Nonlinear dynamical systems for machine learning

Modern machine learning models are brittle: they are accurate in the narrow spaces where much data is already available but break easily in regions where data is not abundant. There are many contributions to the brittleness, but the inability of the underlying systems to distinguish correlation from causation and the presumption of a static world are particularly important. We investigate the potential of a variety of nonlinear dynamical systems with complex behavior (feedback loops, adaptation, chaos) to be adapted as machine learning algorithms capable of making predictions on systems that change dynamically.

People

Jorge A. Muñoz

BS UTEP '07. MS, PhD Caltech '09, '13

Carlos Cuellar Rodriguez

Physics and Mathematics