System, method, and program for augmenting training data used for machine learning
View Patent ↗The problem to be solved is to provide a system and the like for augmenting supervisory data while maintaining the relationship among a plurality supervisory data used for machine learning. The present disclosure provides a system for augmenting supervisory data used for machine learning, the system including an obtaining means that obtains a plurality of supervisory data, a first processing means that derives a covariance matrix from the plurality of supervisory data, a second processing means that decomposes the covariance matrix, and a third processing means that applies a random number to the decomposed matrix.
1 . A computer implemented method of generating, by machine learning, a pain classifier for classifying pain that an object being estimated has based on a brainwave of the object being estimated, wherein the pain classifier is generated to distinguish whether the pain requires therapy, and the computer includes a memory, a processor and an interface connected to an electroencephalograph and a stimulation application unit for applying a plurality of levels of stimulation intensities to the object, comprising:
a) the step of creating computerized visual analog scale (COVAS) templates that expresses subjective evaluation of pain by healthy people by averaging COVAS data respectively corresponding to the plurality of levels of stimulation intensities applied to the healthy people by the stimulation application unit, wherein the stimulation intensities vary from first level to second level and include an intensity corresponding to the pain that requires therapy, wherein the healthy people feel no pain by the first level and feel a pain by the second level;
b) the step of stimulating, by the stimulation application unit, the object being estimated with the plurality of levels of stimulation intensities respectively corresponding to the COVAS templates;
c) the step of obtaining, by the electroencephalograph, brainwave data or analysis data thereof of the object being estimated respectively corresponding to the stimulation intensities;
d) the step of augmenting, by the processor, brainwave data or analysis data thereof of the object being estimated with the corresponding COVAS templates, comprising:
i) deriving a covariance matrix from brainwave data or analysis data thereof of the object being estimated;
ii) decomposing the covariance matrix; and
iii) applying a random number to the decomposed matrix;
e) the step of plotting, by the processor, the stimulation intensities or subjective pain sensation levels labeled by the COVAS templates each corresponding to the stimulation intensity and the augmented brainwave data or analysis data thereof to fit to a pain function to obtain a pain function specific to the object being estimated; and
f) the step of, when regression coefficient of the fitting to the specific pain function is equal to or more than what is predetermined, identifying, by the processor, a pain classifier for dividing a pain level into at least two or more sub-levels based on the specific pain function, wherein the sub-levels include a level corresponding to the pain that requires therapy.
2 . An apparatus generating a pain classifier, by machine learning, for classifying pain that an object being estimated has based on a brainwave of the object being estimated wherein the pain classifier is generated to distinguish whether the pain requires therapy, comprising:
a processor;
a stimulation application unit for applying a plurality of levels of stimulation intensities to the object; and
an interface, wherein the processor includes:
a computerized visual analog scale (COVAS) template creating unit for creating COVAS templates that express subjective evaluation of pain by healthy people by averaging COVAS data respectively corresponding to the plurality of levels of stimulation intensities applied to the healthy people by the stimulation application unit, wherein the stimulation intensities vary from first level to second level and include an intensity corresponding to the pain that requires therapy, wherein the healthy people feel no pain by the first level and feel a pain by the second level;
a stimulation unit connected through the interface to the stimulation application unit for stimulating the object being estimated with the plurality of levels of stimulation intensities respectively corresponding to the COVAS templates;
a brainwave data obtaining unit connected through the interface to an electroencephalograph for obtaining brainwave data or analysis data thereof the object being estimated respectively corresponding to the stimulation intensities;
an augmentation unit for augmenting the brainwave data or analysis data thereof of the object being estimated with the corresponding COVAS templates, wherein the augmentation unit is configured to:
i) derive a covariance matrix from brainwave data or analysis data thereof of the object being estimated;
ii) decompose the covariance matrix; and
iii) apply a random number to the decomposed matrix; and
a pain classifier generation unit for plotting the stimulation intensities or subjective pain sensation levels labeled by the COVAS templates each corresponding to the stimulation intensity and the augmented brainwave data or analysis data thereof to fit to a pain function to obtain a pain function specific to the object being estimated and identifying a pain classifier for dividing a pain level into at least two or more sub-levels based on the specific pain function, wherein the sub-levels include a level corresponding to the pain that requires therapy.