There are many good reasons for the peaked interest in decision and inference from multi-agent distributed learners, especially in this day and age when the word “network” has become commonplace whether one is referring to social networks, power networks, biological networks or other types of networks. Some of these reasons have to do with the benefits of cooperation among agents in terms of improved performance and improved resilience to failure. Other reasons deal with privacy and secrecy considerations where agents may not be comfortable sharing their data with remote fusion centers. In other situations, the data may already be available in dispersed locations, as happens with cloud computing. One may also be interested in learning through data mining from Big Data sets.
The field of network science deals with issues related to the aggregation, processing, and diffusion of information over graphs linking a multitude of agents. While interactions among agents can be studied from the perspective of cluster formations, degrees of connectivity, and small-world effects, it is the possibility of having agents interact dynamically with each other, and influence each other’s behavior, that opens up a plethora of notable possibilities. For example, examination of how local interactions influence global behavior can lead to a broader understanding of how localized interactions in the social sciences, life sciences, and system sciences influence the evolution of the respective networks.
In this lecture, we provide new insights into the learning behavior of networked agents and describe some interesting patterns of behavior on how information flows over graphs. For strongly-connected graphs, we will explain how all agents are able to learn the desired true state within the same accuracy level even when different agents are subjected to heterogeneous levels of information. In contrast, over weakly-connected graphs, which are useful in modeling interactions over some social networks, a leader-follower relationship develops with some agents dictating the behavior of other agents regardless of the local clues that are sensed by these other agents. The findings clarify how asymmetries in the exchange of information over graphs play a critical role in defining the cognitive abilities of the networked system.