1: \begin{abstract}

2:   This paper presents a detailed theoretical analysis of the three

3:   stochastic approximation proximal gradient algorithms proposed in

4:   our companion paper \cite{vidal:et:al:2019a} to set regularization

5:   parameters by marginal maximum likelihood estimation. We prove the convergence

6:   of a more general stochastic approximation scheme that includes the

7:   three algorithms of \cite{vidal:et:al:2019a} as special

8:   cases. This includes

9:   asymptotic and non-asymptotic convergence results with natural and

10:   easily verifiable conditions, as well as explicit bounds on the

11:   convergence rates. Importantly, the theory is also general in that it can be applied to other intractable

12:   optimisation problems. A main novelty of the work is that the stochastic gradient estimates of our scheme

13:   are constructed from inexact proximal Markov chain Monte Carlo samplers. This allows the use of samplers that

14:   scale efficiently to large problems and for which we have precise

15:   theoretical guarantees.

16: 	\end{abstract}

17: