Multi-scale Dynamic Network Graph (MUSING) for Network-based Information Exploitation

As previously dispersed information or entities become more and more connected, they form increasingly complicated networks and result in complex interactions. The benefit of an interconnected network is that since the information in the network is related in some fashion, information can be exploited to extract latent behaviors or trends of the network which would otherwise be missed if the information is viewed in isolation. We propose the Multi-scale Dynamic Network Graph (MUSING) model to encode, infer and predict the status of dynamic networks by fusing distributed observations (networked data) in the presence of noise or uncertainty. The model structure takes advantage of natural clustering of many networks to facilitate its temporal evolution. The architecture of the scale and time interactions in the MUSING model is particularly amenable to efficient propagation of information. Furthermore, due to the scale nodes in the model, large-scale behaviors and trends of the network are readily available, which offers additional insight into the network status, in addition to the individual nodes.