Analyzing the Theoretical Performance of Information Sharing

Abstract

Individuals in large, heterogeneous teams will commonly produce sensor data that is likely useful to some other members of the team, but it is not precisely known to whom the information is useful. Some recent work has shown that randomly propagating the information performed surprisingly well, compared to infeasible optimal approaches. This chapter extends that work by looking at how the relative performance of random information passing algorithms scales with the size of the team. Additionally, the chapter looks at how random information passing performs when sensor data is noisy, so that individuals need multiple pieces of data to reach a conclusion, and the underlying situation is dynamic, so individuals need new information over time. Results show that random information passing is broadly effective, although relative performance is lower in some situations.