Title: Randomized local model order reduction
Applications that require multiple simulation requests or a real-time simulation response are ubiquitous in science and engineering. However, using standard methods such as finite elements is often prohibitive for such tasks. Model order reduction approaches, in which the equation that models the considered phenomena is (approximately) solved in a carefully chosen subspace of the high-dimensional discretization space, have been developed to tackle such situations.
In this talk we show how to use local reduced models within domain decomposition methods to compute fast approximations for large-scale applications. For the efficient construct of the local reduced models we employ randomized methods used in data science, compressed sensing, and deep learning.