttons, and / or a touch panel, etc., and is used to input various operation instructions. The output device 1008 outputs the calculation results.

[0063] The CPU 1004 may be another processor such as a DSP (Digital Signal Processor), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array).

[0064] <Effects of the embodiment> According to the present embodiment, the network optimization device 100, the network optimization system 1, or the network optimization method that optimizes a network using a large-scale language model can handle problems that are difficult to formulate.

[0065] For example, with this embodiment, even in the case of a realistic network optimization problem in which the problem setting is complex or the problem is ill-posed due to a lack of information, it is expected that a more appropriate solution can be found by iterative search based on the network model 108 while utilizing given known information.

[0066] Summary of Embodiments This specification discloses at least the network optimization device, network optimization system, and network optimization method of the following paragraphs: (1) A network optimization device comprising: a generation unit that executes a generation process by inputting a prompt including known information about a network and a problem setting for the network optimization into a large-scale language model to generate solution candidates for the network optimization; and an analysis unit that executes an analysis process that analyzes a performance value of the network using one or more of the solution candidates and the network or a network model corresponding to the network, wherein the network optimization device updates the prompt based on an analysis result of the analysis unit and executes the generation process and the analysis process again. (2) The network optimization device described in paragraph 1, which repeatedly executes the prompt update, the generation process, and the analysis process until a predetermined termination condition is satisfied, and outputs the solution candidate with the best network performance as the solution for the network optimization. (3) The network optimization device described in paragraph 1 or 2, wherein the generation unit updates the prompt using the analysis result of the analysis process and a large-scale language model. (4) A network optimization method, in which a computer performs a generation process in which a prompt including known information about a network and a problem setting for the network optimization is input into a large-scale language model to generate candidate solutions for the network optimization; an analysis process in which a performance value of the network is analyzed using one or more of the candidate solutions and the network or a network model corresponding to the network; and an analysis process in which the prompt is updated based on the analysis results of the analysis process, and the generation process and the analysis process are executed again.(5) A network optimization system comprising: a generation unit that executes a generation process by inputting a prompt including known information about a network and a problem setting for the network optimization into a large-scale language model to generate solution candidates for the network optimization; and an analysis unit that executes an analysis process to analyze a performance value of the network using one or more of the solution candidates and the network or a network model corresponding to the network, wherein the prompt is updated based on an analysis result of the analysis unit, and the generation process and the analysis process are executed again.

[0067] Although the present embodiment has been described above, the present invention is not limited to such a specific embodiment, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

[0068] REFERENCE SIGNS LIST 1 Network optimization system 100 Network optimization device 101 Input unit 102 Generation unit 103 Analysis unit 104 Output unit 106 LLM (solution candidate generation) 107 Experimental result DB 108 Network model 109 LLM (analysis statement creation) 500 Computer

Claims

1. A network optimization device comprising: a generation unit that executes a generation process in which a prompt including known information about a network and a problem setting for network optimization is input into a large-scale language model to generate solution candidates for the network optimization; and an analysis unit that executes an analysis process in which the performance values ​​of the network are analyzed using one or more of the solution candidates and the network or a network model corresponding to the network, wherein the prompt is updated based on the analysis results of the analysis unit, and the generation process and the analysis process are executed again.

2. The network optimization device according to claim 1, wherein the prompt update, the generation process, and the analysis process are repeatedly executed until a predetermined termination condition is met, and the solution candidate with the best network performance is output as the solution for the network optimization.

3. The network optimization device according to claim 1 or 2, wherein the generation unit updates the prompts using the analysis results of the analysis process and a large-scale language model.

4. A network optimization method in which a computer executes a generation process in which a prompt including known information about the network and a problem setting for network optimization is input into a large-scale language model to generate solution candidates for the network optimization; and an analysis process in which performance values ​​of the network are analyzed using one or more of the solution candidates and the network or a network model corresponding to the network; the computer updates the prompt based on the analysis results of the analysis process, and executes the generation process and the analysis process again.

Images