Modular Structure Generation and Its Application to Feature Extraction

R. Kamimura (Japan)


: mutual information maximization, modularinformation, unit information, competition


In this paper, we propose a new method to generate modu lar structures. In the method, the number of elements, that is, the number of competitive units is gradually increased. To control a process of module generation, we introduce two kinds of information, that is, unit and modular informa tion. Unit information represents information content ob tained by individual elements in all modules. On the other hand, modular information is information content obtained by each module. We try to increase both types of informa tion simultaneously. We applied our method to two classi fication problems: random data classification and chemical data classification. In both cases, we observed that modular structures were automatically generated. In addition, in the chemical data application, we could see that interpretable connection weights could be generated.

Important Links:

Go Back