Stochastic proximal-gradient algorithms for penalized mixed models

Motivated by penalized likelihood maximization in complex models, we study optimization problems where neither the function to optimize nor its gradient has an explicit expression, but its gradient can be approximated by a Monte Carlo technique. We propose a new algorithm based on a stochastic approximation of the proximal-gradient (PG) algorithm. This new algorithm, named stochastic approximation PG (SAPG) is the combination of a stochastic gradient descent step which—roughly speaking—computes a smoothed approximation of the gradient along the iterations, and a proximal step. The choice of the step size and of the Monte Carlo batch size for the stochastic gradient descent step in SAPG is discussed. Our convergence results cover the cases of biased and unbiased Monte Carlo approximations. While the convergence analysis of some classical Monte Carlo approximation of the gradient is already addressed in the literature (see Atchadé et al. in J Mach Learn Res 18(10):1–33, 2017), the convergence analysis of SAPG is new. Practical implementation is discussed, and guidelines to tune the algorithm are given. The two algorithms are compared on a linear mixed effect model as a toy example. A more challenging application is proposed on nonlinear mixed effect models in high dimension with a pharmacokinetic data set including genomic covariates. To our best knowledge, our work provides the first convergence result of a numerical method designed to solve penalized maximum likelihood in a nonlinear mixed effect model.