Intelligent Information Sharing for Localized, Non-Stationary Phenomena

Information sharing is important in agent-based sensing, especially when only a small subset of a large team of agents can directly observe local environment phenomena. Moreover, sharing is further complicated in non-stationary environments, where changes in the phenomena over time require the team to collectively revise their beliefs as the phenomena change. In this paper, we first analytically and empirically demonstrate the difficulty inherent in sharing information and revising beliefs over time about localized, non-stationary phenomena, uncovering the inertia-based Institutional Memory Problem. Subsequently, we propose two solutions for addressing this problem: 1) a change detection and response algorithm, and 2) a forgetting-based solution. We test our solutions under several network structures to verify the efficacy of our approaches and evaluate their robustness in the presence of faulty and/or malicious agents injecting incorrect information into the team.