Modeling And Defending Against Synergetic Attacks In Wildlife Protection Games

Illegal poaching is a severe international issue which leads to destruction of ecosys-tems and might even cause extinction of species.As the population of endangered species keeps dwindling,optimizing defender's anti-poaching strategies with limited security resources is receiving researchers' attention.Stackelberg security games,which appropriately model interactions between attackers and defenders,have been success-fully deployed in many real-life scenarios.While some traditional Stackelberg security games make the assumptions that attackers are perfectly rational and attack individu-ally,these assumptions are not practical in wildlife protection.To combat illegal poach-ing more effectively,we delve into the model designing and defender strategy gener-ating in a wildlife protection game,taking the possible cooperation among attackers into consideration.The major contributions in this paper are summarized as follows:?We introduce the basic concepts and researching actuality of Stackelberg secu-rity game,and construct a cooperation-enabled and multi-round wildlife protection game addressing potential synergetic attacks.?To predict attackers' choices on targets in the wildlife protection game,we pro-pose a novel adaptive human behavior model based on human attackers'bounded rationality and adaptive nature.This model defines the attackers' "Inclination" at each target,associates attackers' choices with the linear combination of 3 features of the targets,and presents the probability that a given target is attacked.Based on the model,we formulate the optimization problem generating the optimal defender strategies.?We propose an approximation-based method that solves the non-convex and non-linear optimization problem generating the optimal defender strategies.This method decomposes the original problem into 4 subproblems and makes a series of variable substitutions.After introducing two sets of piecewise linear functions,we trans-form the subproblem into a mixed integer quadratically constrained quadratic pro-gram(MIQCQP)problem which could be numerically solved using any MIQCQP solver.? To evaluate the effectiveness of our model and solution,we conduct experiments in which we invite players with various education backgrounds to play the poaching game we design.With 25000 attacks recorded,we show our solution effectively helps the defender to deal with poachers who might cooperate.