1: \begin{abstract}

2: We present a framework for asynchronously solving convex optimization problems over networks of agents

3: which are augmented by the presence of a centralized cloud computer. 

4: This framework uses a Tikhonov-regularized primal-dual approach

5: in which the agents update the system's primal variables and the cloud updates

6: its dual variables. 

7: To minimize coordination requirements placed upon the system,

8: the times of

9: communications and computations among the agents are allowed to be arbitrary, provided

10: they satisfy mild conditions. Communications

11: from the agents to the cloud are likewise carried out without any coordination in their timing. 

12: However, we require that the cloud keep the dual variable's value synchronized across

13: the agents, and a counterexample is provided that demonstrates that 

14: this level of synchrony is indeed necessary for convergence. Convergence rate estimates are provided

15: in both the primal and dual spaces, and simulation results are presented that demonstrate

16: the operation and convergence of the proposed algorithm. 

17: \end{abstract}

18: