While it has been mathematically proven that Projective Splitting (PS) algorithms can converge in parallel and distributed computing settings, to-date, it appears there were no open-source implementations of the full algorithm with asynchronous computing capabilities. This note fills this gap by providing a Julia implementation of the asynchronous PS algorithm of Eckstein and Combettes for solving fully nonsmooth convex optimization problems. Our methodology includes interoperability with existing packages within the Julia ecosystem, and we also document observations from running this algorithm asynchronously on problems in image processing and machine learning.
Article
Loading...