ANALYSIS DEVICE, METHOD, AND PROGRAM

To analyze what kind of mapping of input/output is mainly performed, in the whole hidden layer of a neural network.SOLUTION: A relation analysis unit 30 calculates the strength of the relation of each of combinations of the dimension of input data and the unit of a neural network, and calculates the...

Full description

Saved in:
Bibliographic Details
Main Authors WATANABE CHIHIRO, HIRAMATSU KAORU, KASHINO KUNIO
Format Patent
LanguageEnglish
Japanese
Published 07.11.2019
Subjects
Online AccessGet full text

Cover

Loading…
More Information
Summary:To analyze what kind of mapping of input/output is mainly performed, in the whole hidden layer of a neural network.SOLUTION: A relation analysis unit 30 calculates the strength of the relation of each of combinations of the dimension of input data and the unit of a neural network, and calculates the strength of the relation of each of the combinations of the unit and the dimension of output data. A role analysis unit 32 calculates the relation between a prescribed number of types of roles and the unit, and the relation among the prescribed number of types of roles, the dimension of the input data and the dimension of the output data, based on the strength of the relation of each of the combinations of the dimension of the input data and the unit of the neural network, and the strength of the relation of each of the combinations of the unit and the dimension of the output data.SELECTED DRAWING: Figure 1 【課題】ニューラルネットワークの隠れ層全体で、主にどのような入出力のマッピングが行われているかを解析することができる。【解決手段】関係解析部30が、前記入力データの次元と前記ニューラルネットワークのユニットとの組み合わせの各々の関係の強さを算出し、前記ユニットと前記出力データの次元との組み合わせの各々の関係の強さを算出する。役割解析部32が、前記入力データの次元と前記ニューラルネットワークのユニットとの組み合わせの各々の関係の強さ、及び前記ユニットと前記出力データの次元との組み合わせの各々の関係の強さに基づいて、所定の種類数の役割と前記ユニットとの関係性、並びに前記所定の種類数の役割と前記入力データの次元及び前記出力データの次元との関係性を算出する。【選択図】図1
Bibliography:Application Number: JP20180088177