WO2022130460A1 - 学習装置、学習方法、異常検知装置、異常検知方法、及びコンピュータ読み取り可能な記録媒体 - Google Patents
学習装置、学習方法、異常検知装置、異常検知方法、及びコンピュータ読み取り可能な記録媒体 Download PDFInfo
- Publication number
- WO2022130460A1 WO2022130460A1 PCT/JP2020/046553 JP2020046553W WO2022130460A1 WO 2022130460 A1 WO2022130460 A1 WO 2022130460A1 JP 2020046553 W JP2020046553 W JP 2020046553W WO 2022130460 A1 WO2022130460 A1 WO 2022130460A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- feature vector
- data
- mapping
- subspace
- learning
- Prior art date
Links
- 230000005856 abnormality Effects 0.000 title claims description 95
- 238000001514 detection method Methods 0.000 title claims description 93
- 238000000034 method Methods 0.000 title claims description 66
- 239000013598 vector Substances 0.000 claims abstract description 193
- 238000013507 mapping Methods 0.000 claims abstract description 144
- 238000012549 training Methods 0.000 claims abstract description 16
- 230000002159 abnormal effect Effects 0.000 claims description 76
- 230000008569 process Effects 0.000 claims description 27
- 238000003860 storage Methods 0.000 description 19
- 238000010586 diagram Methods 0.000 description 16
- 238000012986 modification Methods 0.000 description 12
- 230000004048 modification Effects 0.000 description 12
- 230000006870 function Effects 0.000 description 10
- 238000004891 communication Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000001052 transient effect Effects 0.000 description 5
- 230000007704 transition Effects 0.000 description 5
- 230000002547 anomalous effect Effects 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000010365 information processing Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 241000196324 Embryophyta Species 0.000 description 1
- 241000282414 Homo sapiens Species 0.000 description 1
- 238000000342 Monte Carlo simulation Methods 0.000 description 1
- 240000007594 Oryza sativa Species 0.000 description 1
- 235000007164 Oryza sativa Nutrition 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000005401 electroluminescence Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000013604 expression vector Substances 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000003973 irrigation Methods 0.000 description 1
- 230000002262 irrigation Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 235000009566 rice Nutrition 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000010865 sewage Substances 0.000 description 1
- 238000005309 stochastic process Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
- H04L63/1408—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic by monitoring network traffic
- H04L63/1425—Traffic logging, e.g. anomaly detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
Definitions
- the present invention relates to an anomaly detection device and anomaly detection method for learning parameters used for mapping, a learning device, a learning method, and detecting anomalies based on the mapping results, and further, a learning device, a learning method, and anomalies.
- the present invention relates to a computer-readable recording medium that records a detection device and a program for realizing an abnormality detection method.
- packets flowing through the control system network are monitored and unauthorized control is performed.
- a technique for detecting abnormal data generated by a procedure is disclosed.
- Non-Patent Document 1 describes that the feature vector of normal data and the feature vector of abnormal data are separated by mapping the feature vector of normal data among the input data inside a hypersphere characterized by a center and a radius.
- the technology to be used is disclosed.
- a neural network is learned by using a deep support vector data description (Deep Support Vector Data Description: Deep S VDD), normal data is stored inside the hypersphere as much as possible, and the volume of the hypersphere is reduced. It is minimized.
- Deep S VDD Deep S VDD
- Non-Patent Document 1 when normal data and abnormal data are mapped by the technique shown in Non-Patent Document 1, a large amount of abnormal data may be mapped inside the hypersphere.
- the system state also includes a transient state in which the system state transitions.
- anomaly detection which learns parameters for mapping so that normal data and abnormal data are accurately separated, accurately detects anomalies based on learning devices, learning methods, and mapping results. It is an object of the present invention to provide an apparatus, an abnormality detection method, and a computer-readable recording medium.
- the learning device in one aspect is It is included in the mapping model for mapping the feature vector generated based on the normal data input as training data to the area set based on the preset subspace and the distance from the subspace. It is characterized by having a learning unit for learning a first parameter for generating the feature vector and a second parameter for adjusting the distance.
- the abnormality detection device on one side is The input data acquired from the target system is input to the mapping model, and the feature vector generated based on the input data is set in the area set based on the preset subspace and the distance from the subspace.
- Mapping part and mapping part A determination unit that determines the feature vector as abnormal based on the result of the mapping, It is characterized by having.
- the learning method in one aspect is It is included in the mapping model for mapping the feature vector generated based on the normal data input as training data to the area set based on the preset subspace and the distance from the subspace. It is characterized by having a learning step of learning a first parameter for generating the feature vector and a second parameter for adjusting the distance.
- the abnormality detection method in one aspect is The input data acquired from the target system is input to the mapping model, and the feature vector generated based on the input data is set in the area set based on the preset subspace and the distance from the subspace.
- a computer-readable recording medium on which a program in one aspect is recorded may be used. It is included in the mapping model for mapping the feature vector generated based on the normal data input as training data to the area set based on the preset subspace and the distance from the subspace. It is characterized by recording a program including an instruction to execute a learning step for learning a first parameter for generating the feature vector and a second parameter for adjusting the distance.
- a computer-readable recording medium on which a program in one aspect is recorded is provided.
- the input data acquired from the target system is input to the mapping model, and the feature vector generated based on the input data is set in the area set based on the preset subspace and the distance from the subspace.
- normal data and abnormal data can be mapped so as to be separated accurately, and abnormalities can be detected accurately based on the mapping results.
- FIG. 1 is a diagram for explaining an example of a learning device.
- FIG. 2 is a diagram for explaining the mapping of feature vectors.
- FIG. 3 is a diagram for explaining an example of a system having an abnormality detection device.
- FIG. 4 is a diagram for explaining an example of the operation of the learning device.
- FIG. 5 is a diagram for explaining an example of the operation of the abnormality detection device.
- FIG. 6 is a diagram for explaining an example of a system having an abnormality detection device.
- FIG. 7 is a diagram for explaining an example of the operation of the abnormality detection device.
- FIG. 8 is a block diagram showing an example of a computer that realizes the learning device and the abnormality detection device according to the first embodiment, the first modification, and the second embodiment.
- the system having the learning device and the anomaly detection device described in the embodiment is used to monitor packets flowing through the network of the control system in order to prevent attacks on the control system. ..
- the learning device generates a model that accurately separates and maps normal data and abnormal data generated by an illegal control procedure.
- the anomaly detection device detects anomalies using the model generated by the learning device.
- Non-Patent Document 1 a neural network that maps different inputs to different points is used.
- mapping to different points is that if mapping to the same point is allowed, the feature vector of normal data and the feature vector of abnormal data may all be mapped to the same point, so abnormal data cannot be detected. Is.
- the input pattern increases according to the system state, so that the number of points to which normal data is mapped increases as the input pattern increases.
- the radius of the hypersphere must be increased in order to accommodate the points of the feature vectors of all the different normal data in the hypersphere.
- Non-Patent Document 1 learning is performed using normal data, but learning is not performed using abnormal data. Therefore, the point corresponding to the feature vector of abnormal data is in the entire space. Evenly distributed.
- the radius of the hypersphere is increased due to (1) and (2), the feature vector of the abnormal data that is evenly distributed in the entire space due to (3) is that of the hypersphere. It is easy to map inside.
- Non-Patent Document 1 when there are a plurality of system states, it becomes difficult to accurately separate normal data and abnormal data even if the technique shown in Non-Patent Document 1 is used.
- the system state also has a transient system state during the state transition period.
- the transient normal data during the state transition and the normal data before and after the state transition are clustered as the same set, so that a single hypersphere also contains multiple system states.
- the radius of the hypersphere becomes large, and it becomes difficult to accurately separate normal data and abnormal data.
- the inventor derives a model that is a meaningful product that cannot be generated by human beings, which accurately separates and maps the feature vector of normal data and the feature vector of abnormal data in the monitoring of the control system. I came to do. As a result, it is possible to accurately detect anomalies occurring in the control system based on the result of feature vector mapping using this model.
- FIG. 1 is a diagram for explaining an example of a learning device.
- the learning device 10 shown in FIG. 1 is a device for learning a model for mapping the feature vectors of normal data and abnormal data acquired from the network of the control system into a subspace. Further, as shown in FIG. 1, the learning device 10 has a learning unit 11 and a selection unit 12.
- the learning device 10 is equipped with, for example, a programmable device such as a CPU (Central Processing Unit) or FPGA (Field-Programmable Gate Array), a GPU (Graphics Processing Unit), or one or more of them.
- Information processing devices such as circuits, server computers, personal computers, and mobile terminals.
- Traffic data Event series
- sensor data time series
- the traffic data and the sensor data may be stored in a storage device such as a database or a server computer by using a data collection device connected to the control system, for example.
- the control system is a system used for public or public facilities such as power plants, power networks, communication networks, roads, railways, ports, airports, water and sewage systems, irrigation facilities, hydraulic facilities, etc.
- the event series represents a series of events that occur when the control system controls the target. That is, the event series represents the order of events that occur when the target is controlled.
- the event is, for example, a control command, a state transition event, a notification event, or the like.
- Traffic data is data that includes a packet and a set of packet reception dates and times.
- the header field of the packet includes, for example, a source / destination MAC (MediaAccessControl) address, an IP (InternetProtocol) address, a port number, a version, and the like.
- the payload of the packet contains, for example, the type of application, associated device ID, control value, state value, and the like.
- the traffic data may include packet statistics.
- the time series represents a series of process values measured by the sensor. That is, the time series represents the order of process values that occur when the target is controlled.
- the process values are, for example, continuous values such as velocity, position, temperature, pressure, and flow velocity, and discrete values representing switch switching. If the process value is controlled by an illegal control procedure, the control system falls into an abnormal state, and the process value also becomes an abnormal value.
- the feature vector can be paraphrased as, for example, a feature quantity, a latent vector, an expression vector, an expression, an embedding, a low-dimensional vector, a mapping to a feature space, a mapping to an expression space, a mapping to a latent space (projection), and the like.
- the learning unit 11 extracts the feature vector of the normal data from the training data and learns the mapping model used to map the feature vector of the normal data to the normal region. After that, the learning unit 11 stores the learned mapping model in the storage device 20.
- the learning unit 11 first acquires the subspace selection information regarding the subspace from the selection unit 12. Next, the learning unit 11 sets the subspaces required for model learning based on the subspace selection information, and ends the preparation for model learning.
- the subspace is, for example, a hyperspherical surface, a quadric hypersurface (for example, a hyperelliptic surface, a hyperboloid surface, etc.), a torus, or a hyperplane.
- a hyperspherical surface for example, a hyperspherical surface, a quadric hypersurface (for example, a hyperelliptic surface, a hyperboloid surface, etc.), a torus, or a hyperplane.
- the subspace may be a part of any one of a hypersphere, a quadric surface, a torus, and a hyperplane.
- the subspace may be a union that combines a plurality of one or more of a hypersphere, a quadric surface, a torus, and a hyperplane.
- the union also includes disjoint unions (direct sums).
- the subspace may be a product set in which a plurality of one or more of a hypersphere, a quadric surface, a torus, and a hyperplane are combined.
- the subspace selection information includes information representing the selected subspace.
- Information that represents the selected subspace includes, for example, the number of dimensions of the selected subspace, the radius of the hypersphere, the coefficients of the quadratic hypersurface, the ellipticity of the hyperelliptic surface, and the affine conversion parameter that specifies the inclination of the hyperplane. Is.
- mapping model for example, a linear model, a neural network, a kernel model, a logistic model, a probability distribution regression, a stochastic process regression, a hierarchical Bayes model, an RNN (Recurrent Neural Network), a Transformer, or the like may be used.
- learning method for example, a generalized inverse matrix, a gradient descent method, a Monte Carlo method, or the like may be used.
- the learning unit 11 acquires normal data input as training data.
- the training data includes, for example, time series, audio, image, video, and relationship data (for example, presence / absence and strength of friendship between people, presence / absence and strength of correlation between data, and inclusion) in addition to event series data. Data such as whether or not there is a relationship) and action history may be used.
- the learning unit 11 inputs the normal data input as training data into the model, generates a feature vector of the normal data, and learns the model for mapping the generated feature vector of the normal data to the normal region. do.
- the learning unit 11 generates a first parameter included in the model used for generating a feature vector and a second parameter used for adjusting the distance from the subspace by learning. ..
- the normal area is an area set based on a preset subspace and a distance from the subspace (distance from a surface), and is obtained by learning.
- FIG. 2 is a diagram for explaining the mapping of feature vectors.
- the conventional hypersphere mapping will be described.
- the feature vector of the normal data is inside the hypersphere 22 of FIG. Not only (black circle: ⁇ ) but also the feature vector of abnormal data (white circle: ⁇ ) is mapped.
- the subspace mapping model 23 shown in FIG. 2 is a model trained using the selected torus when the torus is selected as the subspace. Then, when the input data shown in FIG. 2 is input to the learned subspace mapping model 23, when the input data is normal data, the normal region 24 (near the submanifold) in FIG. 2 is characterized by the normal data. Vectors (black circles: ⁇ ) are mapped. When the input data is abnormal data, the feature vector (white circle: ⁇ ) of the abnormal data is not mapped to the normal region 24.
- mapping model can be represented by a loss function such as Equation 1.
- the subspace is not limited to the hypersphere.
- the learning unit 11 learns to use the first parameter included in the loss function (model) of Equation 1 to generate the feature vector and the second parameter used to adjust the distance from the subspace. Learn with parameters.
- the center point may be set in advance, but the center point may be learned as a third parameter.
- the first parameter used to generate the feature vector the second parameter used to adjust the distance of the normal region from the subspace, and the third parameter to specify a part of the subspace. Since it can be set by learning, the work related to parameter adjustment can be reduced.
- the feature vectors of the normal data are dispersed so as not to be concentrated near the same point in the normal region.
- the feature vectors of normal data can be evenly distributed in the direction along the subspace.
- the feature vector of the normal data can be stored inside a small distance from the subspace, and the volume of the normal region can be reduced by making the normal region very thin. Therefore, the feature vector of the abnormal data can be hard to be confused with the normal region. That is, the feature vector of normal data and the feature vector of abnormal data can be separated accurately.
- the feature vector of normal data is mapped to a normal region with a small volume around a curved subspace such as a hyperspherical surface or a super quadratic curved surface, the relationship between the normal region and the feature vector of normal data is normal. From the relationship between the region and the feature vector of the abnormal data, the feature vector of the normal data in the transient state connecting between the two normal states and the feature vector of the abnormal data located between the two normal states. It becomes easy to separate from.
- mapping of feature vectors of anomalous data located between the two normal states depends on the structure of the mapping model, such as a neural network, but is often two on a curved subspace corresponding to the two normal states. It is mapped on a straight line (geodesic line) connecting points. Therefore, the feature vector of the anomalous data located between the two normal states is not mapped on the curved subspace, but on the outside of the normal region.
- the selection unit 12 selects a subspace as described above.
- the selection unit 12 has, as a subspace, at least a hypersphere, a quadric hypersurface (for example, a hyperelliptical surface, a hyperboloid, etc.), a torus, a hyperplane, a part thereof, a union or a product set thereof. Select one of the above.
- select the subspace for determining the normal area As a selection method, a method of allowing the user to select a subspace by displaying a plurality of subspaces on the screen can be considered.
- a subspace suitable for the control system may be determined in advance by experiments, simulations, machine learning, or the like.
- the selection unit 12 outputs the subspace selection information to the learning unit 11 after any of the subspaces is selected by the user.
- FIG. 3 is a diagram showing an example of a system having an abnormality detection device.
- the system in the first embodiment includes a learning device 10, a storage device 20, an abnormality detection device 30, and an output device 40.
- the abnormality detection device 30 has a mapping unit 31, a determination unit 32, and an output information generation unit 33.
- the abnormality detection device 30 is, for example, a programmable device such as a CPU or FPGA, a GPU, or an information processing device such as a circuit, a server computer, a personal computer, a mobile terminal, or the like equipped with any one or more of them. be.
- the output device 40 acquires the output information described later, which has been converted into an outputable format by the output information generation unit 33, and outputs the generated image, sound, and the like based on the output information.
- the output device 40 is, for example, an image display device using a liquid crystal display, an organic EL (ElectroLuminescence), or a CRT (CathodeRayTube). Further, the image display device may include an audio output device such as a speaker.
- the output device 40 may be a printing device such as a printer.
- the mapping unit 31 inputs the input data acquired from the target control system to the model and maps the feature vector of the input data.
- the mapping unit 31 first acquires input data from a control system or a storage device (not shown).
- the input data includes, for example, time series, audio, image, video, and relationship data (whether or not there is a friendship between people, whether or not there is a correlation between the data, and whether or not there is a correlation, in addition to the event series and time series data. Data such as presence / absence of inclusion relationship) and action history may be used.
- the mapping unit 31 inputs the input data to the mapping model and extracts the feature vector based on the learned mapping model.
- the feature vector is represented by, for example, a set of n (1 or more) real numbers.
- mapping unit 31 outputs the mapping result information representing the mapping result to the determination unit 3.
- the mapping result is, for example, an image as shown in the mapping of the invention of FIG.
- the mapping result information is information having identification information for identifying the feature vector of each input data, mapping position information indicating the position (point) of the feature vector, and distance information indicating the distance between the point and the normal region.
- the determination unit 32 determines that the feature vector is abnormal based on the mapping result. Specifically, the determination unit 32 first acquires the mapping result information from the mapping unit 31.
- the determination unit 32 detects the feature vector mapped outside the normal region based on the mapping result information. Among the feature vectors, the determination unit 32 determines that the feature vector mapped to the normal region is the feature vector of the normal data, and determines that the feature vector mapped outside the normal region is the feature vector of the abnormal data.
- the determination unit 32 outputs the determination result information having the determination result to the output information generation unit 33.
- the determination result information includes, for example, information such as a feature vector of the input data and a determination result indicating whether the input data is normal or abnormal.
- the determination result information may include, for example, a log or the like.
- the determination result may be provided with not only two values of normal and abnormal but also a plurality of abnormal levels.
- the determination unit 32 may output the determination result information to another analysis engine.
- the output information generation unit 33 acquires information such as determination result information and input data, and generates output information converted into a format that can be output to the output device 40.
- the output information is information for causing the output device 40 to output at least the determination result.
- Modification 1 Modification 1 will be described. In the first modification, another determination method of the determination unit 32 will be described.
- the model for mapping the feature vector to the normal region is not necessarily the model learned using the data acquired by operating the control system. Even if the model is trained using the data acquired by operating the control system, there may be a large time difference between the time of learning and the time of operation of using the model. Moreover, overfitting can occur even when there is little time lag.
- the threshold value used to absorb this error is set in advance. Specifically, the determination unit 32 compares a threshold value preset based on the normal region with the distance between the normal region and the feature vector, and determines whether or not the distance is equal to or greater than the threshold value.
- the threshold value may be obtained by experiment or simulation. For example, it is desirable to set the threshold value so that the false positive rate is 1 [%] or less. However, the false positive rate is not limited to 1 [%].
- FIG. 4 is a diagram for explaining an example of the operation of the learning device.
- FIG. 5 is a diagram for explaining an example of the operation of the abnormality detection device. In the following description, the figures will be referred to as appropriate.
- the learning method and the abnormality detection method are implemented by operating the learning device and the abnormality detection device. Therefore, the description of the learning method and the abnormality detection method in the first embodiment is replaced with the following operation description of the learning device and the abnormality detection device.
- the selection unit 12 selects a subspace for determining a normal region (step A1). Specifically, in step A1, the selection unit 12 has at least a supersphere, a quadratic hyperboloid (for example, a hyperboloid, a hyperboloid, etc.), a torus, a hyperplane, a part thereof, and these as subspaces. Select either one of the sum set or the product set of, and output the subspace selection information regarding the subspace.
- a quadratic hyperboloid for example, a hyperboloid, a hyperboloid, etc.
- the learning unit 11 acquires the subspace selection information regarding the subspace from the selection unit 12 (step A2). Next, the learning unit 11 sets the subspace required for model learning based on the subspace selection information, and ends the preparation for model learning (step A3).
- the learning unit 11 acquires normal data input as training data (step A4).
- the learning unit 11 inputs the normal data input as training data into the model, generates a feature vector of the normal data, and learns the model for mapping the generated feature vector of the normal data to the normal region. (Step A5).
- step A5 the learning unit 11 uses the first parameter included in the model for generating the feature vector and the second parameter used for adjusting the distance from the subspace. Generated by learning.
- step A6: Yes when the learning device 10 acquires an instruction to end the learning process (step A6: Yes), the learning device 10 ends the learning process.
- step A6: No the process proceeds to step A1 and the process is continued.
- the mapping unit 31 acquires input data from a control system or a storage device (not shown) (step B1).
- the mapping unit 31 inputs the input data to the mapping model and extracts the feature vector based on the learned mapping model (step B2).
- the feature vector is represented by, for example, a set of n (1 or more) real numbers.
- mapping unit 31 outputs the mapping result information representing the mapping result to the determination unit 3.
- the mapping result is, for example, an image as shown in the mapping of the invention of FIG.
- the determination unit 32 acquires the mapping result information from the mapping unit 31 (step B3). Next, the determination unit 32 detects the feature vector mapped outside the normal region based on the mapping result information (step B4).
- the determination unit 32 determines that the feature vector mapped to the normal region is the feature vector of the normal data, and determines that the feature vector mapped outside the normal region is the feature vector of the abnormal data.
- the determination unit 32 outputs the determination result information having the determination result to the output information generation unit 33.
- the determination unit 32 may determine the feature vector of the normal data and the feature vector of the abnormal data based on the threshold value described in the modification 1.
- the determination result may be provided not only with two values of normal and abnormal, but also with a plurality of abnormal levels.
- the determination unit 32 may output the determination result information to another analysis engine.
- the output information generation unit 33 acquires information such as determination result information and input data, and generates output information converted into a format that can be output to the output device 40 (step B5). Next, the output information generation unit 33 outputs the output information to the output device 40 (step B6).
- step B7: Yes when the abnormality detection device 30 acquires an instruction to end the abnormality detection process (step B7: Yes), the abnormality detection device 30 ends the abnormality detection process.
- step B7: No the process proceeds to step B1 and the process is continued.
- the first and second parameters and the third parameter can be set by learning, so that the work related to the parameter adjustment can be reduced.
- the feature vectors of the normal data are dispersed so as not to be concentrated near the same point in the normal region.
- the feature vectors of normal data can be evenly distributed in the direction along the subspace.
- the feature vector of the normal data can be stored inside a small distance from the subspace, and the volume of the normal region can be reduced by making the normal region very thin. Therefore, the feature vector of the abnormal data can be hard to be confused with the normal region. That is, the feature vector of normal data and the feature vector of abnormal data can be separated accurately.
- the feature vector of normal data is mapped to a normal region with a small volume around a curved subspace such as a hyperspherical surface or a super quadratic curved surface, the relationship between the normal region and the feature vector of normal data is normal. From the relationship between the region and the feature vector of the abnormal data, the feature vector of the normal data in the transient state connecting between the two normal states and the feature vector of the abnormal data located between the two normal states. It becomes easy to separate from.
- the program in the first embodiment and the first modification of the present invention may be a program that causes a computer to execute steps A1 to A6 shown in FIG. 4 and a program that causes a computer to execute steps B1 to B7 shown in FIG.
- the learning device and learning method By installing this program on a computer and executing it, the learning device and learning method, and the abnormality detection device and abnormality detection method in this embodiment can be realized.
- the computer processor functions as a learning unit 11, a selection unit 12, a mapping unit 31, a determination unit 32, and an output information generation unit 33, and performs processing.
- each computer may function as any of a learning unit 11, a selection unit 12, a mapping unit 31, a determination unit 32, and an output information generation unit 33, respectively.
- FIG. 6 is a diagram showing an example of a system having an abnormality detection device.
- an autoencoder is used for abnormality detection.
- the system according to the second embodiment includes an abnormality detection device 70, a learning device 10, a storage device 20, and an output device 40.
- the abnormality detection device 70 includes a mapping unit 31, an output information generation unit 33, a determination unit 71, and an autoencoder 72.
- the determination unit 71 determines the abnormality of the feature vector by using the reconstruction error in addition to the mapping result.
- the determination unit 71 first acquires mapping result information from the mapping unit 31. Next, the determination unit 71 inputs the feature vector of the input data to the autoencoder 72 and acquires the reconstructed data corresponding to the input data.
- the determination unit 71 generates reconstruction error information representing the difference between the input data and the data corresponding to the input data reconstructed from the feature vector of the input data.
- Reconstruction error information is output as one or more real values by, for example, calculating the squared error or cross entropy.
- the determination unit 71 determines whether the input data is normal or abnormal based on the mapping result, as in the determination unit 32 described above (see the first embodiment and the first modification). One judgment). Further, the determination unit 71 determines whether the input data is normal or abnormal according to the difference included in the reconstruction error information (second determination).
- the determination unit 71 determines that the input data is normal when both the first determination and the second determination are normal. If both the first determination and the second determination are abnormal, it is determined that the input data is abnormal. Further, when either the first determination or the second determination is abnormal, the determination unit 71 determines that the input data is abnormal.
- the determination unit 71 determines the distance between the feature vector of the input data and the subspace in the normal region based on the mapping result, as in the determination unit 32 described above (see the first embodiment and the first modification). Calculate the weighted sum with the difference included in the reconstruction error information. The weighted sum represents the degree of abnormality of the input data.
- the determination unit 71 sets an abnormality determination threshold value for the weighted sum in advance, as in the determination unit 32 described above, and determines that the input data is normal when the weighted sum is below the threshold value. Further, the determination unit 71 determines that the input data is abnormal when the weighted sum exceeds the threshold value.
- the determination unit 71 outputs the determination result information having the determination result to the output information generation unit 33.
- the autoencoder 72 learns by inputting the feature vector of normal data in the learning phase. Further, the parameters generated by the learning of the autoencoder 72 may be stored in a storage device provided in the abnormality detection device 70, or may be stored in a storage device other than the abnormality detection device 70.
- the autoencoder 72 When the autoencoder 72 is learned using the feature vector of normal data, if the input data is normal data, the autoencoder 72 can restore the input data. On the other hand, when abnormal data is input to the autoencoder 72, the autoencoder 72 cannot reflect the feature vector of the abnormal data.
- the input data and the output data of the autoencoder 72 are compared, and if there is a large difference, it can be determined that the input data has abnormal data.
- mapping model and the learning of the autoencoder 72 may be performed in parallel or separately.
- FIG. 7 is a diagram for explaining an example of the operation of the abnormality detection device. In the following description, the figures will be referred to as appropriate. Further, in the second embodiment, the abnormality detection method is implemented by operating the abnormality detection device. Therefore, the description of the abnormality detection method in the second embodiment is replaced with the following operation description of the abnormality detection device.
- the mapping unit 31 acquires input data from a control system or a storage device (not shown) (step B1). Next, the mapping unit 31 inputs the input data to the mapping model and extracts the feature vector based on the learned mapping model (step B2). Next, the mapping unit 31 outputs the mapping result information representing the mapping result to the determination unit 71.
- the determination unit 71 acquires the mapping result information from the mapping unit 31 (step B3).
- the determination unit 71 detects the feature vector mapped outside the normal region based on the mapping result information (step B4). Alternatively, the distance from the subspace in the normal region to the feature vector is calculated.
- the determination unit 71 determines whether the input data is normal or abnormal based on the mapping result (first determination). ). The determination unit 71 outputs the determination result information having the determination result to the output information generation unit 33.
- the determination unit 71 inputs the feature vector of the input data to the autoencoder 72 and acquires the reconstructed data corresponding to the input data (step C1).
- the determination unit 71 generates reconstruction error information representing the difference between the input data and the data corresponding to the input data reconstructed from the feature vector of the input data (step C2).
- the determination unit 71 further determines whether the input data is normal or abnormal according to the difference included in the reconstruction error information (second determination) (step C3).
- the determination unit 71 determines that the input data is normal when both the first determination and the second determination are normal (step C4). If both the first determination and the second determination are abnormal, it is determined that the input data is abnormal. Further, when either the first determination or the second determination is abnormal, the determination unit 71 determines that the input data is abnormal.
- the determination unit 71 represents a reconstruction error representing the difference between the distance from the subspace in the normal region to the feature vector of the input data and the data corresponding to the input data reconstructed from the feature vector of the input data. Calculate the weighted sum of the information. Further, when the weighted sum exceeds a predetermined threshold value, the input data is determined to be abnormal.
- the determination unit 71 outputs the determination result information having the determination result to the output information generation unit 33.
- the output information generation unit 33 acquires information such as determination result information and input data, and generates output information converted into a format that can be output to the output device 40 (step B5). Next, the output information generation unit 33 outputs the output information to the output device 40 (step B6).
- step B7: Yes when the abnormality detection device 30 acquires an instruction to end the abnormality detection process (step B7: Yes), the abnormality detection device 30 ends the abnormality detection process.
- step B7: No the process proceeds to step B1 and the process is continued.
- the program according to the second embodiment of the present invention may be any program that causes a computer to execute steps B1 to B4, C1 to C4, and B5 to B7 shown in FIG. 7.
- the computer processor functions as a mapping unit 31, a determination unit 71, an output information generation unit 33, and an autoencoder 72, and performs processing.
- each computer may function as any of a mapping unit 31, a determination unit 71, an output information generation unit 33, and an autoencoder 72, respectively.
- FIG. 8 is a block diagram showing an example of a computer that realizes the learning device and the abnormality detection device according to the first embodiment, the first modification, and the second embodiment.
- the computer 110 includes a CPU (Central Processing Unit) 111, a main memory 112, a storage device 113, an input interface 114, a display controller 115, a data reader / writer 116, and a communication interface 117. And. Each of these parts is connected to each other via a bus 121 so as to be capable of data communication.
- the computer 110 may include a GPU or FPGA in addition to or in place of the CPU 111.
- the CPU 111 expands the program (code) in the present embodiment stored in the storage device 113 into the main memory 112, and executes these in a predetermined order to perform various operations.
- the main memory 112 is typically a volatile storage device such as a DRAM (Dynamic Random Access Memory).
- the program in the present embodiment is provided in a state of being stored in a computer-readable recording medium 120.
- the program in the present embodiment may be distributed on the Internet connected via the communication interface 117.
- the recording medium 120 is a non-volatile recording medium.
- the storage device 113 include a semiconductor storage device such as a flash memory in addition to a hard disk drive.
- the input interface 114 mediates data transmission between the CPU 111 and an input device 118 such as a keyboard and mouse.
- the display controller 115 is connected to the display device 119 and controls the display on the display device 119.
- the data reader / writer 116 mediates the data transmission between the CPU 111 and the recording medium 120, reads the program from the recording medium 120, and writes the processing result in the computer 110 to the recording medium 120.
- the communication interface 117 mediates data transmission between the CPU 111 and another computer.
- the recording medium 120 include a general-purpose semiconductor storage device such as CF (CompactFlash (registered trademark)) and SD (SecureDigital), a magnetic recording medium such as a flexible disk, or a CD-.
- CF CompactFlash (registered trademark)
- SD Secure Digital
- magnetic recording medium such as a flexible disk
- CD- CompactDiskReadOnlyMemory
- optical recording media such as ROM (CompactDiskReadOnlyMemory).
- the learning device and the abnormality detection device in the present embodiment can also be realized by using the hardware corresponding to each part instead of the computer in which the program is installed. Further, the learning device and the abnormality detection device may be partially realized by a program and the rest may be realized by hardware.
- mapping model It is included in the mapping model for mapping the feature vector generated based on the normal data input as training data to the area set based on the preset subspace and the distance from the subspace.
- a learning device having a learning unit that learns a first parameter for generating the feature vector and a second parameter for adjusting the distance.
- Appendix 2 The learning device according to Appendix 1, As the subspace, it has a selection unit that selects at least one of a supersphere, a superelliptical surface, a hyperboloid, a torus, a hyperplane, a part thereof, and a union or product set thereof. Learning device.
- Appendix 3 The learning device according to Appendix 1 or 2.
- a learning device having an autoencoder that inputs a feature vector of the normal data and reconstructs the input data corresponding to the feature vector.
- mapping model The input data acquired from the target system is input to the mapping model, and the feature vector generated based on the input data is set in the area set based on the preset subspace and the distance from the subspace.
- mapping part and mapping part A determination unit that determines the feature vector as abnormal based on the result of the mapping, Anomaly detection device with.
- the abnormality detection device described in Appendix 4 is an abnormality detection device that determines a feature vector mapped outside the region as an abnormality.
- the abnormality detection device according to Appendix 4 or 5. It has an autoencoder that inputs the feature vector of the normal data and reconstructs the input data corresponding to the feature vector.
- the determination unit calculates a reconstruction error representing the difference between the input data and the data reconstructed by inputting the feature vector of the input data into the autoencoder, and the mapping result and the reconstruction error.
- An abnormality detection device that determines an abnormality in the feature vector based on the above.
- the abnormality detection device according to any one of Supplementary note 4 to 6.
- the input data is an abnormality detection device including any one of network traffic data in the system and sensor data output from the sensor.
- mapping model for mapping the feature vector generated based on the normal data input as training data to the area set based on the preset subspace and the distance from the subspace.
- a learning method having a learning step of learning a first parameter for generating the feature vector and a second parameter for adjusting the distance.
- Appendix 9 The learning method described in Appendix 8 As the subspace, it has a selection step of selecting at least one of a supersphere, a superelliptic surface, a hyperboloid, a torus, a hyperplane, a part thereof, and a union or product set thereof. Learning method.
- Appendix 10 The learning method described in Appendix 8 or 9, A learning method having an auto-encoding step in which a feature vector of the normal data is input and the input data corresponding to the feature vector is reconstructed.
- mapping model 11 The input data acquired from the target system is input to the mapping model, and the feature vector generated based on the input data is set in the area set based on the preset subspace and the distance from the subspace. Mapping, mapping steps, A determination step in which the feature vector is determined to be abnormal based on the result of the mapping, and Anomaly detection method with.
- Appendix 12 The abnormality detection method described in Appendix 11 An abnormality detection method for determining a feature vector mapped outside the region as an abnormality in the determination step.
- Appendix 13 The abnormality detection method according to Appendix 11 or 12.
- An auto-encoding step that inputs the feature vector of the normal data and reconstructs the input data corresponding to the feature vector.
- a reconstruction error representing the difference between the input data and the data reconstructed by inputting the feature vector of the input data is calculated, and the feature vector is calculated based on the reconstruction error due to the reconstruction.
- the abnormality detection method according to any one of the appendices 11 to 13.
- the input data is an abnormality detection method including any one of network traffic data in the system and sensor data output from a sensor.
- mapping model for mapping the feature vector generated based on the normal data input as training data to the area set based on the preset subspace and the distance from the subspace.
- a computer-readable record recording the program, including instructions to perform a learning step, learning a first parameter for generating the feature vector and a second parameter for adjusting the distance.
- Appendix 16 The computer-readable recording medium according to Appendix 15, wherein the recording medium is readable.
- the program is on the computer Execute a selection step to select at least one of a supersphere, a superelliptical surface, a hyperboloid, a torus, a hyperplane, a part thereof, and a sum set or a product set thereof as the subspace.
- a computer-readable recording medium recording a program, including instructions to cause it.
- Appendix 17 A computer-readable recording medium according to Appendix 15 or 16.
- the program is on the computer A computer-readable recording medium recording a program, including an instruction to execute an autoencoding step that inputs a feature vector of the normal data and reconstructs the input data corresponding to the feature vector.
- mapping model On the computer
- the input data acquired from the target system is input to the mapping model, and the feature vector generated based on the input data is set in the area set based on the preset subspace and the distance from the subspace.
- Appendix 19 The computer-readable recording medium according to Appendix 18, wherein the recording medium is readable.
- Appendix 20 A computer-readable recording medium according to Appendix 18 or 19.
- the program is on the computer It contains an instruction to execute an auto-encode step that inputs a feature vector of normal data and reconstructs the input data corresponding to the feature vector.
- a reconstruction error representing the difference between the input data and the data reconstructed by inputting the feature vector of the input data is calculated, and based on the mapping result and the reconstruction error,
- the computer-readable recording medium according to any one of the appendices 18 to 20.
- the input data is a computer-readable recording medium including any one of network traffic data in the system and sensor data output from a sensor.
- normal data and abnormal data can be mapped so as to be accurately separated, and an abnormality can be detected accurately based on the mapping result.
- the present invention is useful in fields where control system monitoring is required.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
Description
あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する、学習部
を有することを特徴とする。
対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングする、マッピング部と、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する、判定部と、
を有することを特徴とする。
あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する、学習ステップ
を有することを特徴とする。
対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングする、マッピングステップと、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する、判定ステップと、
を有することを特徴とする。
あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する、学習ステップ
を実行させる命令を含むプログラムを記録していることを特徴とする。
対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングする、マッピングステップと、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する、判定ステップと、
を実行させる命令を含むプログラムを記録していることを特徴とする。
実施形態で説明する学習装置と異常検知装置を有するシステム(同一の技術分野に属するシステム)は、制御システムへの攻撃を防止するために、制御システムのネットワークを流れるパケットを監視するために用いられる。
図1を用いて、本実施形態1における学習装置の構成について説明する。図1は、学習装置の一例を説明するための図である。
図1に示す学習装置10は、制御システムのネットワークから取得した正常データと異常データの特徴ベクトルを、部分空間にマッピングをするためのモデルを学習する装置である。また、図1に示すように、学習装置10は、学習部11と、選択部12とを有する。
図2は、特徴ベクトルのマッピングを説明するための図である。まず、従来の超球体マッピングについて説明する。非特許文献1に示されているような、超球体のマッピングモデル21に、図2に示す入力データ(トラフィックデータ)を入力すると、図2の超球体22の内部には、正常データの特徴ベクトル(黒丸:●)だけでなく、異常データの特徴ベクトル(白丸:○)もマッピングされる。
例えば、部分空間として超球面が選択された場合、モデルは、数1のような損失関数により表すことができる。ただし、部分空間は超球面に限定されるものではない。
続いて、図3を用いて、本実施形態1における異常検知装置30の構成を具体的に説明する。図3は、異常検知装置を有するシステムの一例を示す図である。
学習装置10、記憶装置20については、既に説明をしたので説明を省略する。
マッピング部31は、対象の制御システムから取得した入力データをモデルに入力し、入力データの特徴ベクトルをマッピングする。
変形例1について説明する。変形例1では、判定部32の他の判定方法について説明する。
次に、実施形態1における学習装置と異常検知装置の動作について、図4、図5を用いて説明する。図4は、学習装置の動作の一例を説明するための図である。図5は、異常検知装置の動作の一例を説明するための図である。以下の説明においては、適宜図を参照する。また、本実施形態1では、学習装置と異常検知装置を動作させることによって、学習法と異常検知方法が実施される。よって、本実施形態1における学習方法と異常検知方法の説明は、以下の学習装置と異常検知装置の動作説明に代える。
図4に示すように、選択部12は、正常領域を決定するための部分空間を選択する(ステップA1)。具体的には、ステップA1において、選択部12は、部分空間として、少なくとも超球面、二次超曲面(例えば、超楕円面、超双曲面など)、トーラス、超平面、これらの一部、これらの和集合又は積集合、のうちのいずれか一つを選択し、部分空間に関する部分空間選択情報を出力する。
図5に示すように、マッピング部31は、制御システム又は記憶装置(不図示)から入力データを取得する(ステップB1)。
以上のように実施形態1、変形例1によれば、第一、第二のパラメータ及び第三のパラメータを学習により設定できるので、パラメータの調整に係る作業を削減できる。
本発明の実施形態1、変形例1におけるプログラムは、コンピュータに、図4に示すステップA1からA6を実行させるプログラム、図5に示すステップB1からB7を実行させるプログラムであればよい。
図6を用いて、実施形態2における異常検知装置の構成について説明する。図6は、異常検知装置を有するシステムの一例を示す図である。実施形態2では、異常検知にオートエンコーダを用いた例について説明する。
図6に示すように、実施形態2におけるシステムは、異常検知装置70と、学習装置10と、記憶装置20と、出力装置40とを有する。異常検知装置70は、マッピング部31と、出力情報生成部33と、判定部71と、オートエンコーダ72とを有する。
判定部71は、マッピングの結果に加え、再構成誤差を用いて、特徴ベクトルの異常を判別する。
次に、本発明の実施形態2における異常検知装置の動作について図7を用いて説明する。図7は、異常検知装置の動作の一例を説明するための図である。以下の説明においては、適宜図を参照する。また、本実施形態2では、異常検知装置を動作させることによって、異常検知方法が実施される。よって、本実施形態2における異常検知方法の説明は、以下の異常検知装置の動作説明に代える。
以上のように本実施形態2によれば、更に、実施形態1よりも異常検知の精度を向上させることができる。
本発明の実施形態2におけるプログラムは、コンピュータに、図7に示すステップB1からB4、C1からC4、B5からB7を実行させるプログラムであればよい。このプログラムをコンピュータにインストールし、実行することによって、本実施形態における異常検知装置と異常検知方法を実現することができる。この場合、コンピュータのプロセッサは、マッピング部31、判定部71、出力情報生成部33、オートエンコーダ72として機能し、処理を行なう。
ここで、実施形態1、変形例1、実施形態2におけるプログラムを実行することによって、学習装置及び異常検知装置を実現するコンピュータについて図8を用いて説明する。図8は、実施形態1、変形例1、実施形態2における学習装置及び異常検知装置を実現するコンピュータの一例を示すブロック図である。
以上の実施形態に関し、更に以下の付記を開示する。上述した実施形態の一部又は全部は、以下に記載する(付記1)から(付記21)により表現することができるが、以下の記載に限定されるものではない。
あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する、学習部
を有する学習装置。
付記1に記載の学習装置であって、
前記部分空間として、少なくとも超球面、超楕円面、超双曲面、トーラス、超平面、これらの一部、これらの和集合又は積集合、のうちのいずれか一つを選択する、選択部
を有する学習装置。
付記1又は2に記載の学習装置であって、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコーダ
を有する学習装置。
対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングする、マッピング部と、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する、判定部と、
を有する異常検知装置。
付記4に記載の異常検知装置であって、
前記判定部は、前記領域外にマッピングされた特徴ベクトルを異常と判定する
異常検知装置。
付記4又は5に記載の異常検知装置であって、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコーダを有し、
前記判定部は、前記入力データと、前記オートエンコーダに前記入力データの特徴ベクトルを入力して再構成したデータとの差分を表す再構成誤差を算出し、前記マッピングの結果と前記再構成誤差とに基づいて、前記特徴ベクトルの異常を判別する
異常検知装置。
付記4から6のいずれか一つに記載の異常検知装置であって、
前記入力データは、前記システムにおけるネットワークのトラフィックデータ、センサから出力されるセンサデータのうちのいずれか一つを含む
異常検知装置。
あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する、学習ステップ
を有する学習方法。
付記8に記載の学習方法であって、
前記部分空間として、少なくとも超球面、超楕円面、超双曲面、トーラス、超平面、これらの一部、これらの和集合又は積集合、のうちのいずれか一つを選択する、選択ステップ
を有する学習方法。
付記8又は9に記載の学習方法であって、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコードステップ
を有する学習方法。
対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングする、マッピングステップと、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する、判定ステップと、
を有する異常検知方法。
付記11に記載の異常検知方法であって、
前記判定ステップにおいて、前記領域外にマッピングされた特徴ベクトルを異常と判定する
異常検知方法。
付記11又は12に記載の異常検知方法であって、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコードステップと、
前記判定ステップにおいて、前記入力データと、前記入力データの特徴ベクトルを入力して再構成したデータとの差分を表す再構成誤差を算出し、前記再構成による再構成誤差に基づいて、前記特徴ベクトルの異常を判別する
異常検知方法。
付記11から13のいずれか一つに記載の異常検知方法であって、
前記入力データは、前記システムにおけるネットワークのトラフィックデータ、センサから出力されるセンサデータのうちのいずれか一つを含む
異常検知方法。
コンピュータに、
あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する、学習ステップ
を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。
付記15に記載のコンピュータ読み取り可能な記録媒体であって、
前記プログラムが、前記コンピュータに、
前記部分空間として、少なくとも超球面、超楕円面、超双曲面、トーラス、超平面、これらの一部、これらの和集合又は積集合、のうちのいずれか一つを選択する、選択ステップ
を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。
付記15又は16に記載のコンピュータ読み取り可能な記録媒体であって、
前記プログラムが、前記コンピュータに、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコードステップ
を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。
コンピュータに、
対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングする、マッピングステップと、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する、判定ステップと、
を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。
付記18に記載のコンピュータ読み取り可能な記録媒体であって、
前記判定ステップにおいて、前記領域外にマッピングされた特徴ベクトルを異常と判定する
コンピュータ読み取り可能な記録媒体。
付記18又は19に記載のコンピュータ読み取り可能な記録媒体であって、
前記プログラムが、前記コンピュータに、
正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコードステップを実行させる命令を含み、
前記判定ステップにおいて、前記入力データと、前記入力データの特徴ベクトルを入力して再構成したデータとの差分を表す再構成誤差を算出し、前記マッピングの結果と前記再構成誤差とに基づいて、前記特徴ベクトルの異常を判別する
コンピュータ読み取り可能な記録媒体。
付記18から20のいずれか一つに記載のコンピュータ読み取り可能な記録媒体であって、
前記入力データは、前記システムにおけるネットワークのトラフィックデータ、センサから出力されるセンサデータのうちのいずれか一つを含む
コンピュータ読み取り可能な記録媒体。
11 学習部
12 選択部
20 記憶装置
30 異常検知装置
31 マッピング部
32 判定部
33 出力情報生成部
40 出力装置
70 異常検知装置
71 判定部
72 オートエンコーダ
110 コンピュータ
111 CPU
112 メインメモリ
113 記憶装置
114 入力インターフェイス
115 表示コントローラ
116 データリーダ/ライタ
117 通信インターフェイス
118 入力機器
119 ディスプレイ装置
120 記録媒体
121 バス
Claims (21)
- あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する、学習手段
を有する学習装置。 - 請求項1に記載の学習装置であって、
前記部分空間として、少なくとも超球面、超楕円面、超双曲面、トーラス、超平面、これらの一部、これらの和集合又は積集合、のうちのいずれか一つを選択する、選択手段
を有する学習装置。 - 請求項1又は2に記載の学習装置であって、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコーダ
を有する学習装置。 - 対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングする、マッピング手段と、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する、判定手段と、
を有する異常検知装置。 - 請求項4に記載の異常検知装置であって、
前記判定手段は、前記領域外にマッピングされた特徴ベクトルを異常と判定する
異常検知装置。 - 請求項4又は5に記載の異常検知装置であって、
正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する、オートエンコーダを有し、
前記判定手段は、前記入力データと、前記オートエンコーダに前記入力データの特徴ベクトルを入力して再構成したデータとの差分を表す再構成誤差を算出し、前記マッピングの結果と前記再構成誤差とに基づいて、前記特徴ベクトルの異常を判別する
異常検知装置。 - 請求項4から6のいずれか一つに記載の異常検知装置であって、
前記入力データは、前記システムにおけるネットワークのトラフィックデータ、センサから出力されるセンサデータのうちのいずれか一つを含む
異常検知装置。 - あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する
学習方法。 - 請求項8に記載の学習方法であって、
前記部分空間として、少なくとも超球面、超楕円面、超双曲面、トーラス、超平面、これらの一部、これらの和集合又は積集合、のうちのいずれか一つを選択する
学習方法。 - 請求項8又は9に記載の学習方法であって、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する
学習方法。 - 対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングし、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する
異常検知方法。 - 請求項11に記載の異常検知方法であって、
前記判定において、前記領域外にマッピングされた特徴ベクトルを異常と判定する
異常検知方法。 - 請求項11又は12に記載の異常検知方法であって、
正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成し、
前記判定において、前記入力データと、前記入力データの特徴ベクトルを入力して再構成したデータとの差分を表す再構成誤差を算出し、前記再構成による再構成誤差に基づいて、前記特徴ベクトルの異常を判別する
異常検知方法。 - 請求項11から13のいずれか一つに記載の異常検知方法であって、
前記入力データは、前記システムにおけるネットワークのトラフィックデータ、センサから出力されるセンサデータのうちのいずれか一つを含む
異常検知方法。 - コンピュータに、
あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、訓練データとして入力された正常データに基づいて生成された特徴ベクトルをマッピングするためのマッピングモデルに含まれる、前記特徴ベクトルを生成するための第一のパラメータと、前記距離を調整するための第二のパラメータとを学習する
処理を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。 - 請求項15に記載のコンピュータ読み取り可能な記録媒体であって、
前記プログラムが、前記コンピュータに、
前記部分空間として、少なくとも超球面、超楕円面、超双曲面、トーラス、超平面、これらの一部、これらの和集合又は積集合、のうちのいずれか一つを選択する
処理を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。 - 請求項15又は16に記載のコンピュータ読み取り可能な記録媒体であって、
前記プログラムが、前記コンピュータに、
前記正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する
処理を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。 - コンピュータに、
対象のシステムから取得した入力データをマッピングモデルに入力し、あらかじめ設定された部分空間と前記部分空間からの距離とに基づいて設定される領域に、前記入力データに基づいて生成された特徴ベクトルをマッピングし、
前記マッピングの結果に基づいて特徴ベクトルを異常と判定する
処理を実行させる命令を含む、プログラムを記録しているコンピュータ読み取り可能な記録媒体。 - 請求項18に記載のコンピュータ読み取り可能な記録媒体であって、
前記領域外にマッピングされた特徴ベクトルを異常と判定する
コンピュータ読み取り可能な記録媒体。 - 請求項18又は19に記載のコンピュータ読み取り可能な記録媒体であって、
前記プログラムが、前記コンピュータに、
正常データの特徴ベクトルを入力し、当該特徴ベクトルに対応する入力データを再構成する処理を実行させる命令を含み、
前記入力データと、前記入力データの特徴ベクトルを入力して再構成したデータとの差分を表す再構成誤差を算出し、前記マッピングの結果と前記再構成誤差とに基づいて、前記特徴ベクトルの異常を判別する
コンピュータ読み取り可能な記録媒体。 - 請求項18から20のいずれか一つに記載のコンピュータ読み取り可能な記録媒体であって、
前記入力データは、前記システムにおけるネットワークのトラフィックデータ、センサから出力されるセンサデータのうちのいずれか一つを含む
コンピュータ読み取り可能な記録媒体。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/046553 WO2022130460A1 (ja) | 2020-12-14 | 2020-12-14 | 学習装置、学習方法、異常検知装置、異常検知方法、及びコンピュータ読み取り可能な記録媒体 |
US18/265,346 US20240039940A1 (en) | 2020-12-14 | 2020-12-14 | Learning apparatus, learning method, anomaly detection apparatus, anomaly detection method, and computer-readable recording medium |
JP2022569335A JPWO2022130460A5 (ja) | 2020-12-14 | 学習装置、学習方法、異常検知装置、異常検知方法、及びプログラム |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/046553 WO2022130460A1 (ja) | 2020-12-14 | 2020-12-14 | 学習装置、学習方法、異常検知装置、異常検知方法、及びコンピュータ読み取り可能な記録媒体 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022130460A1 true WO2022130460A1 (ja) | 2022-06-23 |
Family
ID=82057403
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2020/046553 WO2022130460A1 (ja) | 2020-12-14 | 2020-12-14 | 学習装置、学習方法、異常検知装置、異常検知方法、及びコンピュータ読み取り可能な記録媒体 |
Country Status (2)
Country | Link |
---|---|
US (1) | US20240039940A1 (ja) |
WO (1) | WO2022130460A1 (ja) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002190025A (ja) * | 2000-12-22 | 2002-07-05 | Fuji Xerox Co Ltd | パターン識別方法および装置 |
JP2016049055A (ja) * | 2014-08-29 | 2016-04-11 | 株式会社エヌテック | 微生物検査装置の検証方法、微生物検査装置における検証装置及びプログラム |
JP2018049355A (ja) * | 2016-09-20 | 2018-03-29 | 株式会社東芝 | 異常検知装置、学習装置、異常検知方法、学習方法、異常検知プログラム、および学習プログラム |
-
2020
- 2020-12-14 US US18/265,346 patent/US20240039940A1/en active Pending
- 2020-12-14 WO PCT/JP2020/046553 patent/WO2022130460A1/ja active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002190025A (ja) * | 2000-12-22 | 2002-07-05 | Fuji Xerox Co Ltd | パターン識別方法および装置 |
JP2016049055A (ja) * | 2014-08-29 | 2016-04-11 | 株式会社エヌテック | 微生物検査装置の検証方法、微生物検査装置における検証装置及びプログラム |
JP2018049355A (ja) * | 2016-09-20 | 2018-03-29 | 株式会社東芝 | 異常検知装置、学習装置、異常検知方法、学習方法、異常検知プログラム、および学習プログラム |
Non-Patent Citations (1)
Title |
---|
CHALAPATHY, RAGHAVENDRA ET AL.: "Anomaly Detection using One-Class Neural Networks", ARXIV, 11 January 2019 (2019-01-11), XP055945722, Retrieved from the Internet <URL:https://arxiv.org/pdf/1802.06360v2.pdf> [retrieved on 20210309] * |
Also Published As
Publication number | Publication date |
---|---|
JPWO2022130460A1 (ja) | 2022-06-23 |
US20240039940A1 (en) | 2024-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP7240071B2 (ja) | 無人航空機のサイバー攻撃検出、位置特定、および中和 | |
JP6757978B2 (ja) | 異常トラフィックを探知する方法および装置 | |
Sayed et al. | Deep and transfer learning for building occupancy detection: A review and comparative analysis | |
US11693924B2 (en) | System and method for maintenance recommendation in industrial networks | |
CN110853033B (zh) | 基于帧间相似度的视频检测方法和装置 | |
US11687787B2 (en) | Learning and detection method of neural network model for flame determination, and apparatus performing the same | |
TW201832138A (zh) | 圖像識別方法及裝置 | |
CN111652290B (zh) | 一种对抗样本的检测方法及装置 | |
Samek et al. | The convergence of machine learning and communications | |
US20080201278A1 (en) | Method and Apparatus for Automatic Online Detection and Classification of Anomalous Objects in a Data Stream | |
US20210037044A1 (en) | Resilient estimation for grid situational awareness | |
Chen et al. | Deep learning regressors for quantitative steganalysis | |
US10909460B2 (en) | Intelligent data curation | |
Bodström et al. | State of the art literature review on network anomaly detection with deep learning | |
CN114330544A (zh) | 一种业务流量异常检测模型建立方法及异常检测方法 | |
Kodali et al. | The value of summary statistics for anomaly detection in temporally evolving networks: A performance evaluation study | |
CN110618854B (zh) | 基于深度学习与内存镜像分析的虚机行为分析系统 | |
WO2022130460A1 (ja) | 学習装置、学習方法、異常検知装置、異常検知方法、及びコンピュータ読み取り可能な記録媒体 | |
Wang et al. | Flowadgan: Adversarial learning for deep anomaly network intrusion detection | |
CN115567224A (zh) | 一种用于检测区块链交易异常的方法及相关产品 | |
Kim et al. | Revitalizing self-organizing map: Anomaly detection using forecasting error patterns | |
Díaz et al. | Learning latent functions for causal discovery | |
WO2021152689A1 (ja) | イベント分析支援装置、方法及び記録媒体 | |
WO2023156251A1 (en) | Anomaly detection | |
Tamboli et al. | Various techniques used in building intrusion detection system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20965848 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18265346 Country of ref document: US |
|
ENP | Entry into the national phase |
Ref document number: 2022569335 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20965848 Country of ref document: EP Kind code of ref document: A1 |