COG-PETS: Code Genertion for Parameter Estimation in Time Series


Christopher Kumar Anand
Jacques Carette
Andrew Thomas Curtis
David Miller


Abstract

We extend previous work on symbolic code generation for efficient solvers in the domain of image and signal processing, to take advantage of recurrence relations for efficiently generating model function values corresponding to a regularly-sampled time series. First- and zeroth-order recurrences are symbolically identified by Maple. Two example applications to Magnetic Resonance Spectroscopy and Relaxometry are described, and we show two orders of magintude accelerations for the resulting solvers.