One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh...
Transcript of One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh...
![Page 1: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/1.jpg)
One Equilibrium Is Not Enough: Computing Game-Theoretic
Sol tions to Act StrategicallSolutions to Act Strategically0, 0 -1, 2 0, 0 -1, 1
Vincent Conitzer-1, 1 0, 0
2, 2 -1, 0
1, -1 -5, -5
1, 1 3, 0
Vincent ConitzerDuke University
My wonderful co-authors (alphabetically):
2, 2 1, 0
-7, -8 0, 0
1, 1 3, 0
0, 0 2, 1My wonderful co authors (alphabetically):
Krzysztof Apt, CWI Amsterdam. Sayan Bhattacharya, Duke. Craig Boutilier, U. Toronto. Andrew Davenport, IBM Research. Jonathan Derryberry,
CMU. Bruce Donald, Duke. Joseph Farfel, Duke. Nikesh Garera, Johns Hopkins. Andrew Gilpin, CMU. Mingyu Guo, Liverpool. Erik Halvorson,
Duke. Paul Harrenstein, TU Munich. Ryo Ichimura, Kyushu U. Nicole Immorlica, Northwestern. Atsushi Iwasaki, Kyushu U. Kamal Jain, MSR.
Manish Jain, USC. Jayant Kalagnanam, IBM Research. Christopher Kiekintveld, UT El Paso. Dmytro Korzhyk, Duke. Jerome Lang, U. Paris-
Dauphine. Joshua Letchford, Duke. Vangelis Markakis, Athens U. Econ. and Business. Kohki Maruono, Kyushu U. Kamesh Munagala, Duke.
Yoshifusa Omori, Kyushu U. Naoki Ohta, Kyushu U. Ron Parr, Duke. Michal Pechoucek, Prague TU. Ariel Procaccia, Harvard U. Daniel Reeves,
Yahoo! Research Matthew Rognlie MIT Jeff Rosenschein Hebrew U Yuko Sakurai Kyushu U Tuomas Sandholm CMU Paolo Santi IITYahoo! Research. Matthew Rognlie, MIT. Jeff Rosenschein, Hebrew U. Yuko Sakurai, Kyushu U. Tuomas Sandholm, CMU. Paolo Santi, IIT
CNR. Yasufumi Satoh, Kyushu U. Peng Shi, MIT. Milind Tambe, USC. Taiki Todo, Kyushu U. Ondrej Vanek, Prague TU. Liad Wagman, Illinois
Institute of Technology. Toby Walsh, NICTA and UNSW. Mathijs de Weerdt, TU Delft. Lirong Xia, Duke. Zhengyu Yin, USC. Makoto Yokoo,
Kyushu U. Michael Zuckerman, Hebrew U.
![Page 2: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/2.jpg)
Multiple entities with different interests
How can AI help?
![Page 3: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/3.jpg)
Multiple entities with different interests
Auctions
How can AI help?
![Page 4: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/4.jpg)
Multiple entities with different interests
AuctionsRating/voting systems
How can AI help?
![Page 5: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/5.jpg)
Multiple entities with different interests
AuctionsRating/voting systems
Kidney exchanges
How can AI help?Kidney exchanges
![Page 6: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/6.jpg)
Multiple entities with different interests
AuctionsRating/voting systems
Kidney exchanges
How can AI help?Kidney exchanges
Prediction markets
![Page 7: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/7.jpg)
Multiple entities with different interests
AuctionsRating/voting systems
Kidney exchanges
How can AI help?Kidney exchanges
Prediction marketsDonation matching
![Page 8: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/8.jpg)
Multiple entities with different interests
Security
AuctionsRating/voting systems
Kidney exchanges
How can AI help?Kidney exchanges
Prediction marketsDonation matching
![Page 9: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/9.jpg)
Multiple entities with different interests
THIS TALKSecurity
THIS TALKAuctions
Rating/voting systems
Kidney exchanges
How can AI help?Kidney exchanges
Prediction marketsDonation matching
![Page 10: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/10.jpg)
Multiple entities with different interests
THIS TALKSecurity
THIS TALKAuctions
Rating/voting systemsoverview: C., CACM March
2010
Kidney exchanges
2010
How can AI help?Kidney exchanges
Prediction marketsDonation matching
![Page 11: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/11.jpg)
Closer to home…
![Page 12: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/12.jpg)
Closer to home…G l iGame playing
![Page 13: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/13.jpg)
Closer to home…G l iGame playing
![Page 14: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/14.jpg)
Closer to home…G l iGame playing
![Page 15: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/15.jpg)
Closer to home…G l iGame playing
![Page 16: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/16.jpg)
Closer to home…G l iGame playingMultiagent systems
Goal: Blocked(Room0)Blocked(Room0)
Goal: Clean(Room0)
![Page 17: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/17.jpg)
![Page 18: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/18.jpg)
![Page 19: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/19.jpg)
Some microeconomic theory tools for AIGAME THEORY
2, 2 -1, 0
Dima Korzhyk -7, -8 0, 0
Josh Letchford
![Page 20: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/20.jpg)
Some microeconomic theory tools for AIGAME THEORY SOCIAL CHOICE
A>B>C2, 2 -1, 0
A>B>CB>A>CC>B>A
Lirong XiaDima Korzhyk -7, -8 0, 0 ↓B wins
Josh Letchford
![Page 21: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/21.jpg)
Some microeconomic theory tools for AIGAME THEORY SOCIAL CHOICE
A>B>C2, 2 -1, 0
A>B>CB>A>CC>B>A
Lirong XiaDima Korzhyk -7, -8 0, 0 ↓B wins
MECHANISM DESIGNJosh Letchford
42v1=42v2=30
20→ 1 wins,
pays 30
Liad WagmanMingyu Guov3=20 pays 30
![Page 22: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/22.jpg)
Some microeconomic theory tools for AIGAME THEORY SOCIAL CHOICE
A>B>C2, 2 -1, 0
A>B>CB>A>CC>B>A
Lirong XiaDima Korzhyk -7, -8 0, 0 ↓B wins
THIS TALK
MECHANISM DESIGNJosh Letchford
42v1=42v2=30
20→ 1 wins,
pays 30
Liad WagmanMingyu Guov3=20 pays 30
![Page 23: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/23.jpg)
Penalty kick example
![Page 24: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/24.jpg)
Penalty kick example
probability .6
probability .4
![Page 25: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/25.jpg)
Penalty kick example
action
probability .6
action
probability .4
![Page 26: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/26.jpg)
Penalty kick example
probability .7
probability .3
action
probability .6
action
probability .4
![Page 27: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/27.jpg)
Penalty kick example
probability .7
probability .3
probability 1
action
probability .6
action
probability .4
![Page 28: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/28.jpg)
Penalty kick example
probability .7
probability .3
action
probability 1
action
probability .6
action
probability .4
![Page 29: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/29.jpg)
Penalty kick example
probability .7
probability .3
action
probability 1
Is this a action
probability .6“rational” outcome? If not, what
action
probability .4 is?
![Page 30: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/30.jpg)
Penalty kick(also known as: matching pennies)
L R
0 0 -1 1L
L R
0, 0 1, 1-1 1 0 0
L
R 1, 1 0, 0R
![Page 31: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/31.jpg)
Penalty kick(also known as: matching pennies)
L R.5 .5
0 0 -1 1L
L R
5 0, 0 1, 1-1 1 0 0
L
R
.5
5 1, 1 0, 0R.5
![Page 32: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/32.jpg)
Security example
BCN terminal 2A BCN terminal 2B
![Page 33: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/33.jpg)
Security example
BCN terminal 2A BCN terminal 2B
actionaction
actionaction
![Page 34: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/34.jpg)
Security gamey g
2A 2B
0, 0 -1, 22A
-1, 1 0, 02B
![Page 35: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/35.jpg)
Security gamey g
2A 2B
0, 0 -1, 22A.33
-1, 1 0, 02B.67
![Page 36: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/36.jpg)
Recent deployments in security
• Tambe’s TEAMCORE group at USC• Tambe s TEAMCORE group at USC
• Airport security• Where should checkpoints, canine units, etc. be deployed?
• Deployed at LAX and another US airport, being evaluated for p y p , gdeployment at all US airports
• Federal Air MarshalsFederal Air Marshals
• Coast Guard
• …
![Page 37: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/37.jpg)
“Should I buy an SUV?”(also known as the Prisoner’s Dilemma)(also known as the Prisoner s Dilemma)
purchasing + gas cost
cost: 5
cost: 3
![Page 38: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/38.jpg)
“Should I buy an SUV?”(also known as the Prisoner’s Dilemma)(also known as the Prisoner s Dilemma)
purchasing + gas cost accident cost
cost: 5 cost: 5 cost: 5
cost: 3 cost: 8 cost: 2
cost: 5 cost: 5
![Page 39: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/39.jpg)
“Should I buy an SUV?”(also known as the Prisoner’s Dilemma)(also known as the Prisoner s Dilemma)
purchasing + gas cost accident cost
cost: 5 cost: 5 cost: 5
cost: 3 cost: 8 cost: 2
cost: 5 cost: 5
-10, -10 -7, -11
-11, -7 -8, -8
![Page 40: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/40.jpg)
“Should I buy an SUV?”(also known as the Prisoner’s Dilemma)(also known as the Prisoner s Dilemma)
purchasing + gas cost accident cost
cost: 5 cost: 5 cost: 5
cost: 3 cost: 8 cost: 2
cost: 5 cost: 5
-10, -10 -7, -11
-11, -7 -8, -8
![Page 41: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/41.jpg)
“Should I buy an SUV?”(also known as the Prisoner’s Dilemma)(also known as the Prisoner s Dilemma)
purchasing + gas cost accident cost
cost: 5 cost: 5 cost: 5
cost: 3 cost: 8 cost: 2
cost: 5 cost: 5Computational aspects of dominance: Gilboa, Kalai, Zemel Math of
-10, -10 -7, -11Kalai, Zemel Math of
OR ‘93; C. & SandholmEC ’05, AAAI’05;
Brandt Brill Fischer-11, -7 -8, -8
Brandt, Brill, Fischer, Harrenstein TOCS ‘11
![Page 42: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/42.jpg)
“Chicken”• Two players drive cars towards each other
• If one player goes straight that player wins• If one player goes straight, that player wins
• If both go straight, they both die
S D
D S
0 0 1 1D S
0, 0 -1, 1D
1, -1 -5, -5S
![Page 43: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/43.jpg)
Nash equilibrium [Nash ‘50]q [ ]
![Page 44: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/44.jpg)
Nash equilibrium [Nash ‘50]q [ ]
![Page 45: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/45.jpg)
Nash equilibrium [Nash ‘50]q [ ]
![Page 46: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/46.jpg)
Nash equilibrium [Nash ‘50]q [ ]
![Page 47: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/47.jpg)
Nash equilibrium [Nash ‘50]q [ ]
• A profile (= strategy for each player) so that no player wants to deviateplayer wants to deviate
![Page 48: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/48.jpg)
Nash equilibrium [Nash ‘50]q [ ]
• A profile (= strategy for each player) so that no player wants to deviateplayer wants to deviate
D S
0, 0 -1, 1D
1, -1 -5, -5S
![Page 49: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/49.jpg)
Nash equilibrium [Nash ‘50]q [ ]
• A profile (= strategy for each player) so that no player wants to deviateplayer wants to deviate
D S
0, 0 -1, 1D
1, -1 -5, -5S
• This game has another Nash equilibrium in g qmixed strategies – both play D with 80%
![Page 50: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/50.jpg)
The presentation game
Pay attention (A)
Do not pay attention (NA)
Put effort into presentation (E) 2, 2 -1, 0
Do not put effort into presentation (NE) -7, -8 0, 0
![Page 51: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/51.jpg)
The presentation game
Pay attention (A)
Do not pay attention (NA)
Put effort into presentation (E) 2, 2 -1, 0
Do not put effort into presentation (NE) -7, -8 0, 0
• Pure-strategy Nash equilibria: (E, A), (NE, NA)
![Page 52: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/52.jpg)
The presentation game
Pay attention (A)
Do not pay attention (NA)
Put effort into presentation (E) 2, 2 -1, 0
Do not put effort into presentation (NE) -7, -8 0, 0
• Pure-strategy Nash equilibria: (E, A), (NE, NA)
• Mixed-strategy Nash equilibrium:
![Page 53: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/53.jpg)
The presentation game
Pay attention (A)
Do not pay attention (NA)
Put effort into presentation (E) 2, 2 -1, 0
Do not put effort into presentation (NE) -7, -8 0, 0
• Pure-strategy Nash equilibria: (E, A), (NE, NA)
• Mixed-strategy Nash equilibrium:
((4/5 E 1/5 NE) (1/10 A 9/10 NA))((4/5 E, 1/5 NE), (1/10 A, 9/10 NA))
![Page 54: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/54.jpg)
The presentation game
Pay attention (A)
Do not pay attention (NA)
Put effort into presentation (E) 2, 2 -1, 0
Do not put effort into presentation (NE) -7, -8 0, 0
• Pure-strategy Nash equilibria: (E, A), (NE, NA)
• Mixed-strategy Nash equilibrium:
((4/5 E 1/5 NE) (1/10 A 9/10 NA))((4/5 E, 1/5 NE), (1/10 A, 9/10 NA))– Utility -7/10 for presenter, 0 for audience
![Page 55: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/55.jpg)
Modeling and representing games2, 2 -1, 0
-7 -8 0 0
THIS TALK(unless specified -7, -8 0, 0
normal-form games
specified otherwise)
![Page 56: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/56.jpg)
Modeling and representing games2, 2 -1, 0
-7 -8 0 0
THIS TALK(unless specified -7, -8 0, 0
normal-form games
specified otherwise)
Bayesian gamesBayesian games
![Page 57: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/57.jpg)
Modeling and representing games2, 2 -1, 0
-7 -8 0 0
THIS TALK(unless specified -7, -8 0, 0
normal-form games
specified otherwise)
extensive-form games
Bayesian gamesBayesian games
![Page 58: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/58.jpg)
Modeling and representing games2, 2 -1, 0
-7 -8 0 0
THIS TALK(unless specified -7, -8 0, 0
normal-form games
specified otherwise)
extensive-form games
Bayesian gamesBayesian gamesstochastic games
![Page 59: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/59.jpg)
Modeling and representing games2, 2 -1, 0
-7 -8 0 0
THIS TALK(unless specified -7, -8 0, 0
normal-form games
specified otherwise)
extensive-form games
Bayesian gamesBayesian gamesstochastic games
hi lgraphical games[Kearns, Littman, Singh UAI’01]
![Page 60: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/60.jpg)
Modeling and representing games2, 2 -1, 0
-7 -8 0 0
THIS TALK(unless specified -7, -8 0, 0
normal-form games
specified otherwise)
extensive-form games
Bayesian gamesBayesian gamesstochastic games
hi laction-graph games
[L B & T h l IJCAI’03graphical games[Kearns, Littman, Singh UAI’01]
[Leyton-Brown & Tennenholtz IJCAI’03[Bhat & Leyton-Brown, UAI’04]
[Jiang, Leyton-Brown, Bhat GEB’11]
![Page 61: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/61.jpg)
Modeling and representing games2, 2 -1, 0
-7 -8 0 0
THIS TALK(unless specified -7, -8 0, 0
normal-form games
specified otherwise)
extensive-form games
Bayesian gamesBayesian gamesstochastic games
hi laction-graph games
[L B & T h l IJCAI’03graphical games[Kearns, Littman, Singh UAI’01]
[Leyton-Brown & Tennenholtz IJCAI’03[Bhat & Leyton-Brown, UAI’04]
[Jiang, Leyton-Brown, Bhat GEB’11] MAIDs [Koller & Milch. IJCAI’01/GEB’03]
![Page 62: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/62.jpg)
Computing a single Nash equilibrium“Together with factoring, the complexity of finding a Nash equilibrium is in my opinion th t i t t t tithe most important concrete open question
on the boundary of P today.”
Christos Papadimitriou, STOC’01STOC’01
![Page 63: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/63.jpg)
Computing a single Nash equilibrium“Together with factoring, the complexity of finding a Nash equilibrium is in my opinion th t i t t t tithe most important concrete open question
on the boundary of P today.”
Christos Papadimitriou, STOC’01
• PPAD complete to compute one Nash equilibrium even in a
STOC’01
• PPAD-complete to compute one Nash equilibrium, even in a two-player game [Daskalakis, Goldberg, Papadimitriou STOC’06 Ch & D FOCS’06]STOC’06; Chen & Deng FOCS’06] • still holds for FPTAS / smoothed poly [Chen, Deng, Teng FOCS‘06]
![Page 64: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/64.jpg)
Computing a single Nash equilibrium“Together with factoring, the complexity of finding a Nash equilibrium is in my opinion th t i t t t tithe most important concrete open question
on the boundary of P today.”
Christos Papadimitriou, STOC’01
• PPAD complete to compute one Nash equilibrium even in a
STOC’01
• PPAD-complete to compute one Nash equilibrium, even in a two-player game [Daskalakis, Goldberg, Papadimitriou STOC’06 Ch & D FOCS’06]STOC’06; Chen & Deng FOCS’06] • still holds for FPTAS / smoothed poly [Chen, Deng, Teng FOCS‘06]
![Page 65: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/65.jpg)
Computing a single Nash equilibrium“Together with factoring, the complexity of finding a Nash equilibrium is in my opinion th t i t t t tithe most important concrete open question
on the boundary of P today.”
Christos Papadimitriou, STOC’01
• PPAD complete to compute one Nash equilibrium even in a
STOC’01[’91]
• PPAD-complete to compute one Nash equilibrium, even in a two-player game [Daskalakis, Goldberg, Papadimitriou STOC’06 Ch & D FOCS’06]STOC’06; Chen & Deng FOCS’06] • still holds for FPTAS / smoothed poly [Chen, Deng, Teng FOCS‘06]
![Page 66: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/66.jpg)
Computing a single Nash equilibrium“Together with factoring, the complexity of finding a Nash equilibrium is in my opinion th t i t t t tithe most important concrete open question
on the boundary of P today.”
Christos Papadimitriou, STOC’01
• PPAD complete to compute one Nash equilibrium even in a
STOC’01[’91]
• PPAD-complete to compute one Nash equilibrium, even in a two-player game [Daskalakis, Goldberg, Papadimitriou STOC’06 Ch & D FOCS’06]STOC’06; Chen & Deng FOCS’06] • still holds for FPTAS / smoothed poly [Chen, Deng, Teng FOCS‘06]
• Is one Nash equilibrium all we need to know?
![Page 67: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/67.jpg)
A useful reduction (SAT → game) [C & Sandholm IJCAI’03 Games and Economic Behavior ‘08][C. & Sandholm IJCAI 03, Games and Economic Behavior 08]
(Earlier reduction with weaker implications: Gilboa & Zemel GEB ‘89)Formula: (x1 or -x2) and (-x1 or x2)Solutions: x =true x =trueSolutions: x1=true,x2=true
x1=false,x2=false
![Page 68: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/68.jpg)
A useful reduction (SAT → game) [C & Sandholm IJCAI’03 Games and Economic Behavior ‘08][C. & Sandholm IJCAI 03, Games and Economic Behavior 08]
(Earlier reduction with weaker implications: Gilboa & Zemel GEB ‘89)Formula: (x1 or -x2) and (-x1 or x2)Solutions: x =true x =trueSolutions: x1=true,x2=true
x1=false,x2=falseGame: x1 x2 +x1 -x1 +x2 -x2 (x1 or -x2) (-x1 or x2) default
x1 -2,-2 -2,-2 0,-2 0,-2 2,-2 2,-2 -2,-2 -2,-2 0,1x2 -2,-2 -2,-2 2,-2 2,-2 0,-2 0,-2 -2,-2 -2,-2 0,1
+x1 -2,0 -2,2 1,1 -2,-2 1,1 1,1 -2,0 -2,2 0,1-x1 -2,0 -2,2 -2,-2 1,1 1,1 1,1 -2,2 -2,0 0,1+x2 -2,2 -2,0 1,1 1,1 1,1 -2,-2 -2,2 -2,0 0,1-x2 -2,2 -2,0 1,1 1,1 -2,-2 1,1 -2,0 -2,2 0,1
(x or -x ) 2 2 2 2 0 2 2 2 2 2 0 2 2 2 2 2 0 1(x1 or -x2) -2,-2 -2,-2 0,-2 2,-2 2,-2 0,-2 -2,-2 -2,-2 0,1(-x1 or x2) -2,-2 -2,-2 2,-2 0,-2 0,-2 2,-2 -2,-2 -2,-2 0,1default 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 ε, ε
![Page 69: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/69.jpg)
A useful reduction (SAT → game) [C & Sandholm IJCAI’03 Games and Economic Behavior ‘08][C. & Sandholm IJCAI 03, Games and Economic Behavior 08]
(Earlier reduction with weaker implications: Gilboa & Zemel GEB ‘89)Formula: (x1 or -x2) and (-x1 or x2)Solutions: x =true x =trueSolutions: x1=true,x2=true
x1=false,x2=falseGame: x1 x2 +x1 -x1 +x2 -x2 (x1 or -x2) (-x1 or x2) default
x1 -2,-2 -2,-2 0,-2 0,-2 2,-2 2,-2 -2,-2 -2,-2 0,1x2 -2,-2 -2,-2 2,-2 2,-2 0,-2 0,-2 -2,-2 -2,-2 0,1
+x1 -2,0 -2,2 1,1 -2,-2 1,1 1,1 -2,0 -2,2 0,1-x1 -2,0 -2,2 -2,-2 1,1 1,1 1,1 -2,2 -2,0 0,1+x2 -2,2 -2,0 1,1 1,1 1,1 -2,-2 -2,2 -2,0 0,1-x2 -2,2 -2,0 1,1 1,1 -2,-2 1,1 -2,0 -2,2 0,1
(x or -x ) 2 2 2 2 0 2 2 2 2 2 0 2 2 2 2 2 0 1(x1 or -x2) -2,-2 -2,-2 0,-2 2,-2 2,-2 0,-2 -2,-2 -2,-2 0,1(-x1 or x2) -2,-2 -2,-2 2,-2 0,-2 0,-2 2,-2 -2,-2 -2,-2 0,1default 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 ε, ε
• Every satisfying assignment (if there are any) corresponds to an equilibrium with utilities 1, 1
• Exactly one additional equilibrium with utilities ε, ε that always exists
![Page 70: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/70.jpg)
Some algorithm families for computing Nash ilib i f 2 l l fequilibria of 2-player normal-form games
L k Himage from von Stengel
Lemke-Howson [J. SIAM ‘64]Exponential time due to Savani & von Stengel [FOCS’04 / Econometrica’06]
![Page 71: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/71.jpg)
Some algorithm families for computing Nash ilib i f 2 l l fequilibria of 2-player normal-form games
L k HSearch over supports / MIP
[Dickhaut & Kaplan Mathematica J ‘91]image from von Stengel
Lemke-Howson [J. SIAM ‘64]Exponential time due to Savani & von Stengel [FOCS’04 / Econometrica’06]
[Dickhaut & Kaplan, Mathematica J. 91] [Porter, Nudelman, Shoham AAAI’04 / GEB’08]
[Sandholm, Gilpin, C. AAAI’05]
![Page 72: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/72.jpg)
Some algorithm families for computing Nash ilib i f 2 l l fequilibria of 2-player normal-form games
L k HSearch over supports / MIP
[Dickhaut & Kaplan Mathematica J ‘91]image from von Stengel
Lemke-Howson [J. SIAM ‘64]Exponential time due to Savani & von Stengel [FOCS’04 / Econometrica’06]
[Dickhaut & Kaplan, Mathematica J. 91] [Porter, Nudelman, Shoham AAAI’04 / GEB’08]
[Sandholm, Gilpin, C. AAAI’05]
Special cases / subroutines[C & Sandholm AAAI’05 AAMAS’06; Benisch[C. & Sandholm AAAI 05, AAMAS 06; Benisch,
Davis, Sandholm AAAI’06 / JAIR’10; Kontogiannis & Spirakis APPROX’11; Adsul,
Garg, Mehta, Sohoni STOC’11; …]
![Page 73: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/73.jpg)
Some algorithm families for computing Nash ilib i f 2 l l fequilibria of 2-player normal-form games
L k HSearch over supports / MIP
[Dickhaut & Kaplan Mathematica J ‘91]image from von Stengel
Lemke-Howson [J. SIAM ‘64]Exponential time due to Savani & von Stengel [FOCS’04 / Econometrica’06]
[Dickhaut & Kaplan, Mathematica J. 91] [Porter, Nudelman, Shoham AAAI’04 / GEB’08]
[Sandholm, Gilpin, C. AAAI’05]
Special cases / subroutines[C & Sandholm AAAI’05 AAMAS’06; Benisch
Approximate equilibria[Brown ’51 / C. ’09 / Goldberg, Savani, Sørensen,
Ventre ’11; Althöfer ‘94 Lipton Markakis Mehta ‘03[C. & Sandholm AAAI 05, AAMAS 06; Benisch, Davis, Sandholm AAAI’06 / JAIR’10;
Kontogiannis & Spirakis APPROX’11; Adsul, Garg, Mehta, Sohoni STOC’11; …]
Ventre 11; Althöfer 94, Lipton, Markakis, Mehta 03,Daskalakis, Mehta, Papadimitriou ‘06, ‘07, Feder, Nazerzadeh, Saberi ‘07, Tsaknakis & Spirakis ‘07,
Spirakis ‘08, Bosse, Byrka, Markakis ‘07, …]
![Page 74: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/74.jpg)
Sidestepping the problemsSidestepping the problems
![Page 75: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/75.jpg)
Sidestepping the problemsSidestepping the problems
(one solution concept is not enough…?)( p g )
![Page 76: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/76.jpg)
Nash is not optimal if one player can commit
1, 1 3, 00, 0 2, 1
von Stackelberg
![Page 77: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/77.jpg)
Nash is not optimal if one player can commit
1, 1 3, 0U i N h
0, 0 2, 1Unique Nash equilibrium
von Stackelberg
![Page 78: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/78.jpg)
Nash is not optimal if one player can commit
1, 1 3, 0U i N h
0, 0 2, 1Unique Nash equilibrium
• Suppose the game is played as follows: von Stackelberg
![Page 79: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/79.jpg)
Nash is not optimal if one player can commit
1, 1 3, 0U i N h
0, 0 2, 1Unique Nash equilibrium
• Suppose the game is played as follows: von Stackelberg
– Player 1 commits to playing one of the rows,
![Page 80: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/80.jpg)
Nash is not optimal if one player can commit
1, 1 3, 0U i N h
0, 0 2, 1Unique Nash equilibrium
• Suppose the game is played as follows: von Stackelberg
– Player 1 commits to playing one of the rows,
– Player 2 observes the commitment and then chooses a columnPlayer 2 observes the commitment and then chooses a column
![Page 81: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/81.jpg)
Nash is not optimal if one player can commit
1, 1 3, 0U i N h
0, 0 2, 1Unique Nash equilibrium
• Suppose the game is played as follows: von Stackelberg
– Player 1 commits to playing one of the rows,
– Player 2 observes the commitment and then chooses a columnPlayer 2 observes the commitment and then chooses a column
• Optimal strategy for player 1: commit to Down
![Page 82: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/82.jpg)
Commitment to mixed strategiesg
1, 1 3, 0, ,
0, 0 2, 1
![Page 83: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/83.jpg)
Commitment to mixed strategiesg
1, 1 3, 0.49 , ,
0, 0 2, 1.51
![Page 84: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/84.jpg)
Commitment to mixed strategiesg
0 1
1, 1 3, 0.49 , ,
0, 0 2, 1.51
![Page 85: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/85.jpg)
Commitment to mixed strategiesg
0 1
1, 1 3, 0.49 , ,
0, 0 2, 1.51
– Sometimes also called a Stackelberg (mixed) strategy
![Page 86: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/86.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
![Page 87: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/87.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
![Page 88: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/88.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
![Page 89: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/89.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
![Page 90: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/90.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
![Page 91: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/91.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
![Page 92: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/92.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
![Page 93: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/93.jpg)
Observing the defender’s distribution in security
BCN terminal 2A
BCN terminal 2B
observeMo Tu We Th Fr Sa
This argument is not uncontroversial… [Pita, Jain, Tambe, Ordóñez, Kraus AIJ’10; Korzhyk, Yin, Kiekintveld, C., Tambe JAIR’11; Korzhyk, C., Parr AAMAS’11]
![Page 94: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/94.jpg)
Computing the optimal mixed strategy to commit to
[C & Sandholm EC’06 von Stengel & Zamir GEB’10][C. & Sandholm EC 06, von Stengel & Zamir GEB 10]
Slide 7
![Page 95: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/95.jpg)
Computing the optimal mixed strategy to commit to
[C & Sandholm EC’06 von Stengel & Zamir GEB’10][C. & Sandholm EC 06, von Stengel & Zamir GEB 10]
• Separate LP for every column c*:p y
Slide 7
![Page 96: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/96.jpg)
Computing the optimal mixed strategy to commit to
[C & Sandholm EC’06 von Stengel & Zamir GEB’10][C. & Sandholm EC 06, von Stengel & Zamir GEB 10]
• Separate LP for every column c*:p y
maximize Σr pr uR(r, c*)
subject tosubject to
for all c, Σr pr uC(r, c) ≤ Σr pr uC(r, c*)
Σr pr = 1
Slide 7
![Page 97: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/97.jpg)
Computing the optimal mixed strategy to commit to
[C & Sandholm EC’06 von Stengel & Zamir GEB’10][C. & Sandholm EC 06, von Stengel & Zamir GEB 10]
• Separate LP for every column c*:p y
maximize Σr pr uR(r, c*)
subject tosubject to
for all c, Σr pr uC(r, c) ≤ Σr pr uC(r, c*)
Σr pr = 1 distributional constraint
Slide 7
![Page 98: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/98.jpg)
Computing the optimal mixed strategy to commit to
[C & Sandholm EC’06 von Stengel & Zamir GEB’10][C. & Sandholm EC 06, von Stengel & Zamir GEB 10]
• Separate LP for every column c*:p y
maximize Σr pr uR(r, c*)
subject tosubject to
for all c, Σr pr uC(r, c) ≤ Σr pr uC(r, c*) follower optimality
Σr pr = 1 distributional constraint
Slide 7
![Page 99: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/99.jpg)
Computing the optimal mixed strategy to commit to
[C & Sandholm EC’06 von Stengel & Zamir GEB’10][C. & Sandholm EC 06, von Stengel & Zamir GEB 10]
• Separate LP for every column c*:p y
maximize Σr pr uR(r, c*)
subject to
leader utility
subject to
for all c, Σr pr uC(r, c) ≤ Σr pr uC(r, c*) follower optimality
Σr pr = 1 distributional constraint
Slide 7
![Page 100: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/100.jpg)
Other nice properties of commitment to mixed strategies
![Page 101: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/101.jpg)
Other nice properties of commitment to mixed strategies
0, 0 -1, 1
• Agrees w. Nash in zero-sum games0, 0 1, 1
-1, 1 0, 0
![Page 102: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/102.jpg)
Other nice properties of commitment to mixed strategies
0, 0 -1, 1
• Agrees w. Nash in zero-sum games0, 0 1, 1
-1, 1 0, 0
• Leader’s payoff at least as good as p y gany Nash eq. or even correlated eq. (von Stengel & Zamir [GEB ‘10]; see also C
≥(von Stengel & Zamir [GEB 10]; see also C.
& Korzhyk [AAAI ‘11], Letchford & C. [draft])
![Page 103: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/103.jpg)
Other nice properties of commitment to mixed strategies
0, 0 -1, 1
• Agrees w. Nash in zero-sum games0, 0 1, 1
-1, 1 0, 0
• Leader’s payoff at least as good as p y gany Nash eq. or even correlated eq. (von Stengel & Zamir [GEB ‘10]; see also C
≥(von Stengel & Zamir [GEB 10]; see also C.
& Korzhyk [AAAI ‘11], Letchford & C. [draft])
• No equilibrium selection problem0, 0 -1, 1
1, -1 -5, -5
![Page 104: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/104.jpg)
Some other work on commitment in unrestricted games
2 2 1 02, 2 -1, 0
-7, -8 0, 0l fnormal-form games
learning to commit [Letchford, C., Munagala SAGT’09] uncertain observability [Korzhyk, C., Parr AAMAS’11]
correlated strategies [C. & Korzhyk, AAAI’11]
![Page 105: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/105.jpg)
Some other work on commitment in unrestricted games
2 2 1 02, 2 -1, 0
-7, -8 0, 0l fnormal-form games
learning to commit [Letchford, C., Munagala SAGT’09] uncertain observability [Korzhyk, C., Parr AAMAS’11]
correlated strategies [C. & Korzhyk, AAAI’11]
commitment in Bayesian games[C. & Sandholm EC’06; Paruchuri, Pearce, Marecki, Tambe, [ ; , , , ,
Ordóñez, Kraus AAMAS’08; Letchford, C., Munagala SAGT’09; Pita, Jain, Tambe, Ordóñez, Kraus AIJ’10; Jain,
Kiekintveld, Tambe AAMAS’11]
![Page 106: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/106.jpg)
Some other work on commitment in unrestricted games
2 2 1 02, 2 -1, 0
-7, -8 0, 0l fnormal-form games
learning to commit [Letchford, C., Munagala SAGT’09] uncertain observability [Korzhyk, C., Parr AAMAS’11]
correlated strategies [C. & Korzhyk, AAAI’11] extensive-form games [Letchford & C., EC’10]
commitment in Bayesian games[C. & Sandholm EC’06; Paruchuri, Pearce, Marecki, Tambe, [ ; , , , ,
Ordóñez, Kraus AAMAS’08; Letchford, C., Munagala SAGT’09; Pita, Jain, Tambe, Ordóñez, Kraus AIJ’10; Jain,
Kiekintveld, Tambe AAMAS’11]
![Page 107: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/107.jpg)
Some other work on commitment in unrestricted games
2 2 1 02, 2 -1, 0
-7, -8 0, 0l fnormal-form games
learning to commit [Letchford, C., Munagala SAGT’09] uncertain observability [Korzhyk, C., Parr AAMAS’11]
correlated strategies [C. & Korzhyk, AAAI’11] extensive-form games [Letchford & C., EC’10]
commitment in Bayesian games[C. & Sandholm EC’06; Paruchuri, Pearce, Marecki, Tambe, stochastic games[ ; , , , ,
Ordóñez, Kraus AAMAS’08; Letchford, C., Munagala SAGT’09; Pita, Jain, Tambe, Ordóñez, Kraus AIJ’10; Jain,
Kiekintveld, Tambe AAMAS’11]
gongoing work with Korzhyk,
Letchford, Parr
![Page 108: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/108.jpg)
Security resource allocation games
[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe AAMAS’09]
![Page 109: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/109.jpg)
Security resource allocation games
• Set of targets T
[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe AAMAS’09]
g
t1
t2t3
t5t4
![Page 110: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/110.jpg)
Security resource allocation games
• Set of targets T
[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe AAMAS’09]
g
• Set of security resources available to the defender (leader)
t1
1 t2
t3
2 t5t4
![Page 111: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/111.jpg)
Security resource allocation games
• Set of targets T
[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe AAMAS’09]
g
• Set of security resources available to the defender (leader)
• Set of schedules• Set of schedules
st1
1
s1
s2
t2t3
22
s3
t5t4
![Page 112: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/112.jpg)
Security resource allocation games
• Set of targets T
[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe AAMAS’09]
g
• Set of security resources available to the defender (leader)
• Set of schedules• Set of schedules
• Resource can be assigned to one of the schedules in
st1
1
s1
s2
t2t3
22
s3
t5t4
![Page 113: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/113.jpg)
Security resource allocation games
• Set of targets T
[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe AAMAS’09]
g
• Set of security resources available to the defender (leader)
• Set of schedules• Set of schedules
• Resource can be assigned to one of the schedules in
• Attacker (follower) chooses one target to attack
st1
1
s1
s2
t2t3
22
s3
t5t4
![Page 114: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/114.jpg)
Security resource allocation games
• Set of targets T
[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe AAMAS’09]
g
• Set of security resources available to the defender (leader)
• Set of schedules• Set of schedules
• Resource can be assigned to one of the schedules in
• Attacker (follower) chooses one target to attack
• Utilities: if the attacked target is defended,
otherwise
• st1
1
s1
s2
t2t3
22
s3
t5t4
![Page 115: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/115.jpg)
Game-theoretic properties of security resource allocation games [Korzhyk, Yin, Kiekintveld, C., Tambe JAIR’11]
![Page 116: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/116.jpg)
Game-theoretic properties of security resource allocation games [Korzhyk, Yin, Kiekintveld, C., Tambe JAIR’11]
For the defender:• For the defender: Stackelberg strategies are also Nash strategies– minor assumption needed
– not true with multiple attacks
![Page 117: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/117.jpg)
Game-theoretic properties of security resource allocation games [Korzhyk, Yin, Kiekintveld, C., Tambe JAIR’11]
For the defender:• For the defender: Stackelberg strategies are also Nash strategies– minor assumption needed
– not true with multiple attacks
• Interchangeability property for• Interchangeability property for Nash equilibria (“solvable”) 1, 2 1, 0 2, 2• no equilibrium selection problem
• still true with multiple attacks 1, 1 1, 0 2, 1
[Korzhyk, C., Parr IJCAI’11 – poster W. 3:30pm, talk F. 10:30am]
0, 1 0, 0 0, 1
![Page 118: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/118.jpg)
Game-theoretic properties of security resource allocation games [Korzhyk, Yin, Kiekintveld, C., Tambe JAIR’11]
For the defender:• For the defender: Stackelberg strategies are also Nash strategies– minor assumption needed
– not true with multiple attacks
• Interchangeability property for• Interchangeability property for Nash equilibria (“solvable”) 1, 2 1, 0 2, 2• no equilibrium selection problem
• still true with multiple attacks 1, 1 1, 0 2, 1
[Korzhyk, C., Parr IJCAI’11 – poster W. 3:30pm, talk F. 10:30am]
0, 1 0, 0 0, 1
![Page 119: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/119.jpg)
Game-theoretic properties of security resource allocation games [Korzhyk, Yin, Kiekintveld, C., Tambe JAIR’11]
For the defender:• For the defender: Stackelberg strategies are also Nash strategies– minor assumption needed
– not true with multiple attacks
• Interchangeability property for• Interchangeability property for Nash equilibria (“solvable”) 1, 2 1, 0 2, 2• no equilibrium selection problem
• still true with multiple attacks 1, 1 1, 0 2, 1
[Korzhyk, C., Parr IJCAI’11 – poster W. 3:30pm, talk F. 10:30am]
0, 1 0, 0 0, 1
![Page 120: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/120.jpg)
Scalability in security gamest1
1s1
s2
1
t2t3
2s3
t5t4
basic model[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe
AAMAS’09; Korzhyk, C., Parr, AAAI’10; Jain, Kardeş, Kiekintveld, Ordóñez, Tambe
AAAI’10; Korzhyk C Parr IJCAI’11]AAAI 10; Korzhyk, C., Parr, IJCAI 11]
![Page 121: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/121.jpg)
Scalability in security gamest1
1s1
s2
1
t2t3
2s3
t5t4
basic modelgames on graphs
(usually zero-sum)[H l C P IJCAI’09 T i Yi[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe
AAMAS’09; Korzhyk, C., Parr, AAAI’10; Jain, Kardeş, Kiekintveld, Ordóñez, Tambe
AAAI’10; Korzhyk C Parr IJCAI’11]
[Halvorson, C., Parr IJCAI’09; Tsai, Yin, Kwak, Kempe, Kiekintveld, Tambe AAAI’10; Jain, Korzhyk, Vaněk, C.,
Pěchouček, Tambe AAMAS’11]; ongoing AAAI 10; Korzhyk, C., Parr, IJCAI 11] , ]; g gwork with Letchford, Vorobeychik
![Page 122: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/122.jpg)
Scalability in security gamest1
1s1
s2
1
t2t3
2s3
t5t4
basic modelgames on graphs
(usually zero-sum)[H l C P IJCAI’09 T i Yi[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe
AAMAS’09; Korzhyk, C., Parr, AAAI’10; Jain, Kardeş, Kiekintveld, Ordóñez, Tambe
AAAI’10; Korzhyk C Parr IJCAI’11]
[Halvorson, C., Parr IJCAI’09; Tsai, Yin, Kwak, Kempe, Kiekintveld, Tambe AAAI’10; Jain, Korzhyk, Vaněk, C.,
Pěchouček, Tambe AAMAS’11]; ongoing
Techniques:
AAAI 10; Korzhyk, C., Parr, IJCAI 11] , ]; g gwork with Letchford, Vorobeychik
![Page 123: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/123.jpg)
Scalability in security gamest1
1s1
s2
1
t2t3
2s3
t5t4
basic modelgames on graphs
(usually zero-sum)[H l C P IJCAI’09 T i Yi[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe
AAMAS’09; Korzhyk, C., Parr, AAAI’10; Jain, Kardeş, Kiekintveld, Ordóñez, Tambe
AAAI’10; Korzhyk C Parr IJCAI’11]
[Halvorson, C., Parr IJCAI’09; Tsai, Yin, Kwak, Kempe, Kiekintveld, Tambe AAAI’10; Jain, Korzhyk, Vaněk, C.,
Pěchouček, Tambe AAMAS’11]; ongoing
Techniques:
AAAI 10; Korzhyk, C., Parr, IJCAI 11] , ]; g gwork with Letchford, Vorobeychik
t li /i tDefender utility
Marginal probability
compact linear/integer programs
Distributional constraints
Marginal probability of t* being defended (?)
Attacker optimality
![Page 124: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/124.jpg)
Scalability in security gamest1
1s1
s2
1
t2t3
2s3
t5t4
basic modelgames on graphs
(usually zero-sum)[H l C P IJCAI’09 T i Yi[Kiekintveld, Jain, Tsai, Pita, Ordóñez, Tambe
AAMAS’09; Korzhyk, C., Parr, AAAI’10; Jain, Kardeş, Kiekintveld, Ordóñez, Tambe
AAAI’10; Korzhyk C Parr IJCAI’11]
[Halvorson, C., Parr IJCAI’09; Tsai, Yin, Kwak, Kempe, Kiekintveld, Tambe AAAI’10; Jain, Korzhyk, Vaněk, C.,
Pěchouček, Tambe AAMAS’11]; ongoing
Techniques:
AAAI 10; Korzhyk, C., Parr, IJCAI 11] , ]; g gwork with Letchford, Vorobeychik
t li /i tDefender utility
Marginal probability
compact linear/integer programsstrategy generation
Distributional constraints
Marginal probability of t* being defended (?)
Attacker optimality
![Page 125: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/125.jpg)
In summary: AI pushing at some of the boundaries of game theory
learning in games
behavioral (humans
game theory
playing games)
AI work in game theorycomputation
AI work in game theory
representationconceptual
(e.g., equilibrium selection)
representation
![Page 126: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/126.jpg)
Funding
Any opinions, conclusions or
recommendations are mine and do not
necessarily reflect the views of the
funding agencies
![Page 127: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/127.jpg)
Academic family
T S dh lVic Lesser (UMass)
Tuomas Sandholm(CMU)
Kate LarsonMike Benisch
Barbara Grosz
meAndrew Gilpin Kate Larson
(Waterloo)Mike Benisch
Avi Pfeffer
Liad Wagman (Ph.D. 2009 Illi i
Mingyu Guo (Ph D 2010
Lirong Xia (Ph D d
Josh Letchford(Ph D d )
Dima Korzhyk(Ph D d ) 2009, now at Illinois
Inst. of Tech.)(Ph.D. 2010, now at U. Liverpool)
(Ph.D. student, starting postdoc
at Harvard)
(Ph.D. student)(Ph.D. student)
![Page 128: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/128.jpg)
AI at Duke
Ron Parr Carlo TomasiRon Parr Carlo Tomasi Bruce DonaldAlex Hartemink
SECONDARY CS FACULTY
C&T book(s?)
SECONDARY CS FACULTY
meSayan MukherjeeSilvia Ferrari Sayan MukherjeeSilvia Ferrari
Don Loveland
office! D207
Mauro MaggioniUwe Ohler Alan Biermann
PROFESSORES EMERITI
![Page 129: One Equilibrium Is Not Enough: Computing Game-Theoretic ...conitzer/CT_slides_split.pdf · Josh Letchford v 1=42 v ... • A profile (= strategy for each player) so that no player](https://reader033.fdocuments.net/reader033/viewer/2022050421/5f90ea79227eef03d500acb7/html5/thumbnails/129.jpg)
Family