Transfer Learning based on the structure of Feature Space

2015 Fiscal Year Final Research Report

• Project/Area Number: 24300049
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: Nara Women's University (2014-2015); Hokkaido University (2012-2013)
• Principal Investigator: YOSHIDA Tetsuya 奈良女子大学, 生活環境科学系, 教授 (80294164)
• Co-Investigator(Kenkyū-buntansha): IMAI Hideyuki 北海道大学, 情報科学研究科, 教授 (10213216)
• Project Period (FY): 2012-04-01 – 2016-03-31
• Keywords: 転移学習 / 特徴空間

Outline of Final Research Achievements

In order to cope with increasing quantity and variety of data, it is important to develop information technology which enables effective reuse of learned knowledge. We have developed a transfer learning method based on the structure of feature space. In the developed method, features are learned from the given data in the source domain, and a graph is constructed based on the learned features. Then, the method tries to preserve the structure of feature space as much as possible between the source domain and the target domain in transfer learning. Under the framework of optimization learning, we have developed a transfer learning algorithm based on multiplicative update rules. The algorithm has been implemented as a prototype system, and experiments over the prototype system were conducted over several benchmark datasets. The results of the experiments indicate the effectiveness of the developed transfer learning method.

Free Research Field

知能情報学