1: \begin{abstract}

2: We analyze the adaptive first order algorithm AMSGrad, for

3: solving a constrained stochastic optimization problem with a weakly

4: convex objective. We prove the $\mathcal{\tilde O}(t^{-1/4})$ rate of convergence for

5: the norm of the gradient of Moreau envelope, which is the standard

6: stationarity measure for this class of problems.  It matches the known

7: rates that adaptive algorithms enjoy for the specific case of

8: unconstrained smooth stochastic optimization.  Our analysis

9: works with mini-batch size of $1$, constant first and second order

10: moment parameters, and possibly unbounded optimization domains.

11: Finally, we illustrate the applications and extensions of our results to specific

12: problems and algorithms.

13: \end{abstract}

14: