Complex systems are usually defined by emergence: interactions of the elements on a fine scale lead to global patterns of behavior. Between the finest and the coarsest levels of detail, how can we learn about the intermediate-scale mechanisms that drive emergence? In this talk I consider the process of learning from the perspective of statistical mechanics, or the study of a large number of low-probability events, for three example systems. For ship design problems in Naval Engineering, I show that statistical mechanics doesn't just solve an optimization problem, but reveals the solution entropy of the problem space and discovers novel phenomena such as phase transitions and emergent localization of design elements. For human learning of semantic networks from textbooks, I predict the shape of mental models formed by the subjects under the dual pressures of finite effort and mental errors. Most recently, I use data-first approaches to extract the effective landscape for sensor placement and to identify nonlinear dynamical equations governing the data. Throughout these examples, statistical mechanics highlights the continuous nature of emergence across scales.

Dr. Andrei A. Klishin is a Postdoctoral Scholar at the AI Institute in Dynamic Systems at University of Washington. His undergraduate education in physics was between Belarusian State University and Massachusetts Institute of Technology. He got his Physics PhD from University of Michigan, where he worked on statistical mechanics of naval ship design problems. He then was a postdoc at University of Pennsylvania working on human learning in networks and gender bias in citations of research papers.