1: \begin{abstract}

2: \emph{Noisy particle gradient descent} (NPGD) is an algorithm to minimize convex functions over the space of measures that include an entropy term. In the many-particle limit, this algorithm is described by a \emph{Mean-Field Langevin} dynamics---a generalization of the Langevin dynamic with a non-linear drift---which is our main object of study. Previous work have shown its convergence to the unique minimizer via non-quantitative arguments.

3: %

4: We prove that this dynamics converges at an exponential rate, under the assumption that a certain family of Log-Sobolev inequalities holds. 

5: %

6: This assumption holds for instance for the minimization of the risk of certain two-layer neural networks, where NPGD is equivalent to standard noisy gradient descent.

7: %

8: We also study the annealed dynamics, and show that for a noise decaying at a logarithmic rate, the dynamics converges in value to the global minimizer of the unregularized objective function.

9: %

10: \end{abstract}

11: