Title: Intelligent Agents: Technology and Applications Game Theory and Multiagent Systems
1Intelligent Agents Technology and
ApplicationsGame Theory and Multi-agent Systems
- IST 597B
- Spring 2003
- John Yen
2Strict Dominance
Strictly dominates another strategy
if
3Strict Dominance
- Repeat
- Remove all strategies (rows) of player A that are
strictly dominated by another row.
- Remove all strategies of player B (columns) that
are strictly dominated by another column.
- Until no more rows or columns to remove
4Problem 1
- Apply the strict dominance strategy to find the
equilibrium point of the following game.
- Can you always find an equilibrium point for a
zero-sum game using this strategy?
- Can you always find an equilibrium point for a
non-zero game using this strategy?
5Why not?
- The equilibrium point considers each players
choices at a particular strategy pairs (without
considering other choices of the opponent). The
strict dominance considers all strategies of the
opponent. - The equilibrium point is a balance point
reached by the two players not necessarily an
optimal point.
6An N-player Prisoners Dilemma
- Farmer
- Strategy Restricted production
- Strategy Full production
- If all adopts , price is higher,
individually better
- If all adopts , price is lower, individually
worse
- For a particular farmer, full production leads to
higher profit (since price is assumed to be not
affected by one player), regardless of the
strategies of other players. - Full production strictly dominates restricted
production.
7What will the farmers do?
- Greedy initially, predict the strategies of other
farmers, adapt if needed.
- Cant control the weather
- Maximize their productions
- Build houses
- Merge small farms to improve productivity
- Switch planting plan
- Form coalitions to control production and price
8Problem 2 How will farmers behave in the real
world?
- They learn from their experience (i.e., the game
is repeated every year).
Farmers form a coalition.
Government regulates to set constraints on
price/production.
9(Static) Coalition Formation
- Given a set of agents A, a value v for each
subsets of A, Find a coalition structure (i.e.,
partition of A), such that the overall utility is
maximized. - Distribute the pay off to members of the
coalition.
10Pareto-optimal payoff distribution
- No agent is better off in any other valid payoff
distribution for the given coalition structure.
11Stable Coalition Configuration
- No agent has an incentive to leave its coalition
due to its assigned payoff.
- Different characteristics and criterions of
stability define different solution space for
cooperative games.
12Nash Equilibrium
Strategies of player 1
Strategies of player 2
is a Nash Equilibrium if
Neither player has motivation to change
unilaterally from its strategy.
If all players are selfish and non-cooperating
they would eventually adopt strategies that form
a Nash Equilibrium.
13Prisons Dilemma
Prisoner 1
Temporal Repetion of the Prisoners Dilemma
Repeated trials and receive payoffs after each
trial.
14Non-zero Game
Jill
Ballet
Boxing
(-1, -1)
(2, 1)
Boxing
Jack
(1, 2)
(-1, -1)
Ballet
Disclosing strategy is beneficial.
what are the equilibrium point?
Multiple equilibrium points with different total
returns?