Incentive-Based Control of Asynchronous Best-Response Dynamics on Binary Decision Networks

Various populations of interacting decision-making agents can be modeled by asynchronous best-response dynamics, or equivalently, linear threshold dynamics. Building upon recent convergence results in the absence of control, we now consider how such a network can be efficiently driven to a desired equilibrium state by offering payoff incentives or rewards for using a particular strategy, either uniformly or targeted to individuals. We begin by showing that strategy changes are monotone following an increase in payoffs in coordination games, and that the resulting equilibrium is unique. Based on these results, for the case when a uniform incentive is offered to all agents, we show how to compute the optimal incentive using a binary search algorithm. When different incentives can be offered to each agent, we propose a new algorithm to select which agents should be targeted based on maximizing a ratio between the cascading effect of a strategy switch by each agent and the incentive required to cause the agent to switch. Simulations show that this algorithm computes near-optimal targeted incentives for a wide range of networks and payoff distributions in coordination games and can also be effective for anti-coordination games