EXPLORING BEHAVIORAL PATTERNS IN COMPLEX ADAPTIVE SYSTEMS
Andrii Cherniak · D-Scholarship@Pitt (University of Pittsburgh) · 2014
Many phenomenons in real world can be characterized as complex adaptive systems (CAS). We are surrounded with a huge number of communicating and interacting agents. Some of those agents may be capable of learning and adapting to new situation, trying to achieve their goals. E-commerce, social media, cloud computing, transportation network and real-time ride sharing, supply chain are a few examples of CAS. These are the systems which surround us in every day’s life, and naturally we want to make sense of those systems and optimize systems’ behavior or optimize our behavior around those systems. Given the complexity of these systems, we want to find a set of simplified patterns out of the seeming chaos of interactions in a CAS, and provide more manageable means of analysis for such systems. In my thesis I consider a few example problems from different domains: modeling human behavior during fire evacuation, detection of notable transitions in data streams, modeling finite resource sharing on a computational cluster with many clients, and predicting buyer behavior on the marketplace. These (and other) seemingly different problems demonstrate one important similarity: complex semi-repetitive or semi-similar behavior. This semi-repetitive behavior poses a challenge to model such processes. This challenge comes for two major reasons: 1 ) state-space explosion and sparsity of data 2 ) critical transitions and precision of process modeling I show, that the analysis of smilingly different CAS coming from different domains, can be performed by following the same recipe.