Keywords: Functional Regression, Convex Optimization, ADMM, High-Dimensional Regression
This work presents a functional augmented alternating directional method of multipliers, ADMM, for fitting function-on-scalar regression models with a large number of scalar predictors. Applications in genetics and bioinformatics have motivated recent work on statistical inference for such models, however, development of the computational tools remains relatively underdeveloped. We establish the convergence of our algorithm over a broad class of problems by assuming the outcomes lie in a real separable Hilbert space. We demonstrate the computational advantages of our approach in a numerical study, and the methods are available to users as an R package, with a backend written in C++, or in Matlab.