Learning data generation method, inference model generation method, arithmetic unit, and computer program

Abstract

To prevent a reduction in the accuracy of inference of an inference model that infers biological information, even in an environment in which a sufficient number of time-series data are difficult to be secured for observation target parameters.SOLUTION: A learning data TR is generated, which is used for machine learning of an inference model 23 that infers biological information. An original data set OR is received, which includes a plurality of observation values acquired from a living body at a plurality of time points for observation target parameters for obtaining the biological information. A first interpolation data set IT1 is generated by interpolating at least one value of the observation target parameters with a first method in a period of the plurality of time points. A second interpolation data set IT2 is generated by interpolating the at least one value of the observation target parameters in the period of the plurality of time points with a second method different from the first method. The learning data TR is generated to include the first interpolation data set IT1 and the second interpolation data set IT2.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a method for generating training data by a computing device, the training data being used in machine learning of an inference model for inferring biological information. The present disclosure also relates to the inference model generating method, the computing device, and a computer program executable by a processor of the computing device. [Background technology]

[0002] Inference models generated using machine learning technology to replace or support clinical judgments of medical professionals in clinical settings are known. For example, Non-Patent Document 1 discloses a model that infers specific biological information based on time series data of observed parameters acquired from a living body. Examples of specific biological information include future values ​​of the observed parameters (used as a time series prediction model), in-hospital mortality probability (used as a classification model), and remaining stay time in an intensive care unit (ICU) (used as a regression model). Non-Patent Document 1 reports that the inference model shows excellent performance for unobserved parameters that are frequently acquired, such as heart rate, while the inference model shows poor performance for unobserved parameters that are intermittently acquired (for which a sufficient number of time series data cannot be secured), such as white blood cell count. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] G. Harerimana, et al. "A Multi-Headed Transformer Approach forPredicting the Patient's Clinical Time-Series Variables From Charted VitalSigns", IEEE Access (Volume 10), 105993-106004, October 3, 2022 Summary of the Invention [Problem to be solved by the invention]

[0004] It is necessary to suppress the deterioration of the inference accuracy of an inference model that infers bioinformation even in an environment where it is difficult to secure a sufficient number of time-series data for observed parameters. [Means for solving the problem]

[0005] One embodiment of the present disclosure provides a method for generating learning data to be used in machine learning of an inference model for inferring biometric information by a computing device, the method comprising: receiving an original data set including a first observation value acquired from a living body at a first time point and a second observation value acquired from the living body at a second time point different from the first time point for an observable parameter for obtaining the biological information; generating a first interpolated data set by interpolating values ​​of at least one of the observed parameters in a time period between the first time point and the second time point using a first method; generating a second interpolated data set by interpolating values ​​of at least one of the observed parameters during a time period between the first time point and the second time point using a second method different from the first method; The training data is generated so as to include the first interpolated data set and the second interpolated data set.

[0006] One embodiment of the present disclosure provides a computing device that generates learning data to be used in machine learning of an inference model that infers biometric information, the computing device comprising: an interface that receives an original data set including a first observation value obtained from a living body at a first time point and a second observation value obtained from the living body at a second time point different from the first time point for a measurable parameter for obtaining the biological information; a processor for generating the training data to include a first interpolated data set and a second interpolated data set; It is equipped with the first interpolated data set is generated by interpolating values ​​of at least one of the observed parameters in a time period between the first time point and the second time point using a first method; The second interpolated data set is generated by interpolating values ​​of at least one of the observed parameters during a time period between the first time point and the second time point in a second manner different from the first manner.

[0007] One embodiment of the present disclosure provides a computer program executable by a processor of a computing device that generates learning data to be used in machine learning of an inference model that infers biological information, the computer program comprising: When executed, the computing device: receiving an original data set including a first observation value acquired from a living body at a first time point and a second observation value acquired from the living body at a second time point different from the first time point for an observable parameter for obtaining the biological information; generating a first interpolated data set by interpolating values ​​of at least one of the observed parameters in a time period between the first time point and the second time point using a first method; generating a second interpolated data set by interpolating values ​​of at least one of the observed parameters during a time period between the first time point and the second time point using a second method different from the first method; The training data is generated so as to include the first interpolated data set and the second interpolated data set.

[0008] In general, the more the amount of data (the number of observed values) included in the learning data, the more the deterioration of the inference accuracy of the generated inference model can be suppressed. On the other hand, for example, in clinical medical practice, armpit temperature is generally measured once every few hours, and it may be difficult to secure time series data including a sufficient number of observed values ​​for the observed parameters. However, according to the configurations according to the above-mentioned embodiments, even if the number of observed values ​​included in the original data set is small, the amount of data included in the learning data can be inflated through the interpolation process. In addition, since the learning data is generated so as to include a plurality of interpolated data sets with different interpolation methods, it is possible to suppress the bias of the inflating tendency that may occur due to relying on a specific interpolation method. Therefore, it is possible to suppress the deterioration of the inference accuracy of the inference model that infers biological information even in an environment where it is difficult to secure a sufficient number of time series data for the observed parameters.

[0009] Therefore, one example embodiment provided by the present disclosure is a method for generating the inference model by a computing device using learning data generated by the above method, comprising: performing supervised learning in which an observed value of the observable parameter acquired after the second time point is used as a correct answer; The apparatus is configured to infer, as the biological information, an unobserved value of the observable parameter in response to an input of a plurality of observed values ​​acquired at different times for the observable parameter.

[0010] One embodiment of the present disclosure provides a method for generating the inference model by a computing device using learning data generated by the above method, comprising the steps of: performing supervised learning in which a correct answer is whether or not an event related to the observable parameter has occurred after the second time point; The apparatus is configured to infer the occurrence probability of the event as the biological information in response to an input of a plurality of observed values ​​of the observable parameter acquired at different times.

[0011] One embodiment of the present disclosure provides a method for generating the inference model by a computing device using learning data generated by the above method, comprising the steps of: performing supervised learning in which an observed value of an observable parameter other than the observable parameter acquired after the second time point is used as a correct answer; The apparatus is configured to infer, as the biological information, an observed value of the other observable parameter when a plurality of observed values ​​of the observable parameter are input and acquired at different times. [Brief description of the drawings]

[0012] [Figure 1] 1 illustrates an example of the functional configuration of an inference system according to one embodiment. [Diagram 2] 2 shows an example of a processing flow performed in the inference system of FIG. 1. [Diagram 3] 3 shows an example of a process of changing the weighting coefficients in FIG. 2. [Figure 4] 3 shows another example of the process of changing the weighting coefficients in FIG. 2. [Diagram 5] 2 shows another example of the processing flow performed in the inference system of FIG. 1. [Figure 6] 2 shows another example of the processing flow performed in the inference system of FIG. 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] The embodiments will be described in detail with reference to the accompanying drawings. In each drawing, the scale is appropriately changed so that each element to be described can be recognized.

[0014] 1 illustrates an example of a functional configuration of an inference system 1 according to an embodiment. The inference system 1 includes an inference device 2.

[0015] The inference device 2 according to this example is a device that predicts the value of armpit temperature after (in the future) a certain point in time based on time series data TS of armpit temperature acquired from a subject S before that point in time. Armpit temperature is an example of an observed parameter. Other examples of observed parameters include respiratory rate obtained by visually observing chest movement, Glasgow Coma Scale (GCS) obtained by visually confirming the state of consciousness, urine volume measured in a urine collection bag, index values ​​obtained through blood tests (red blood cell count, white blood cell count, blood lactate level, arterial blood oxygen partial pressure, etc.), and blood pressure obtained by non-invasive measurement. The future value of the observed parameter is an example of biological information.

[0016] As used herein, the term "time series of an observable parameter" refers to the change over time in the values ​​of an observable parameter obtained at multiple time points, which may or may not be spaced apart by a regular interval.

[0017] The inference device 2 includes an input interface 21. The input interface 21 is configured as a hardware interface capable of receiving the above-mentioned time-series data TS. The time-series data TS may be input from an appropriate sensor or measuring device attached to the subject S, or may be input via an appropriate input device and user operation. The time-series data TS may be in the form of analog data or digital data, depending on the specifications of the input source. When the time-series data TS is in the form of analog data, the input interface 21 includes an appropriate conversion circuit including an A / D converter.

[0018] The inference device 2 includes a processor 22 and an inference model 23. The inference model 23 is a computer algorithm that can be executed by the processor 22. The inference model 23 is configured to receive the time series data TS as input and to output a predicted value of the armpit temperature of the subject S in the future as an inference result.

[0019] The inference device 2 includes an output interface 24. The processor 22 is configured to input the time series data TS received by the input interface 21 to the inference model 23, and to output, from the output interface 24, the biometric information data PH corresponding to the inference result output from the inference model 23.

[0020] The inference system 1 includes an output device 3. The output device 3 is configured to make the bioinformation data PH recognizable by a user. Examples of the output device 3 include a display device, a printing device, an audio output device, a vibration generating device, and a data transmitting device. Therefore, the user can recognize the predicted value of the armpit temperature of the subject S in the future through the output device 3.

[0021] The output interface 24 of the inference device 2 is configured as a hardware interface capable of outputting the biometric information data PH. The biometric information data PH may be in the form of analog data or digital data, depending on the specifications of the output device 3. When the biometric information data PH is in the form of analog data, the output interface 24 is provided with an appropriate conversion circuit including a D / A converter.

[0022] The inference system 1 includes an inference model generating device 4 and a learning data generating device 5. An inference model 23 of the inference device 2 is generated through machine learning executed by the inference model generating device 4. The learning data generating device 5 generates learning data TR used for the machine learning of the inference model 23. Details of the machine learning will be described later. Each of the inference model generating device 4 and the learning data generating device 5 is an example of a computing device.

[0023] The learning data generating device 5 includes an input interface 51. The input interface 51 is configured as a hardware interface capable of receiving the original data set OR. The original data set OR according to this example includes data corresponding to a time series of armpit temperatures acquired from a living body. In other words, the original data set OR includes data corresponding to a plurality of observed values ​​of armpit temperatures acquired from a living body at a plurality of time points.

[0024] The original data set OR may be input from a suitable sensor or measuring device attached to the living body, or may be input through a suitable input device and a user's operation. The original data set OR may be in the form of analog data or digital data, depending on the specifications of the input source. When the original data set OR is in the form of analog data, the input interface 51 includes a suitable conversion circuit including an A / D converter.

[0025] The training data generating device 5 includes a processor 52 and an output interface 53. The processor 52 is configured to generate training data TR based on the original data set OR and output the data from the output interface 53.

[0026] A method for generating the learning data TR executed by the processor 52 will be described with reference to FIG.

[0027] In the illustrated example, the original data set OR includes an observation v1 of armpit temperature acquired at time t1, an observation v2 of armpit temperature acquired at time t2, and an observation v3 of armpit temperature acquired at time t3. The number of observations included in the original data set OR can be determined arbitrarily.

[0028] The processor 52 is configured to generate a number of interpolated data sets in different ways based on the original data set OR. In this example, a first interpolated data set IT1, a second interpolated data set IT2 and a third interpolated data set IT3 are generated.

[0029] The first interpolated data set IT1 includes three interpolated armpit temperature values ​​for each of the time periods between time t1 and time t2 and between time t2 and time t3. The interpolation method is based on the Last Observation Carried Forward (LOCF) method, in which the interpolation is performed such that an observation value at a certain time is maintained until the next observation value is obtained.

[0030] The second interpolated data set IT2 includes three interpolated armpit temperature values ​​for each of the time periods between t1 and t2 and between t2 and t3. The interpolation method is based on linear interpolation, where a value corresponding to an observation at any time point lying on a straight line connecting two observations is interpolated.

[0031] The third interpolated data set IT3 is obtained by interpolating three values ​​of the armpit temperature in each of the time periods between t1 and t2 and between t2 and t3. The interpolation method is based on spline interpolation, in which a value corresponding to an observation at any time point is interpolated, which is located on a polynomial curve connecting two observations.

[0032] The processor 52 generates the training data TR so as to include the first interpolation data set IT1, the second interpolation data set IT2, and the third interpolation data set IT3, and outputs the training data TR from the output interface 53.

[0033] If the interpolation method is different, the number of the plurality of interpolated data sets generated by the processor 52 may be determined arbitrarily. Any of the interpolated data sets selected from the plurality of generated interpolated data sets may be an example of a first interpolated data set. Similarly, any of the other interpolated data sets selected from the plurality of interpolated data sets may be an example of a second interpolated data set.

[0034] Note that the expression "different interpolation methods" used in this specification is not intended to refer only to cases where the types of interpolation methods are different. As an example, an interpolation data set generated by spline interpolation using a cubic polynomial and an interpolation data set generated by spline interpolation using a quintic polynomial are interpreted as "different interpolation methods" even though they are the same type of "spline interpolation". As another example, multiple interpolation data sets that use the same type of interpolation method but have different numbers of values ​​interpolated in the same period are also interpreted as "different interpolation methods".

[0035] In general, the more the amount of data included in the learning data TR (the number of observed values ​​in this example) is, the more the degradation of the inference accuracy of the generated inference model 23 can be suppressed. On the other hand, there may be cases where it is difficult to secure time series data including a sufficient number of observed values ​​for the observed parameters, such as the armpit temperature in this example. However, according to the configuration of this embodiment, even if the number of observed values ​​included in the original data set OR is small, the amount of data included in the learning data TR can be inflated through the interpolation process. In addition, since the learning data TR is generated so as to include multiple interpolated data sets with different interpolation methods, it is possible to suppress the bias of the inflating tendency that may occur due to relying on a specific interpolation method. Therefore, even in an environment where it is difficult to secure a sufficient number of time series data for the observed parameters, it is possible to suppress the degradation of the inference accuracy of the inference model that infers biological information.

[0036] In particular, in this embodiment, a plurality of interpolated data sets are generated using different types of interpolation methods, which can further reduce the tendency toward padding.

[0037] The processor 52 may be configured to assign a weighting factor to each of the generated interpolated data sets. The weighting factor corresponds to the magnitude of the influence on the machine learning. In other words, when the amount of data included in two interpolated data sets is the same, the interpolated data set to which a larger weighting factor is assigned has a larger influence on the machine learning.

[0038] In this embodiment, a first weighting coefficient w1 is assigned to the first interpolated data set IT1. Similarly, a second weighting coefficient w2 is assigned to the second interpolated data set IT2, and a third weighting coefficient w3 is assigned to the third interpolated data set IT3. In this case, the processor 52 is configured to assign at least two types of weighting coefficients. The first weighting coefficient w1 may be different from each of the second weighting coefficient w2 and the third weighting coefficient w3, or may be the same as either the second weighting coefficient w2 or the third weighting coefficient w3. In other words, the multiple interpolated data sets include at least two interpolated data sets to which different weighting coefficients are assigned.

[0039] The weighting coefficients may be adjusted by adjusting the loss value in a loss function used during machine learning, or by adjusting the amount of data finally included in the training data TR. For example, if it is known in advance that spline interpolation is more accurate (has a smaller error from the actual measured value) than LOCF interpolation, adjustments may be made so that the amount of data included in the training data TR from the third interpolation data set IT3 is greater than the amount of data included in the training data TR from the first interpolation data set IT1.

[0040] This configuration makes it possible to realize machine learning that takes into account the different accuracy levels between the interpolation methods, thereby reducing the impact of an interpolation data set generated using an interpolation method with a relatively low accuracy on the inference accuracy of the generated inference model 23.

[0041] 3 shows another example of how weighting coefficients are assigned by the processor 52. In this example, the weighting coefficients are changed according to the number of observations per unit time contained in the original data set OR.

[0042] In Fig. 3, a first interpolated data set IT1 is generated using LOCF interpolation based on the original data set OR. The white circles shown in the first interpolated data set IT1 represent the true values ​​of the observed values. In other words, the length of the dashed lines extending in the vertical direction corresponds to the magnitude of the error of the interpolated value from the true value.

[0043] On the other hand, a first interpolated data set IT1' is generated using LOCF interpolation based on the original data set OR'. The number of observed values ​​per unit time included in the original data set OR' is greater than that of the original data set OR. The magnitude of error of the interpolated values ​​from the true values ​​in the first interpolated data set IT1' is smaller than that of the first interpolated data set IT1.

[0044] The processor 52 may be configured to modify the weighting coefficients assigned to the interpolated data set generated using LOCF interpolation such that the smaller the number of observations per unit time contained in the original data set OR, the smaller the weighting coefficients assigned to the interpolated data set generated using linear or spline interpolation, which are relatively more accurate.

[0045] For example, if the time interval between the most frequent observations that can be assumed is ΔT and the time interval between the actual observations is Δt, the processor 52 calculates a weighting factor w L and a weighting factor w given to the interpolated data set generated using spline interpolation. S may be determined based on the following formula: w L =ΔT / Δt w S =1-(ΔT / Δt)

[0046] With this configuration, when the observations contained in the original data set OR are sparse, it is possible to suppress the impact that an interpolated data set generated using a method with relatively low interpolation accuracy may have on machine learning, thereby reducing the inference accuracy of the inference model 23.

[0047] 4 shows another example of how weighting coefficients are assigned by the processor 52. In this example, the weighting coefficients are changed according to the variation range of the observed values ​​included in the original data set OR.

[0048] In Fig. 4, a first interpolated data set IT1 is also generated using LOCF interpolation based on the original data set OR. The white circles shown in the first interpolated data set IT1 represent the true values ​​of the observed values. In other words, the length of the dashed lines extending in the vertical direction corresponds to the magnitude of the error of the interpolated value from the true value.

[0049] On the other hand, a first interpolated data set IT1" is generated using LOCF interpolation based on the original data set OR". The original data set OR" contains a wider variation range of observed values ​​than the original data set OR. The magnitude of error of the interpolated values ​​from the true values ​​in the first interpolated data set IT1" is smaller than that of the first interpolated data set IT1.

[0050] The processor 52 may be configured to modify the weighting coefficients applied to the interpolated data set using LOCF interpolation such that the greater the variation in the observations contained in the original data set OR, the smaller the weighting coefficients applied to the interpolated data set generated using linear or spline interpolation, which are relatively accurate.

[0051] With this configuration, when the range of variation of the observation values ​​contained in the original data set OR is large, it is possible to suppress the impact that an interpolated data set generated using a method with relatively low interpolation accuracy may have on machine learning, thereby reducing the inference accuracy of the inference model 23.

[0052] The weighting coefficients may be changed in consideration of the variation range of the observed value for the same observed parameter, or may be changed between different types of observed parameters if the inference model 23 has multiple observed parameters as input. For example, blood pressure, respiratory rate, heart rate, etc. have a relatively large variation range, whereas body temperature, index values ​​obtained through blood tests, etc. generally have a relatively small variation range.

[0053] As another example, the performance of the inference model 23 generated by training the training data TR generated using a specific first weighting coefficient w1, a second weighting coefficient w2, and a third weighting coefficient w3 is p(w1, w2, w3), where w1+w2+w3=1, and the initial values ​​of the weighting coefficients are w1=w2=w3=1 / 3.

[0054] The processor 42 calculates the following equation: (p i The new weighting coefficient distribution is adopted as the one that maximizes the value of (p i Repeat until the value of (p'-p) falls below ε, where ε is a sufficiently small value. Δ is an arbitrary value that represents the amount of change between the new weighting coefficients and the original weighting coefficients. p 1 '(w1+Δ,w2-Δ / 2,w3-Δ / 2) p 2 '(w1-Δ / 2,w2+Δ,w3-Δ / 2) p 3 '(w1-Δ / 2,w2-Δ / 2,w3+Δ)

[0055] 1, the inference model generating device 4 includes an input interface 41. The input interface 41 is configured to receive the training data TR generated as described above.

[0056] The inference model generating device 4 includes a processor 42 and an output interface 43. The processor 42 is configured to generate an inference model 23 by training a neural network using the training data TR, and output the inference model 23 from the output interface 43. The inference model 23 output from the output interface 43 is implemented in the inference device 2.

[0057] In the example shown in Fig. 2, the processor 42 uses the training data TR to perform supervised learning in which the observed value of the armpit temperature acquired after time t3 in the original data set OR is taken as the correct answer. The inference model 23 thus generated is configured as a time series prediction model that infers an unobserved value v0 of the armpit temperature in response to input of time series data TS on the armpit temperature (multiple observed values ​​acquired from the subject S at different times). The unobserved value v0 of the armpit temperature is an example of biological information.

[0058] FIG. 5 shows another example of machine learning executed by the processor 42. The processor 42 executes supervised learning using the learning data TR, in which the correct answer is whether or not an event requiring an antipyretic analgesic has occurred after time t3 in the original data set OR. An event requiring an antipyretic analgesic is an example of an event related to the observed parameter. The inference model 23 generated by this is configured as a classification model that infers the occurrence probability of an event requiring an antipyretic analgesic for input of time series data TS (a plurality of observed values ​​obtained from the subject S at different times) regarding armpit temperature. The occurrence probability of an event requiring an antipyretic analgesic is an example of biological information.

[0059] The occurrence probability of an event related to an observable parameter may be inferred only at a specific time point, or may form time-series data including a plurality of inferred values ​​as illustrated in Fig. 5. The inferred occurrence probability value may or may not be used for visualization processing.

[0060] Other examples of biological information whose occurrence probability is inferred by a classification model include the probability of an event requiring respiratory management using a ventilator for an input of time series data including intermittently observed respiratory rate, heart rate, arterial blood oxygen saturation, etc., the probability of developing acute kidney injury for an input of time series data including regularly measured urine volume, and the probability of in-hospital mortality for an input of time series data including intermittently observed respiratory rate, heart rate, blood pressure, etc.

[0061] FIG. 6 shows another example of machine learning executed by the processor 42. The processor 42 executes supervised learning using the training data TR, in which the observed value of deep body temperature acquired after time t3 in the original data set OR is the correct answer. Deep body temperature is another example of an observed parameter. The inference model 23 generated by this is configured as a regression model that infers the observed value of deep body temperature for input of time series data TS about armpit temperature (multiple observed values ​​acquired from the subject S at different times). The observed value of deep body temperature is an example of biological information.

[0062] The observed values ​​of the other observable parameters may be inferred only at a particular time point, or may form a time series including multiple inferred values ​​as illustrated in Figure 6. The inferred observed values ​​may or may not be subjected to visualization processing.

[0063] Other examples of observed parameters whose observed values ​​are inferred by regression models include cardiac index for input of time series data including intermittently observed heart rate, non-invasive blood pressure, and other observed values; blood creatinine for input of time series data including periodically measured urine volume; and remaining stay in an intensive care unit (ICU) for input of time series data including periodically measured blood test results and intermittently observed body temperature, respiratory rate, heart rate, and blood pressure.

[0064] 1, the inference system 1 may include an evaluation system 6. The evaluation system 6 is configured to evaluate the performance of the inference model 23 implemented in the inference device 2. As an example, the evaluation system 6 may have an appropriate configuration for receiving user feedback on the inference result output by the inference model 23. The feedback may take the form of a score or a comment.

[0065] The input interface 51 of the training data generating device 5 can be configured to be able to accept feedback data FB. The feedback data FB includes, as information, the breakdown of weighting coefficients employed in the inference model 23 that has been highly evaluated through the evaluation system 6.

[0066] With this configuration, an inference model 23 that exhibits higher suitability to the user's usage environment can be generated through re-learning based on an evaluation of an inference model 23 that has been implemented in an inference device 2 and put into operation.

[0067] The processor 22 of the inference device 2, the processor 42 of the inference model generating device 4, and the processor 52 of the learning data generating device 5, each having the various functions described above, may be realized by a general-purpose microprocessor operating in cooperation with a general-purpose memory. Examples of the general-purpose microprocessor include a CPU, an MPU, and a GPU. Examples of the general-purpose memory include a ROM and a RAM. In this case, the ROM may store a computer program that executes the above-mentioned processing. The ROM is an example of a non-transient computer-readable medium that stores a computer program. The general-purpose microprocessor specifies at least a part of the program stored on the ROM, expands it on the RAM, and executes the above-mentioned processing in cooperation with the RAM. The computer program may be pre-installed in the general-purpose memory, or may be downloaded from an external server via a communication network and then installed in the general-purpose memory. In this case, the external server is an example of a non-transient computer-readable medium that stores a computer program.

[0068] Each of the processors 22, 42, and 52 may be realized by a dedicated integrated circuit capable of executing the above-mentioned computer program, such as a microcontroller, an ASIC, or an FPGA. In this case, the above-mentioned computer program is pre-installed in a storage element included in the dedicated integrated circuit. The storage element is an example of a computer-readable medium storing a computer program. Each of the processors 22, 42, and 52 may also be realized by a combination of a general-purpose microprocessor and a dedicated integrated circuit.

[0069] The configurations referred to above are merely examples for facilitating understanding of the present disclosure. Each configuration example may be appropriately modified or combined with other configuration examples within the scope of the present disclosure.

[0070] In the above embodiment, the inference model generating device 4 and the learning data generating device 5 are described as devices independent of each other. In this case, the input interface 41 and the output interface 53 are configured as hardware interfaces. However, both devices may be realized as different functional modules in a single device. In this case, both interfaces may be realized as software interfaces.

[0071] In the above embodiment, the inference device 2 and the inference model generation device 4 are described as devices independent of each other. In this case, the input interface 21 and the output interface 43 are configured as hardware interfaces. However, both devices may be realized as different functional modules in a single device. In this case, both interfaces may be realized as software interfaces.

[0072] The inference model 23 does not necessarily have to be installed in the inference device 2. Although not shown in the figure, the inference device 2 may be communicatively connected to an external server device via a communication network. In this case, the inference model 23 may be installed in the external server device.

[0073] In the above embodiment, the inference model generating device 4 generates the inference model 23 through machine learning using a neural network. However, the inference model 23 may be generated through other machine learning algorithms. Examples of other machine learning algorithms include linear regression, logistic regression, decision tree, random forest, and support vector machine.

[0074] Each of the configurations listed below also constitutes part of this disclosure. Item 1: A method for generating learning data to be used in machine learning of an inference model for inferring biometric information by a computing device, comprising: receiving an original data set including a first observation value acquired from a living body at a first time point and a second observation value acquired from the living body at a second time point different from the first time point for an observable parameter for obtaining the biological information; generating a first interpolated data set by interpolating values ​​of at least one of the observed parameters in a time period between the first time point and the second time point using a first method; generating a second interpolated data set by interpolating values ​​of at least one of the observed parameters during a time period between the first time point and the second time point using a second method different from the first method; generating the training data to include the first interpolated data set and the second interpolated data set; How to generate training data. Item 2: The first method and the second method have different types of interpolation methods. 2. A method for generating learning data according to item 1. Item 3: assigning a weighting coefficient to the first interpolated data set and the second interpolated data set corresponding to the magnitude of the influence on the machine learning; the weighting factor assigned to the first interpolated data set and the weighting factor assigned to the second interpolated data set are different; 3. A method for generating learning data according to item 1 or 2. Item 4: The weighting coefficients are changed according to the number of observations per unit time included in the original data set. Item 3. A method for generating learning data according to item 3. Item 5: The weighting coefficients are changed according to the variation range of the observed values ​​included in the original data set. Item 3. A method for generating learning data according to item 3. Item 6: The weighting coefficients are changed based on the performance evaluation results of the inference model. 6. A method for generating learning data according to any one of items 3 to 5. Item 7: A method for generating the inference model by a computing device using learning data generated by the method according to any one of items 1 to 6, performing supervised learning in which an observed value of the observable parameter acquired after the second time point is used as a correct answer; and inferring an unobserved value of the observable parameter as the biological information in response to an input of a plurality of observed values ​​of the observable parameter obtained at different times. How to generate an inference model. Item 8: A method for generating the inference model by a computing device using learning data generated by the method according to any one of items 1 to 6, performing supervised learning in which a correct answer is whether or not an event related to the observable parameter has occurred after the second time point; and inferring a probability of occurrence of the event as the biological information for a plurality of input observation values ​​of the observable parameter obtained at different times. How to generate an inference model. Item 9: A method for generating the inference model by a computing device using learning data generated by the method according to any one of items 1 to 6, performing supervised learning in which an observed value of an observable parameter other than the observable parameter acquired after the second time point is used as a correct answer; and inferring, as the biological information, an observed value of the other observable parameter when a plurality of observed values ​​of the observable parameter are inputted at different times. How to generate an inference model. [Explanation of symbols]

[0075] 23: inference model, 4: inference model generating device, 5: learning data generating device, 51: input interface, 52: processor, IT1: first interpolation data set, IT2: second interpolation data set, IT3: third interpolation data set, OR: original data set, TR: learning data, v0: unobserved value, w1: first weighting coefficient, w2: second weighting coefficient, w3: third weighting coefficient

Claims

1. A method for generating learning data to be used in machine learning of an inference model for inferring biometric information by a computing device, comprising: receiving an original data set including a first observation value acquired from a living body at a first time point and a second observation value acquired from the living body at a second time point different from the first time point for an observable parameter for obtaining the biological information; generating a first interpolated data set by interpolating values ​​of at least one of the observed parameters in a time period between the first time point and the second time point using a first method; generating a second interpolated data set by interpolating values ​​of at least one of the observed parameters during a time period between the first time point and the second time point using a second method different from the first method; generating the training data to include the first interpolated data set and the second interpolated data set; How to generate training data.

2. The first method and the second method have different types of interpolation methods. The method for generating learning data according to claim 1 .

3. assigning a weighting coefficient to the first interpolated data set and the second interpolated data set corresponding to the magnitude of the influence on the machine learning; the weighting factor assigned to the first interpolated data set and the weighting factor assigned to the second interpolated data set are different; The method for generating learning data according to claim 1 .

4. The weighting coefficients are changed according to the number of observations per unit time included in the original data set. The method for generating learning data according to claim 3 .

5. The weighting coefficients are changed according to the variation range of the observed values ​​included in the original data set. The method for generating learning data according to claim 3 .

6. The weighting coefficients are changed based on the performance evaluation results of the inference model. The method for generating learning data according to claim 3 .

7. A method for generating the inference model by a computing device using learning data generated by the method according to any one of claims 1 to 6, comprising: performing supervised learning in which an observed value of the observable parameter acquired after the second time point is used as a correct answer; and inferring an unobserved value of the observable parameter as the biological information in response to an input of a plurality of observed values ​​of the observable parameter obtained at different times. How to generate an inference model.

8. A method for generating the inference model by a computing device using learning data generated by the method according to any one of claims 1 to 6, comprising: performing supervised learning in which a correct answer is whether or not an event related to the observable parameter has occurred after the second time point; and inferring a probability of occurrence of the event as the biological information for a plurality of input observation values ​​of the observable parameter obtained at different times. How to generate an inference model.

9. A method for generating the inference model by a computing device using learning data generated by the method according to any one of claims 1 to 6, comprising: performing supervised learning in which an observed value of an observable parameter other than the observable parameter acquired after the second time point is used as a correct answer; and inferring, as the biological information, an observed value of the other observable parameter when a plurality of observed values ​​of the observable parameter are inputted at different times. How to generate an inference model.

10. A computing device that generates learning data to be used in machine learning of an inference model that infers biometric information, an interface that receives an original data set including a first observation value obtained from a living body at a first time point and a second observation value obtained from the living body at a second time point different from the first time point for a measurable parameter for obtaining the biological information; a processor for generating the training data to include a first interpolated data set and a second interpolated data set; Equipped with the first interpolated data set is generated by interpolating values ​​of at least one of the observed parameters in a time period between the first time point and the second time point using a first method; the second interpolated data set is generated by interpolating values ​​of at least one of the observed parameters during a time period between the first time point and the second time point using a second method different from the first method. Computing device.

11. A computer program executable by a processor of a computing device that generates learning data to be used in machine learning of an inference model that infers bioinformation, When executed, the computing device: receiving an original data set including a first observation value acquired from a living body at a first time point and a second observation value acquired from the living body at a second time point different from the first time point for an observable parameter for obtaining the biological information; generating a first interpolated data set by interpolating values ​​of at least one of the observed parameters in a time period between the first time point and the second time point using a first method; generating a second interpolated data set by interpolating values ​​of at least one of the observed parameters during a time period between the first time point and the second time point using a second method different from the first method; generating the training data to include the first interpolated data set and the second interpolated data set; Computer program.

Images