1: \begin{abstract}: We consider in this paper the problem of sampling a

2:   high-dimensional probability distribution $\pi$ having a density

3:   \wrt\ the Lebesgue measure on $\rset^d$, known up to a normalization

4:   factor $x \mapsto \pi(x)= \rme^{-U(x)}/\int_{\rset^d} \rme^{-U(y)} \rmd y$.

5:   Such problem naturally occurs for example in Bayesian inference and

6:   machine learning.  Under the assumption that $U$ is continuously

7:   differentiable, $\nabla U$ is globally Lipschitz and $U$ is strongly

8:   convex, we obtain non-asymptotic bounds for the convergence to

9:   stationarity in Wasserstein distance of order $2$ and total

10:   variation distance of the sampling method based on the Euler

11:   discretization of the Langevin stochastic differential equation, for

12:   both constant and decreasing step sizes. The dependence on the

13:   dimension of the state space of the obtained bounds is studied to

14:   demonstrate the applicability of this method.  The convergence of an

15:   appropriately weighted empirical measure is also investigated and

16:   bounds for the mean square error and exponential deviation

17:   inequality are reported for functions which are measurable and bounded.

18:    An illustration to Bayesian inference for binary regression is presented.

19: \end{abstract}

20: