El Farol Bar Problem
- Agents try to decide which night of the week they should go
to El Farol Bar.
- Each agent makes an independent decision at the beginning of
the week.
- The agents have the best time if there are c agents in the
bar that night.
- Lets say the agents use reinforcement learning to learn
which night is best. If we, as system designers, want to
maximize the sum of all agent utilities, what reward function
should we use?
- The agent gets a reward based on how many agents where
there on the night it attended?
- On the nights he did not attend?
- Something else?
- COllective INtelligence (COIN) tries to answer this question
for the general case.
José M. Vidal
.
1 of 13