PROBLEM TO BE SOLVED: To efficiently obtain high-quality learning data which greatly improves the accuracy of a model by data expansion. An expansion device provides predetermined data from an external data set based on a metric calculated from the degree of similarity between the target data set and the external data set and the degree of uniformity of the external data set. Select a set. The extension device makes it easy for the generator to be identified as genuine by the classifier when the target class is specified for the generator, and the probability of belonging to multiple external classes by the classifier is non-zero and even. The generative model is trained so as to generate the data calculated in. The extension device specifies the target class in the generator of the trained generative model and generates data. The expansion device extracts data identified as fake by the classifier from the generated data as expansion data. The expansion device assigns a target label to the extracted expansion data. [Selection diagram] Fig. 1
展开▼