Techniques for dynamic behavioral analysis and modeling have recently become an increasingly researched topic.
In essence, they aim to understand the mechanics of a set of variables over time, allowing for prediction of future
data, anomaly or change detection, or estimation of a latent variable. Much of this research has focused on the
sequential analysis of individual tracks of data - for example, in multi-target tracking (MTT). In recent years,
massive amounts of behavioral and usage data have become available due to the proliferation of online services
and their large users bases. The data from these applications can not be said to be monolithically generated -
there are many processes and activities occurring simultaneously. However, it also cannot be said that this data
consists of a set of independently running processes, as there are often strong correlations among subsets of the
variables. Therefore we have a potentially large set of loosely coupled entities that can be modeled neither as
a single, large process, or a large set of individual processes. "Static" applications, e.g. rating predictors for
recommender systems, have greatly exploited entity to entity correlations through processes such as collaborative
filtering. In this paper, we present a probabilistic model for loosely coupled and correlated dynamic data sets
and techniques for making inference about the model. Experimental results are presented using data gathered
from instrumented wireless access points around a college campus.