Teaching data extending device, teaching data extending method, and program
A teaching data extending device includes: a relationship acquiring unit that obtains a relationship between a plurality of features included in each of a plurality of teaching data; a feature selecting unit that selects any one or more of the plurality of features based on the relationship; and a teaching data extending unit that generates, for one or more teaching data, new teaching data in which a value of the feature selected by the feature selecting unit is replaced with a value of the feature in another teaching data classified in a same class.
1. A teaching data extending device comprising:
at least one memory storing instructions; and
at least one processor configured to execute the instructions to:
obtain a relationship between a plurality of features included in each of a plurality of teaching data;
select any one or more of the plurality of features based on the relationship; and
generate new teaching data by replacing a value of a selected feature included in first teaching data among the plurality of teaching data with a value of a selected feature included in second teaching data among the plurality of teaching data, a class of the second teaching data being a same as a class of the first teaching data.
2. The teaching data extending device according to claim 1 , wherein the at least one processor is configured to execute the instructions to extract a plurality of features from input data,
the at least one memory comprises a memory that stores label information representing a class in which a classification target represented by the input data is classified,
the at least one processor is configured to execute the instructions to:
generate real teaching data by associating the plurality of extracted features and the stored label information; and
obtain the relationship by obtaining a relationship between a plurality of features included in a plurality of the real teaching data.
3. The teaching data extending device according to claim 2 , wherein the input data is at least either of an image and a plurality of images that are consecutive in time.
4. The teaching data extending device according to claim 2 , wherein extracting the plurality of features comprises extracting the features that have been designed to identify the classification target from the input data.
5. The teaching data extending device according to claim 2 , wherein extracting the plurality of features comprises designing a feature using a neural network and extracting the designed feature.
6. The teaching data extending device according to claim 1 , wherein selecting any one or more of the plurality of features comprises selecting the feature based on at least one of an uncorrelatedness and an independence between the features.
7. The teaching data extending device according to claim 1 , further comprising:
a display that displays the generated teaching data; and
an interface that accepts a user operation that instructs a feature to be replaced by another feature,
wherein selecting any one or more of the plurality of features comprises selecting any one or more of the plurality of features according to the user operation.
8. The teaching data extending device according to claim 1 ,
wherein the at least one processor is configured to perform machine learning of a classifier by using at least the generated teaching data.
9. A teaching data extending method performed by a computer and comprising:
obtaining a relationship between a plurality of features included in each of a plurality of teaching data;
selecting any one or more of the plurality of features based on the relationship; and
generating new teaching data by replacing a value of a selected feature included in first teaching data among the plurality of teaching data with a value of a selected feature included in second teaching data among the plurality of teaching data, a class of the second teaching data being a same as a class of the first teaching data.
10. A non-transitory computer readable recording medium storing a program for causing a computer to execute:
obtaining a relationship between a plurality of features included in each of a plurality of teaching data;
selecting any one or more of the plurality of features based on the relationship; and
generating new teaching data by replacing a value of a selected feature included in first teaching data among the plurality of teaching data with a value of a selected feature included in second teaching data among the plurality of teaching data, a class of the second teaching data being a same as a class of the first teaching data.