Acceleration of the alternating least squares algorithm for multivariate analysis of non-metric data

2015 Fiscal Year Final Research Report

• Project/Area Number: 24500353
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Statistical science
• Research Institution: Okayama University of Science
• Principal Investigator: Kuroda Masahiro 岡山理科大学, 総合情報学部, 教授 (90279042)
• Co-Investigator(Kenkyū-buntansha): MORI YUICHI 岡山理科大学, 総合情報学部, 教授 (80230085) ; SAKAKIHARA MICHIO 岡山理科大学, 総合情報学部, 教授 (70215614) ; NAKAGAWA SHIGEKAZU 倉敷芸術科学大学, 産業科学技術学部, 教授 (90248203) ; ADACHI KOHEI 大阪大学, 人間科学研究科, 教授 (60299055) ; IIZUKA MASAYA 岡山大学, アドミッションセンター, 教授 (60322236)
• Project Period (FY): 2012-04-01 – 2016-03-31
• Keywords: 交互最小二乗法 / 収束の加速 / ベクターイプシロン法 / 非計量主成分分析 / 非計量因子分析

Outline of Final Research Achievements

We developed acceleration methods of the alternating least squares (ALS) algorithms for nonlinear principal component analysis (NL-PCA) and factor analyses (NL-FA) of non-metric data. The vector epsilon accelerator is utilized for speeding up the convergence of the ALS algorithm. Numerical experiments demonstrated that the accelerated ALS algorithm provides less expensive computational cost than ordinary ALS algorithm in NL-PCA and NL-FA and greatly reduces the number of iterations and computational time of the ALS algorithm. These speeds of the acceleration algorithm are about 3 to 4 times faster than those of ordinary ALS algorithm.

Free Research Field

計算機統計学