Seminar What information theory contributes to modeling and inference of evolving complex systems

Speaker: Amos Golan (American University)

Abstract:

Modeling and inference are central to all areas of sciences. Unfortunately, the data we have for modeling and inference is insufficient and surrounded by deep uncertainty and noise, especially when working on evolving, complex systems, such as social, behavioral, biological, ecological, and economic systems, or with data emerging from such complex systems. In such cases of modeling and inference with insufficient data, there are multiple solutions that satisfy that information (this is called an underdetermined, or partially identified, problem), so which one should we choose? Information theory provides a way to deal with such insufficiency and complexity. It tells us which one of the feasible solutions to choose and provides a way to handle deep uncertainty and accommodate model ambiguities associated with the data. It also provides new insights into basic modeling and a different way of thinking about solving such problems and nesting models in terms of the information they use. It offers a different way of solving inferential problems and developing theories that cannot be solved with conventional methods without imposing additional structure or assumptions, or when improved data cannot be found or collected. It is statistically robust, easy to model and computationally and statistically efficient. Though information-theoretic inference provides us with a general framework for inference (I call it, info-metrics), the exact specification is problem-specific. In this talk I will discuss the basic idea of info-metrics via a number of graphical representations of the theory and will then provide several theoretical and empirical examples. I will also discuss the way some other traditional approaches fit within that framework (such as partial identification and misspecification) and some of the benefits of combining classical and information-theoretic econometric modeling.