A Statistical Physics Approach to Dynamical Inference
Hosted by the Department of Computer & Information Science
The tools of statistical physics are broadly applicable across the sciences. While initially developed to study heat engines in the 19th century, statistical physics is now used to describe not only equilibrium properties of matter in physics but is also widely used in image reconstruction, statistical inference and machine learning. Here we wish to extend the work of E.T. Jaynes and show how statistical physics (MaxEnt in particular) can go beyond describing equilibria, it can also be used to infer dynamical models. In doing so, we will show what dynamical models naturally arise from the data given this strategy and also discuss why using strategies other than MaxEnt can lead to modeling biases.
Dr. Presse received his B.Sc. from McGill University in Montreal in Chemistry, his
Ph.D. from MIT in Chemical Physics from 2003-2008, and his Postdoc in Biophysics with Ken Dill @ UCSF from 2008-2012.