Inference device, inference method, and inference program

Abstract

The present invention keeps constant the data length of time-series data input to a learning model. This inference device comprises: an acquisition unit that acquires time-series data until prescribed inference conditions are met; an adjustment unit that, if the data length of the acquired time-series data exceeds a prescribed data length, extracts data having the prescribed data length from the acquired time-series data; and an inference unit that inputs the extracted time-series data to the learning model and acquires inference results from the learning model.

Description

Inference apparatus, inference method and inference program

[0001] The present invention relates to an inference apparatus, an inference method and an inference program.

[0002] In recent years, machine learning has been used in various fields. Patent Document 1 describes acquiring heartbeat data of a predetermined number of data points, cutting out data centering on the R point of the acquired heartbeat data, and performing binary classification using the cut-out data.

[0003] Japanese Unexamined Patent Publication No. 2020-524065

[0004] When inference is performed by inputting time-series data into a learning model, the time-series data is acquired until an inferable state is reached. Then, by inputting the acquired time-series data into the learning model, an inference result can be obtained from the learning model. Here, the data length of the time-series data acquired until the inferable state is reached may differ each time inference is performed. There is a possibility that inference accuracy may vary depending on the data length of the time-series data input to the learning model. Therefore, it is preferable that the data length of the time-series data input to the learning model is constant.

[0005] One aspect of the disclosed technology is an object of providing an inference apparatus, an inference method and an inference program that can make the data length of time-series data input to a learning model constant.

[0006] One aspect of the disclosed technology is exemplified by the following inference apparatus. The inference apparatus includes: an acquisition unit that acquires time-series data until a predetermined inference condition is satisfied; an adjustment unit that cuts out data of the predetermined data length from the acquired time-series data when the data length of the acquired time-series data is longer than the predetermined data length; and an inference unit that inputs the cut-out time-series data into a learning model and acquires an inference result obtained by the learning model.

[0007] According to this inference device, a predetermined data length is extracted from the acquired time-series data, thereby adjusting the data length of the data input to the learning model to the predetermined data length. Therefore, this inference device can keep the data length of the time-series data input to the learning model constant.

[0008] The inference device may further have the following features: The adjustment unit extracts data of a predetermined length from the time series data so that the end of the acquired time series data is included. The time series data is acquired until the predetermined inference conditions are met. In other words, the data at the end of the time series data is considered important data for satisfying the predetermined inference conditions. By extracting data from the time series data so that the end of the time series data is included, the inference device can input data containing data important for satisfying the predetermined inference conditions into the learning model. The adjustment unit may also extract data of a predetermined length from the time series data so that the beginning of the acquired time series data is included.

[0009] Furthermore, the inference device may include: an acquisition unit that acquires time-series data until predetermined inference conditions are met; an adjustment unit that, if the data length of the acquired time-series data is insufficient to a predetermined data length, adjusts the acquired time-series data by imputing missing values ​​in the insufficient portion of the acquired time-series data so that the data length of the imputed time-series data becomes the predetermined data length; and an inference unit that inputs the imputed time-series data into a learning model and acquires the inference result from the learning model.

[0010] With this type of inference device, missing values ​​are imputed, and the data length of the data input to the learning model is adjusted to the predetermined data length. Therefore, this inference device can keep the data length of the time-series data input to the learning model constant.

[0011] The inference device may further have the following features. The adjustment unit imputes the missing values ​​at the end of the acquired time series data. When the missing values ​​are stored in this way, the data length of the data input to the learning model will be from the beginning of the data to where the missing values ​​appear. Therefore, it becomes easier for the learning model to acquire the data length of the imputed time series data as a feature. The adjustment unit may also impute the missing values ​​at the beginning of the acquired time series data. By imputing the missing values ​​at the beginning of the acquired time series data, the end portion of the time series data can be placed at the end of the data input to the learning model, regardless of the data length of the time series data. Therefore, when features appear at the end of the time series data, an improvement in the inference accuracy regarding the measurement target can be expected.

[0012] This inference device may further have the following features: The missing value is "0". When the missing value is set to "0", the result of the calculation in the part of the learning model in which the missing value is imputed becomes "0". Therefore, the decrease in the inference accuracy of the learning model due to the missing value acting as noise is suppressed.

[0013] The inference device may further include the following features: a cuff for compressing the area to be measured of a person being measured, a first determination unit, and a second determination unit. The acquisition unit controls the compression by the cuff to acquire time-series data of the pulse wave, the first determination unit determines whether the time-series data satisfies the predetermined inference conditions, and if it determines that the predetermined inference conditions are met, it terminates the acquisition of the time-series data by the acquisition unit, the second determination unit determines whether the data length of the time-series data acquired up to the time termination by the first determination unit matches the predetermined data length, and if the second determination unit determines that the data length of the time-series data does not match the predetermined data length, the adjustment unit adjusts the data length of the time-series data to match the predetermined data length, and the inference unit acquires the blood pressure of the person being measured. With such an inference device, the data length of the data input to the learning model is adjusted to the predetermined data length, so that the influence of variations in the data length of the data input to the learning model on the accuracy of blood pressure inference is suppressed.

[0014] The technology of disclosure can also be understood from the perspective of inference methods and inference programs.

[0015] According to the disclosed technology, the data length of the time-series data input to the learning model can be kept constant.

[0016] Figure 1 shows an example of a blood pressure monitor according to the embodiment. Figure 2 shows an example of a processing block for the blood pressure monitor body according to the embodiment. Figure 3 shows an example of a process to adjust the data length by extracting data from the acquired time-series data by the adjustment unit, according to the embodiment. Figure 4 shows an example of a process to adjust the data length by imputing the missing parts of the acquired time-series data by the adjustment unit with missing values, according to the embodiment. Figure 5 shows an example of a processing flow for blood pressure measurement processing by the blood pressure monitor according to the embodiment. Figure 6 shows another example of a process to adjust the data length by extracting data from the acquired time-series data by the adjustment unit. Figure 7 shows another example of a process to adjust the data length by imputing the missing parts of the acquired time-series data by the adjustment unit with missing values.

[0017] <Application Examples> An example of the application of the present invention will now be described. An example of the application of the present invention is the blood pressure monitor 1 illustrated in Figures 1 and 2. In the blood pressure monitor 1, the inference unit 105 inputs time-series data of cuff pressure and pulse wave acquired by the acquisition unit 101 into the learning model 106 to measure the blood pressure of the person being measured.

[0018] In the blood pressure inference using the learning model 106, the data length of the input time series data can be considered a feature. If the data length of the time series data input to the learning model 106 is not constant, the accuracy of blood pressure inference using the learning model 106 may decrease.

[0019] Therefore, in this application example, the adjustment unit 104 adjusts the data length of the time series data acquired by the acquisition unit 101 until a predetermined inference condition is met. For example, if the data length of the time series data is longer than a predetermined data length, the adjustment unit 104 adjusts the data length by extracting data of the predetermined data length from the acquired time series data. Also, if the data length of the acquired time series data is shorter than a predetermined data length, the adjustment unit 104 adjusts the data length by supplementing the missing values ​​for the portion of the time series data that is insufficient for the predetermined data length, thereby increasing the data length. As a result, according to this application example, the data length of the time series data input to the learning model 106 can be kept constant. The predetermined inference condition is an example of a "predetermined inference condition".

[0020] <Embodiments> Embodiments will be described below with reference to the drawings. Figure 1 is a diagram showing an example of a blood pressure monitor 1 according to an embodiment. The blood pressure monitor 1 is a device that measures the blood pressure of a person by wrapping a cuff 16 around the area to be measured and compressing the area with the cuff 16. The area to be measured can be, for example, the upper arm.

[0021] Figure 1 shows an example of the hardware configuration of a blood pressure monitor 1. The blood pressure monitor 1 comprises a blood pressure monitor body 10 and a cuff 16. The blood pressure monitor body 10 is an information processing device comprising a Central Processing Unit (CPU) 11, a main memory unit 12, an auxiliary memory unit 13, a display 14, a connection unit 15, a cuff 16, and a connection bus B1. The CPU 11, main memory unit 12, auxiliary memory unit 13, display 14, and connection unit 15 are interconnected by the connection bus B1.

[0022] The CPU 11 is also called a microprocessor unit (MPU) or processor. The CPU 11 is not limited to a single processor and may be in a multiprocessor configuration. A single CPU 11 connected via a single socket may also have a multicore configuration. At least a portion of the processing performed by the CPU 11 may be performed by a processor other than the CPU 11, such as a dedicated processor like a Digital Signal Processor (DSP), Graphics Processing Unit (GPU), numerical processor, vector processor, or image processing processor. At least a portion of the processing performed by the CPU 11 may also be performed by an integrated circuit (IC) or other digital circuit. Furthermore, at least a portion of the CPU 11 may include analog circuits. Integrated circuits include Large Scale Integrated Circuits (LSIs), Application Specific Integrated Circuits (ASICs), and Programmable Logic Devices (PLDs). PLDs include, for example, Field-Programmable Gate Arrays (FPGAs). The CPU 11 may be a combination of a processor and an integrated circuit. Such combinations are called, for example, a microcontroller unit (MCU), a System-on-a-chip (SoC), a system LSI, or a chipset. In the blood pressure monitor 1, the CPU 11 deploys a program stored in the auxiliary storage unit 13 to the work area of ​​the main storage unit 12 and controls peripheral devices through program execution. This allows the blood pressure monitor 1 to perform processing that matches a predetermined purpose. The main memory unit 12 and the auxiliary memory unit 13 are recording media that can be read by the CPU 11.

[0023] The main memory unit 12 is exemplified as a memory unit that is directly accessed by the CPU 11. The main memory unit 12 includes Random Access Memory (RAM) and Read Only Memory (ROM).

[0024] The auxiliary storage unit 13 stores various programs and data on a recording medium in a read-write manner. The auxiliary storage unit 13 is also called an external storage device. The auxiliary storage unit 13 stores the Operating System (OS), various programs, various tables, etc. The auxiliary storage unit 13 may, for example, be part of a cloud system, which is a group of computers on a network.

[0025] The auxiliary storage unit 13 is, for example, an Erasable Programmable ROM (EPROM), a Solid State Drive (SSD), a Hard Disk Drive (HDD), etc.

[0026] The display 14 displays data processed by the CPU 11 and data stored in the main memory 12. The display 14 is, for example, a Liquid Crystal Display (LCD), a Plasma Display Panel (PDP), an inorganic Electroluminescence (EL) panel, or an organic EL panel. The display 14 may also have a touch panel superimposed on it, for example, to detect touch operations by the user's finger. By superimposing a touch panel on the display 14, the smartphone 100 can provide the user with an intuitive operating environment.

[0027] The connection part 15 is an interface for connecting the cuff 16 to the blood pressure monitor body 10. The cuff 16 is connected to the blood pressure monitor body 10 via the connection part 15. The blood pressure monitor body 10 controls the compression of the area to be measured using the cuff 16 via the connection part 15.

[0028] The cuff 16 is a component that is wrapped around the area to be measured. The cuff 16 is made of a flexible material such as cloth. The cuff 16 is equipped with a cuff pressure sensor 17 for detecting cuff pressure. Cuff pressure includes, for example, the pressure exerted by the cuff 16 on the area to be measured and pressure vibrations caused by the pulsation of blood flow in the arteries of the area to be measured.

[0029] <Processing Block of Blood Pressure Monitor Body 10> Figure 2 shows an example of the processing block of the blood pressure monitor body 10 according to the embodiment. The blood pressure monitor body 10 includes an acquisition unit 101, a first determination unit 102, a second determination unit 103, an adjustment unit 104, an inference unit 105, and a learning model 106. The blood pressure monitor body 10 performs processing as each part of the blood pressure monitor body 10, such as the acquisition unit 101, the first determination unit 102, the second determination unit 103, the adjustment unit 104, the inference unit 105, and the learning model 106, by having the CPU 11 execute a computer program that has been loaded into the main memory unit 12 in an executable manner.

[0030] The acquisition unit 101 controls the compression applied to the measurement site by the cuff 16 and acquires the time-series change of cuff pressure detected by the cuff pressure sensor 17. The acquisition unit 101 also acquires the time-series change of pulse wave from the acquired time-series change of cuff pressure. Hereinafter in this specification, the time-series change of cuff pressure and the time-series change of pulse wave acquired by the acquisition unit 101 will also be referred to as acquired time-series data.

[0031] The first determination unit 102 determines whether the acquired time-series data satisfies predetermined inference conditions. If the first determination unit 102 determines that the predetermined inference conditions are met, it terminates the acquisition of time-series changes in cuff pressure and pulse wave by the acquisition unit 101. The predetermined inference conditions are, for example, conditions indicating that the inference unit 105 can estimate the blood pressure of the person being measured from the acquired time-series data. For example, the predetermined inference conditions include the time when the amplitude of the pulse wave exceeds the maximum value and then falls below a certain percentage of the maximum value of the pulse wave. Another example of the predetermined inference conditions is when the amplitude of the pulse wave exceeds the maximum value, several more pulse waves are detected, and the maximum value of the pulse wave is determined.

[0032] The second determination unit 103 compares whether the data length of the acquired time-series data acquired up to the time of termination by the first determination unit 102 matches a predetermined data length pre-stored in the auxiliary storage unit 13. The second determination unit 103 then determines whether the data length of the acquired time-series data matches the predetermined data length, is longer than the predetermined data length, or is shorter than the predetermined data length. The predetermined data length may be appropriately determined during the design of the blood pressure monitor 1 based on the blood pressure inference accuracy of the inference unit 105 and the learning model 106. Alternatively, the predetermined data length may be the size specified as the data length of the input data accepted by the learning model 106. The data length is indicated by time, for example, 10 seconds or 1 minute.

[0033] If the second determination unit 103 determines that the data length of the acquired time series data does not match a predetermined data length, the adjustment unit 104 adjusts the data length of the acquired time series data so that it matches the predetermined data length.

[0034] The adjustment unit 104 adjusts the data length of the acquired time series data by, for example, extracting a predetermined amount of data from the acquired time series data if the data length of the acquired time series data is longer than a predetermined data length. Figure 3 is a diagram showing an example of the process by which the adjustment unit 104 adjusts the data length by extracting data from the acquired time series data D1 in an embodiment. In the example of Figure 3, the acquisition unit 101 starts acquiring the time series change of cuff pressure from time T1, and at time T2, the first determination unit 102 determines that a predetermined inference condition is met. As illustrated in the upper part of Figure 3, the data length W2 of the acquired time series data D1 is longer than the predetermined data length W1.

[0035] In such cases, the adjustment unit 104 extracts data from the acquired time series data D1 by a predetermined data length W1 so that the data at time T2, which is the end of the acquired time series data D1, is included. As a result of the extraction by the adjustment unit 104, the extracted data D2 illustrated in the lower part of Figure 3 is obtained. Since the extracted data D2 was extracted from the acquired time series data D1 so that the data at time T2, which is the end of the acquired time series data D1, is included, the beginning portion of the acquired time series data D1 is not included. Also, the data length of the extracted data D2 matches the predetermined data length W1.

[0036] Furthermore, the adjustment unit 104 adjusts the data length of the acquired time series data by filling in the missing values ​​for the portion that is insufficient compared to the predetermined data length W1 of the acquired time series data, for example, if the data length of the acquired time series data is shorter than a predetermined data length. Figure 4 is a diagram showing an example of the process by which the adjustment unit 104 adjusts the data length of the acquired time series data D11 by filling in the missing portion with missing values ​​in an embodiment. In the example of Figure 4, the acquisition unit 101 starts acquiring the time series change of cuff pressure from time T11, and at time T12, the first determination unit 102 determines that the predetermined inference conditions are met. As illustrated in the upper part of Figure 4, the data length W12 of the acquired time series data D11 is shorter than the predetermined data length W1.

[0037] In such cases, the adjustment unit 104 interpolates the missing value in the missing portion R1 of the acquired time series data D11 (the portion from time T12 onwards, and from time T11 until a predetermined data length W1 is satisfied). In other words, the missing value is interpolated at the end of the acquired time series data D11. The missing value can be any value. For example, if "0" is adopted as the missing value, the absence of the missing value as noise during inference by the inference unit 105 is suppressed, and consequently, the decrease in the inference accuracy of the inference unit 105 is suppressed. Therefore, in this embodiment, "0" is adopted as the missing value.

[0038] As a result of imputing missing values, the imputed data D13, illustrated in the lower part of Figure 4, is obtained. The imputed data D13 includes the acquired time series data D11 and the imputed data D12, which is imputed by the missing values. The data length of the imputed data D13 is the sum of the data length W12 of the acquired time series data D11 and the data length W13 of the imputed data D12. The data length of the imputed data D13 matches a predetermined data length W1. Hereinafter in this specification, data whose data length has been adjusted to a predetermined data length W1 by the adjustment unit 104 through extraction of acquired time series data or imputation of missing values ​​will also be referred to as adjusted data.

[0039] Returning to Figure 2, the inference unit 105 inputs the adjusted data, whose data length has been adjusted by the adjustment unit 104, into the learning model 106, and obtains blood pressure values ​​such as the subject's minimum and maximum blood pressure values ​​as an inference result from the learning model 106. The inference unit 105 outputs the blood pressure values ​​obtained from the learning model 106 to, for example, the display 14.

[0040] The learning model 106 is a learning model that outputs diastolic blood pressure and systolic blood pressure values ​​when time-series changes in cuff pressure and pulse wave are input. The learning model 106 is constructed, for example, by supervised machine learning with time-series changes in cuff pressure and pulse wave as explanatory variables and diastolic blood pressure and systolic blood pressure values ​​as target variables.

[0041] <Processing Flow> Figure 5 shows an example of the processing flow for blood pressure measurement by the blood pressure monitor 1 according to this embodiment. Hereinafter, an example of the processing flow for blood pressure measurement by the blood pressure monitor 1 will be described with reference to Figure 5.

[0042] In step S1, the acquisition unit 101 controls the compression applied to the measurement site by the cuff 16 and acquires the time-series changes in cuff pressure and pulse wave.

[0043] In step S2, the first determination unit 102 determines whether or not the acquired time-series data obtained in step S1 satisfies a predetermined inferable condition. If the predetermined inferable condition is satisfied (YES in step S2), the process proceeds to step S3. If the predetermined inferable condition is not satisfied (NO in step S2), the process of step S1 is repeated, whereby the acquisition unit 101 continues acquiring the time-series change of cuff pressure and the time-series change of the pulse wave.

[0044] In step S3, the first determination unit 102 terminates the acquisition of the time-series change of cuff pressure and the time-series change of the pulse wave by the acquisition unit 101.

[0045] In step S4, the second determination unit 103 compares the data length of the acquired time-series data obtained up to step S3 with a predetermined data length. If the data length of the acquired time-series data is longer than the predetermined data length ("longer" in step S4), the process proceeds to step S5. If the data length of the acquired time-series data is shorter than the predetermined data length ("shorter" in step S4), the process proceeds to step S6. If the data length of the acquired time-series data matches the predetermined data length ("matched" in step S4), the process proceeds to step S7.

[0046] In step S5, the adjustment unit 104 adjusts the data length of the acquired time-series data by cutting out data of a predetermined data length from the acquired time-series data.

[0047] In step S6, the adjustment unit 104 adjusts the data length of the acquired time-series data by supplementing missing values for the part that is insufficient with respect to the predetermined data length.

[0048] In step S7, the inference unit 105 inputs the adjusted data whose data length has been adjusted in step S5 or step S6, or the acquired time-series data when it is determined in step S4 that the data length of the acquired time-series data matches the predetermined data length, into the learning model 106, and acquires the blood pressure value of the subject. The inference unit 105 may output the acquired blood pressure value to the display 14.

[0049] <Effects of the Embodiment> In inference by the learning model 106, for example, the data length of the input data can also be one of the features. Therefore, if there is variation in the data length of the input data, it may affect the inference accuracy of the learning model 106. According to this embodiment, the data length of the data input to the learning model 106 is adjusted to a predetermined data length. In this embodiment, for example, if the data length of the acquired time series data is longer than the predetermined data length, the data length of the acquired time series data is adjusted by extracting data of the predetermined data length from the acquired time series data. Also, if the data length of the acquired time series data is shorter than the predetermined data length, the data length of the acquired time series data is adjusted by interpolating missing values ​​for the portion that is insufficient to the predetermined data length. Therefore, the impact of variations in the data length of the data input to the learning model 106 on the inference accuracy of the learning model 106 is suppressed.

[0050] In this embodiment, when data of a predetermined length is extracted from the acquired time series data, data of a predetermined length W1 is extracted from the acquired time series data D1 so that the data at the end of the acquired time series data D1 is included. The data near the end of the acquired time series data D1 is considered important data for inference in the inference unit 105 using the learning model 106 because it is data from a timing that satisfies predetermined inference conditions. Therefore, according to this embodiment, it is possible to suppress the deletion of important data when extracting data from the acquired time series data D1.

[0051] In this embodiment, when imputing missing values ​​in the acquired time-series data D11, "0" is used as the missing value. By using "0" as the missing value, the absence of missing values ​​as noise during inference by the inference unit 105 is suppressed, and consequently, the decrease in the inference accuracy of the inference unit 105 is suppressed.

[0052] <Modification> In the embodiment described above, if the data length of the acquired time series data D1 is longer than a predetermined data length, the adjustment unit 104 extracts the data so that the last data of the acquired time series data D1 is included. However, the adjustment unit 104 may, for example, extract the data so that the first data of the acquired time series data D1 is included (for example, the data at time T1 in Figure 3).

[0053] Figure 6 shows another example of the process of adjusting the data length by extracting data from the acquired time series data D1 by the adjustment unit 104. In the example in Figure 6, the data is extracted so that it includes the data at time T1, which is the beginning of the acquired time series data D1, resulting in extracted data D22. Depending on the measurement target, it is possible that inputting extracted data D22 instead of extracted data D2 into the learning model may improve the inference accuracy using the learning model.

[0054] Furthermore, in the design stage of the blood pressure monitor 1, if the data length of the acquired time-series data is longer than a predetermined data length, the adjustment unit 104 may employ a method of extraction by comparing the inference accuracy when the inference unit 105 performs inference using the extracted data D2 and when the inference unit 105 performs inference using the extracted data D22, and generating extracted data with higher inference accuracy.

[0055] In the embodiments described above, the adjustment unit 104 imputed missing values ​​at the end of the acquired time series data D11 when the data length of the acquired time series data was shorter than a predetermined data length W1. However, the adjustment unit 104 may also imputate missing values ​​at the beginning of the acquired time series data D11. When missing values ​​are imputed at the beginning, the end portion of the acquired time series data D11 will be located at the end portion of the data after imputation. That is, in both cases—data in which missing values ​​were imputed because it was shorter than the predetermined data length W1, and data in which missing values ​​were not imputed because it matched the predetermined data length W1—the end portion of the acquired time series data D11 will be placed at the end of the data to be input to the learning model 106. In other words, regardless of the data length of the acquired time series data D11, the end portion of the acquired time series data D11 can be placed at the end of the data to be input to the learning model 106. Therefore, in cases where features appear in the end portion of the acquired time series data D11, an improvement in the inference accuracy regarding the measurement target can be expected.

[0056] Figure 7 shows another example of the process by which the adjustment unit 104 adjusts the data length by imputing the missing portion of the acquired time series data D11 with missing values. In the example in Figure 7, missing values ​​are imputed for the missing portion R11 (the portion before the data at time T11) that precedes the beginning of the acquired time series data D11. The imputed data D33, exemplified in the lower part of Figure 4, in which the missing values ​​are stored, includes the acquired time series data D11 and the imputed data D32 which is imputed by the missing values. The data length of the imputed data D33 matches a predetermined data length W1. Depending on the measurement target, it is conceivable that inputting the imputed data D33 instead of the imputed data D13 into the learning model may improve the inference accuracy using the learning model. Furthermore, in inference using a learning model, if the data length is an important feature, it is considered that using the imputed data D13 will improve the inference accuracy because it becomes easier to acquire the data length from time T11 to time T12 when the missing values ​​begin.

[0057] Furthermore, in the design stage of the blood pressure monitor 1, if the data length of the acquired time-series data is shorter than a predetermined data length, the adjustment unit 104 may employ a method of interpolation by comparing the inference accuracy when the inference unit 105 performs inference using the interpolated data D13 and when the inference unit 105 performs inference using the interpolated data D33, and generating interpolated data with higher inference accuracy.

[0058] In the embodiment described above, the blood pressure of the subject was inferred by the blood pressure monitor 1, but the inference target of this embodiment is not limited to blood pressure. The technique of keeping the data length of the data input to the learning model constant in this embodiment is also effective for various inference targets other than blood pressure.

[0059] The embodiments and variations disclosed above can be combined in any way.

[0060] <Computer-readable recording medium> An information processing program that enables a computer or other machine or device (hereinafter referred to as "computer, etc.") to perform any of the above functions can be recorded on a computer-readable recording medium. By having the computer, etc. read and execute the program on this recording medium, it can provide that function.

[0061] Here, a recording medium that can be read by a computer refers to a recording medium that stores information such as data and programs through electrical, magnetic, optical, mechanical, or chemical means and can be read by a computer. Examples of such recording media that can be removed from a computer include flexible disks, magneto-optical disks, Compact Disc Read Only Memory (CD-ROM), Compact Disc-Recordable (CD-R), Compact Disc-ReWritable (CD-RW), Digital Versatile Disc (DVD), Blu-ray Disc (BD), Digital Audio Tape (DAT), 8mm tape, flash memory, external hard disk drives, and Solid State Drives (SSDs). Furthermore, there are internal hard disk drives, SSDs, and ROMs as recording media fixed to computers and other devices.

[0062] <Note 1> An inference device (1) comprising: an acquisition unit (101) that acquires time-series data until predetermined inference conditions are met; an adjustment unit (104) that, if the data length of the acquired time-series data is greater than the predetermined data length, extracts data of the predetermined data length from the acquired time-series data; and an inference unit (105) that inputs the extracted time-series data to a learning model (106) and obtains an inference result from the learning model (106). <Note 2> The inference device (1) according to Note 1, wherein the adjustment unit (104) extracts data of the predetermined data length from the time-series data so that the end of the acquired time-series data is included. <Note 3> The inference device (1) according to Note 1, wherein the adjustment unit (104) extracts data of the predetermined data length from the time-series data so that the beginning of the acquired time-series data is included. <Note 4> An inference device (1) comprising: an acquisition unit (101) that acquires time series data until predetermined inference conditions are met; an adjustment unit (102) that, if the data length of the acquired time series data is insufficient to a predetermined data length, adjusts the acquired time series data so that the data length of the time series data after the adjustment becomes the predetermined data length by imputing missing values ​​in the insufficient portion of the acquired time series data; and an inference unit (105) that inputs the imputed time series data to a learning model (106) and acquires the inference result by the learning model (106). <Note 5> The inference device (1) according to Note 4, wherein the adjustment unit (104) imputes the missing values ​​at the end of the acquired time series data. <Note 6> The inference device (1) according to Note 4, wherein the adjustment unit imputes the missing values ​​at the beginning of the acquired time series data. <Note 7> The inference device (1) according to any one of Notes 4 to 6, wherein the missing value is "0".<Note 8> The inference device (1) according to Note 1 or 4 further comprises a cuff for compressing a part of a person to be measured, a first determination unit, and a second determination unit, wherein the acquisition unit (101) controls the compression by the cuff (16) to acquire time-series data of pulse waves, the first determination unit (102) determines whether the time-series data satisfies the predetermined inference conditions, and if it determines that the predetermined inference conditions are satisfied, terminates the acquisition of the time-series data by the acquisition unit, the second determination unit (103) determines whether the data length of the time-series data acquired up to the time termination by the first determination unit (102) matches the predetermined data length, the adjustment unit (104) adjusts the data length of the time-series data to match the predetermined data length if the second determination unit (103) determines that the data length of the time-series data does not match the predetermined data length, and the inference unit (105) acquires the blood pressure of the person to be measured. <Note 9> An inference method comprising: an acquisition process in which a computer acquires time-series data until predetermined inference conditions are met; an adjustment process in which, if the data length of the acquired time-series data is greater than the predetermined data length, data equivalent to the predetermined data length is extracted from the acquired time-series data; and an inference process in which the extracted time-series data is input into a learning model and an inference result is obtained by the learning model. <Note 10> An inference program that causes a computer to perform an acquisition process in which it acquires time-series data until predetermined inference conditions are met; an adjustment process in which, if the data length of the acquired time-series data is greater than the predetermined data length, data equivalent to the predetermined data length is extracted from the acquired time-series data; and an inference process in which the extracted time-series data is input into a learning model and an inference result is obtained by the learning model.<Note 11> An inference method comprising: an acquisition process in which a computer acquires time-series data until predetermined inference conditions are met; an adjustment process in which, if the data length of the acquired time-series data is insufficient to a predetermined data length, the acquired time-series data is supplemented with missing values ​​to fill in the missing portion of the acquired time-series data so that the data length of the supplemented time-series data becomes the predetermined data length; and an inference process in which the supplemented time-series data is input into a learning model and an inference result is obtained by the learning model. <Note 12> An inference program that causes a computer to perform an acquisition process in which it acquires time-series data until predetermined inference conditions are met; an adjustment process in which, if the data length of the acquired time-series data is insufficient to a predetermined data length, the acquired time-series data is supplemented with missing values ​​to fill in the missing portion of the acquired time-series data so that the data length of the supplemented time-series data becomes the predetermined data length; and an inference process in which the supplemented time-series data is input into a learning model and an inference result is obtained by the learning model.

[0063] 1. Blood pressure monitor 10. Blood pressure monitor body 11. CPU 12. Main memory unit 13. Auxiliary memory unit 14. Display 15. Connection unit 16. Cuff 17. Cuff pressure sensor 101. Acquisition unit 102. First determination unit 103. Second determination unit 104. Adjustment unit 105. Inference unit 106. Learning model B1. Connection bus

Claims

1. An inference device comprising: an acquisition unit that acquires time-series data until predetermined inference conditions are met; an adjustment unit that, if the data length of the acquired time-series data is greater than a predetermined data length, extracts data of the predetermined data length from the acquired time-series data; and an inference unit that inputs the extracted time-series data into a learning model and obtains an inference result from the learning model.

2. The inference apparatus according to claim 1, wherein the adjustment unit extracts data of a predetermined data length from the time series data so that the end of the acquired time series data is included.

3. The inference apparatus according to claim 1, wherein the adjustment unit extracts data of a predetermined data length from the time series data so that the beginning of the acquired time series data is included.

4. An inference device comprising: an acquisition unit that acquires time series data until predetermined inference conditions are met; an adjustment unit that, if the data length of the acquired time series data is insufficient to a predetermined data length, adjusts the acquired time series data by imputing missing values ​​for the insufficient portion of the acquired time series data so that the data length of the imputed time series data becomes the predetermined data length; and an inference unit that inputs the imputed time series data into a learning model and acquires the inference result from the learning model.

5. The inference device according to claim 4, wherein the adjustment unit imparts the missing values ​​to the end of the acquired time-series data.

6. The inference device according to claim 4, wherein the adjustment unit imparts the missing values ​​to the beginning of the acquired time-series data.

7. The inference device according to any one of claims 4 to 6, wherein the missing value is "0".

8. An inference device according to claim 1 or 4, further comprising a cuff for compressing a part of a person to be measured, a first determination unit, and a second determination unit, wherein the acquisition unit controls the compression by the cuff to acquire time-series data of a pulse wave, the first determination unit determines whether the time-series data satisfies predetermined inference conditions, and if it determines that the predetermined inference conditions are satisfied, terminates the acquisition of the time-series data by the acquisition unit, the second determination unit determines whether the data length of the time-series data acquired up to the time of termination by the first determination unit matches a predetermined data length, the adjustment unit adjusts the data length of the time-series data to match the predetermined data length if the second determination unit determines that the data length of the time-series data does not match the predetermined data length, and the inference unit acquires the blood pressure of the person to be measured.

9. An inference method comprising: an acquisition process in which a computer acquires time-series data until predetermined inference conditions are met; an adjustment process in which, if the data length of the acquired time-series data is greater than a predetermined data length, data equivalent to the predetermined data length is extracted from the acquired time-series data; and an inference process in which the extracted time-series data is input into a learning model and an inference result is obtained by the learning model.

10. An inference program that causes a computer to perform an acquisition process to acquire time-series data until predetermined inference conditions are met; an adjustment process to extract data of the predetermined length from the acquired time-series data if the data length of the acquired time-series data is greater than the predetermined data length; and an inference process to input the extracted time-series data into a learning model and obtain an inference result from the learning model.

11. An inference method comprising: an acquisition process in which a computer acquires time-series data until predetermined inference conditions are met; an adjustment process in which, if the data length of the acquired time-series data is insufficient to a predetermined data length, the computer adjusts the acquired time-series data by imputing missing values ​​for the insufficient portion of the acquired time-series data so that the data length of the imputed time-series data becomes the predetermined data length; and an inference process in which the imputed time-series data is input into a learning model and an inference result is obtained by the learning model.

12. An inference program that causes a computer to perform the following: an acquisition process to acquire time series data until predetermined inference conditions are met; an adjustment process to adjust the acquired time series data so that the data length after interpolation becomes the predetermined data length, by interpolating missing values ​​for the missing portion of the acquired time series data if the data length of the acquired time series data is less than the predetermined data length; and an inference process to input the interpolated time series data into a learning model and obtain an inference result from the learning model.

Images