Formation of information representation in self-organizing neural networks

2017 Fiscal Year Final Research Report

• Project/Area Number: 26330280
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Soft computing
• Research Institution: University of Miyazaki
• Principal Investigator: Date Akira 宮崎大学, 工学部, 准教授 (60322707)
• Co-Investigator(Renkei-kenkyūsha): KURATA Koji 琉球大学, 工学部, 教授 (40170071)
• Project Period (FY): 2014-04-01 – 2018-03-31
• Keywords: 神経回路モデル / 数理モデル / 自己組織化 / 数理脳科学 / ニューラルネットワーク / 学習 / 情報表現

Outline of Final Research Achievements

Interest in the study of deep learning (i.e., feedforward neural networks trained by error backpropagation) has grown remarkably in the last several years. Many researchers have emphasized the black box nature of neural networks. However, the fundamental issues of why and how they work well have not been clearly understood. Although it is not easy to understand the effect of pretraining the network by unsupervised learning before supervised learning, the unsupervised learning must play an important role. The model of unsupervised learning or self-organizing model usually have two dimensional structure, and it can have high dimensional structure from the view point of discovering the salient features. Here we have examined the various unsupervised algorithms appropriate for learning using both computer simulation and mathematical analysis.

Free Research Field

知能情報工学