Pattern recognition as combinatorial optimization: basic and applied research

2015 Fiscal Year Final Research Report

• Project/Area Number: 25330200
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Perceptual information processing
• Research Institution: Nagasaki University
• Principal Investigator: SAKAI Tomoya 長崎大学, 工学研究科, 准教授 (30345003)
• Co-Investigator(Renkei-kenkyūsha): KIYASU Senya 長崎大学, 工学研究科, 教授 (20234388)
• Research Collaborator: MIYAHARA Sueharu
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: パターン認識 / 機械学習 / 信号処理 / 圧縮センシング / スパースモデリング

Outline of Final Research Achievements

Present research has established "combinatorial pattern recognition", which formulates the pattern recognition as a problem of finding an optimal combination of basic concepts or classes that concisely explains a given pattern. The combinatorial problem can be translated into convex optimization in which sparsity, low-rankness, co-occurrence of features, etc, are quantified as convex functions. In the applied research, the combinatorial pattern recognition has enabled, e.g., Chinese character detection based on typical combinations of lines, lung-sound recognition via signal separation into combinations of normal and adventitious sound components, and video understanding by combinations of egomotion and object motion.

Free Research Field

計算機科学