Name
Abstract | ||
|---|---|---|
We present several new algorithms for multiagent reinforcementlearning. A commonfeatureof these algorithms is a parameterized, structured represen- tation of a policy or value function. This structure is leveraged in an approach we call coordinated re- inforcement learning, by which agents coordinate both their action selection activities and their pa- rameter updates. Within the limits of our para- metric representations, the agents will determine a jointly optimal action without explicitly consid- ering every possible action in their exponentially large joint action space. Our methods differ from many previous reinforcement learning approaches to multiagent coordination in that structured com- munication and coordination between agents ap- pears at the core of both the learning algorithm and the execution architecture. Our experimental re- sults, comparing our approach to other RL meth- ods, illustrate both the quality of the policies ob- tained and the additional benefits of coordination. |
Year | Venue | Keywords |
|---|---|---|
2002 | ICML | coordinated reinforcement learning,action selection,reinforcement learning,value function |
Field | DocType | ISBN |
Architecture,Parameterized complexity,Structured communication,Computer science,Q-learning,Bellman equation,Parametric statistics,Artificial intelligence,Action selection,Machine learning,Reinforcement learning | Conference | 1-55860-873-7 |
Citations | PageRank | References |
80 | 4.77 | 13 |
Authors | ||
3 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Carlos Guestrin | 1 | 9220 | 488.92 |
| Michail G. Lagoudakis | 2 | 1164 | 79.51 |
| Ronald Parr | 3 | 2428 | 186.85 |