Research Plan as Part
of Knowledge Discovery from Huge Data Base;
Discovery Science -- 1998 Grant-in-Aid for Scientific Research on Priority Area
Research Plan as Part
of Knowledge Discovery from Huge Data Base;
1. Organization
Main Investigator
| Name | Hiroshi Motoda |
| Affiliation | Institute of Scientific and Industrial Research Osaka University |
| Title | Professor |
| Professional Area | Artificial Intelligence |
| Last Graduated | University of Tokyo |
| Year Graduated | 1965 |
| Degree | PhD. |
Research Collaborator
| Name | Takashi Washio |
| Affiliation | Institute of Scientific and Industrial Research Osaka University |
| Title | Associate Professor |
| Professional Area | Artificial Intelligence |
| Last Graduated | Tohoku University |
| Year Graduated | 1983 |
| Degree | PhD. |
Research Collaborator
| Name | Tadashi Horiuchi |
| Affiliation | Institute of Scientific and Industrial Research Osaka University |
| Title | Research Associate |
| Professional Area | Artificial Intelligence |
| Last Graduated | Kyoto University |
| Year Graduated | 1992 |
| Degree | MsC. |
2. Project Name
Development of Law Discovery System
3. Objective
Finding an underlying relationship that characterizes the system behavior from experimental data of an unknown system is called scientific law discovery. This is a field of long history where physicists have challenged to seek for the truth for many years. Many of the attempts that try to automate law discovery by machine have no sound ground to justify that the derived relationship is indeed the first principle, rather only derive candidates of experimental fitting formulae, and are far from the level of practical use for complicated phenomena. In this project we explore sound criteria by which to derive an algorithm that can deduce a true relationship of high generality by searching for various relations from experimental data, and develop a practical law discovery system. We further attempt to carry out experiments to derive laws from noisy data and evaluate its performance.
4. Research Plan
The first year (1998):
We investigate the major processing components, develop an algorithm and build a proto-type system. We pursue how far we can go without use of domain specific knowledge. The key idea is to use mathematical constraints that the scale types of the measured variables impose.
The second year (1999):
We apply the proto-type system to different problems including those used in the past research, evaluate its performance, compare with the past results, and show that the system can indeed discover laws for a large problem for which no past attempts have succeeded. We further extend the system so that it can handle a system that can only be described by multiple equations.
The last year (2000):
We summarize the results obtained so far and compile them into a practical law discovery system. In particular we focus on testing its performance over various problems of large scale, get feedbacks from these experiments and improve the algorithm if necessary. The end product is a law discovery system that can be used in practice for problems that are encountered in many domains such as physics, economics, cognitive psychology and social studies.