ICML
2003
IEEE
14 years 5 months ago
2003
IEEE
The paper explores a very simple agent design method called Q-decomposition, wherein a complex agent is built from simpler subagents. Each subagent has its own reward function and...