Probing for “slow” variables using diffusion maps

Benjamin Sonday, Princeton University


Many problems in science and engineering behave spatially or temporally as if there were multiple scales. Using ideas from machine learning, we probe for these scales by looking at the spectrum of diffusion operators on the data sets generated by these scientific problems. The resulting low-dimensional description allows us to do things like projective coarse integration, free energy calculations, bifurcation analysis, etc.

Abstract Author(s): Benjamin Sonday, Yannis Kevrekidis