Numba-accelerated parameter estimation for artificial pancreas applications

General-purpose computing on graphics processing units (GPU) has gained more attention over the last years in scientific computing. Tasks like population-based optimizations are widely used and can be efficiently parallelized. While Python is considered generally “slow”, it is one of the most popular programming languages. In this study, we evaluate the acceleration possibilities of a parameter estimation problem based on optimization utilizing the Numba library on the central processing unit (CPU) and on the GPU. Runtimes indicate that Numba achieves high runtime improvements with little modifications in the syntax. On the other hand, utilization of the GPU is beneficial when high parallelization can be guaranteed and using cloud computing platforms.