Estimation device, estimation method, and estimation program

The estimation device employs a CVAE model to accurately estimate configuration information for communication equipment and software by calculating and updating parameters, addressing the limitations of conventional methods in estimating communication-related configuration information.

WO2024247003A1PCT designated stage expired Publication Date: 2024-12-05NT T INC
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2023/019766
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-26
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Conventional methods, such as Wireshark, face challenges in accurately estimating configuration information of communication-related equipment and software due to the lack of defined rules and unreliable vendor-provided information, especially when Software Bill of Materials (SBOM) is absent or incorrect.

Method used

An estimation device utilizing a Conditional Variational Autoencoder (CVAE) model to output likelihood information regarding the joint distribution of communication and configuration features, enabling accurate estimation of configuration information by calculating and updating model parameters to satisfy specific conditions.

Benefits of technology

Enables high-accuracy estimation of configuration information for communication devices and software, even in the absence of SBOM, by searching for constituent features that meet likelihood conditions, thereby improving vulnerability management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2023019766_05122024_PF_FP_ABST
    Figure JP2023019766_05122024_PF_FP_ABST
Patent Text Reader

Abstract

An estimation device (10) includes a search unit (152) and an output control unit (154). The search unit (152) uses a model that outputs information relating to a likelihood regarding a joint distribution of a communication feature amount that represents a state of communication and a configuration feature amount that represents a configuration in the communication to search for such a configuration feature amount that the magnitude of the likelihood satisfies a condition when this configuration feature amount is input to the model together with an observed communication feature amount. For example, the search unit (152) uses a constrained probability density estimation model as the model to search for the configuration feature amount. The output control unit (154) outputs the configuration feature amount retrieved by the search unit (152).
Need to check novelty before this filing date? Find Prior Art

Description

Estimation device, estimation method, and estimation program

[0001] The present invention relates to an estimation device, an estimation method, and an estimation program.

[0002] There is a growing trend to utilize configuration information such as Software Bill of Materials (SBOM) for vulnerability management of devices and software.

[0003] Wireshark is also known as a tool that has the function of estimating internal configuration information of devices and software from communication information (see, for example, Non-Patent Document 1).

[0004] wireshark, [online], [Retrieved May 22, 2023], Internet (https: / / www.wireshark.org / ) Conditonal VAE, [online], [Retrieved May 22, 2023], Internet (https: / / proceedings.neurips.cc / paper / 2014 / hash / d523773c6b194f37b938d340d5d02232-Abstract.html)

[0005] However, the conventional techniques have a problem in that it is difficult to accurately estimate the configuration information of communication-related devices and software.

[0006] For example, devices and software may not be provided with configuration information such as SBOM, and the configuration information provided by the vendor may not always be correct.

[0007] Furthermore, Wireshark described in Non-Patent Document 1 is a rule-based estimation method, so it is difficult for Wireshark to estimate configuration information for which rules for estimation are not defined.

[0008] In order to solve the above-mentioned problems and achieve the object, the estimation device is characterized by having: a calculation unit that calculates a likelihood regarding the joint distribution of an observed communication feature and a configuration feature, using a model that outputs likelihood information regarding the joint distribution of the communication feature representing a state of communication and a configuration feature representing a configuration in the communication; and an update unit that updates parameters of the model so that the magnitude of the likelihood satisfies a condition.

[0009] The estimation device is also characterized by having: a search unit that uses a model that outputs likelihood information regarding the joint distribution of a communication feature that represents a state of communication and a configuration feature that represents a configuration in the communication, and searches for a configuration feature that satisfies a condition when the configuration feature is input to the model together with an observed communication feature; and an output control unit that outputs the configuration feature searched for by the search unit.

[0010] According to the present invention, configuration information of devices and software related to communication can be estimated with high accuracy.

[0011] FIG. 1 is a diagram illustrating an example of the configuration of an estimation device according to a first embodiment. FIG. 2 is a diagram illustrating a model. FIG. 3 is a flowchart illustrating the flow of a learning process. FIG. 4 is a flowchart illustrating the flow of a search process. FIG. 5 is a diagram illustrating the results of an experiment. FIG. 6 is a diagram illustrating the results of an experiment. FIG. 7 is a diagram illustrating an example of a computer that executes a program.

[0012] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] Exemplary embodiments of an estimation device, an estimation method, and an estimation program according to the present application will be described in detail below with reference to the accompanying drawings. However, the present invention is not limited to the following exemplary embodiments.

[0013] [Configuration of First Embodiment] The configuration of an estimation device according to a first embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the configuration of an estimation device according to the first embodiment.

[0014] As shown in FIG. 1 , the estimation device 10 includes a communication unit 11 , an input unit 12 , an output unit 13 , a storage unit 14 , and a control unit 15 .

[0015] The communication unit 11 performs data communication with other devices via a network. For example, the communication unit 11 is a network interface card (NIC).

[0016] The input unit 12 receives input of data. For example, the input unit 12 is an interface connected to input devices such as a mouse and a keyboard.

[0017] The output unit 13 outputs data. The output unit 13 is an interface connected to an output device such as a display and a speaker.

[0018] The storage unit 14 is a storage device such as a hard disk drive (HDD), a solid state drive (SSD), an optical disk, etc. Note that the storage unit 14 may also be a data-rewritable semiconductor memory such as a random access memory (RAM), a flash memory, or a non-volatile static random access memory (NVSRAM). The storage unit 14 stores an operating system (OS) and various programs executed by the estimation device 10.

[0019] The storage unit 14 stores model information 141. The model information 141 is parameters for constructing a machine learning model. For example, the model information 141 is parameters such as weights and biases of a neural network.

[0020] The model of this embodiment will now be described with reference to Fig. 2. Fig. 2 is a diagram for explaining the model.

[0021] 2 is a Conditional Variational Autoencoder (CVAE) (see, for example, Non-Patent Document 2). However, the model of this embodiment is not limited to the CVAE, and may be any model that can be regarded as a constrained probability density estimation model or a constrained generative model (for example, a Conditional GAN, a Conditional Flow Model, a Conditional Diffusion Model, etc.). Furthermore, the model of this embodiment may be a model in which the same number of probability density estimation models or generative models as the number of constraint combination patterns are combined into one.

[0022] x in Figure 2 i is a communication feature. A communication feature is a feature obtained by converting a communication log or the like into a feature. For example, a communication feature is flow information or a sampled packet itself. The flow information includes an IP address, a port number, a protocol, the number of packets, a communication capacity, etc.

[0023] y in Figure 2 i is a configuration feature. It is a feature obtained by converting information acquired from a device or software that performs communication. For example, the configuration feature is the source code of the program that performs communication, an operation log, setting information, a source port number, etc. In particular, the configuration feature may be an SBOM. Furthermore, the configuration feature functions as a label.

[0024] The model includes a transform encoder, an estimation encoder, and an estimation decoder, all of which may be neural networks.

[0025] The conversion encoder uses the structural feature y i into a format that can be input to the estimation encoder and the estimation decoder. The conversion encoder will be described in detail later.

[0026] The estimation encoder and the estimation decoder constitute an autoencoder. However, the estimation encoder and the estimation decoder in the CVAE include a communication feature x i Not only the structural feature y i is entered.

[0027] Here, VAE and CVAE will be described. VAE is a method for estimating the probability density of high-dimensional data using a deep neural network (DNN). Probability density estimation is a method for estimating the probability density of high-dimensional data using a large amount of given data {x i} n From the distribution p that is thought to have generated the data θ This is a process of estimating (x) (θ is a parameter).

[0028] Furthermore, CVAE is a model that allows labels to be assigned to VAE, and it can be used to generate a large amount of labeled data {x i, y i} n Using this, we can estimate the distribution p θ (x|y) is estimated. Generally, x is a wide variety of data, and y is a fixed low-dimensional data.

[0029] Returning to FIG. 2, the model of this embodiment, which is a CVAE, is i and the structural feature y i p(x i |y i ) is the structural feature y i Communication feature x under i is a numerical value of the probability (likelihood) of

[0030] The estimation device 10 of this embodiment performs model learning and estimation (search) of configuration information using the learned model.

[0031] In training, the model uses communication features x i and the structural feature y i From the likelihood p(x i |y i ) is optimized. That is, through the optimization of θ, the trained model will output a high probability when an appropriate pair of x and y is given.

[0032] On the other hand, in the search, the parameter θ is fixed and the given communication feature x i A structural feature y is searched for that will cause the model to output a high probability for

[0033] The control unit 15 controls the entire estimation device 10. The control unit 15 is, for example, an electronic circuit such as a central processing unit (CPU), a micro processing unit (MPU), or a graphics processing unit (GPU), or an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0034] The control unit 15 also has an internal memory for storing programs that define various processing procedures and control data, and executes each process using the internal memory. The control unit 15 also functions as various processing units when various programs are run. For example, the control unit 15 has a calculation unit 151, a search unit 152, an update unit 153, and an output control unit 154.

[0035] The calculation unit 151 calculates the likelihood regarding the joint distribution of the observed communication features and composition features using a model that outputs likelihood information regarding the joint distribution of communication features that represent the state of communication and composition features that represent the composition of the communication.

[0036] In particular, the calculation unit 151 inputs communication features and configuration features into the model during learning and calculates likelihood. The update unit 153 updates the parameters of the model so that the magnitude of the calculated likelihood satisfies a condition (e.g., so that it is maximized). The update unit 153 updates the model information 141.

[0037] Furthermore, the calculation unit 151 uses a conversion encoder to calculate the structural feature y i is converted into a format that can be input to the CVAE. Usually, CVAE training is performed on a fixed-length vector representation. For example, the structural feature y i When the input source is a source code, the conversion encoder uses CodeBERT or the like to convert the structural feature y i is converted into a fixed-length vector and then input to the estimation encoder or the estimation decoder that constitutes the CVAE. In this case, the conversion encoder converts the source code, which has no fixed length, into a fixed-length abstract representation.

[0038] In addition, the structural feature y i If the dimension of y is huge, the calculation unit 151 applies a dimension reduction method or performs feature embedding, etc., to calculate the component feature y i For example, the dimension of the structural feature y i If the above is a list of OSS (Open Source Software) used by the device, the number of dimensions is the number of all OSSs (or as many as possible) in the world, which can be enormous.

[0039] In addition, the communication feature x i is data with no fixed length, such as a packet itself, the communication feature x i may also be converted into a fixed vector by BERT, Word2Vec, or the like and then input to the estimation encoder.

[0040] The search unit 152 uses a model that outputs likelihood information regarding the joint distribution of communication features that represent the state of communication and configuration features that represent the configuration of communication, and searches for configuration features whose magnitude of likelihood satisfies a condition when input to the model together with the observed communication features.

[0041] The search unit 152 fixes the model parameters and then searches the observed communication feature x i is input to the model, and the configuration feature y is optimized to maximize the likelihood, so that the input communication feature x i For example, the search unit 152 can perform the search using a general optimization method such as Newton's method.

[0042] The output control unit 154 outputs information indicating the structure feature value y obtained by the search performed by the search unit 152 .

[0043] [Processing Flow of First Embodiment] Fig. 3 is a flowchart showing the flow of the learning process. As shown in Fig. 3, the estimation device 10 first inputs a combination of observed communication information and configuration information into a model (step S101). Then, the estimation device 10 calculates likelihood using the model (step S102).

[0044] Here, the estimation device 10 updates the parameters of the model so that the likelihood condition is satisfied (for example, so that the likelihood is maximized) (step S103).

[0045] 4 is a flowchart showing the flow of the search process. As shown in FIG. 4, first, the estimation device 10 inputs observed communication information into a model (step S201).

[0046] The estimation device 10 then searches for configuration information that satisfies the likelihood condition (e.g., maximizes the likelihood) when input to the model (step S202), and then outputs the search result (step S203).

[0047] [Effects of the First Embodiment] As described above, the estimation device 10 includes a search unit 152 and an output control unit 154. The search unit 152 uses a model that outputs likelihood information regarding the joint distribution of communication features that represent the state of communication and configuration features that represent the configuration of communication, and searches for configuration features whose magnitude of likelihood satisfies a condition when input to the model together with observed communication features. For example, the search unit 152 searches for configuration features using a constrained probability density estimation model as the model. The output control unit 154 outputs the configuration features searched for by the search unit 152.

[0048] As a result, if the model has been trained, the estimation device 10 can accurately estimate configuration information of communication-related devices and software.

[0049] For example, consider a case where the configuration feature is a list of OSSs included in software, and the model has already learned the list of OSSs and communication feature values ​​(e.g., flow information) of software that performs normal communication.

[0050] In this case, the estimation device 10 obtains a list of OSS corresponding to the observed communication feature by searching. If the obtained OSS list differs from the OSS list of software performing normal communication, it is detected that the communication that generated the communication feature is abnormal.

[0051] The search unit 152 searches for configuration features using a model that outputs likelihood information regarding the joint distribution of flow information, which is a communication feature, and SBOM, which is a configuration feature. This enables the estimation device 10 to estimate SBOM even if the equipment and software do not have SBOM.

[0052] The calculation unit 151 calculates likelihoods regarding the joint distribution of observed communication features and configuration features using a model that outputs likelihood information regarding the joint distribution of communication features that represent the state of communication and configuration features that represent the configuration of communication. The update unit 153 updates the parameters of the model so that the magnitude of the likelihood satisfies a condition. This allows the estimation device 10 to learn the model.

[0053] [Experiment] An experiment conducted to confirm the effects of this embodiment will be described. In the experiment, the conversion encoder had a part that performed one-hot processing and a one-layer NN. The configuration feature is a list of protocols, i.e., a list of srcports (source port numbers) set in the device. Furthermore, the communication feature is information obtained by excluding srcports from the flow information.

[0054] The model has been trained using the configuration features and communication features of three webcams with communication capabilities and one sensor with communication capabilities. The srcports used by the webcams are "1935", "123", "80", and "443". The srcports used by the sensor are "1883" and "80".

[0055] 5 shows the results when the estimation device 10 inputs the communication features of the RTSP communication used by the webcamera into the trained model and searches for the srcport, which is a component feature. Note that the srcport of the webcamera in RTSP communication is "1935." The "Value" column is the vector searched for as the component feature.

[0056] As shown in Figure 5, the value for port number "1935" is 1, which is the largest, and the values ​​for other port numbers are 0 or nearly 0. The larger the value of the constituent feature, the greater the confidence in the search result. This shows that the estimation device 10 is able to correctly estimate the constituent feature.

[0057] 6 shows the results when the estimation device 10 inputs the communication features of the communication used by the webcam into the trained model and searches for the srcport, which is a configuration feature. Note that the srcports of the webcam in the communication used by the webcam are "1935," "123," "80," and "443."

[0058] In this case too, the values ​​of the elements in the vector of the configuration features corresponding to the port numbers “1935,” “123,” “80,” and “443” are large, which indicates that the estimation device 10 has correctly estimated the configuration features.

[0059] [System Configuration, etc.] The components of each device shown in the figures are conceptual functional units and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown, and all or part of the devices can be functionally or physically distributed or integrated in any unit depending on various loads, usage conditions, etc. Furthermore, all or any part of the processing functions performed by each device can be realized by a CPU and a program analyzed and executed by the CPU, or can be realized as hardware using wired logic.

[0060] Furthermore, among the processes described in this embodiment, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using a known method.In addition, the information including the processing procedures, control procedures, specific names, various data and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified.

[0061] [Program] In one embodiment, the estimation device 10 can be implemented by installing a program that executes the above-described learning process as package software or online software on a desired computer. For example, by causing an information processing device to execute the above-described program, the information processing device can function as the estimation device 10. The information processing device referred to here includes desktop and notebook personal computers. In addition, information processing devices also include mobile communication terminals such as smartphones, mobile phones, and PHS (Personal Handyphone Systems), as well as slate terminals such as PDAs (Personal Digital Assistants).

[0062] The estimation device 10 may also be implemented as a server device that provides a service related to the above processing to a client terminal device used by a user. For example, the server device may be implemented as a server device that receives a domain name to be detected as an input and outputs a determination result as to whether the domain name is malicious. In this case, the server device may be implemented as a web server or as a cloud that provides a service related to the above processing by outsourcing.

[0063] 7 is a diagram showing an example of a computer that executes a program. The computer 1000 includes, for example, a memory 1010 and a CPU 1020. The computer 1000 also includes a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0064] The memory 1010 includes a read-only memory (ROM) 1011 and a random access memory (RAM) 1012. The ROM 1011 stores a boot program such as a basic input / output system (BIOS). The hard disk drive interface 1030 is connected to a hard disk drive 1090. The disk drive interface 1040 is connected to a disk drive 1100. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to a mouse 1110 and a keyboard 1120, for example. The video adapter 1060 is connected to a display 1130, for example.

[0065] The hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. That is, a program that defines each process of the estimation device 10 is implemented as a program module 1093 in which computer-executable code is written. The program module 1093 is stored, for example, in the hard disk drive 1090. For example, the program module 1093 for executing processes similar to those of the functional configuration of the estimation device 10 is stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced with an SSD (Solid State Drive).

[0066] Furthermore, setting data used in the processing of the above-described embodiment is stored as program data 1094, for example, in the memory 1010 or the hard disk drive 1090. Then, the CPU 1020 reads the program module 1093 or the program data 1094 stored in the memory 1010 or the hard disk drive 1090 into the RAM 1012 as necessary, and executes the processing of the above-described embodiment.

[0067] The program module 1093 and program data 1094 may not necessarily be stored in the hard disk drive 1090, but may also be stored in a removable storage medium and read by the CPU 1020 via the disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (such as a local area network (LAN) or a wide area network (WAN)). The program module 1093 and program data 1094 may then be read by the CPU 1020 from the other computer via the network interface 1070.

[0068] REFERENCE SIGNS LIST 10 Estimation device 11 Communication unit 12 Input unit 13 Output unit 14 Storage unit 15 Control unit 141 Model information 151 Calculation unit 152 Search unit 153 Update unit 154 Output control unit

Claims

1. An estimation device comprising: a search unit that uses a model that outputs likelihood information regarding the joint distribution of a communication feature that represents a state of communication and a configuration feature that represents a configuration in the communication, and searches for a configuration feature whose magnitude of the likelihood satisfies a condition when input to the model together with an observed communication feature; and an output control unit that outputs the configuration feature searched for by the search unit.

2. The estimation device according to claim 1, characterized in that the search section searches for the constituent features using a constrained probability density estimation model as the model.

3. The estimation device according to claim 1, characterized in that the search unit searches for the constituent features using the model that outputs information on the likelihood regarding the joint distribution of flow information, which is the communication feature, and SBOM, which is the constituent feature.

4. An estimation device comprising: a calculation unit that calculates a likelihood regarding a joint distribution of observed communication features and configuration features using a model that outputs information on the likelihood regarding the joint distribution of communication features that represent a state of communication and configuration features that represent a configuration in the communication; and an update unit that updates parameters of the model so that the magnitude of the likelihood satisfies a condition.

5. An estimation method, characterized in that a computer executes the following steps: a search process using a model that outputs likelihood information regarding the joint distribution of communication features that represent a state of communication and configuration features that represent a configuration in the communication, to search for configuration features whose magnitude of the likelihood satisfies a condition when input to the model together with the observed communication features; and an output control process that outputs the configuration features found by the search process.

6. An estimation method, characterized in that a computer executes the following steps: a calculation step of calculating a likelihood regarding a joint distribution of observed communication features and configuration features, using a model that outputs information on the likelihood regarding the joint distribution of communication features that represent a state of communication and configuration features that represent a configuration in the communication; and an update step of updating parameters of the model so that the magnitude of the likelihood satisfies a condition.

7. An estimation program that causes a computer to execute the following steps: a search step of using a model that outputs likelihood information regarding the joint distribution of communication features that represent a state of communication and configuration features that represent a configuration in the communication, to search for configuration features whose magnitude of the likelihood satisfies a condition when input to the model together with observed communication features; and an output control step of outputting the configuration features searched for by the search step.

8. An estimation program that causes a computer to execute the following steps: a calculation step of calculating a likelihood regarding a joint distribution of observed communication features and configuration features, using a model that outputs information on the likelihood regarding the joint distribution of communication features that represent a state of communication and configuration features that represent a configuration in the communication; and an update step of updating parameters of the model so that the magnitude of the likelihood satisfies a condition.

Citation Information

Patent Citations

  • Runtime detection of database protocol metadata anomalies in database client connections

    US20220038428A1

  • Learning device, estimation device, learning method, and learning program

    WO2021100184A1