Learning-type signal separation method and learning-type signal separation device

1. A learning-type signal separation method performed using a model formulation unit, which performs learning processing based on a training-use signal including a specific component, and a training-use signal not including the specific component, the training-use signals including a common characteristic,

wherein the learning-type signal separation method includes:

classifying and acquiring the training-use signal including the specific component and the training-use signal not including the specific component;

generating learned data by causing the model formulation unit to perform learning processing based on the training-use signals and information indicating whether or not the specific component is included in the training-use signal, to generate a data series signal in which the specific component has been separated and removed from a data series of the training-use signal;

acquiring any signal including the common characteristic; and

generating, based on the acquired any signal and the generated learned data, a data series signal in which the specific component has been separated and removed from a data series of the any signal,

wherein the model formulation unit is a GAN (Generative Adversarial Network) including a classification unit, a generator, and a learning processing unit that causes the classification unit and the generator to perform learning,

the learning-type signal separation method further comprises:

providing, by the learning processing unit, taking the training-use signal as an input signal, the input signal and information indicating whether or not the specific component is included in the input signal to the classification unit and the generator; and

causing the classification unit to perform learning such that the specific component is included in the input signal, while causing the generator to perform learning so as to cause the classification unit to classify that the specific component is not included in the input signal, at the time an output signal of the generator is provided as the input signal of the classification unit;

in the generating, based on signal separation model data and the acquired any signal that is measured at a single point, the data series signal in which the specific component indicating a frequency characteristic has been separated and removed from the data series of the any signal are generated, the signal separation model data being learned data from the generator among the generated learned data,

wherein the learning-type signal separation method further comprises:

while causing the generator to perform learning such that the classification unit outputs a classification result indicating that the the specific component and a component newly added by the generator are not included in the input signal at the time the output signal output by the generator resulting from providing the input signal including the specific component to the generator is provided to the classification unit as the input signal,

causing the classification unit to perform learning such that the classification unit outputs the classification result indicating that the specific component or the component newly added is included in the input signal at the time the output signal output by the generator after the learning resulting from providing the input signal including the specific component to the generator after the learning is provided to the classification unit as the input signal; and

while causing the generator to perform learning such that the same data series signal as the data series of the input signal is generated when the specific component is not included in the input signal provided to the generator and to perform learning such that the classification unit outputs the classification result indicating that the specific component and the component newly added are not included in the input signal at the time the output signal output by the generator resulting from providing the input signal not including the specific component to the generator is provided to the classification unit as the input signal,

causing the classification unit to perform learning such that the classification unit outputs the classification result indicating that the specific component or the component newly added is included in the input signal at the time the output signal output by the generator after the learning resulting from providing the input signal not including the specific component to the generator after the learning is provided to the classification unit as the input signal, and

the component newly added indicates a difference component between the output signal output by the generator resulting from providing the input signal to the generator, and the input signal.

2. The learning-type signal separation method according to claim 1 wherein processes by the learning processing unit further comprises:

causing the classification unit to perform learning such that the classification unit outputs a classification result indicating that the specific component is included when the specific component is included in the input signal, and outputs a classification result indicating that the specific component is not included in the input signal when the specific component is not included in the input signal;

causing the classification unit to perform learning such that the classification unit outputs a classification result indicating that the specific component is included in the input signal at the time an output signal output by the generator resulting from providing the input signal including the specific component to the generator is provided to the classification unit as the input signal; and

causing the generator to perform learning such that the classification unit outputs a classification result indicating that the specific component is not included in the input signal at the time an output signal output by the generator resulting from providing the input signal including the specific component to the generator is provided to the classification unit as the input signal.

3. The learning-type signal separation method according to claim 1 , wherein the training-use signal, the any signal, and the data series signal are multivariate signals constituted by a plurality of additive components, or a univariate signal.

4. The learning-type signal separation method according to claim 1 , wherein the training-use signal, the any signal, and the data series signal are measured biological signals, 5 and the specific component is an unnecessary component added to a biological signal.

5. The learning-type signal separation method according to claim 1 wherein the data series is a waveform or a frequency.

6. A learning-type signal separation device including:

at least one memory configured to store instructions; and

at least one processor configured to execute the instruction to:

acquire and classifies a signal as a training-use signal including a specific component or a training-use signal not including the specific component, the training-use signals including a common characteristic;

perform, based on the training-use signal acquired by a training signal acquisition unit and information indicating whether or not the specific component is included in the training-use signal, learning that generates a data series signal in which the specific component has been separated and removed from a data series of the training-use signal, and generate learned data;

acquire any signal including the common characteristic; and

generate, based on the any signal acquired by a test signal acquisition unit and the generated learned data, a data series signal in which the specific component has been separated and removed from a data series of the any signal,

wherein the at least one processor is configured to execute the instructions to constitute a GAN (Generative Adversarial Network) including a classification unit, a generator, and a learning processing unit that causes the classification unit and the generator to perform learning,

the learning processing unit is configured to:

provide taking the training-use signal as an input signal, the input signal and information indicating whether or not the specific component is included in the input signal to the classification unit and the generator;

cause the classification unit to perform learning such that the specific component is included in the input signal, while causing the generator to perform learning so as to cause the classification unit to classify that the specific component is not included in the input signal, at the time an output signal of the generator is provided as the input signal of the classification unit;

wherein, based on signal separation model data and the acquired any signal that is measured at a single point, the data series signal in which the specific component indicating a frequency characteristic has been separated and removed from the data series of the any signal are generated, the signal separation model data being learned data from the generator among the generated learned data,

wherein the learning processing unit is further configured to:

while cause the generator to perform learning such that the classification unit outputs a classification result indicating that the the specific component and a component newly added by the generator are not included in the input signal at the time the output signal output by the generator resulting from providing the input signal including the specific component to the generator is provided to the classification unit as the input signal,

cause the classification unit to perform learning such that the classification unit outputs the classification result indicating that the specific component or the component newly added is included in the input signal at the time the output signal output by the generator after the learning resulting from providing the input signal including the specific component to the generator after the learning is provided to the classification unit as the input signal; and

while cause the generator to perform learning such that the same data series signal as the data series of the input signal is generated when the specific component is not included in the input signal provided to the generator and to perform learning such that the classification unit outputs the classification result indicating that the specific component and the component newly added are not included in the input signal at the time the output signal output by the generator resulting from providing the input signal not including the specific component to the generator is provided to the classification unit as the input signal,

cause the classification unit to perform learning such that the classification unit outputs the classification result indicating that the specific component or the component newly added is included in the input signal at the time the output signal output by the generator after the learning resulting from providing the input signal not including the specific component to the generator after the learning is provided to the classification unit as the input signal, and

the component newly added indicates a difference component between the output signal output by the generator resulting from providing the input signal to the generator, and the input signal.