Towards a Science of Security Games : Key Algorithmic Principles, Deployed Systems, Research Challenges Milind Tambe Helen N. and Emmett H. Jones Professor in Engineering University of Southern California with: Current/former PhD students/postdocs: Other collaborators: Fernando Ordonez (USC & U Chile), Richard John Bo An, Matthew Brown, Francesco Delle Fave, Fei (USC) , David Kempe (USC), Shaddin Dughmi Fang, Benjamin Ford, William Haskell, Manish Jain, (USC) & Albert Jiang, Debarun Kar, Chris Kiekintveld, Rajiv Maheswaran, Janusz Marecki, Praveen Paruchuri, Craig Boutilier (Toronto), Jeff Brantingahm Jonathan Pearce,James Pita, Thanh Nguyen, Yundi (UCLA), Vince Conitzer (Duke), Sarit Kraus (BIU, Qian, Eric Shieh, Jason Tsai, Pradeep Varakantham, Israel), Andrew Lemieux (NCSR),Kevin Leyton- Haifeng Xu, Amulya Yadav, Rong Yang, Zhengyu Brown (UBC), M. Pechoucek (CTU, Czech R), Ariel Yin, Chao Zhang Procaccia (CMU), Tuomas Sandholm (CMU), Martin Short (GATech), Y. Vorobeychik (Vanderbilt), …. 1 /59
Global Challenges for Security: Game Theory for Security Resource Optimization 2 /59
Example Model: Stackelberg Security Games Adversary Security allocation: Targets have weights Adversary surveillance Defender Target Target #1 #2 Target #1 4, -3 -1, 1 Target #2 -5, 5 2, -1 3 /59
Example Model: Stackelberg Security Games Adversary Security allocation: Targets have weights Adversary surveillance Defender Target Target #1 #2 Target #1 4, -3 -1, 1 Target #2 -5, 5 2, -1 4 /59
Example Model: Stackelberg Security Games Adversary Security allocation: Targets have weights Adversary surveillance Defender Target Target #1 #2 Target #1 4, -3 -1, 1 Target #2 -5, 5 2, -1 5 /59
Stackelberg Security Games Security Resource Optimization: Not 100% Security Random strategy: Adversary Increase cost/uncertainty to attackers Stackelberg game: Defender commits to mixed strategy Adversary conducts surveillance; responds Stackelberg Equilibrium: Optimal random? Target Target Defender #1 #2 Target #1 4, -3 -1, 1 Target #2 -5, 5 2, -1 6 /59
Research Contributions: Game Theory for Security Behavioral game Computational game theory: theory: • Exploit human • behavior Massive games models + P lanning under uncertainty, learning… Computational Game Theory in the Field 7 /59
Applications: Deployed Security Assistants Ports & port traffic US Coast Guard Airports, access roads & flights TSA, Airport Police Urban transport LA Sheriff’s/TSA Singapore Police Environment US Coast Guard, WWF, WCS… 8/59
Key Lessons: Security Games Decision aids based on computational game theory in daily use Optimize limited security resources against adversaries Applications yield research challenges: Science of security games Scale-up: Incremental strategy generation & Marginals Uncertainty: Integrate MDPs, Robustness, Quantal response Current applications (wildlife security): Interdisciplinary challenge Global challenges: Merge planning/learning & security games 9 /59
Outline: “Security Games” Research (2007 -Now) Environment Trains Ports Roads Flights Airports 2007 2009 2011 2012 2013 2013- Evaluation I: Scale up? Handle uncertainty? Evaluation II: Real-world deployments (Patience) Publications: AAMAS, AAAI, IJCAI… 2007 onwards 10 /59
Airport Security: Mapping to Stackelberg Games ARMOR: LAX (2007) GUARDS: TSA (2011) • 6 plots against LAX GLASGOW 6/30/07 11 /59
ARMOR Operation [2007] Generate Detailed Defender Schedule Pita Paruchuri Target #1 Target #2 Defender #1 2, -1 -3, 4 Defender #2 -3, 3 3, -2 Mixed Integer Program Pr(Canine patrol, 8 AM @Terminals 2,5,6) = 0.17 Pr(Canine patrol, 8 AM @ Terminals 3,5,7) = 0.33 …… Canine Team Schedule, July 28 Term 1 Term 2 Term 3 Term 4 Term 5 Term 6 Term 7 Term 8 8 AM Team1 Team3 Team5 9 AM Team1 Team2 Team4 12 /59 10 AM Team3 Team5 Team2
ARMOR MIP [2007] Generate Mixed Strategy for Defender Pita Paruchuri Target #1 Target #2 Defender #1 2, -1 -3, 4 Defender #2 -3, 3 3, -2 Maximize defender max R x q ij i j expected utility i X j Q Defender mixed . . 1 s t x i strategy i 1 q Adversary response j Q j Adversary best 0 ( ) ( 1 ) a C x q M ij i j response 13 /59 i X
ARMOR Payoffs [2007] Previous Research Provides Payoffs in Security Game Domains Target #1 Target #2 Defender #1 2, -1 -3, 4 Defender #2 -3, 3 3, -2 Maximize defender max R x q ij i j expected utility i X j Q 14 /59
ARMOR MIP [2007] Solving for a Single Adversary Type Term #1 Term #2 𝑦 𝑗 Defend#1 2, -1 -3, 4 Defend#2 -3, 1 3, -3 ARMOR…throws a digital cloak of invisibility…. Maximize defender max R x q ij i j expected utility i X j Q . . 1 s t x Defender strategy i i 1 q Adversary strategy j Q j Adversary best 0 ( ) ( 1 ) a C x q M ij i j response i X 15 /59
IRIS: Federal Air Marshals Service [2009] Scale Up Number of Defender Strategies Strategy 1 Strategy 2 Strategy 3 Strategy 1 Strategy 2 Strategy 3 Strategy 1 Strateg y 1 Strategy 2 Strategy 3 Strateg y 2 Strategy 4 Strateg y 3 Strategy 5 Strategy 6 Strateg y 4 Strateg y 5 Strateg y 6 1000 Flights, 20 air marshals: 10 41 combinations ARMOR out of memory Not enumerate all combinations: Branch and price: Incremental strategy generation 16 /59
IRIS: Scale Up Number of Defender Strategies [2009] Small Support Set for Mixed Strategies Small support set size: 1000 flights, 20 air marshals: • Most xi variables zero 10 41 combinations max , R x q Attack Attack Attack Attack x q ij i j … 1 2 1000 i X j Q … 1,2,3.. 5,-10 4,-8 -20,9 x123=0.0 . . 1 , 1 s t x q i j 1,2,4.. 5,-10 … 4,-8 -20,9 x124=0.239 i j Q 1,3,5.. 5,-10 … x135=0.0 -9,5 -20,9 0 ( ) ( 1 ) a C x q M ij i j … i X 10 41 rows x378=0.123 … [ 0 ... 1 ], { 0 , 1 } x q i j 17 /59
IRIS: Incremental Strategy Generation Exploit Small Support Jain Kiekintveld Master Attack 1 Attack 2 Attack … Attack 6 Slave (LP Duality Theory) … 1,2,4 5,-10 4,-8 -20,9 Best new pure strategy: Minimum cost network flow Attack 1 Attack 2 Attack … Attack 6 Target 7 Target 3 Resource Sink … 1,2,4 5,-10 4,-8 -20,9 … … … 3,7,8 -8, 10 -8,10 -8,10 Attack 1 Attack 2 Attack Attack 6 Converge: 500 rows … 1,2,4 5,-10 4,-8 -20,9 GLOBAL NOT 10 41 … 3,7,8 -8, 10 -8,10 -8,10 OPTIMAL … 18 /59
IRIS: Deployed FAMS (2009-) Significant change in operations “… in 2011, the Military Operations Research Society selected a University of Southern California project with FAMS on randomizing flight schedules for the prestigious Rist Award …” - R. S. Bray (TSA) Transportation Security Subcommittee US House of Representatives 2012 19 /59
Networks: Mumbai Police Checkpoints[2013]* Jain 150 edges; 2 Checkpoints 150-choose-2 strategies *With V Conitzer 20 /59
Networks: Mumbai Police Checkpoints[2013]* Incremental Strategy Generation Jain Double oracle: Converge to a global optimal Path #1 Path #2 Path #3 Defender oracle Checkpoint 5, -5 -1, 1 -2, 2 Path #1 Path #2 strategy #1 Path #1 Path #2 Checkpoint 5, -5 -1, 1 Checkpoint -5, 5 1, -1 -2, 2 strategy #1 Checkpoint 5, -5 -1, 1 strategy #2 strategy #1 Checkpoint -5, 5 2, -1 Attacker oracle strategy #2 Path #1 Path #2 Path #3 Checkpoint 5, -5 -1, 1 -2, 2 strategy #1 Checkpoint -5, 5 1, -1 -2, 2 strategy #2 *With V Conitzer 21 /59
Double Oracle[2013] Incremental Strategy Generation: Exploit Small Support Jain 150 edges; 2 Checkpoints Only six candidate edges for checkpoints 22 /59
Mumbai Police Checkpoints[2013] Results of Scale-up 20416 Roads,15 checkpoints: 20 min 23 /59
Double Oracle: Social, Cyber Networks[2013] Incremental Strategy Generation Tsai Social networks: Cyber networks: e.g., counter-insurgency Sources Links Targets Intermediate Nodes 24 /59
Outline: “Security Games” Research Environment Trains Ports Roads Flights Airports 2007 2009 2011 2012 2013 2013- 25 /59
Port Security Threat Scenarios US Ports: $3.15 trillion economy Attack on a ferry USS Cole after suicide attack French oil tanker hit by small boat 26 /59
PROTECT: Randomized Patrol Scheduling [2013] Port Protection (Scale-up) and Ferries (Continuous Space/time) 27 /59
PROTECT: Randomized Patrol Scheduling [2013] Port Protection (Scale-up) and Ferries (Continuous Space/time) 28 /59
Ferries: Scale-up with Mobile Resources & Moving Targets Transition Graph Representation 10 min 5 min 15 min A A, 5 min A, 10 min A, 15 min B B, 5 min B, 10 min B, 15 min C C, 5 min C, 10 min C, 15 min 29 /59
Recommend
More recommend
