A study of image processing methods to improve explainability and redesign through shallow layer learning

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K11865
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: Tottori University
• Principal Investigator: Iwai Yoshio 鳥取大学, 工学研究科, 教授 (70294163)
• Co-Investigator(Kenkyū-buntansha): 西山 正志 鳥取大学, 工学研究科, 教授 (20756449)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 浅層学習 / 説明可能性 / 再設計可能性 / 機械学習

Outline of Final Research Achievements

The purpose of this study is to realize an image recognition method that can be easily redesigned while improving the explainability of the learning results by removing the deep network structure and performing machine learning with a shallow network structure, while maintaining the feature of integrated learning that deep learning possesses. In particular, to improve the explainability and redesignability, it is necessary to replace the feature extraction process with a conventional feature extraction process using a deep network structure, which is realized by deep learning. Therefore, we extended the conventional feature extraction process and constructed a learnable feature extractor by parameterizing it, and built a prototype discriminator with a high learning capacity based on a redesignable shallow network structure and a learnable feature extractor that contributes to recognition performance.

Free Research Field

画像認識

Academic Significance and Societal Importance of the Research Achievements

本研究が実現することにより，計算機が何故そのような判断を下したかの説明を行いやすくなる．また，その説明を受けて，浅いネットワーク構造のため理解がしやすく，再設計を行うことが可能なネットワーク構造を構築することが出来る．