Convergence of forward-backward splitting algorithms with errors

Forward-backward splitting algorithms are among the most popular optimization methods to solve a wide class of signal processing and machine learning problems, often abstracted into a convex composite optimization problem, where the objective function is the sum of a smooth and a nonsmooth component. In this talk I will present convergence results for this class of algorithms, focusing on the influence of computational errors. I will discuss in particular the case of stochastic errors.