Abstract | ||
---|---|---|
Imitation is actively being studied as an effec- tive means of learning in multi-agent environ- ments. It allows an agent to learn how to act well (perhaps optimally) by passively observ- ing the actions of cooperative teachers or other more experienced agents its environment. We propose a straightforward imitation mechanism called model extraction that can be integrated easily into standard model-based reinforcement learning algorithms. Roughly, by observing a mentor with similar capabilities, an agent can ex- tract information about its own capabilities in un- visited parts of state space. The extracted infor- mation can accelerate learning dramatically. We illustrate the benefits of model extraction by inte- grating it with prioritized sweeping, and demon- strating improved performance and convergence through observation of single and multiple men- tors. Though we make some stringent assump- tions regarding observability, possible interac- tions and common abilities, we briefly comment on extensions of the model that relax these. |
Year | Venue | Keywords |
---|---|---|
1999 | ICML | imitation,multi-agent systems,reinforcement learning,implicit imitation,multiagent reinforcement learning,state space,standard model,multi agent system |
Field | DocType | ISBN |
Convergence (routing),Observability,Computer science,Imitation,Artificial intelligence,Model extraction,Error-driven learning,State space,Machine learning,Reinforcement learning | Conference | 1-55860-612-2 |
Citations | PageRank | References |
24 | 1.67 | 17 |
Authors | ||
2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Bob Price | 1 | 481 | 31.72 |
Craig Boutilier | 2 | 6864 | 621.05 |