| Co-Investigator(Kenkyū-buntansha) |
FURUZUKI Takayuki Waseda University, Graduate School of Information, Production and Systems, Associate Professor (50294905)
MABU Shingo Waseda University, Graduate School of Information, Production and Systems, Assistant Professor (70434321)
SHIMADA Kaoru Waseda University, Graduate School of Information, Production and Systems, Lecturer (20454100)
江口 徹 早稲田大学, 理工学術院, 助手 (10386724)
|
|---|
| Budget Amount *help |
¥14,820,000 (Direct Cost: ¥13,800,000、Indirect Cost: ¥1,020,000)
Fiscal Year 2007: ¥4,420,000 (Direct Cost: ¥3,400,000、Indirect Cost: ¥1,020,000)
Fiscal Year 2006: ¥5,100,000 (Direct Cost: ¥5,100,000)
Fiscal Year 2005: ¥5,300,000 (Direct Cost: ¥5,300,000)
|
|---|
| Research Abstract |
Genetic Algorithm (GA), a method which imitates nature and living things, was proposed as a model which explains the adaptive process of the system of nature. In addition, Genetic Programming (GP) was proposed to deal with knowledge expression, programs, concept trees and so on. Many algorithms of these kinds of evolutionary computations have been developed and applied to real world problems. However, these methods represent their solutions using strings or tree structures, so the abilities of representing solutions and the evolution are not enough in terms of the system modeling and the optimization. Therefore, Genetic Network Programming was proposed and its effectiveness was confirmed. In this research, the following extensions and the real world applications of GNP are studied.
1. Extensions of GNP
Combination of learning and evolution, function localized GNP, GNP with Macro nodes, GNP with Symbiotic learning and evolution, Variable-size GNP
2. Applications of GNP
Elevator Group Supervisory Control Systems, Data Mining, Stock Trading Model In GNP, judgment nodes and processing nodes are connected with directed links with each other. The graph structure contributes to reusing nodes, good expression ability, simplicity of understanding the algorithms and good performance of evolution. In addition, unlike Finite Automata, GNP uses only the necessary information at the current time to judge the situation, so GNP can be evolved under Partially Observable Markov Decision Process. As a result, the applicable field of GNP can be extended.
|
