Title: Training dynamical systems to predict their input
Abstract: Evolved systems seem to predict their environment. Even bacteria can implicitly predict future concentrations of scarce sugar or antibiotics, and emerging evidence suggests that even our retinae are able to predict what we see. How? We explore some basic design principles for what causes a system to predict its input, finishing with a call to arms for mathematicians to develop a better framework for understanding input-dependent dynamical systems or recurrent networks.
Prof. Marzen is Assistant Professor of Physics at Claremont’s Keck Science Department.