Dr Patrick HU received his BE and ME from Xi’an Jiaotong Univ in 1985 and 1988 respectively, and Ph.D from The University of Auckland, New Zealand in 2001. Before that he worked as a lecturer, an electrical engineer, group leader, and director of technical development section in Xi’an, China. He received the University of Auckland Ph.D scholarship, the Asian 2000 Foundation Scholarship for a four-month study visit to the National University of Singapore, and an industrial scholarship from Wampfler AG, Germany. Now he works in the Department of Electrical and Electronic Engineering, the University of Auckland, conducting teaching/research and supervising postgraduates at both ME and Ph.D levels. He is also a senior IEEE member and co-chairman of the New Zealand Joint IEEE Power Engineering and Power Electronics Society. His research interests include wireless/contactless power transfer and application of power electronics in power systems.

 

    Linear Discriminant Analysis (LDA) has been widely researched to implement various computation intelligences, such as pattern recognition, data mining, bioinformatics and robotics. However, learning LDA in real world confronts difficulties in different application scenarios. Starting from classic LDA, this talk introduces a series of recent LDA developments, where an LDA model is enabled to be learned either in one batch session, or incrementally by ILDA through instance-space merging, or through LDA eigenspace merging; In multi-agent background, LDA can be learned cooperatively by a number of agents with knowledge sharing in-between each other; In a special case, a created LDA can be renovated by LDA splitting with a minimum processing on the raw data instance; Even In some physical limited environment such as the remote space, LDA can be actively learned by an independent agent on fewer selected curiosity instances, or by a multiple of agents in a competitive and cooperative learning manner.