| |
|
| |
f=0127
IEEE CIS > Technical Activities
Adaptive Dynamic Programming and Reinforcement Learning TC
(1) Officers
 | Marco Wiering, Chair (2010) Department of Artificial Intelligence University of Groningen Padualaan 14, De Uithof Groningen, Nijenborgh 9 9700AK, Netherlands phone: +31(0)50-3636956 fax: +31(0)50-3636687 email: m.a.wiering .a_t. rug.nl www: www.ai.rug.nl/~mwiering |
(2) Members
 | Marcus Hutter Research School of Information Sciences and Engineering Australian National University Corner of North and Daley Road Canberra ACT 0200, Australia phone: +61(0)2 612 51605 fax: +61(0)2 612 58651 email: marcus.hutter .a_t. anu.edu.au www: www.hutter1.net |
 | Jagannathan Sarangapani Electrical and Computer Engineering Missouri university of Science and Technology 1870 Miner Circle Rolla, MO 65409, USA phone: (573)341-6775 fax: (573)341-4532 email: sarangap .a_t. mst.edu www: web.mst.edu/~sarangap/ |
 | Stefan Schaal Computer Science, Neuroscience , and Biomedical Engineering University of Southern California 3710 S. McClintock Ave Los Angeles, California 90089-2905, USA phone: 310 740 1976 fax: 213 740 1510 email: sschaal .a_t. usc.edu www: www-clmc.usc.edu/~sschaal/ |
 | Athanasios V. Vasilakos Department of Computer and Telecommunications Eng University of Western Macedonia,Greece Krinis 3 N.Erythraia, Greece 14671, Greece phone: +30 6977449705 email: vasilako .a_t. ath.forthnet.gr |
 | Ganesh Kumar Venayagamoorthy Real-Time Power and Intelligent Systems Laboratory Missouri University of Science and Technology Rolla Missouri, USA email: gkumar .a_t. ieee.org |
 | Marco Wiering Department of Artificial Intelligence University of Groningen Padualaan 14, De Uithof Groningen, Nijenborgh 9 9700AK, Netherlands phone: +31(0)50-3636956 fax: +31(0)50-3636687 email: m.a.wiering .a_t. rug.nl www: www.ai.rug.nl/~mwiering |
 | Donald C. Wunsch II Dept. of Electrical and Computer Engineering Missouri University of Science & Technology 301 W. 16th St, 131 EECH Rolla MO 65409, USA phone: 573-341-4521 fax: 573-341-4532 email: wunsch .a_t. ieee.org www: www.linkedin.com/in/wunsch |
(3) Task Forces
3.1 Important applications of ADP and RL
Analysis of power grid cascades and other cascades in man-made systems, analysis of evolution of epidemics, evolution of the stock market, optimal trading of commodities. There are also a large numbers of applications in economics, management science and other areas, and often they are less abstract and theoretical and tries to make actual, measurable, concrete contributions to improve decision making by firms, governments and other organizations.
3.2 Reinforcement Learning and Function Approximation
In ADP and RL, we need a function approximator to represent the learned function, either the value, or the policy, or a model of the dynamics. Tools used for such approximation includes neural networks and many others. There are also issues on how to represent a state in order to achieve the best learning curve. Many issues are intertwined here, ranging from fundamental issues, to algorithmic ones, and practical ones.
3.3 Robot Reinforcement Learning
 | Stefan Schaal, Vice Chair Computer Science, Neuroscience , and Biomedical Engineering University of Southern California 3710 S. McClintock Ave Los Angeles, California 90089-2905, USA phone: 310 740 1976 fax: 213 740 1510 email: sschaal .a_t. usc.edu www: www-clmc.usc.edu/~sschaal/ |
Efficient self-improvement by trial and error is a key ability to allow robots to
adapt to their environment and to learn new tricks. Reinforcement learning
offers some of the most general tools in order to fomulate such robot learning
problems while robotics is in theory a natural application domain for reinforcement
learning. However, most reinforcement learning methods cannot be applied
straightforwardly in robotics as the real-world constraints of the domain
are creating exceedingly complex scenarios. On the other hand, robotics offers
an enormous source of inspirations to reinforcement learning, Hence, it is essential
to bring the insights from robotics into reinforcement learning and to create
domain-appropriate reinforcement learning methods for robotics.
|
|