When Security Games Go Green Suraj Nair Outline

When Security Games Go Green Suraj Nair

Outline q Green Security Game q Planning algorithms q Planning and learning q Results 2

Green Security Domains: Protecting Fish and Wildlife 3

Features Green security games q Generalized Stackelberg assumption q Repeated and frequent attacks q Significant amounts data q Attacker bounded rationality q Limited surveillance/planning 4

Green Security Game Model n Jan Feb Mar Apr May 0. 3 0. 4 0. 1 0. 2 0. 5 0 0. 1 0. 7 0. 2 0. 3 0. 4 0. 5 0. 6 0. 1 0. 6 0. 7 0. 1 0. 2 0. 3 0. 2 0 0. 9 0. 4 0. 5 0. 6 0. 3 0. 1 0. 2 0. 3 0. 4 0. 5 0. 3 0. 4 0. 1 0. 3 0. 4 0. 5 0. 1 0. 2 0. 3 0. 4 0. 5 0. 1 0. 6 0. 3 0. 4 0. 2 0. 4 5

Green Security Game Model n Jan Feb Mar Apr May 6

Green Security Game Model n 7

Outline q Green Security Game Model q Planning algorithms q Planning and Learning q Results 8

Planning n N N S Jan 80% 20% Feb 20% 80% S 9

x n Jan Feb Planning Mar Apr May ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 10

Plan. Ahead-M q q Look ahead M steps: find an optimal strategy for current round as if it is the Mth last round of the game Sliding window of size M. Example with M=2 Jan Feb Mar Apr May 11

Plan. Ahead-M q Mathematical program 12

Fixed. Sequence-M n Jan A Feb B Mar A Apr B May A 13

Fixed. Sequence-M 14

Outline q Green Security Game Model q Planning algorithms q Planning and Learning q Results 15

Planning and Learning n Learn parameters in attackers’ bounded rationality model from attack data n Previous work n n Apply Maximum Likelihood Estimation (MLE) n May lead to highly biased results Proposed learning algorithm n Calculate posterior distribution for each data point 16

Planning and Learning 17

General Framework of Green Security Game Start New Round Defender plans her strategy Learn from data: Improve model Local guards executes patrols Poachers attack targets 18

Outline q Green Security Game Model q Planning algorithms q Planning and Learning q Results 19

Experimental Results Planning n Baseline: FS-1 (Stackelberg), PA-1 (Myopic) n Attacker respond to last round strategy, 10 Targets, 4 Patrollers Baseline 20

Experimental Results Planning and Learning n Baseline: Maximum Likelihood Estimation (MLE) Solution Quality Runtime 21

Thank you! 22