Establishment of Self-information Control Mechanism for Machine Learning

2014 Fiscal Year Final Research Report

• Project/Area Number: 25540094
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Intelligent informatics
• Research Institution: University of Tsukuba
• Principal Investigator: SAKUMA Jun 筑波大学, システム情報系, 准教授 (90376963)
• Co-Investigator(Kenkyū-buntansha): KAMISHIMA Toshihiro 産業技術総合研究所, ヒューマンライフテクノロジー研究部門, 研究員 (50356820)
• Project Period (FY): 2013-04-01 – 2015-03-31
• Keywords: プライバシー / 機械学習 / データマイニング / セキュリティ

Outline of Final Research Achievements

The neutrality of predictions made by machine learning is measured by dependency between values of a specified attribute and predictions. In this research, we developed a framework that controls dependency between prediction and a specified attribute values to achieve fairness, privacy protection, and prevention of discrimination of predictions. One of the significant results of our study is the generalization analysis of neutrality, in which we proved that generalization neutrality can be probabilistically upper-bounded by O(1/√n) for unseen examples.

Free Research Field

プライバシー