Machine learning device, machine learning method, and machine learning program

Abstract

To provide a machine learning technique capable of improving classification performance in continuous small-shot learning.SOLUTION: In a machine learning device 100 for continuously learning data of a new class whose number is smaller than that of data of a basic class, a feature extraction part 10 is preliminarily learned by using the data of the basic class, and inputs the data of the new class, and outputs the feature vector of the data of the new class. The weight calculating unit 20 receives the feature vector of the data of the new class and calculates the classification weight of the new class. The graph model 30 outputs a reconstructed classification weight by adapting and reconstructing the calculated classification weight of the new class and the classification weights of all the previously learned classes. When the graph model 30 performs meta learning of a dependency relationship between a basic class and a new class, the graph model 30 sets the degree of dependency of the new class to be lower than that of the basic class.SELECTED DRAWING: Figure 11

Images