| Project/Area Number |
10680372
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | Tokyo Institute of Technology |
|---|
Principal Investigator |
YAMAMURA Masayuki Tokyo Institute of Technology, Interdisciplinary Graduate School of Science and Engineering, Associate Professor, 大学院・総合理工学研究科, 助教授 (00220442)
|
|---|
| Project Period (FY) |
1998 – 1999
|
| Project Status |
Completed (Fiscal Year 1999)
|
| Budget Amount *help |
¥3,300,000 (Direct Cost: ¥3,300,000)
Fiscal Year 1999: ¥1,200,000 (Direct Cost: ¥1,200,000)
Fiscal Year 1998: ¥2,100,000 (Direct Cost: ¥2,100,000)
|
|---|
| Keywords | reinforcement learning / Bayesian network / stochastic gradient method / Kepera robot simularot / lifelong learning / bidirectional AntNet / multiagent reinforcement learning / traffic signal control / Life long Learning / タスク連結 |
|---|
| Research Abstract |
Results of this project consists of following three groups;
(1) Reinforcement learning on Bayesian network
We derived a set of propagation rules for the stochastic gradient method, which is a kind of reinforcement learning methods, from belief propagation rules of Bayesian network. We also applied it for robot navigation tasks on Kepera robot simulator to incorporate a priori knowledge such as a map.
(2) Lifelong reinforcement learning
We extended the framework of lifelong learning into reinforcement learning. Since a lifelong agent faces multiple tasks which share some invariant properties, previous experiences would help performing future tasks. We confirmed its effects on robot navigation tasks. We also derived some mathematical theorem in continuous world.
(3) Multiagent reinforcement learning in open world
We tried to explore new frontier of reinforcement learning into multiagent systems. We analized dynamical behavior of distributed adaptive controler for traffic signal systems. We also proposed bidirectional AntNet for adaptive routing for computer networks, and realized better performance than existing works.
|
