1: \begin{abstract}

2: We study the kinetic mean field Langevin dynamics

3: under the functional convexity assumption of the mean field energy functional.

4: Using hypocoercivity,

5: we first establish the exponential convergence of the mean field dynamics

6: and then show the corresponding \(N\)-particle system converges

7: exponentially in a rate uniform in \(N\) modulo a small error.

8: Finally we study the short-time regularization effects of the dynamics

9: and prove its uniform-in-time propagation of chaos property

10: in both the Wasserstein and entropic sense.

11: Our results can be applied to the training of

12: two-layer neural networks with momentum

13: and we include the numerical experiments.

14: \end{abstract}

15: