Informatics Seminar (Perspective in Informatics 4B) 2010 - 2011
November 5 (Fri), 16:30 - 18:00
- Place:
Lecture Hall 1, Faculty of Engineering Bldg. No.10, Main Campus
- Title:
Kernel design for discrete data structures
- Speaker:
Kilho Shin (Hyogo University)
- Abstract:
Kernel multivariate analysis is widely used as an effective methodology for analyzing and learning from data. On the other hand, many of
practically important data such as proteins, Web contents and so forth are
represented in a form of discrete structures. Thus, researching design
methods of kernels particularly for discrete data structures is indeed
important, we have some interesting results in the literature including
widely known Haussler’s convolution kernel and Binet-Cauchy kernel.
Recently, we have succeeded in generalizing these two kernels to a large
extent. In addition, we have started a theoretical research on computability
of kernels for descrete data structures on the basis of the dynamic
programming methodology. In this lecture, beginning with an elementary
explanation of the kernel method, we
will illustrate some research results in the literature, and then
will report our recent progress.
Kyoto University
> Graduate School of Informatics
> International Courses
> Informatics Seminar