Information processing device, information processing method, and program

Abstract

Provided is an information processing device, wherein a provision means provides a shared search model to a plurality of terminals. A filter acquisition means acquires, from each terminal, a filter configured by using a characteristic amount obtained using the search model. A data acquisition means acquires, from each terminal, object time-series data which is the object of state estimation. An estimation means estimates, on the basis of object time-series data and a filter, the state of a terminal which output object time-series data.

Description

[Technical field]

[0001] The present disclosure relates to processing data obtained at multiple terminal devices. [Background technology]

[0002] There is a known method for searching information distributed and stored in multiple terminal devices using an index prepared for each terminal. For example, Patent Document 1 discloses a method for identifying a data block in which the search target data is likely to exist by referring to a hierarchical divided transposed Bloom filter when the search target data is given. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2012-252658 A Summary of the Invention [Problem to be solved by the invention]

[0004] One objective of the present disclosure is to estimate the state of each terminal by using time-series data obtained at the terminal. [Means for solving the problem]

[0005] According to one aspect of the present disclosure, there is provided an information processing device, A providing means for providing a common search model to a plurality of terminals; From each terminal, using the search model From time series data showing the specific state of the device A filter acquisition means for acquiring a filter configured with the acquired feature quantity; A data acquisition means for acquiring target time series data to be subjected to state estimation from each of the terminals; The apparatus further comprises an estimation means for estimating a state of a terminal that outputs the target time series data, based on the target time series data and the filter.

[0006] In another aspect of the present disclosure, a computer-implemented information processing method includes: Providing a common search model for multiple devices, From each terminal, using the search model From time series data showing the specific state of the device A filter composed of the obtained features is obtained, Acquire target time series data to be subjected to state estimation from each of the terminals; A state of the terminal that outputs the target time series data is estimated based on the target time series data and the filter.

[0007] In yet another aspect of the disclosure, a program includes: Providing a common search model for multiple devices, From each terminal, using the search model From time series data showing the specific state of the device A filter composed of the obtained features is obtained, Acquire target time series data to be subjected to state estimation from each of the terminals; The computer is caused to execute a process of estimating a state of a terminal that outputs the target time series data, based on the target time series data and the filter. Effect of the Invention

[0008] According to the present disclosure, the state of each terminal can be estimated using time series data obtained at the terminal. [Brief description of the drawings]

[0009] [Figure 1] The basic concept of this embodiment will be described. [Diagram 2] 1 shows an outline of a method for the server to estimate the state of each terminal. [Diagram 3] This section describes how to create filters on each device. [Figure 4] A search method using the generated filter is shown below. [Diagram 5] FIG. 2 is a block diagram showing the hardware configuration of a server and a terminal according to the first embodiment. [Figure 6]FIG. 2 is a block diagram showing the functional configuration of a server and a terminal. [Figure 7] 13 is a flowchart of a filter creation process. [Figure 8] 13 is a flowchart of a state estimation process. [Figure 9] 1 shows an application example of the first embodiment. [Figure 10] Another application example of the first embodiment will be described. [Figure 11] Another application example of the first embodiment will be described. [Figure 12] Another application example of the first embodiment will be described. [Figure 13] FIG. 11 is a block diagram showing a functional configuration of an information processing apparatus according to a second embodiment. [Figure 14] 10 is a flowchart of a process performed by an information processing device according to a second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, preferred embodiments of the present disclosure will be described with reference to the drawings. <Basic concept> First, the basic concept of the embodiment will be described. In this embodiment, the states of multiple terminals are managed on the server (cloud) side. FIG. 1 shows the basic concept of this embodiment. A server and multiple terminals are configured to be able to communicate with each other. Each terminal performs various processes and acquires time-series data relating to the state and status of the processes. For example, each terminal performs measurements using a sensor or the like to acquire time-series data. Note that the time-series data is not limited to measurement data from a sensor, and may be images captured by a camera or audio collected by a microphone.

[0011] In each terminal, instead of storing the time series data as it is, the time series data is converted into features and stored in a database (hereinafter referred to as "DB"). Specifically, an analysis model is prepared for each terminal, and the acquired time series data is converted into features using the analysis model and stored in the DB. As a result, in each terminal, the features of the time series data obtained during the execution of processing are accumulated in the DB.

[0012] The server can know the state of each terminal by referring to the features stored in the DB of each terminal. Now, assume that an abnormal state occurs in one terminal X, and the server wants to check whether a similar abnormal state occurs in other terminals. In this case, the server checks whether time series data similar to the time series data acquired in the abnormal state of terminal X occurs in other terminals. If similar time series data is acquired in another terminal Y, the server can determine that the same abnormal state as terminal X also occurs in terminal Y.

[0013] In this case, the server searches each terminal for time series data detected in the abnormal state of terminal X. Specifically, the server first transmits the time series data of the abnormal state to each terminal. Each terminal converts the time series data of the abnormal state into a feature using an analysis model, and determines whether the same feature as the feature is stored in the DB. If the same feature is found in the DB, it is determined that a similar abnormal state has occurred in that terminal. On the other hand, if the same feature is not found in the DB, it is determined that a similar abnormal state has not occurred in that terminal. Then, each terminal transmits the determination result to the server. In this way, the server can check whether a similar abnormal state has occurred in each terminal. Note that in the above example, an abnormal state is detected, but the server can detect a specific state of each terminal, not limited to an abnormal state.

[0014] As mentioned above, the server can check the status of each terminal by searching the time series data on each terminal. However, this method generates a processing load by converting the time series data on multiple terminals into features and searching the DB. Also, in the case of an abnormal state, the time required to detect the state can be a problem.

[0015] Therefore, in this embodiment, a filter is generated using a common search model in each terminal, and the filters generated in each terminal are collected and stored in a server. The server then uses the filters collected from each terminal to estimate whether each terminal is in a specific state. This eliminates the need to transmit time-series data from the server to each terminal and perform searches on each terminal, making it possible to reduce processing load and shorten processing time.

[0016] FIG. 2 shows an outline of a method of estimating the state of each terminal by the server. First, the server distributes a common (identical) search model to the terminals. In the following, taking terminal A as an example, the server distributes the search model to terminal A (step S11). Terminal A inputs the time series data acquired by terminal A into the search model and generates filter A (step S12). The method of generating a filter will be described in detail later. The time series data used to create filter A is the same as that converted into features by the analysis model and stored in the DB, so filter A serves as an index of the features stored in the DB. Then, terminal A uploads the created filter A to the server (step S13). Similarly, other terminals also generate filters using the common search model distributed from the server and upload them to the server. As a result, the filters of each terminal are collected in the server.

[0017] Now, when detecting a terminal in the same state as the state in which certain time series data was obtained, the server uses the search model to generate a search bit string from the time series data (step S14). Specifically, the server inputs the time series data into the search model to convert it into features, and generates a search bit string based on the features. The server then compares the search bit string with the filter of each terminal, and estimates the terminal (hereinafter also referred to as a "matching terminal") that corresponds to the filter including the search bit string (step S15). The matching terminal is estimated to be in the same state as the terminal from which the time series data was obtained.

[0018] In this way, in this embodiment, a filter corresponding to an index of time series data in each terminal is collected in advance in the server, and the filter is used to estimate whether the target time series data is in each terminal, so that the state of each terminal can be estimated basically by processing only on the server side. Note that, based on the state estimation result on the server side, further time series data may be sent to the compatible terminal for confirmation or the like, and collated with the feature amount in the DB.

[0019] Next, the filter will be described in detail. FIG. 3 shows a method for creating a filter in each terminal. In this embodiment, a Bloom filter is created in each terminal. A search model distributed from a server to each terminal is a model for creating a Bloom filter. Specifically, time series data is input to the search model in each terminal. In FIG. 3, the time series data is a value X (X1, X2, ..). The search model applies k predetermined hash functions to the input value X, and generates an m-bit feature (hash code) using each hash function. As a result, k m-bit features are obtained for each value X.

[0020] Next, the search model combines the k m-bit features obtained into one m-bit bit string. Specifically, the search model ORs the k bit strings to generate one m-bit bit string. In this way, the search model generates one m-bit bit string for each value X. The search model then ORs the m-bit bit strings obtained for each value X to create a filter.

[0021] Figure 4 shows a search method using the filter thus generated. As shown in Figure 4(A), at the time of search, the server inputs the time series data to be searched (hereinafter also referred to as "target time series data") into the search model and obtains features (hash codes) using k hash functions. When there is one target time series data, the server uses the feature as the search bit string. On the other hand, when there are multiple target time series data, the server calculates the OR of the features of each target time series data to obtain the search bit string. The server then compares the search bit string with the filters of each terminal.

[0022] For simplicity, the filter and the search bit string are assumed to be 6 bits. The server judges whether the search bit string is included in the filter. Here, the fact that the search bit string is included in the filter means that all bits that are "1" in the search bit string are "1" in the filter. As shown in FIG. 4(B), when the search bit string is included in the filter, the server judges that there is a possibility of existence, that is, that the target time series data is likely to exist in the terminal. On the other hand, as shown in FIG. 4(C), when the search bit string is not included in the filter, the server judges that there is no possibility of existence, that is, that the target time series data is not included in the terminal. Note that in the Bloom filter, false positives occur but false negatives do not occur. Therefore, the server does not need to perform further processing such as confirmation for a terminal that is judged to have no possibility of existence using the filter.

[0023] First Embodiment Next, a first embodiment will be described. [Hardware configuration] Fig. 5 is a block diagram showing the hardware configuration of the server 100 and the terminal 200 of the first embodiment. Fig. 5(A) shows the hardware configuration of the server 100. As shown in Fig. 5(A), the server 100 includes a communication unit 111, a processor 112, a memory 113, a recording medium 114, a database (DB) 115, a display unit 116, and an input unit 117.

[0024] The communication unit 111 transmits and receives data to and from each terminal 200. Specifically, the communication unit 111 transmits a search model to each terminal 200, and receives filters from each terminal 200.

[0025] The processor 112 is a computer such as a CPU (Central Processing Unit) and controls the entire server 100 by executing a program prepared in advance. The processor 112 may be a GPU (Graphics Processing Unit) or an FPGA (Field-Programmable Gate Array). The processor 112 executes a state estimation process described later.

[0026] The memory 113 is composed of a read only memory (ROM), a random access memory (RAM), etc. The memory 113 is also used as a working memory while the processor 112 is executing various processes.

[0027] The recording medium 114 is a non-volatile and non-temporary recording medium such as a disk-shaped recording medium or a semiconductor memory, and is configured to be detachable from the server 100. The recording medium 114 records various programs executed by the processor 112. When the server 100 executes various processes, the programs recorded in the recording medium 114 are loaded into the memory 113 and executed by the processor 112. The DB 115 stores target time-series data to be subjected to state estimation, a search model, filters collected from each terminal 200, and the like.

[0028] The display unit 116 is, for example, a liquid crystal display device, and displays the results of the state estimation by the server 100. The input unit 117 is an input device such as a mouse, a keyboard, or a touch panel, and is operated by an operator when giving or inputting necessary instructions to the server 100.

[0029] 5(B) shows a hardware configuration of the terminal 200. As shown in FIG. 5(B), the terminal 200 includes a communication unit 211, a processor 212, a memory 213, a sensor 214, and a database (DB) 215.

[0030] The communication unit 211 transmits and receives data to and from the server 100. Specifically, the communication unit 211 receives a search model from the server 100, and transmits a filter created by the terminal 200 to the server 100.

[0031] The processor 212 is a computer such as a CPU, and controls the entire terminal 200 by executing a program prepared in advance. The processor 212 may be a GPU or an FPGA. The processor 212 executes a filter creation process described below. The memory 213 is composed of a ROM, a RAM, etc. The memory 213 is also used as a working memory while the processor 212 is executing various processes.

[0032] The sensor 214 generates time-series data related to the processing executed by the terminal 200. Note that the time-series data may be generated using a camera, a microphone, or the like instead of a sensor. The DB 215 stores the time-series data generated by the sensor 214, features obtained by converting the time-series data using an analysis model, filters generated by the terminal 200, and the like.

[0033] [Function configuration] 6 is a block diagram showing the functional configuration of the server 100 and the terminal 200. The server 100 includes a data input unit 121, a feature amount calculation unit 122, a terminal information management unit 123, a state estimation unit 124, and a confirmation request unit 125. The terminal 200 includes a data input unit 221, a filter creation unit 222, an analysis processing unit 223, and a server cooperation unit 224 in addition to the sensor 214 and DB 215.

[0034] First, a configuration on the terminal 200 side will be described. The sensor 214 continuously generates time-series data related to the processing performed by the terminal 200, and outputs the data to the data input unit 221. The data input unit 221 acquires the time-series data from the sensor 214, and outputs the data to the filter creation unit 222 and the analysis processing unit 223.

[0035] The analysis processing unit 223 performs regular analysis processing. That is, the analysis processing unit 223 converts the input time series data into feature quantities using a previously prepared analysis model, and stores the feature quantities in the DB 215. As a result, the time series data measured by the sensor 214 is converted into feature quantities and accumulated in the DB 215. The analysis processing unit 223 can perform necessary management for each terminal 200, such as issuing a warning when the feature quantities converted from the time series data correspond to a predetermined abnormal state.

[0036] The filter creation unit 222 creates a filter to be provided to the server 100. The filter creation unit 222 converts each time-series data input from the data input unit 221 into a bit string (hash code) using the search model distributed from the server 100, creates a filter by combining the bit strings of multiple time-series data, and outputs the filter to the server collaboration unit 224. The server collaboration unit 224 receives the search model from the server 100 and transmits the filter created by the filter creation unit 222 to the server 100. Note that the terminal 200 periodically creates a filter using the filter creation unit 222 and transmits it to the server 100, so that the latest filter created by the terminal 200 is always held in the server 100.

[0037] Next, the configuration of the server 100 side will be described. Target time series data to be subjected to state estimation is input to the data input unit 121. For example, when detecting a specific state in each terminal 200, target time series data corresponding to the state is input to the data input unit 121. The data input unit 121 outputs the input target time series data to the feature calculation unit 122.

[0038] The feature calculation unit 122 calculates features from the input target time series data using the search model. Specifically, the feature calculation unit 122 calculates m-bit features from the target time series data using k hash functions as described above, generates a search bit string, and outputs it to the state estimation unit 124.

[0039] The terminal information management unit 123 distributes a search model to each terminal 200. In addition, the terminal information management unit 123 manages information of each terminal 200. In addition, the terminal information management unit 123 stores filters collected from each terminal 200.

[0040] The state estimation unit 124 acquires the search bit string from the feature calculation unit 122 and the filter of each terminal 200 from the terminal information management unit 123, and estimates the state of each terminal. Specifically, the state estimation unit 124 compares the search bit string with the filter of each terminal 200. If the search bit string is included in the filter, that is, if the search bit string matches the filter, the state estimation unit 124 estimates that the target time series data may exist in the terminal 200 corresponding to the filter. On the other hand, if the search bit string does not match the filter, the state estimation unit 124 estimates that the target time series data may not exist in the terminal 200 corresponding to the filter. The state estimation unit 124 estimates that the terminal 200 in which the target time series data may exist has once fallen into a specific state indicated by the input time series data. In this way, by using the filters of each terminal 200 collected in advance, the server 100 can estimate that each terminal 200 is or has been in a specific state indicated by the time series data, using only processing on the server 100 side, i.e., without making any inquiries to each terminal 200.

[0041] The confirmation request unit 125 transmits a confirmation request to the terminal 200 that the state estimation unit 124 estimates to be in a specific state, as necessary. Specifically, the confirmation request unit 125 transmits time series data indicating a specific state to the terminal 200, and requests confirmation that the terminal 200 is in the specific state. When a confirmation request is received from the server 100, the server cooperation unit 224 in the terminal 200 receives the target time series data from the server 100 and outputs it to the analysis processing unit 223. The analysis processing unit 223 converts the target time series data into a feature using an analysis model, and determines whether the same feature is stored in the DB 215. The analysis processing unit 223 outputs the determination result to the server cooperation unit 224, and the server cooperation unit 224 transmits the determination result to the confirmation request unit 125 of the server 100 as a confirmation result in response to the confirmation request. In this way, as necessary, confirmation based on the feature stored in the DB 215 of each terminal 200 can also be performed.

[0042] In the above configuration, the data input unit 121 is an example of a data acquiring means, the state estimating unit 124 is an example of an estimating means, and the terminal information managing unit 123 is an example of a providing means and a filter acquiring means.

[0043] [Filter creation process] Next, a filter creation process in terminal 200 will be described. Fig. 7 is a flowchart of the filter creation process. This process is realized by processor 212 shown in Fig. 5(B) executing a program prepared in advance and operating as each element shown in Fig. 6.

[0044] First, server cooperation unit 224 acquires a search model from server 100 and supplies it to filter creation unit 222 (step S21). Next, filter creation unit 222 acquires time-series data measured by sensor 214 from data input unit 221 (step S22). Next, filter creation unit 222 uses the search model to generate k features (hash codes) from the time-series data (step S23), and combines the k features into one bit string (step S24).

[0045] Next, the filter creation unit 222 determines whether there is other time series data (step S25). The filter creation unit 222 groups a predetermined number of time series data and creates one filter for each group. That is, in step S25, the filter creation unit 222 determines whether the processing has been performed on all time series data belonging to one group.

[0046] If there is other time series data (step S25: Yes), the process returns to step S22, and steps S22 to S24 are repeated. On the other hand, if there is no other time series data (step S25: No), the filter creation unit 222 takes an OR of the obtained multiple bit strings to create a filter (step S26). Then, the server cooperation unit 224 transmits the created filter to the server 100 (step S27).

[0047] In this way, the filter created in each terminal 200 is transmitted to server 100. Note that terminal 200 performs the processes of steps S22 to S27 described above at predetermined time intervals or every time a predetermined amount of time-series data is acquired, updates the filter, and transmits it to server 100. As a result, the latest filter corresponding to each terminal 200 is always stored in terminal information management unit 123 of server 100.

[0048] [State estimation process] Next, a description will be given of the state estimation process in the server 100. Fig. 8 is a flowchart of the state estimation process. This process is realized by the processor 112 shown in Fig. 5(A) executing a program prepared in advance and operating as each element shown in Fig. 6.

[0049] First, target time series data corresponding to a specific state to be detected is input to the data input unit 121, and the data input unit 121 acquires the target time series data (step S31). Next, the feature calculation unit 122 calculates features from the target time series data using a search model, and generates a search bit string (step S32).

[0050] Next, the state estimation unit 124 selects a filter of one terminal in the terminal information management unit 123 (step S33), and determines whether or not the search bit string is included in the filter, i.e., whether or not the search bit string matches the filter (step S34). If the search bit string is included in the filter, the state estimation unit 124 estimates that the target time series data is included in the terminal corresponding to the filter, and if the search bit string is not included in the filter, the state estimation unit 124 estimates that the target time series data is not included in the terminal corresponding to the filter.

[0051] Next, the state estimation unit 124 judges whether or not there is another filter in the terminal information management unit 123 (step S35). If there is another filter (step S35: Yes), the state estimation unit 124 repeats the processes of steps S33 to S34 for the other filter. On the other hand, if there is no other filter (step S35: No), the state estimation unit 124 saves or outputs the estimation result based on each filter obtained in step S34 (step S36). For example, the server 100 can display on the display unit 116 a list of terminals 200 estimated to fall into the specific state to be detected, or transmit a confirmation request as described above to the terminals 200 estimated to fall into the state.

[0052] [Estimation of similar states] In the above example, in the state estimation in the server 100, a feature amount is calculated from the target time series data corresponding to a specific state to generate a search bit string, which is then compared with the filter of each terminal to estimate whether or not there is a state in each terminal that matches the specific state. Alternatively, a state similar to the specific state may be estimated by making the search bit string ambiguous. Specifically, the following method is considered.

[0053] (1) Inverting some bits in the search bit string In the above example, the state estimation unit 124 converts the target time-series data into features using a search model as shown in FIG. 4(A) to generate a search bit string, and then compares the search bit string with a filter as shown in FIGS. 4(B) and 4(C) to estimate whether or not there is a state that matches a specific state (e.g., a normal state or an abnormal state).

[0054] On the other hand, when estimating a state similar to a specific state, the state estimation unit 124 generates a search bit string by inverting one or several bits of the generated search bit string (hereinafter referred to as an "inverted bit string"), and compares the inverted bit string with the filter. Fig. 4(D) shows an example in which one bit of the search bit string is inverted. As a result, the inverted bit string becomes a bit string that indicates not only a specific state, but also a state similar to it, and by comparing this with the filter, it becomes possible to estimate a state similar to the specific state.

[0055] (2) How to split a filter into multiple ones In another method, the state estimation unit 124 divides the filter into multiple filters. For example, as shown in Fig. 4(E), the state estimation unit 124 divides a 7-bit filter into five filters of 3 bits each. In the example of Fig. 4(E), 3 bits out of the 7 bits of the original filter are selected while shifting 1 bit at a time from the left, and the original filter is divided into 5 filters.

[0056] The state estimation unit 124 divides the 7-bit search bit string into five bit strings of three bits each in a similar manner. Then, the state estimation unit 124 compares the divided 3-bit search bit string with the corresponding divided 3-bit filter. If the number of filters obtained by division is "a" and the number of search bit strings that match the filter is "b", the state estimation unit 124 determines that there is a state similar to a specific state when the ratio (b / a) of the number of matches between the search bit string and the filter to the number of divided filters is equal to or greater than a predetermined threshold. For example, assuming that the threshold is 4 / 5 in the above example, the state estimation unit 124 estimates that there is a state similar to a specific state when four or more of the five divided search bit strings match the filters.

[0057] [Example of application] Next, a specific application example of the above embodiment will be described. (1) Analysis of new device data The method of the embodiment can be used to analyze data obtained from a new terminal. Usually, when a new terminal is deployed at a certain location, the analysis model in the terminal is trained using local learning data acquired at that location. However, if a new terminal has just been deployed, there are cases where the local learning data has not been collected sufficiently, and the training for adapting the analysis model to the location is insufficient. In such a case, the target data (time series data) obtained from the new terminal can be simply analyzed using the filters of other existing terminals collected in the server.

[0058] Specifically, as shown in Fig. 9(A), the target data acquired at the new terminal is input to filters of existing terminals 1 to N collected in server 10, and the target data acquired at the new terminal is simply analyzed based on the results. Note that, in a similar manner, data from terminals other than the new terminal that do not have sufficient computational resources can also be simply analyzed using filters collected from other terminals.

[0059] (2) Operation mode estimation The method of the embodiment can be used to estimate the operation state (hereinafter also referred to as "operation mode") of each terminal. Specifically, as illustrated in FIG. 9(B), a filter is generated for each operation mode in each terminal, and collected in a server. For example, in each terminal, filter A is generated using time series data obtained during operation in operation mode A, and filter B is generated using time series data obtained during operation in operation mode B, and so on, generating a filter for each operation mode. In this way, when a feature value of certain time series data is included (matches) filter A, it can be estimated that the time series data is time series data obtained during operation in operation mode A.

[0060] Therefore, when a certain target time series data is obtained, the server applies the data to a group of filters collected from each terminal, and determines which operation mode the filter matches, thereby estimating which operation mode the time series data was obtained in. For example, the server compares certain time series data with filters A to E of each terminal, and counts up the filters that match. Then, if the filter that matches most often is filter C, the server 100 estimates that the time series data is close to operation mode C.

[0061] (3) Anomaly detection The method of the embodiment can be used to detect anomalies in each terminal. Specifically, as shown in Fig. 10, a filter is created in each terminal using time series data acquired when the terminal is operating normally. In this way, time series data acquired when the terminal is in a normal state matches this filter, and this filter can be used as a filter for detecting a normal state (hereinafter, referred to as a "normal state detection filter").

[0062] Therefore, the normal state detection filters created at each terminal are collected in a server, and the server periodically transmits time series data observed at each terminal to the server for monitoring. If the time series data transmitted from a terminal matches the normal state detection filter of that terminal, the server can estimate that the terminal is operating normally. Conversely, if the time series data transmitted from a terminal does not match the normal state detection filter of that terminal, the server can estimate that there may be an abnormality in the terminal. In this case, the server can make necessary inquiries to the terminal. In this way, by collecting normal state detection filters created at each terminal in the server, the server can monitor whether each terminal is operating normally or not.

[0063] (4) Maintenance of anonymity and confidentiality When the time-series data itself has high anonymity or confidentiality, integrating multiple filters enables searches that maintain anonymity. Now, assume that filters 1 to N have been generated in terminals 1 to N, as shown in Fig. 11. After collecting filters 1 to N from terminals 1 to N, the server integrates the multiple filters to create an integrated filter. For example, as shown in Fig. 11, the server integrates filters 1 and 2 to create integrated filter 1-2, and integrates filters 3 and 4 to create integrated filter 3-4.

[0064] When a certain target time series data matches the integrated filter 1-2, the server can estimate that the target time series data was obtained from either terminal 1 or terminal 2, but cannot specify whether the time series data was obtained from terminal 1 or terminal 2. In this way, the server can specify the terminals corresponding to the time series data in groups without specifying whether the time series data was obtained from terminal 1 or terminal 2. This makes it possible to maintain the anonymity and confidentiality of the time series data.

[0065] Fig. 12 shows an example of maintaining the anonymity of time-series data. In this example, a person's heart rate data is used to estimate whether the person has a certain disease (hereinafter referred to as "Disease X"). In this example, heart rate data is used as time-series data, and a feature is generated from the heart rate data of people diagnosed with Disease X to create a filter.

[0066] In one method, a filter is created individually using heart rate data for each person diagnosed with disease X, and the filters are collected. In this case, based on the result of matching certain heart rate data with the filters, the probability that the heart rate data corresponds to disease X can be calculated while maintaining anonymity. For example, as shown in FIG. 12(A), multiple filters are created based on the heart rate data of a person diagnosed with disease X. If certain heart rate data matches 80% of the filters as a result of matching the multiple filters, it can be estimated that the incidence rate of disease X for that heart rate data is 80%.

[0067] In another method, a filter may be created for each person diagnosed with disease X, and filters A and B may be created for each group of a predetermined number of people, as shown in Fig. 12(B). In this case, based on the results of comparing certain heart rate data with these filters, statistical information can be obtained that indicates what percentage of patients with disease X the heart rate data is found in, while maintaining anonymity.

[0068] In yet another method, as shown in Fig. 12(B), filters created individually for each person diagnosed with disease X may be integrated to create one filter C for all people. In this case, based on the result of matching certain heart rate data with the filter, it is possible to obtain information indicating whether the heart rate data is data found in at least one patient with disease X, while maintaining anonymity.

[0069] (5) When there are a huge number of terminals Even when the anonymity and confidentiality of time-series data are not an issue, the technique of integrating filters is effective, for example, when the number of terminals is huge. When the number of terminals is huge, the server needs to compare the time-series data with a huge number of filters to estimate the state of the terminals. Therefore, the filters are grouped into a predetermined number to create an integrated filter. For example, when there are 100 terminals, 10 terminals are grouped together, and the filters of each terminal are integrated in each group, so that 10 integrated filters are obtained. The server first compares the time-series data with the 10 integrated filters, and based on the result, it further compares the time-series data with the filters of each individual terminal belonging to the group as necessary to estimate the state.

[0070] For example, assuming that the filter is the normal state detection filter described above, the server first performs state estimation using 10 integrated filters to identify a group including a terminal in an abnormal state, and then performs state estimation using the normal state detection filters of the 10 terminals included in the group to estimate which terminals are in an abnormal state.

[0071] <Second embodiment>

[0072] 13 is a block diagram showing the functional configuration of an information processing device according to the second embodiment. The information processing device 70 includes a providing unit 71, a filter acquiring unit 72, a data acquiring unit 73, and an estimating unit 74.

[0073] 14 is a flowchart of processing by the information processing device 70 of the second embodiment. The providing means 71 provides a common search model to a plurality of terminals (step S41). The filter acquiring means 72 acquires a filter composed of features obtained using the search model from each terminal (step S42). The data acquiring means 73 acquires target time series data to be subjected to state estimation from each terminal (step S43). The estimating means 74 estimates the state of the terminal that output the target time series data based on the target time series data and the filter (step S44).

[0074] According to the information processing device of the second embodiment, it is possible to estimate the state of each terminal by using time-series data obtained at the terminal.

[0075] A part or all of the above-described embodiments can be described as, but is not limited to, the following supplementary notes.

[0076] (Appendix 1) A providing means for providing a common search model to a plurality of terminals; a filter acquisition means for acquiring, from each terminal, a filter configured with a feature amount obtained by using the search model; A data acquisition means for acquiring target time series data to be subjected to state estimation from each of the terminals; an estimation means for estimating a state of a terminal that outputs the target time series data based on the target time series data and the filter; An information processing device comprising:

[0077] (Appendix 2) The information processing device according to claim 1, wherein the estimation means extracts features from the target time series data using the search model, and estimates the state of each of the terminals based on whether the features are included in the filter.

[0078] (Appendix 3) 3. The information processing device according to claim 2, wherein the filter is formed by a logical sum of features extracted from the target time series data using the search model.

[0079] (Appendix 4) the filter is configured with a feature amount obtained from time-series data indicating a specific state of the terminal by using the search model; The information processing device according to claim 2 or 3, wherein the estimation means estimates that a terminal that outputs the target time series data is in a specific state corresponding to the filter when a feature extracted from the target time series data is included in the filter.

[0080] (Appendix 5) the filter is configured with a feature amount obtained from time-series data indicating an operation mode of the terminal by using the search model; The information processing device according to claim 2 or 3, wherein the estimation means estimates that a terminal that outputs the target time series data is in an operation mode corresponding to the filter when a feature extracted from the target time series data is included in the filter.

[0081] (Appendix 6) the filter is configured with a feature amount obtained from time-series data indicating a normal state or an abnormal state of the terminal by using the search model; The information processing device according to claim 2 or 3, wherein the estimation means estimates that a terminal that outputs the target time series data is in a normal state or an abnormal state when a feature extracted from the target time series data is included in the filter.

[0082] (Appendix 7) The information processing device according to any one of appendix 2 to 6, wherein the estimation means creates an integrated filter by integrating filters acquired from the multiple terminals, and compares features extracted from the target time series data with the integrated filter.

[0083] (Appendix 8) The information processing device according to any one of claims 1 to 7, wherein the estimation means estimates the state of the terminal using a filter in which some bits of the filter are inverted.

[0084] (Claim 9) The information processing device according to any one of claims 1 to 7, wherein the estimation means divides the filter into a plurality of filters and estimates the state of the terminal using the plurality of divided filters.

[0085] (Appendix 10) Providing a common search model for multiple devices, A filter configured with the feature amount obtained by using the search model is obtained from each terminal; Acquire target time series data to be subjected to state estimation from each of the terminals; An information processing method for estimating a state of a terminal that outputs the target time series data based on the target time series data and the filter.

[0086] (Appendix 11) Providing a common search model for multiple devices, A filter configured with the feature amount obtained by using the search model is obtained from each terminal; Acquire target time series data to be subjected to state estimation from each of the terminals; A recording medium having a program recorded thereon that causes a computer to execute a process of estimating the state of a terminal that outputs the target time series data based on the target time series data and the filter.

[0087] Although the present disclosure has been described above with reference to the embodiments and examples, the present disclosure is not limited to the above-mentioned embodiments and examples. Various modifications that can be understood by a person skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. [Explanation of symbols]

[0088] 100 Servers 112, 212 processors 121, 221 Data Entry Section 122 Feature Calculation Unit 123 Terminal Information Management Department 124 State Estimation Unit 125 Confirmation Request Department 200 terminals 222 Filter Creation Department 223 Analysis Processing Unit 224 Server Linkage Department

Claims

1. A providing means for providing a common search model to a plurality of terminals; a filter acquisition means for acquiring, from each terminal, a filter configured with a feature amount obtained from time-series data indicating a specific state of the terminal by using the search model; A data acquisition means for acquiring target time series data to be subjected to state estimation from each of the terminals; an estimation means for estimating a state of a terminal that outputs the target time series data based on the target time series data and the filter; An information processing device comprising:

2. The information processing apparatus according to claim 1 , wherein the estimation means extracts a feature from the target time-series data using the search model, and estimates the state of each of the terminals based on whether the feature is included in the filter.

3. The information processing apparatus according to claim 2 , wherein the filter is formed by a logical sum of feature amounts extracted from the target time-series data using the search model.

4. The information processing device according to claim 2 or 3, wherein the estimation means estimates that, when a feature extracted from the target time series data is included in the filter, the terminal that output the target time series data is in a specific state corresponding to the filter.

5. The specific state of the terminal includes an operation mode of the terminal, The information processing device according to claim 2 or 3, wherein the estimation means estimates that, when a feature extracted from the target time series data is included in the filter, the terminal that output the target time series data is in an operating mode corresponding to the filter.

6. The specific state of the terminal includes a normal state or an abnormal state of the terminal, The information processing device according to claim 2 or 3, wherein the estimation means estimates that a terminal that outputs the target time series data is in a normal state or an abnormal state when a feature extracted from the target time series data is included in the filter.

7. The information processing device according to claim 2 , wherein the estimation means creates an integrated filter by integrating filters acquired from the plurality of terminals, and compares the feature amount extracted from the target time series data with the integrated filter.

8. 8. The information processing device according to claim 1, wherein the estimation means estimates the state of the terminal by comparing an inverted bit string obtained by inverting some bits of a search bit string generated by converting the target time series data into a feature using the search model with the filter.

9. The information processing device according to claim 1 , wherein the estimation means divides the filter into a plurality of filters, and estimates the state of the terminal by using the plurality of divided filters.

10. 1. A computer-implemented information processing method, comprising: Providing a common search model for multiple devices, From each terminal, a filter configured with a feature amount obtained from time-series data indicating a specific state of the terminal is obtained using the search model; Acquire target time series data to be subjected to state estimation from each of the terminals; An information processing method for estimating a state of a terminal that outputs the target time series data based on the target time series data and the filter.

11. Providing a common search model for multiple devices, From each terminal, a filter configured with a feature amount obtained from time-series data indicating a specific state of the terminal is obtained using the search model; Acquire target time series data to be subjected to state estimation from each of the terminals; A program that causes a computer to execute a process of estimating the state of a terminal that outputs the target time series data based on the target time series data and the filter.

Images