Network topology reconstruction method, device and equipment based on deep learning

Through a deep learning-based method, network dynamic equations and generative adversarial networks are used to solve the problem of data acquisition in complex network topology reconstruction, and the accurate reconstruction of larger-scale networks is achieved.

CN114997036BActive Publication Date: 2024-11-22NAT UNIV OF DEFENSE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210182108.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-25
Publication Date
2024-11-22
Estimated Expiration
2042-02-25

AI Technical Summary

Technical Problem

The prior art is difficult to effectively reconstruct the topological structure of complex networks under limited data volume, especially when node dynamic time series data acquisition is difficult.

Method used

Using a deep learning-based method, the node state matrix is ​​obtained by obtaining the network dynamics equation, and the deep learning model includes preprocessing layer, generator and discriminator. The generative adversarial network training model is used to predict the probability of edge connection between nodes, which is suitable for comparison Large-scale network topology reconstruction.

Benefits of technology

The deep learning model can better learn the correlation between node state vectors, improve the accuracy and adaptability of network reconstruction, and is suitable for network topology reconstruction scenarios of different scales.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114997036B_ABST
    Figure CN114997036B_ABST
Patent Text Reader

Abstract

The present application relates to a method, device and equipment for network topology reconstruction based on deep learning. The method comprises: obtaining the network dynamics equation of the target network, obtaining the state matrix of the network node through the dynamics equation; marking the state matrix to obtain the training sample of the deep learning model; inputting the training sample into the deep learning model to train the deep learning model; constructing the first loss function for training the discriminator to train the discriminator; constructing the second loss function for training the generator to train the generator; inputting the state matrix of the network node to be predicted into the trained deep learning model to obtain the network topology. The present method can adapt to large-scale network topology reconstruction scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device and equipment for reconstructing network topology based on deep learning. Background Art

[0002] Many systems in the real world, such as ecosystems, transportation systems, and social systems, can be abstracted into complex networks. Each element in the system is abstracted into nodes, and the relationship between elements is abstracted into edges. The edges between nodes constitute the network structure of the system. In addition to certain structural characteristics, the states of elements in the real world often follow certain rules. For example, in an ecosystem, in order to maintain the balance of the entire ecosystem, the number of species of different types will have certain changing rules; for humans, with the development of the economy, the birth rate of humans also has certain changing trends. In order to characterize the changes in the states of each element in different systems, scholars in different fields use different dynamic equations to describe the change process of the system state. After the system is abstracted into a complex network and the state change process of the elements in the system is described by the dynamic equation, the structure of the system is connected with the dynamic process of the system. A system without structure means that the elements in the system are isolated from each other, that is, there is no information exchange between the elements, and there is no dynamic change process in the system. Such a system does not actually exist; similarly, there is no system without interaction between elements. Therefore, there is a close dependence between the structure of the system and the dynamic process of the system.

[0003] The network reconstruction problem refers to the inverse problem of studying the network topology through the dynamic process on the network. This problem has become a hot topic in the field of network science. So far, relevant scholars have done a lot of work on this problem. Earlier related studies include Granger causality methods, compressed sensing methods, and information theory-based methods. Due to the close dependence between network structure and network dynamics, many scholars have studied how to reconstruct the network topology through the dynamic time series of nodes on the network. However, since it is difficult to obtain the complete time series data of nodes in the network, many works have studied the network reconstruction problem under the condition of limited data amount. Summary of the invention

[0004] Based on this, it is necessary to provide a network topology reconstruction method, device and equipment based on deep learning to address the above technical problems.

[0005] A network topology reconstruction method based on deep learning, the method comprising:

[0006] Obtaining a network dynamics equation of the target network, and obtaining a state matrix of the network nodes through the dynamics equation;

[0007] Labeling the state matrix to obtain training samples of a deep learning model;

[0008] The training sample is input into the deep learning model to train the deep learning model; the deep learning model comprises: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training sample to obtain input data; the discriminator discriminates the input data and outputs a first edge prediction result; the generator receives the input data and outputs generated data; the discriminator discriminates the generator to obtain a second edge prediction result;

[0009] Constructing a first loss function for training the discriminator to train the discriminator; the first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data;

[0010] Constructing a second loss function for training the generator to train the generator; the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data;

[0011] The state matrix of the network node to be predicted is input into the trained deep learning model to obtain the network topology.

[0012] In one embodiment, the network dynamics equations include: heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation; and further include: respectively obtaining the heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation, solving the heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation, and obtaining the state value of each node in the target network; using the state value as a vector value to obtain a state matrix of the target network.

[0013] In one embodiment, the method further includes: obtaining a topological structure of a continuous time data reconstruction network, and representing a state set of nodes on the continuous time data reconstruction network as:

[0014]

[0015] in, represents the state value of node i at time t, and T represents the total time length;

[0016] Based on the state set of the two nodes, predict the probability that there is an edge between the two nodes

[0017] Sure Among them, N is the number of network nodes, and A is the adjacency matrix of the network, so as to convert the state set of each node into the state vector of the node.

[0018] In one embodiment, the first loss function and the second loss function are both square loss functions.

[0019] In one embodiment, the generator is a three-layer fully connected neural network with a sigmoid activation layer.

[0020] In one embodiment, the discriminator is a three-layer fully connected neural network with a sigmoid activation layer.

[0021] A network topology reconstruction device based on deep learning, the device comprising:

[0022] A state matrix acquisition module is used to acquire the network dynamics equation of the target network, and obtain the state matrix of the network node through the dynamics equation;

[0023] A training module, used to label the state matrix to obtain training samples of the deep learning model; input the training samples into the deep learning model to train the deep learning model; the deep learning model includes: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training samples to obtain input data; the discriminator discriminates the input data and outputs a first edge prediction result; the generator receives input data and outputs generated data; the discriminator discriminates the generator to obtain a second edge prediction result; constructs a first loss function for training the discriminator and trains the discriminator; the first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data; constructs a second loss function for training the generator and trains the generator; the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data;

[0024] The reconstruction module is used to input the state matrix of the network node to be predicted into the trained deep learning model to obtain the network topology.

[0025] In one embodiment, the network dynamics equations include: heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation; the state matrix acquisition module is also used to respectively acquire the heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation, solve the heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation, and obtain the state value of each node in the target network; use the state value as a vector value to obtain the state matrix of the target network.

[0026] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0027] Obtaining a network dynamics equation of the target network, and obtaining a state matrix of the network nodes through the dynamics equation;

[0028] Labeling the state matrix to obtain training samples of a deep learning model;

[0029] The training sample is input into the deep learning model to train the deep learning model; the deep learning model comprises: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training sample to obtain input data; the discriminator discriminates the input data and outputs a first edge prediction result; the generator receives the input data and outputs generated data; the discriminator discriminates the generator to obtain a second edge prediction result;

[0030] Constructing a first loss function for training the discriminator to train the discriminator; the first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data;

[0031] Constructing a second loss function for training the generator to train the generator; the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data;

[0032] The state matrix of the network node to be predicted is input into the trained deep learning model to obtain the network topology.

[0033] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0034] Obtaining a network dynamics equation of the target network, and obtaining a state matrix of the network nodes through the dynamics equation;

[0035] Labeling the state matrix to obtain training samples of a deep learning model;

[0036] The training sample is input into the deep learning model to train the deep learning model; the deep learning model comprises: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training sample to obtain input data; the discriminator discriminates the input data and outputs a first edge prediction result; the generator receives the input data and outputs generated data; the discriminator discriminates the generator to obtain a second edge prediction result;

[0037] Constructing a first loss function for training the discriminator to train the discriminator; the first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data;

[0038] Constructing a second loss function for training the generator to train the generator; the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data;

[0039] The state matrix of the network node to be predicted is input into the trained deep learning model to obtain the network topology.

[0040] The above-mentioned network topology reconstruction method, device, computer equipment and storage medium based on deep learning, by studying the mechanism of the topological network, expresses the connectivity between nodes in the network by the degree of correlation between state matrices. Therefore, when applying the deep learning model for network reconstruction, the deep learning network can better learn the correlation between state vectors, so that when performing network reconstruction, it can adapt to larger-scale network topology reconstruction scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of a flow chart of a network topology reconstruction method based on deep learning in one embodiment;

[0042] Figure 2 A schematic diagram of a deep learning model process in one embodiment;

[0043] Figure 3 is an experimental result of heat diffusion dynamics on a BA scale-free network in one embodiment; wherein Figure 3 (a) is the ACC indicator result, Figure 3 (b) is the F1 indicator result;

[0044] Figure 4 is an experimental result of mutualism dynamics on a BA scale-free network in one embodiment; wherein Figure 4 (a) is the ACC indicator result, Figure 4 (b) is the F1 indicator result;

[0045] Figure 5 is an experimental result of gene regulation dynamics on a BA scale-free network in one embodiment; wherein Figure 5 (a) is the ACC indicator result, Figure 5 (b) is the F1 indicator result;

[0046] Figure 6 is an experimental result of heat diffusion dynamics on a WS small-world network in one embodiment; wherein Figure 6 (a) is the ACC indicator result, Figure 6 (b) is the F1 indicator result;

[0047] Figure 7 is an experimental result of mutualism dynamics on a WS small-world network in one embodiment; wherein Figure 7 (a) is the ACC indicator result, Figure 7 (b) is the F1 indicator result;

[0048] Figure 8 is an experimental result of gene regulation dynamics on a WS small-world network in one embodiment; wherein Figure 8 (a) is the ACC indicator result, Figure 8 (b) is the F1 indicator result;

[0049] Fig. 9 It is a structural block diagram of a network topology reconstruction device based on deep learning in one embodiment;

[0050] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] In one embodiment, Figure 1 As shown, a network topology reconstruction method based on deep learning is provided, comprising the following steps:

[0053] Step 102, obtaining the network dynamics equation of the target network, and obtaining the state matrix of the network nodes through the dynamics equation.

[0054] There are generally two ways to obtain data. One way is to directly obtain it by some observation method without knowing the network dynamics equation. For example, in a traffic network, sensors can be used to collect the number and speed data of vehicles at each intersection at different times. The other way is to obtain the state data of each node in the network at different times through simulation when the network dynamics equation is known. The present invention mainly considers the latter situation, that is, obtaining dynamic process data on the network through simulation.

[0055] The dynamics on the network can be described by differential equations. This paper considers three continuous-time network dynamics from the fields of physics and biology. The first dynamic is the heat diffusion dynamics, which is expressed as Among them, x i represents the heat value of node i, k i,j is the heat coefficient, A i,jRepresents the value corresponding to node i and node j in the network adjacency matrix, that is, if there is an edge between node i and node j, then A i,j =1, otherwise A i,j = 0. The second dynamics is the mutualism dynamics between species in the ecosystem, and the differential equation is expressed as Among them, x i represents the number of species i, b i The amount of alien species migration in the ecosystem, k i represents species capacity, c i represents the species threshold. The third dynamics is the gene regulation dynamics, and the differential equation is expressed as f is the gene regulation pattern coefficient, and h is the Hill coefficient regulating gene activation.

[0056] Step 104: label the state matrix to obtain training samples for the deep learning model.

[0057] According to the actual existence of edges between nodes in the network, the state vectors corresponding to the two nodes with edges are concatenated and labeled with "1". Correspondingly, the concatenated vectors of the two nodes without edges are labeled with "0". For ease of explanation, the present invention refers to samples labeled "1" as positive samples and samples labeled "0" as negative samples.

[0058] Due to the sparse nature of real networks, that is, the number of nodes with edges in the network is small, if the state vectors between nodes are labeled according to the above method, many negative samples will be obtained, that is, the sample data is unbalanced, and such data has no learning value for the neural network. To address the problem of unbalanced state vector samples between network nodes, the oversampling method is used to balance the sample data by expanding the number of positive samples. The specific approach is: copy the positive samples in the ratio of the number of negative samples to the number of positive samples, and then mix the obtained positive samples with the original negative samples and shuffle the order as the data needed for subsequent neural network training and testing. Take 80% of the mixed data for model training, and the remaining 20% ​​for model testing. Since the state vectors of any two nodes in the network need to be spliced ​​and labeled, the complexity of data processing is

[0059] Step 106: input the training samples into the deep learning model to train the deep learning model.

[0060] The deep learning model includes: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training samples to obtain input data; the discriminator discriminates the input data and outputs the first edge prediction result; the generator receives the input data and outputs the generated data; the discriminator discriminates the generator to obtain the second edge prediction result.

[0061] In this step, the deep learning model is a generative adversarial network. Figure 2 As shown, Figure 2 In the example, samples represents sample input, train represents training, D represents the discriminator, G represents the generator, loss represents the loss, and W represents the loss. N×T Represents the state matrix.

[0062] The specific steps are as follows: Use the preprocessed data to train the discriminator D, and finally obtain the true node connection probability by optimizing the loss function. Then, similarly, the preprocessed data is used again to train the generator G, and the output of the generator G is passed through the loss function in the discriminator D and optimized, and finally the predicted value of the probability of the edge connection between nodes is obtained.

[0063] Step 108, constructing a first loss function for training the discriminator, training the discriminator, constructing a second loss function for training the generator, and training the generator;

[0064] The first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data, and the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data.

[0065] Step 110, input the state matrix of the network node to be predicted into the trained deep learning model to obtain the network topology.

[0066] The above-mentioned network topology reconstruction method based on deep learning studies the mechanism of the topological network and represents the connectivity between nodes in the network through the degree of correlation between state matrices. Therefore, when the deep learning model is applied to reconstruct the network, the deep learning network can better learn the correlation between state vectors, so that when performing network reconstruction, it can adapt to larger-scale network topology reconstruction scenarios.

[0067] In one embodiment, the network dynamics equations include: a heat diffusion dynamics equation, a mutualistic symbiosis dynamics equation, and a gene regulation dynamics equation. The heat diffusion dynamics equation, the mutualistic symbiosis dynamics equation, and the gene regulation dynamics equation are obtained respectively, and the heat diffusion dynamics equation, the mutualistic symbiosis dynamics equation, and the gene regulation dynamics equation are solved to obtain the state value of each node in the target network; the state value is used as a vector value to obtain a state matrix of the target network.

[0068] The specific state value can be calculated by solving the corresponding differential equation using Neural ODE.

[0069] The network reconstruction problem can be divided into two categories. One is the network dynamics reconstruction problem (network dynamics reconstruction) when the network topology is known or not concerned with the network topology, that is, fitting the dynamic process on the network based on certain properties known about the network; the other is the network topology reconstruction problem (network topology reconstruction) when the dynamics on the network is known or not concerned with the network dynamics. The problem studied in the present invention is network topology reconstruction. Generally speaking, network topology reconstruction will utilize the dynamic process on the network, that is, utilize the state change law of each node in the network. In the case of unknown network dynamics, the required dynamic process data can be obtained through observation; in the case of known network dynamics, the change data of the state of each node on the network can be obtained through simulation. Depending on whether the obtained network data is continuous, the network topology reconstruction problem can be divided into a network topology reconstruction problem based on discrete data and a network topology reconstruction problem based on continuous time data. The present invention mainly considers how to reconstruct the network topology structure based on continuous time data on the network.

[0070] In one embodiment, in order to more rigorously describe how to reconstruct the topology of the network based on continuous time data, the state set of nodes on the network is represented as in represents the state value of node i at time t, and T represents the total time length. Then the network topology reconstruction problem can be decomposed into: predicting the probability of an edge between two nodes based on the state set of the two nodes, that is, Where E ij Indicates that there is an edge between node i and node j. Then the network topology reconstruction problem can be transformed into: Wherein, N is the number of network nodes, and A is the adjacency matrix of the network. The state set of each node in the network can be converted into the state vector of the node, and then the problem can be converted into using the correlation method to calculate the correlation between the state vectors of two nodes to predict the probability of the existence of an edge between the two nodes. However, due to the complexity of the network, including the complexity of the network structure and the complexity of the dynamic process on the network, it is difficult to accurately measure the size of the correlation between each node in the network using a certain correlation index. Therefore, the present invention uses a deep learning method to judge the probability of the existence of an edge between two nodes by learning the size of the correlation between the state vectors of two nodes.

[0071] In one embodiment, a three-layer fully connected neural network with a sigmoid activation layer is used as the main architecture of the generator module. By using the loss function in the discriminator to train the training data, the edge prediction results obtained by the data input to the generator are made as close as possible to the edge probability obtained by the discriminator for the real data, and finally the predicted value of the edge probability of the nodes in the network is obtained.

[0072] In one embodiment, similar to the generator, we also use a three-layer fully connected neural network with a sigmoid activation layer as the main architecture of the discriminator module. The purpose of is to determine the probability of whether there is a real edge between different nodes in the training data, that is, if there is an edge between two nodes, then the probability As close to 1 as possible, otherwise, make the probability As close to 0 as possible. The loss function in the discriminator uses the square loss function (Mean Square Error, MSE), and the loss function is optimized using the Adam optimizer.

[0073] When training the model, first, train the discriminator with the processed training data Make the output of the trainer as close to the true value as possible, that is, the output of the discriminator is as consistent as possible with the label value, and use the Adam optimizer to optimize the loss function to obtain the training parameters θ of the discriminator. D Then, the generator is trained using the same training data as the discriminator. And get the training parameters θ of the generator G ,Since the generator and the discriminator share the same loss function, the same training parameters as those for training the discriminator can be used.

[0074] Regarding the above embodiment, experimental verification is performed in the following for BA scale-free networks and WS small-world networks with different node sizes.

[0075] Three dynamic equations are used to generate network node state data. The topological characteristics of the network are shown in Table 1, where N is the number of network nodes, E is the number of network edges, <k>is the average degree of the network, C is the network clustering coefficient, <l>is the average distance length of the network.

[0076] Table 1 Basic topological characteristics of the network

[0077]

[0078] In order to accurately evaluate the accuracy of the experimental results, this paper uses ACC and F1 as indicators to measure the accuracy of network reconstruction. Samples labeled "1" are called positive samples, and samples labeled "0" are called negative samples. The network topology reconstruction problem can be regarded as a classification problem, and the confusion matrix of the classification result can be used to calculate the ACC and F1 indicators of the classification result. The confusion matrix of the classification result is defined as follows:

[0079] Table 2 Confusion matrix

[0080]

[0081]

[0082] The precision P and recall R are defined as follows:

[0083]

[0084]

[0085] The ACC and F1 indicators are defined as follows:

[0086]

[0087]

[0088] On the BA scale-free network and the WS small-world network, the node state data obtained by the three dynamic equations are used for experimental analysis. Figure 3 The experimental results of heat diffusion dynamics on BA scale-free networks are shown, where the horizontal axis is the number of network nodes and the vertical axis is the indicator value. The left figure shows the ACC indicator results obtained on the training data set and the test data set, and the right figure shows the F1 indicator results. From the experimental results, it can be found that the ACC indicator and F1 indicator results on the training data set and the test data set are relatively similar. As the network scale increases, the average score of the ACC indicator and the F1 indicator can be maintained above 0.95.

[0089] Figure 4 and Figure 5 The experimental results of mutualistic symbiosis dynamics and gene regulation dynamics on BA scale-free networks are shown respectively. Similar to the experimental results of heat diffusion dynamics on BA scale-free networks, the ACC and F1 results on the training and test datasets are quite similar, and the average scores of ACC and F1 are around 0.95. Figure 6 , Figure 7 and Figure 8 The experimental results of heat diffusion dynamics, mutualism dynamics and gene regulation dynamics on WS small-world network are shown respectively.

[0090] Table 3 records in detail the experimental results of three different dynamics on BA scale-free networks and WS small-world networks of different scales.

[0091] Table 3 Statistical table of experimental results of three kinds of dynamics on networks of different scales

[0092]

[0093]

[0094] In summary, this paper proposes a deep learning framework for the problem of network topology reconstruction under continuous time data. The framework can well capture the correlation between node state vectors, and uses three dynamics, heat diffusion, mutualism, and gene regulation, to generate continuous time state data of network nodes. Experimental verification has been carried out in BA scale-free networks and WS small-world networks of different scales. The experimental results show that the learning framework proposed in this paper can accurately restore the topological structure of the network and is suitable for large-scale network topology reconstruction scenarios.

[0095] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0096] In one embodiment, Fig. 9 As shown, a network topology reconstruction device based on deep learning is provided, including: a state matrix acquisition module 902, a training module 904 and a reconstruction module 906, wherein:

[0097] A state matrix acquisition module 902 is used to acquire a network dynamics equation of a target network, and obtain a state matrix of a network node through the dynamics equation;

[0098] The training module 904 is used to label the state matrix to obtain a training sample of the deep learning model; input the training sample into the deep learning model to train the deep learning model; the deep learning model includes: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training sample to obtain input data; the discriminator discriminates the input data and outputs a first edge prediction result; the generator receives the input data and outputs the generated data; the discriminator discriminates the generator to obtain a second edge prediction result; constructs a first loss function for training the discriminator and trains the discriminator; the first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data; constructs a second loss function for training the generator and trains the generator; the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data;

[0099] The reconstruction module 906 is used to input the state matrix of the network node to be predicted into the trained deep learning model to obtain the network topology.

[0100] In one embodiment, the network dynamics equations include: heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation; the state matrix acquisition module 902 is also used to respectively acquire the heat diffusion dynamics equation, the mutualism dynamics equation and the gene regulation dynamics equation, solve the heat diffusion dynamics equation, the mutualism dynamics equation and the gene regulation dynamics equation, and obtain the state value of each node in the target network; use the state value as a vector value to obtain the state matrix of the target network.

[0101] In one embodiment, the state matrix acquisition module 902 is further used to acquire the topological structure of the continuous time data reconstruction network, and express the state set of nodes on the continuous time data reconstruction network as:

[0102]

[0103] in, represents the state value of node i at time t, and T represents the total time length;

[0104] Based on the state set of the two nodes, predict the probability that there is an edge between the two nodes

[0105] Sure Among them, N is the number of network nodes, and A is the adjacency matrix of the network, so as to convert the state set of each node into the state vector of the node.

[0106] In one embodiment, the first loss function and the second loss function are both square loss functions.

[0107] In one embodiment, the generator is a three-layer fully connected neural network with a sigmoid activation layer.

[0108] In one embodiment, the discriminator is a three-layer fully connected neural network with a sigmoid activation layer.

[0109] For the specific definition of the network topology reconstruction device based on deep learning, please refer to the definition of the network topology reconstruction method based on deep learning above, which will not be repeated here. Each module in the above-mentioned network topology reconstruction device based on deep learning can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0110] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a network topology reconstruction method based on deep learning is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0111] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0112] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method in the above embodiment when executing the computer program.

[0113] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in the above embodiment are implemented.

[0114] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0115] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.< / l> < / k>

Claims

1. A network topology reconstruction method based on deep learning, characterized in that: The method comprises: Obtaining a network dynamics equation of the target network, and obtaining a state matrix of the network nodes through the dynamics equation; Labeling the state matrix to obtain training samples of a deep learning model; The training sample is input into the deep learning model to train the deep learning model; the deep learning model comprises: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training sample to obtain input data; the discriminator discriminates the input data and outputs a first edge prediction result; the generator receives the input data and outputs generated data; the discriminator discriminates the generator to obtain a second edge prediction result; Constructing a first loss function for training the discriminator to train the discriminator; the first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data; Constructing a second loss function for training the generator to train the generator; the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data; Input the state matrix of the network node to be predicted into the trained deep learning model to obtain the network topology; The network dynamics equations include: heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation; The step of obtaining a network dynamics equation of the target network and obtaining a state matrix of network nodes through the dynamics equation includes: Respectively obtain a heat diffusion kinetic equation, a mutualistic symbiosis kinetic equation, and a gene regulation kinetic equation, and solve the heat diffusion kinetic equation, the mutualistic symbiosis kinetic equation, and the gene regulation kinetic equation to obtain a state value of each node in the target network; Taking the state value as a vector value, obtaining a state matrix of the target network; The method further comprises: The topological structure of the continuous-time data reconstruction network is obtained, and the state set of nodes on the continuous-time data reconstruction network is expressed as: in, Representation Node exist The state value at the moment, Indicates the total length of time; Based on the state set of the two nodes, predict the probability that there is an edge between the two nodes ; Sure ,in, is the number of network nodes, is the adjacency matrix of the network, thereby converting the state set of each node into the state vector of the node.

2. The method according to claim 1, characterized in that Both the first loss function and the second loss function are square loss functions.

3. The method according to claim 1, characterized in that The generator is a three-layer fully connected neural network with a sigmoid activation layer.

4. The method according to claim 1, characterized in that: The discriminator is a three-layer fully connected neural network with a sigmoid activation layer.

5. A network topology reconstruction device based on deep learning, characterized in that: The device comprises: A state matrix acquisition module is used to acquire the network dynamics equation of the target network, and obtain the state matrix of the network node through the dynamics equation; A training module, used to label the state matrix to obtain training samples of a deep learning model; input the training samples into the deep learning model to train the deep learning model; the deep learning model includes: a preprocessing layer, a generator and a discriminator; the preprocessing layer preprocesses the input training samples to obtain input data; the discriminator discriminates the input data and outputs a first edge prediction result; the generator receives input data and outputs generated data; the discriminator discriminates the generator to obtain a second edge prediction result; a first loss function is constructed to train the discriminator and the discriminator is trained; the first loss function calculates the loss of the first edge prediction result and the label corresponding to the input data; a second loss function is constructed to train the generator and the generator is trained; the second loss function calculates the loss of the second edge prediction result and the label corresponding to the input data; The reconstruction module is used to input the state matrix of the network node to be predicted into the trained deep learning model to obtain the network topology; The network dynamics equations include: heat diffusion dynamics equation, mutualism dynamics equation and gene regulation dynamics equation; The state matrix acquisition module is also used to respectively acquire the heat diffusion kinetic equation, the mutualism kinetic equation and the gene regulation kinetic equation, solve the heat diffusion kinetic equation, the mutualism kinetic equation and the gene regulation kinetic equation, and obtain the state value of each node in the target network; use the state value as a vector value to obtain the state matrix of the target network; The state matrix acquisition module is also used to obtain the topological structure of the continuous time data reconstruction network, and the state set of nodes on the continuous time data reconstruction network is expressed as: in, Representation Node exist The state value at the moment, Indicates the total length of time; Based on the state set of the two nodes, predict the probability that there is an edge between the two nodes ; Sure ,in, is the number of network nodes, is the adjacency matrix of the network, thereby converting the state set of each node into the state vector of the node.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.