Learning device, inference device, and learning method

Abstract

The learning device (100) includes a time-series data acquisition unit (11) that acquires time-series data during a specific period in which a plurality of specific events have occurred, a partial data extraction unit (13) that divides the time-series data acquired by the time-series data acquisition unit (11) into a plurality of partial data, and extracts therefrom a plurality of first partial data corresponding to a first period that is a period immediately before the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events, and a learning unit (18) that performs learning based on the plurality of first partial data and the plurality of second partial data, with each of the plurality of first partial data as a positive example and each of the plurality of second partial data as a negative example, and generates a learned model for inferring the probability that the partial data is a positive example based on the input of partial data of other time-series data corresponding to the time-series data.

Description

Technical Field

[0001] The present disclosure relates to a learning device, an inference device, and a learning method.

Background Art

[0002] Conventionally, a computer system has been disclosed that predicts whether an accident or failure may occur in an oil refinery or a chemical processing plant based on time-series data acquired by a plurality of sensors (see, for example, Patent Document 1). This computer system identifies a precursor pattern of an accident or failure from time-series data acquired by a plurality of sensors, and predicts whether an accident or failure may occur based on the time-series data and a dependency graph generated based on the precursor pattern.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, as described above, the computer system described in Patent Document 1 predicts whether an accident or failure may occur in an oil refinery or a chemical processing plant based on time-series data acquired by a plurality of sensors, and has a problem that it cannot infer (predict) the probability of occurrence of a general event.

[0005] The present disclosure has been made based on the recognition of the above problems, and an object thereof is to provide a learning device, an inference device, and a learning method capable of inferring the probability of occurrence of a specific event.

Means for Solving the Problems

[0006] The learning device according to the present disclosure includes a time-series data acquisition unit that acquires time-series data during a specific period in which a plurality of specific events have occurred, and divides the time-series data acquired by the time-series data acquisition unit into a plurality of partial data, and from among them, a plurality of first partial data corresponding to a first period that is the period immediately before the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events are extracted. A partial data extraction unit, and learning is performed based on the plurality of first partial data and the plurality of second partial data, with each of the plurality of first partial data as a positive example and each of the plurality of second partial data as a negative example, and based on the input of partial data of other time-series data corresponding to the time-series data, a learned model for inferring the probability that the partial data is a positive example is generated. A learning unit, and is characterized by including the above.

Effect of the Invention

[0007] The learning device according to the present disclosure can infer the probability that a specific event will occur.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Mode for Carrying Out the Invention

[0009] Hereinafter, embodiments according to the present disclosure will be described in detail with reference to the drawings. Embodiment 1. First, with reference to FIG. 1, the inference system 1 according to Embodiment 1 will be described. FIG. 1 is a block diagram showing the schematic configuration of the inference system 1 according to Embodiment 1. As shown in FIG. 1, the inference system 1 according to Embodiment 1 includes a learning device 100 and an inference device 500, which are connected wirelessly or wired so as to be communicable with each other. The learning device 100 is a device that generates a learned model for inferring the probability that a specific event occurs based on time-series data. Further, the inference device 500 is a device that infers the probability that a specific event occurs using the learned model generated by the learning device 100. Note that the inference device 500 only needs to be able to acquire information from the learning device 100, and the learning device 100 and the inference device 500 may be configured to be able to communicate information via a device or communication line (not shown) with each other.

[0010] The learning device 100 includes a time-series data acquisition unit 11, a specific event information acquisition unit 12, a partial data extraction unit 13, a feature amount extraction unit 14, a learning data generation unit 17, and a learning unit 18.

[0011] The time-series data acquisition unit 11 acquires time-series data during a specific period in which a plurality of specific events have occurred. For example, the time-series data acquisition unit 11 acquires past time-series data that has been empirically known in advance to be correlated with the occurrence of a plurality of specific events during a specific period in which the plurality of specific events have occurred, or that can be empirically inferred to be correlated with the occurrence of the plurality of specific events. For example, generally, although it is unclear what kind of correlation exists between the time-series data of a company's management indicators and an important event of the company as a specific event, it is empirically inferred that there is some correlation. The time-series data acquisition unit 11 acquires, for example, time-series data that can be inferred to be correlated with the occurrence of such specific events. Specifically, the time-series data acquisition unit 11 acquires time-series data indicating the transition of the stock price of a specific company. Note that the time-series data acquired by the time-series data acquisition unit 11 is not limited to this, and may be time-series data during a specific period including the time when a plurality of specific events have occurred, may be data indicating the time transition of economic indicators in a specific market, may be weather data indicating the time transition of the observation results of specific weather information, or may be statistical data indicating the transition of the social situation in a specific society.

[0012] The specific event information acquisition unit 12 acquires information indicating the occurrence time of a specific event that occurred during a specific period indicated by the time-series data acquired by the time-series data acquisition unit 11. For example, when the time-series data acquired by the time-series data acquisition unit 11 is time-series data indicating the monthly change of specific information, the specific event information acquisition unit 12 acquires information indicating the month in which a specific event occurred in the past, and when the time-series data is time-series data indicating the time change of specific information, the specific event information acquisition unit 12 acquires information indicating the time when a specific event occurred in the past. The specific event is an event that is the target of inferring the probability of occurrence by the inference device 500, and may be a business activity such as the announcement of a new product occurring in a company, may be an event such as the establishment or bankruptcy of a company, or may be an event occurring in society.

[0013] The partial data extraction unit 13 extracts partial data from the time-series data acquired by the time-series data acquisition unit 11 based on the information indicating the occurrence time of the specific event acquired by the specific event information acquisition unit 12. Specifically, the partial data extraction unit 13 divides the time-series data acquired by the time-series data acquisition unit 11 into a plurality of partial data, and from among them, a plurality of first partial data corresponding to a first period, which is the period immediately before the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events are extracted. Further, the partial data extraction unit 13 extracts the plurality of first partial data and the plurality of second partial data such that each first period and each second period have the same length. Note that the plurality of first periods and the plurality of second periods may partially overlap with each other. Also, the first period may include the time point of the occurrence of the specific event, or may be such that the influence of the occurrence of the specific event on the time-series data is considered to be sufficiently large, for example, about one acquisition interval of the time-series data, and the end point of the first period may be before the time point of the occurrence of the specific event.

[0014] The feature extraction unit 14 extracts features from each of the plurality of first partial data and the plurality of second partial data extracted by the partial data extraction unit 13. Specifically, the feature extraction unit 14 extracts, from each of the plurality of first partial data and the plurality of second partial data, a feature indicating the degree of association with other partial data among the plurality of partial data composed of the plurality of first partial data and the plurality of second partial data. In other words, the feature extraction unit 14 calculates, for each of the plurality of first partial data and the plurality of second partial data, a feature indicating the degree of association with other partial data among the plurality of partial data composed of the plurality of first partial data and the plurality of second partial data. For example, the feature extraction unit 14 calculates, for each of the plurality of first partial data and the plurality of second partial data, a Discord Score, which is a feature indicating the nearest neighbor distance to a plurality of other partial data, as a feature indicating the degree of association with other partial data.

[0015] The learning data generation unit 17 generates learning data for the learning device 100 to perform learning, using a plurality of first partial data as positive examples and a plurality of second partial data as negative examples, based on the feature amounts extracted by the feature extraction unit 14. Specifically, the learning data generation unit 17 assigns a positive example label to each feature amount extracted from the plurality of first partial data, and assigns a negative example label to each feature amount extracted from each of the plurality of second partial data, for the plurality of feature amounts extracted from the plurality of first partial data and the plurality of second partial data by the feature extraction unit 14, and generates learning data with the labeled plurality of feature amounts as a data set.

[0016] The learning unit 18 performs learning based on the learning data generated by the learning data generation unit 17, and generates a learned model for inferring the probability that the partial data is a positive example based on the input of partial data of other time series data corresponding to the time series data acquired by the time series data acquisition unit 11. For example, the learning unit 18 generates a logistic regression model for inferring the probability that the input partial data is a positive example by logistic regression analysis based on the learning data generated by the learning data generation unit 17.

[0017] The inference device 500 includes a partial data acquisition unit 51, a feature extraction unit 52, an inference unit 53, and a storage unit 54 that stores information.

[0018] The partial data acquisition unit 51 acquires partial data used for performing inference by the learned model generated by the learning unit 18. For example, the partial data acquisition unit 51 acquires other time-series data corresponding to the time-series data acquired by the time-series data acquisition unit 11, divides the acquired other time-series data into a plurality of partial data, and extracts, from among them, third partial data that is partial data of a period to be the inference target and has the same length as the first period and the second period. For example, the partial data acquisition unit 51 acquires, as partial data used for performing inference by the learned model, partial data of the transition of the same data in a period after the specific period of the time-series data acquired by the learning device 100 for generating learning data. Also, for example, the partial data acquisition unit 51 acquires, as other time-series data, time-series data indicating the transition of the same data in a period after the specific period of the time-series data acquired by the learning device 100 for generating learning data.

[0019] Note that the partial data acquisition unit 51 is not limited to acquiring time-series data indicating the transition of the same data in a period after a specific period of the time-series data acquired by the learning device 100 as partial data of other time-series data corresponding to the time-series data acquired by the learning device 100. The partial data acquisition unit 51 may be any time-series data related to the change of the time-series data acquired by the learning device 100 as partial data of other time-series data corresponding to the time-series data acquired by the learning device 100. For example, when the time-series data acquired by the learning device 100 is time-series data indicating the transition of representative values of business indicators of a plurality of companies, the partial data acquisition unit 51 may be configured to acquire time-series data indicating the transition of business indicators of a specific company in the same industry as the plurality of companies. Or when the time-series data acquired by the learning device 100 is time-series data indicating the transition of business indicators of a specific company, the partial data acquisition unit 51 may be configured to acquire time-series data indicating the transition of business indicators of other companies in the same industry as the specific company. Or when the time-series data acquired by the learning device 100 is time-series data indicating the transition of economic indicators of a specific society, the partial data acquisition unit 51 may be configured to acquire time-series data indicating the transition of economic indicators of another society different from the specific society.

[0020] The feature extraction unit 52 extracts features from the partial data acquired by the partial data acquisition unit 51. Specifically, the feature extraction unit 52 extracts, from the partial data acquired by the partial data acquisition unit 51, features indicating the degree of relevance to other partial data among the plurality of partial data composed of a plurality of first partial data and a plurality of second partial data extracted by the partial data extraction unit 13 of the learning device 100. In other words, the feature extraction unit 52 calculates, for the partial data acquired by the partial data acquisition unit 51, features indicating the degree of relevance to other partial data among the plurality of partial data composed of a plurality of first partial data and a plurality of second partial data extracted by the partial data extraction unit 13 of the learning device 100. For example, the feature extraction unit 52 calculates, for the partial data acquired by the partial data acquisition unit 51, a Discord Score, which is a feature indicating the nearest neighbor distance to a plurality of other partial data, as a feature indicating the degree of relevance to other partial data.

[0021] The inference unit 53 infers the probability that the partial data acquired by the partial data acquisition unit 51 is a positive example using the learned model generated by the learning unit 18. In other words, the inference unit 53 infers the probability that the partial data of other time series data corresponding to the time series data is a positive example by the learned model generated by the learning unit 18. Also, in other words, the inference unit 53 inputs the features extracted by the feature extraction unit 52 into the learned model generated by the learning unit 18 to obtain, as an inference result, the probability that the partial data acquired by the partial data acquisition unit 51 is a positive example. The inference unit 53 outputs the inference result to another device such as a display device (not shown) or another computer.

[0022] The storage unit 54 stores information used in various processes performed by the inference device 500 and information indicating the results of various processes performed by the inference device 500. Specifically, the storage unit 54 stores any one or a plurality of the time-series data, partial data, feature amounts acquired by the learning device 100, the learned model generated by the learning device 100, the partial data acquired by the partial data acquisition unit 51, the feature amounts acquired by the feature amount extraction unit 52, and the results of inference by the inference unit 53. Each component of the inference device 500 refers to the information stored in the storage unit 54 when the inference device 500 performs processing.

[0023] Next, with reference to FIGS. 2 and 3, the hardware configuration of the learning device 100 will be described. FIG. 2 is a diagram showing an example of the hardware configuration of the learning device 100, and FIG. 3 is a diagram showing an example different from that of FIG. 2 of the hardware configuration of the learning device 100. For example, as shown in FIG. 2, the learning device 100 is a computer having a processor 100a, a memory 100b, and an I / O port 100c, and is configured such that the processor 100a reads and executes a program stored in the memory 100b.

[0024] Also, for example, as shown in FIG. 3, the learning device 100 is a computer having a processing circuit 100d, which is dedicated hardware, and an I / O port 100c, and executes a program. The processing circuit 100d is configured by, for example, a single circuit, a composite circuit, a programmed processor, a parallel-programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. Each function of the learning device 100 is realized by the processor 100a or the processing circuit 100d, which is dedicated hardware, executing a program. Note that the learning device 100 may have hardware other than those described above, such as a hardware timer.

[0025] Note that since the hardware configuration of the inference device 500 is the same as that of the learning device 100, the description thereof will be omitted.

[0026] Next, with reference to FIGS. 1, 4 to 7, details of the processing performed by the learning device 100 and the processing performed by the inference device 500 will be described. FIG. 4 is a flowchart showing an example of the processing performed by the learning device 100 according to the first embodiment. The processing performed by the learning device 100 shown in FIG. 4 is processing for generating a learned model that infers the probability of a specific event occurring based on time-series data. As shown in FIG. 4, when starting the processing, the learning device 100 first acquires time-series data (step ST01). In this processing, the learning device 100 acquires the time-series data used for generating learning data by the time-series data acquisition unit 11.

[0027] FIG. 5 is a diagram showing a graph of time-series data in a specific period acquired by the learning device 100 according to the first embodiment. As shown in FIG. 5, for example, the time-series data J0 acquired by the learning device 100 is data showing the transition of a value k indicating specific information that changes with the passage of time t.

[0028] After performing the processing of step ST01, the learning device 100 acquires information indicating the occurrence time of a specific event (step ST03). In this processing, the learning device 100 acquires, by the specific event information acquisition unit 12, information indicating the occurrence time of a specific event that occurred in a specific period that is the range of time t when the time-series data J0 was acquired. As shown in FIG. 5, for example, in this processing, the learning device 100 acquires information indicating the occurrence times of specific events E1, E2, E3, and E4 that occurred in the specific period.

[0029] When the learning device 100 performs the process of step ST03, it extracts first partial data and second partial data (step ST04). In this process, the learning device 100 extracts, for example, from the time-series data J0 obtained in the process of step ST01, based on the information indicating the occurrence time of the specific event obtained in the process of step ST02, a plurality of first partial data j1, j2, j3, j4 corresponding to a first period p1 which is the period immediately before the occurrence of each of the plurality of specific events E1 to E4, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period p1 of each of the plurality of specific events, by the partial data acquisition unit 51.

[0030] FIG. 6A is a diagram showing a graph of the first partial data j1 to j4 extracted from the time-series data J0 by the learning device 100 according to Embodiment 1, and FIG. 6B is a diagram showing a graph of the second partial data j5, j6, j7, j8, j9, j10, j11, j12, j13 extracted from the time-series data J0 by the learning device 100 according to Embodiment 1. As shown in FIGS. 6A and 6B, in the process of step ST04, the learning device 100 extracts, for example, from the graph of the time-series data J0 obtained in the process of step ST01, a plurality of first partial data j1 to j4 indicated by partial graphs (partial waveforms) of the time-series data J0, and a plurality of second partial data j5 to j13 indicated by partial graphs (partial waveforms) of the time-series data J0.

[0031] When the learning device 100 performs the process of step ST04, it extracts feature quantities from the first partial data and the second partial data (step ST06). In this process, the learning device 100 extracts feature quantities from each of the plurality of first partial data j1 to j4 and the plurality of second partial data j5 to j13 extracted in the process of step ST04 by the feature quantity extraction unit 14.

[0032] As shown in FIGS. 6A and 6B, in the process of step ST06, the learning device 100 extracts, for example, discord scores 4, 5, 8850, and 3024 as feature amounts from each of a plurality of first partial data j1 to j4, and extracts discord scores 4, 4, 6, 5, 17963, 2410, 11, 24, and 8 as feature amounts from each of a plurality of second partial data j5 to j13.

[0033] When the learning device 100 performs the process of step ST06, it labels the extracted feature amounts to generate learning data (step ST11). In this process, the learning device 100 assigns either a positive example or a negative example label to the plurality of feature amounts extracted in the process of step ST06, and generates learning data using these plurality of feature amounts as a data set.

[0034] When the learning device 100 performs the process of step ST11, it generates a learned model (step ST13). In this process, the learning device 100 generates, for example, a learned model that is a logistic regression model for inferring the probability that the partial data is a positive example based on the input of other partial data corresponding to the first partial data by logistic regression analysis based on the learning data generated in the process of step ST11.

[0035] FIG. 7 is a flowchart showing an example of the process performed by the inference device 500 according to Embodiment 1. The process performed by the inference device 500 shown in FIG. 7 is a process for inferring the probability that a specific event occurs using the learned model generated by the learning device 100. As shown in FIG. 7, when starting the process, the inference device 500 first acquires the learned model (step ST21). In this process, the inference device 500 acquires the learned model generated by the learning device 100 from the learning device 100.

[0036] When the inference device 500 performs the process of step ST21, it acquires time-series data (step ST22). In this process, the inference device 500 acquires, by the partial data acquisition unit 51, other time-series data corresponding to the time-series data acquired by the learning device 100.

[0037] When the inference device 500 performs the process of step ST22, it extracts third partial data (step ST24). In this process, the inference device 500 divides the time-series data acquired in the process of step ST22 into a plurality of partial data, and from among them, acquires, by the partial data acquisition unit 51, the third partial data used for performing inference by the learned model generated by the learning device 100.

[0038] When the inference device 500 performs the process of step ST24, it extracts feature quantities from the third partial data (step ST25). In this process, the inference device 500 extracts, by the feature quantity extraction unit 52, the feature quantities for input to the learned model generated by the learning device 100 from the third partial data acquired in the process of step ST24.

[0039] When the inference device 500 performs the process of step ST25, it inputs the extracted feature quantities to the learned model (step ST26). In this process, the inference device 500 inputs the feature quantities to the learned model generated by the learning device 100 in order to infer the probability that the partial data corresponding to the feature quantities extracted in the process of step ST25 is a positive example.

[0040] When the inference device 500 performs the process of step ST26, it acquires an inference result (step ST27). In this process, the inference device 500 acquires, as the inference result by the learned model generated by the learning device 100, the probability that the third partial data acquired in the process of step ST24 is a positive example.

[0041] When the inference device 500 performs the process of step ST27, it outputs an inference result (step ST28). In this process, the inference device 500 causes, for example, a display device such as a liquid crystal display device communicably connected to the inference device 500 to display the inference result.

[0042] As described above, the learning device 100 according to Embodiment 1 includes a time-series data acquisition unit 11 that acquires time-series data in a specific period in which a plurality of specific events have occurred, and divides the time-series data acquired by the time-series data acquisition unit 11 into a plurality of partial data, and extracts, from among them, a plurality of first partial data corresponding to a first period that is a period immediately before the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events; and a learning unit 18 that performs learning based on the plurality of first partial data and the plurality of second partial data, with each of the plurality of first partial data being a positive example and each of the plurality of second partial data being a negative example, and generates a learned model for inferring the probability that the partial data of other time-series data corresponding to the time-series data is a positive example based on the input of the partial data of the other time-series data.

[0043] With such a configuration, the learning device 100 can generate a learned model for inferring the probability that the partial data of the time-series data input to the learned model is a positive example based on past time-series data and information indicating the time when a specific event occurred in the past. In other words, the learning device 100 can generate a learned model for inferring the probability that a specific event will occur immediately after the partial data of the time-series data input to the learned model based on past time-series data and information indicating the time when a specific event occurred in the past.

[0044] For example, the learning device 100 can generate a learned model that infers the probability of a company going bankrupt in a specific industry in the near future based on time-series data indicating the stock price index of the market during a past specific period and information indicating the timing of past company bankruptcies in a specific industry as a specific event. Also, for example, the learning device 100 can generate a learned model that infers the probability of a specific company announcing a new product in the near future based on time-series data indicating the stock price of the specific company during a past specific period and information indicating the timing of the announcement of the new product of the specific company as a specific event.

[0045] Further, the learning device 100 according to Embodiment 1 includes a learning data generation unit that generates learning data based on a feature amount indicating the degree of association with other partial data included in the plurality of first partial data and the plurality of second partial data for each of the plurality of first partial data and the plurality of second partial data. For example, the learning device 100 according to Embodiment 1 is configured to generate learning data based on the nearest neighbor distance to other partial data as a feature amount indicating the degree of association with other partial data for each of the plurality of first partial data and the plurality of second partial data. Configured in this way, the learning device 100 can generate a learned model in which the characteristics of the partial data are easily reflected in the inference result, and it becomes possible to improve the accuracy of the inference result by the learned model.

[0046] In addition, in Embodiment 1, although the learning device 100 generates learning data for generating a learned model based on feature amounts indicating degrees of association with other partial data extracted from a plurality of first partial data and a plurality of second partial data, the present invention is not limited to this. The learning device may be configured to perform learning based on the plurality of first partial data and the plurality of second partial data, with each of the plurality of first partial data being a positive example and each of the plurality of second partial data being a negative example. For example, the learning device may assign a positive example label to a feature amount indicating the average value (moving average) of each first partial data, assign a negative example label to a feature amount indicating the average value (moving average) of each second partial data, and be configured to perform learning based on learning data having a plurality of labeled feature amounts as a data set. Alternatively, the learning device may assign a positive example label to a feature amount obtained by time differentiation of each first partial data, assign a negative example label to a feature amount obtained by time differentiation of each second partial data, and be configured to perform learning based on learning data having a plurality of labeled feature amounts as a data set. Further, the learning device may be configured to assign a positive example label to the plurality of first partial data, assign a negative example label to the plurality of second partial data, and perform learning based on learning data having the plurality of labeled first partial data and the plurality of second partial data as a data set.

[0047] In addition, in Embodiment 1, although the learning device 100 is configured to generate a learned model that is a logistic regression model, the present invention is not limited to this. The learning device may be configured to generate a learned model that infers the probability that partial data of time-series data is a positive example. For example, the learning device may be configured to generate a learned model using an algorithm other than the logistic regression model, such as Naive Bayes, SVM (Support Vector Machine), Random Forest, or a deep learning algorithm. Alternatively, calibration techniques such as isotonic regression (see Non-Patent Document 1) and temperature scaling (see Non-Patent Document 2) may be used in combination. [Non-Patent Document 1] Zadrozny, Bianca, et al. “Transforming classifier scores into accurate multiclass probability estimates.” Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining. 2002. [Non-Patent Document 2] Hinton, Geoggrey, et al. “Distilling the knowledge in a neural network.” arXiv preprint arXiv:1503.02531 (2015).

[0048] Embodiment 2. Next, with reference to FIGS. 8 to 11, the inference system 2 according to Embodiment 2 will be described. The inference system 2 according to Embodiment 2 has some different configurations from the inference system according to Embodiment 1, but the other configurations are the same. For the same configurations as those in Embodiment 1, the same names and reference numerals as those in Embodiment 1 are given and the description thereof is omitted.

[0049] FIG. 8 is a block diagram showing a schematic configuration of the inference system 2 according to Embodiment 2. As shown in FIG. 8, the inference system 2 according to Embodiment 2 includes a learning device 200 and an inference device 600, which are connected wirelessly or wiredly so as to be communicable with each other. The learning device 200 is a device that generates a learned model for inferring the probability that a specific event occurs based on time-series data. The inference device 600 is a device that infers the probability that a specific event occurs using the learned model generated by the learning device 200.

[0050] The learning device 200 includes a time-series data acquisition unit 11, a specific event information acquisition unit 12, a partial data extraction unit 13, a feature quantity extraction unit 14, a relevance calculation unit 15, a time-series data selection unit 16, a learning data generation unit 17, and a learning unit 18.

[0051] Based on the feature quantity indicating the relevance between a plurality of partial data included in the time-series data acquired by the time-series data acquisition unit 11, the relevance calculation unit 15 extracts a characteristic feature period in a specific period during which the time-series data is acquired by the time-series data acquisition unit 11, and calculates the relevance between the extracted feature period and a plurality of first periods. For example, the relevance calculation unit 15 slides a window with a specific time width from the start point to the end point of the specific period to extract a plurality of partial data, and extracts a feature quantity indicating the relevance between each partial data and other partial data. Examples of the feature quantity indicating the relevance between each partial data and other partial data include a discord score.

[0052] For example, the relevance calculation unit 15 extracts, as the feature period, a period indicating an outlier (outlier waveform) different from other partial data (partial waveforms) in the time-series data. Specifically, the relevance calculation unit 15 extracts, as the feature period, a period in which the extracted feature quantity exceeds a preset threshold value. More specifically, the relevance calculation unit 15 extracts, as one feature period, a period in which the extracted feature quantity continuously exceeds a preset threshold value. Note that the relevance calculation unit 15 constitutes the period extraction unit in Embodiment 1.

[0053] In addition, the relevance calculation unit 15 calculates the relevance between the extracted feature period and a plurality of first periods of the time-series data acquired by the time-series data acquisition unit 11. In other words, the relevance calculation unit 15 calculates the co-occurrence degree, which is the degree to which the extracted feature period and a plurality of first periods of the time-series data acquired by the time-series data acquisition unit 11 co-occur. For example, as the relevance (co-occurrence degree) between the feature period and the first period, the relevance calculation unit 15 uses the recall rate Recall, which is an index indicating the reproducibility of the feature period in the first period, and represents the ratio of the period in which at least a part overlaps with any one of the plurality of first periods.eve Calculate the recall rate. eve It is calculated by the following formula (1). Recall eve =(Number of overlapping periods between the first period and the characteristic period) / (Number of the first periods) ···(1)

[0054] Also, for example, the relevance calculation unit 15 uses, as the relevance between the characteristic period and the first period, the discrimination rate Precision that indicates the proportion of the period in which at least a part overlaps with any one of the first periods among a plurality of characteristic periods, which is an index indicating the discriminability in the first period of the characteristic period. var Calculate the precision rate. var It is calculated by the following formula (2). Precision var =(Number of overlapping periods between the first period and the characteristic period) / (Number of the characteristic periods) ···(2)

[0055] Also, for example, the relevance calculation unit 15 uses, as the relevance between the characteristic period and the first period, the recall rate Recall eve and the discrimination rate Precision var to calculate the harmonic mean D. The harmonic mean D is calculated by the following formula (3). D = 2 × Recall eve × Precision var / (Recall eve + Precision var ) ···(3)

[0056] When a plurality of time series data are acquired by the time series data acquisition unit 11, the time series data selection unit 16 selects any one of the acquired plurality of time series data based on the relevance calculated by the relevance calculation unit 15. For example, when the first time series data and the second time series data are acquired by the time series data acquisition unit 11, the time series data selection unit 16 selects the time series data with a higher relevance calculated by the relevance calculation unit 15 among these time series data.

[0057] For example, the time series data selection unit 16 selects the time series data with high reproducibility calculated by the relevance calculation unit 15 among the first time series data and the second time series data acquired by the time series data acquisition unit 11. Also, for example, the time series data selection unit 16 selects the time series data with high discriminability calculated by the relevance calculation unit 15 among the first time series data and the second time series data acquired by the time series data acquisition unit 11. Also, for example, the time series data selection unit 16 selects the time series data with a high harmonic mean of the reproducibility rate and the discrimination rate calculated by the relevance calculation unit 15 among the first time series data and the second time series data acquired by the time series data acquisition unit 11. Note that the time series data selection unit 16 may be configured to select the time series data with the highest relevance calculated by the relevance calculation unit 15 from the acquired plurality of time series data when a plurality of time series data are acquired by the time series data acquisition unit 11, or may be configured to select any one of the time series data based on the relevance calculated by the relevance calculation unit 15 and other conditions other than the relevance calculated by the relevance calculation unit 15. The time series data selection unit 16 causes the learning data generation unit 17 to generate learning data based on the selected time series data.

[0058] The inference device 600 includes a partial data acquisition unit 61, a feature extraction unit 52, an inference unit 53, and a storage unit 54.

[0059] The partial data acquisition unit 61 acquires partial data used for performing inference by the learned model generated by the learning unit 18. For example, the partial data acquisition unit 61 acquires other time-series data corresponding to the time-series data selected by the time-series data selection unit 16, divides the acquired other time-series data into a plurality of partial data, and extracts, from among them, third partial data that is partial data for a period to be an inference target and has the same length as the first period and the second period. Other features of the partial data acquisition unit 61 are the same as those of the partial data acquisition unit 51 according to the first embodiment, and thus the description thereof is omitted.

[0060] Note that since the hardware configurations of the learning device 200 and the inference device 600 are the same as those of the learning device 100, the description thereof is omitted.

[0061] Next, with reference to FIGS. 8 to 11, details of the processing performed by the learning device 200 and the processing performed by the inference device 600 will be described. FIG. 9 is a flowchart showing an example of the processing performed by the learning device 200 according to the second embodiment. The processing performed by the learning device 200 shown in FIG. 9 is processing for generating a learned model that infers the probability that a specific event occurs based on time-series data. Note that a part of the processing performed by the learning device 200 according to the second embodiment is the same as the processing performed by the learning device 100 according to the first embodiment, and thus the description of the same processing as that in the first embodiment is omitted.

[0062] As shown in FIG. 9, when starting the processing, the learning device 200 first acquires first time-series data and second time-series data (step ST02). In this processing, the learning device 200 acquires the first time-series data and the second time-series data by the time-series data acquisition unit 11 as the time-series data used for generating learning data. Note that in this processing, the learning device 200 may be configured to acquire three or more time-series data including the first time-series data and the second time-series data by the time-series data acquisition unit 11.

[0063] When the learning device 200 performs the process of step ST02, it acquires information indicating the occurrence time of a specific event (step ST03).

[0064] When the learning device 200 performs the process of step ST03, it extracts first partial data and second partial data from the first time-series data and the second time-series data (step ST05). In this process, the learning device 200, based on the information indicating the occurrence time of the specific event acquired in the process of step ST03, from each of the first time-series data and the second time-series data acquired in the process of step ST02, extracts, by the partial data extraction unit 13, a plurality of first partial data corresponding to a first period which is the period immediately before the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events.

[0065] When the learning device 200 performs the process of step ST05, it extracts feature amounts from the first partial data and the second partial data (step ST07). In this process, the learning device 200 extracts, by the feature amount extraction unit 14, feature amounts from each of the plurality of first partial data and the plurality of second partial data extracted in the process of step ST05.

[0066] FIG. 10 is a diagram showing graphs of feature amounts of the first time-series data J1 and the second time-series data J2 in a specific period acquired by the learning device 200 according to Embodiment 1. In FIG. 10, for example, in the graphs showing the feature amounts of the first time-series data J1, it is shown that the feature amounts exceed a preset threshold value th1 for each time-series data in the portions of S11, S12, S13, S14, S15, S16, and S17. Also, in FIG. 10, for example, in the graphs showing the feature amounts of the second time-series data J2, it is shown that the feature amounts exceed a preset threshold value th2 for each time-series data in the portions of S21, S22, S23, S24, and S25.

[0067] When the learning device 200 performs the process of step ST07, it calculates a reproducibility rate indicating the reproducibility of the characteristic period for each of the first time-series data and the second time-series data (step ST08). In this process, for example, the learning device 200 calculates the reproducibility rate, which is the ratio of the period that at least partially overlaps with seven characteristic periods corresponding to S11 to S17 among the eight first periods corresponding to the specific events E11, E12, E13, E14, E15, E16, E17, E18 in the graph of FIG. 10, as 6 / 8.

[0068] When the learning device 200 performs the process of step ST08, it calculates the discriminability of the characteristic period for each of the first time-series data and the second time-series data (step ST09). In this process, for example, the learning device 200 calculates the discrimination rate, which is the ratio of the period that at least partially overlaps with eight first periods among the seven characteristic periods corresponding to S11 to S17 in the graph of FIG. 10, as 1 / 7.

[0069] When the learning device 200 performs the process of step ST09, it calculates the degree of association between the specific event and the characteristic period (step ST10). In this process, for example, the learning device 200 calculates the degree of association between the specific event and the characteristic period by calculating the harmonic mean of the reproducibility rate calculated in the process of step ST08 and the discrimination rate calculated in the process of step ST09.

[0070] When the learning device 200 performs the process of step ST10, it generates learning data by labeling the feature amounts extracted from the time-series data with a high degree of relevance between the specific event and the characteristic period among the first time-series data and the second time-series data (step ST12). As described above, the learning device 200 may be configured to select any one of the plurality of time-series data using the recall rate calculated in the process of step ST08 and the discrimination rate calculated in the process of step ST09 as the degree of relevance between the specific event and the characteristic period. Alternatively, the arithmetic mean (additive mean) of the recall rate calculated in the process of step ST08 and the discrimination rate calculated in the process of step ST09 may be used as the degree of relevance between the specific event and the characteristic period to select any one of the plurality of time-series data. Or it may be configured to select any one of the plurality of time-series data based on other degrees of relevance.

[0071] When the learning device 200 performs the process of step ST12, it generates a learned model (step ST13). The process by which the learning device 200 generates a learned model based on the learning data is the same as that of the learning device 200 according to the first embodiment.

[0072] FIG. 11 is a flowchart showing an example of the process performed by the inference device 600 according to the second embodiment. The process performed by the inference device 600 shown in FIG. 11 is a process for inferring the probability that a specific event occurs using the learned model generated by the learning device 200. As shown in FIG. 11, when starting the process, the inference device 600 first acquires the learned model (step ST21).

[0073] When the inference device 600 performs the process of step ST21, it acquires the time-series data corresponding to the time-series data with a high degree of relevance between the specific event and the characteristic period among the first time-series data and the second time-series data (step ST23). In this process, in order to infer the probability of the occurrence of a specific event using the learned model generated by the learning device 200, the inference device 600 acquires the time-series data corresponding to the time-series data selected based on the degree of relevance between the specific event and the characteristic period among the plurality of time-series data acquired by the learning device 200.

[0074] When the inference device 600 performs the process of step ST24, it extracts the third partial data (step ST24). In this process, the inference device 600 divides the time-series data acquired in the process of step ST23 into a plurality of partial data, and among them, the partial data acquisition unit 61 acquires the third partial data used for performing inference by the learned model generated by the learning device 100.

[0075] When the inference device 600 performs the process of step ST25, it extracts feature quantities from the third partial data (step ST25). When the inference device 600 performs the process of step ST25, it inputs the extracted feature quantities into the learned model (step ST26). When the inference device 600 performs the process of step ST26, it acquires an inference result (step ST27). When the inference device 600 performs the process of step ST27, it outputs the inference result (step ST28).

[0076] As described above, the learning device 200 according to the second embodiment extracts a characteristic characteristic period in a specific period based on the feature quantity indicating the degree of relevance between a plurality of partial data included in the time-series data, and includes a relevance calculation unit 15 that calculates the degree of relevance between the extracted characteristic period and the plurality of first periods. The learning device 200 is configured to perform learning based on the partial data of the time-series data with a high degree of relevance calculated by the relevance calculation unit 15 among the first time-series data and the second time-series data acquired by the time-series data acquisition unit.

[0077] Configured in this way, when the learning device 200 can acquire a plurality of time-series data for inferring the probability of a specific event occurring, the learning device 200 can select time-series data with a high degree of correlation between the characteristic period of the time-series data and the occurrence of the specific event, and generate a learned model based on the selected time-series data. Thereby, the accuracy in inferring the probability of a specific event occurring by the learned model can be improved.

[0078] Further, the learning device 200 according to the second embodiment is configured to calculate the degree of correlation between the characteristic period extracted by the degree-of-correlation calculation unit 15 and the plurality of first periods based on the overlapping ratio between the plurality of first periods and the plurality of characteristic periods extracted by the degree-of-correlation calculation unit 15. Further, the learning device 200 according to the second embodiment is configured to calculate the degree of correlation between the characteristic period extracted by the degree-of-correlation calculation unit 15 and the plurality of first periods based on the overlapping ratio between the plurality of first periods and the plurality of characteristic periods extracted by the degree-of-correlation calculation unit 15. Configured in this way, the learning device 200 can select time-series data suitable for generating a learned model from a plurality of time-series data based on the reproducibility or discriminability of the characteristic period in the time-series data.

[0079] Note that, in the second embodiment, the learning device 200 is configured to perform learning based on the partial data of the time-series data with the highest degree of relevance between the characteristic period and the first period among the acquired plurality of time-series data, and not to perform learning based on the partial data of other time-series data, but it is not limited to this. The learning device may be configured to perform learning based on the partial data of the time-series data selected based on the degree of relevance between the characteristic period and the first period among the acquired plurality of time-series data. For example, the learning device may be configured to perform learning based on the partial data of other time-series data excluding the time-series data with the lowest degree of relevance between the characteristic period and the first period among the acquired plurality of time-series data, or may be configured to perform learning based on the partial data of some of the time-series data selected based on the degree of relevance between the characteristic period and the first period among the acquired plurality of time-series data, or may be configured to perform different weightings on the time-series data with a relatively high degree of relevance and the time-series data with a relatively low degree of relevance between the characteristic period and the first period among the acquired plurality of time-series data, and perform learning based on the partial data of each time-series data.

[0080] Note that the learning device may include some or all of the configurations provided in the inference device, or the inference device may include some or all of the configurations provided in the learning device, or some of the configurations provided in the learning device or the inference device may be provided in another device that is communicably connected to the learning device and the inference device. Also, a part of the configuration of the learning device may have a part of the function of the inference device, or a part of the configuration of the inference device may have a part of the function of the learning device.

[0081] Note that the present disclosure allows for free combinations of the respective embodiments, or modifications of any constituent elements of the respective embodiments, or omissions of any constituent elements in the respective embodiments.

Industrial Applicability

[0082] The information processing apparatus according to the present disclosure can be used, for example, to infer the probability that a specific event that affects the performance of a company occurs based on time series data such as economic indicators.

[0083] Hereinafter, various aspects of the present disclosure will be collectively described as appendices.

[0084] (Appendix 1) A time series data acquisition unit that acquires time series data in a specific period in which a plurality of specific events occur, The time series data acquired by the time series data acquisition unit is divided into a plurality of partial data, and from among them, a plurality of first partial data corresponding to a first period that is the period immediately before the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events are extracted, and a partial data extraction unit, Using each of the plurality of first partial data as a positive example and each of the plurality of second partial data as a negative example, learning is performed based on the plurality of first partial data and the plurality of second partial data, and based on the input of partial data of other time series data corresponding to the time series data, a learned model for inferring the probability that the partial data is a positive example is generated, a learning unit, A learning apparatus characterized by the above. (Appendix 2) For each of the plurality of first partial data and the plurality of second partial data, a learning data generation unit that generates the learning data based on a feature amount indicating the degree of relevance with other partial data included in the plurality of first partial data and the plurality of second partial data, The learning apparatus according to Appendix 1, characterized by the above. (Appendix 3) The learning data generation unit generates the learning data based on the nearest neighbor distance to other partial data as a feature amount indicating the degree of relevance with other partial data for each of the plurality of first partial data and the plurality of second partial data. The learning apparatus according to Appendix 1 or 2, characterized by the above. (Appendix 4) A period extraction unit that extracts a characteristic feature period in the specific period based on a feature amount indicating the degree of association between a plurality of partial data included in the time series data; A degree-of-association calculation unit that calculates the degree of association between the feature period extracted by the period extraction unit and the plurality of first periods, and The learning unit performs learning based on partial data of the time series data with a high degree of association calculated by the degree-of-association calculation unit among the first time series data and the second time series data acquired by the time series data acquisition unit. The learning device according to any one of Appendices 1 to 3, characterized in that. (Appendix 5) The learning unit does not perform learning based on partial data of the time series data with a low degree of association calculated by the degree-of-association calculation unit among the first time series data and the second time series data acquired by the time series data acquisition unit. The learning device according to any one of Appendices 1 to 4, characterized in that. (Appendix 6) The degree-of-association calculation unit calculates the degree of association between the feature period extracted by the period extraction unit and the plurality of first periods based on the ratio of overlap with the plurality of feature periods extracted by the period extraction unit among the plurality of first periods. The learning device according to any one of Appendices 1 to 5, characterized in that. (Appendix 7) The degree-of-association calculation unit calculates the degree of association between the feature period extracted by the period extraction unit and the plurality of first periods based on the ratio of overlap with the plurality of first periods among the plurality of feature periods extracted by the period extraction unit. The learning device according to any one of Appendices 1 to 6, characterized in that. (Appendix 8) A storage unit that stores a learned model generated by learning based on the plurality of first partial data and the plurality of second partial data, with each of the plurality of first partial data corresponding to a first period that is the period immediately before the occurrence of each of the plurality of specific events, which is extracted from time-series data during a specific period in which a plurality of specific events occurred, being regarded as a positive example, and each of the plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events being regarded as a negative example, and for inferring the probability that partial data of other time-series data corresponding to the input time-series data is a positive example. An inference unit that infers the probability that partial data of other time-series data corresponding to the time-series data is a positive example by using the learned model. An inference device characterized by the above. (Appendix 9) A learning method performed by a device including a time-series data acquisition unit, a partial data extraction unit, and a learning unit, wherein the time-series data acquisition unit acquires time-series data during a specific period in which a plurality of specific events occurred; the partial data extraction unit divides the time-series data acquired by the time-series data acquisition unit into a plurality of partial data, and extracts therefrom a plurality of first partial data corresponding to a first period that is the period immediately before the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods that do not overlap with the first period of each of the plurality of specific events; and the learning unit performs learning based on the plurality of first partial data and the plurality of second partial data, with each of the plurality of first partial data being regarded as a positive example and each of the plurality of second partial data being regarded as a negative example, and generates a learned model for inferring the probability that partial data is a positive example based on the input of partial data of other time-series data corresponding to the time-series data. A learning method characterized by the above.

Explanation of Reference Numerals

[0085] 1 Inference system, 2 Inference system, 11 Time-series data acquisition unit, 12 Specific event information acquisition unit, 13 Sub-data extraction unit, 14 Feature quantity extraction unit, 15 Relevance calculation unit (period extraction unit), 16 Time-series data selection unit, 17 Learning data generation unit, 18 Learning unit, 51 Sub-data acquisition unit, 52 Feature quantity extraction unit, 53 Inference unit, 54 Memory unit, 61 Sub-data acquisition unit, 100 Learning device, 100a Processor, 100b Memory, 100c I / O port, 100d Processing circuit, 200 Learning device, 500 Inference device, 600 Inference device, D Harmonic mean, E1 Specific event, E11 Specific event, E12 Specific event, E13 Specific event, E14 Specific event, E15 Specific event, E16 Specific event, E17 Specific event, E18 Specific event, E2 Specific event, E3 Specific event, E4 Specific event, J0 Time-series data, J1 First time-series data, J2 Second time-series data, j1 First sub-data, j10 Second sub-data, j11 Second sub-data, j12 Second sub-data, j13 Second sub-data, j2 First sub-data, j3 First sub-data, j4 First sub-data, j5 Second sub-data, j6 Second sub-data, j7 Second sub-data, j8 Second sub-data, j9 Second sub-data, p1 First period, th1 Threshold value, th2 Threshold value.

Claims

1. A time-series data acquisition unit that acquires time-series data for a specific period in which multiple specific events occurred, A partial data extraction unit divides the time-series data acquired by the time-series data acquisition unit into multiple partial data, and extracts from these multiple partial data a plurality of first partial data corresponding to a first period which is the period immediately preceding the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods which do not overlap with the first period of each of the plurality of specific events. The system includes a learning unit that learns based on the plurality of first partial data and the plurality of second partial data, using each of the plurality of first partial data as a positive example and each of the plurality of second partial data as a negative example, and generates a trained model for inferring the probability that a partial data is a positive example based on input of partial data of other time series data corresponding to the time series data. A learning device characterized by the following features.

2. The system includes a learning data generation unit that generates learning data for each of the plurality of first partial data and the plurality of second partial data based on feature quantities that indicate the degree of association between the plurality of first partial data and the plurality of second partial data included in the plurality of first partial data and the plurality of second partial data. The learning device according to claim 1, characterized by the features described above.

3. The learning data generation unit generates the learning data for each of the plurality of first partial data and the plurality of second partial data, based on the nearest neighbor distance to the other partial data, as a feature quantity indicating the degree of association with the other partial data. The learning device according to feature 2.

4. A period extraction unit extracts characteristic feature periods within a specific period based on feature quantities that indicate the degree of association between multiple subdata included in the time series data, The system includes a correlation calculation unit that calculates the correlation between the feature period extracted by the period extraction unit and the plurality of first periods, The learning unit performs learning based on partial data of the time series data with a high correlation calculated by the correlation calculation unit, from among the first time series data and second time series data acquired by the time series data acquisition unit. A learning device according to any one of claims 1 to 3.

5. The learning unit does not perform learning based on partial data of time series data with low relevance calculated by the relevance calculation unit, among the first and second time series data acquired by the time series data acquisition unit. The learning device according to feature 4.

6. The correlation calculation unit calculates the correlation between the feature periods extracted by the period extraction unit and the multiple first periods based on the proportion of overlap between the multiple first periods and the multiple feature periods extracted by the period extraction unit. The learning device according to feature 4.

7. The correlation calculation unit calculates the correlation between the feature periods extracted by the period extraction unit and the multiple first periods based on the proportion of overlap between the feature periods extracted by the period extraction unit and the multiple first periods. The learning device according to feature 4.

8. A storage unit stores a trained model generated by learning based on the multiple first and second partial data, which is used to infer the probability that a partial data of other time series data corresponding to the input time series data is a positive example. The time series data over a specific period in which multiple specific events occur is divided into multiple partial data, and each of the multiple first partial data corresponding to the first period, which is the period immediately preceding the occurrence of each of the multiple specific events, is treated as a positive example, and each of the multiple second partial data corresponding to the second period, which does not overlap with one of the periods of each of the multiple specific events, is treated as a negative example. The system comprises an inference unit that uses the trained model to infer the probability that a subdata of other time series data corresponding to the time series data is a positive example. An inference device characterized by the following features.

9. A learning method performed by a device comprising a time-series data acquisition unit, a partial data extraction unit, and a learning unit, The time-series data acquisition unit includes the step of acquiring time-series data for a specific period in which multiple specific events occurred, The partial data extraction unit divides the time-series data acquired by the time-series data acquisition unit into a plurality of partial data, and extracts from there a plurality of first partial data corresponding to a first period which is the period immediately preceding the occurrence of each of the plurality of specific events, and a plurality of second partial data corresponding to a plurality of second periods which do not overlap with the first period of each of the plurality of specific events. The learning unit performs learning based on the plurality of first partial data and the plurality of second partial data, with each of the plurality of first partial data being a positive example and each of the plurality of second partial data being a negative example, and generates a trained model for inferring the probability that a partial data is a positive example based on input of partial data of other time series data corresponding to the time series data. A learning method characterized by the following:

Images