Geometrical Information Processing by Hypercomplex-valued Neural Networks

2014 Fiscal Year Final Research Report

• Project/Area Number: 24700227
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Sensitivity informatics/Soft computing
• Research Institution: University of Hyogo
• Principal Investigator: ISOKAWA Teijiro 兵庫県立大学, 工学(系)研究科(研究院), 准教授 (70336832)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 複素ニューラルネットワーク / 超複素数 / 四元数 / 連想記憶 / ホップフィールドネットワーク

Outline of Final Research Achievements

This research project intended to investigate neural network models based on complex and hypercomplex number systems, from the viewpoints of theoretical analysis and applications to practical/engineering problems. For dealing with high-dimensional data, such as color information and body coordinate systems, neural networks with hypercomplex number systems are expected to work more efficiently than conventional (real-valued) neural networks. In the term of this research project, several researches had been conducted, i.e., the analysis of fundamental properties for associative memories based on quaternionic (four dimensional hypercomplex number) Hopfield networks and the application of counting pedestrians from the sequences of scenery images.

Free Research Field

知能情報学