1: \begin{abstract}

2: %Trace-norm constrained optimization is in the heart of large-scale machine learning, with applications including matrix completion, matrix sensing, multi-label multi-class classification, affine rank minimization, phase retrieval problems, and neural network with quadratic activation. 

3: Large-scale machine learning training suffers from two prior challenges, specifically for nuclear-norm constrained problems with distributed systems: the synchronization slowdown due to the straggling workers, and high communication costs.

4: %Solving trace-norm constrained problem with distributed systems involves two major challenges: synchronization slowdown due to the straggling workers and high communication cost. 

5: In this work, we propose an asynchronous Stochastic Frank Wolfe (SFW-asyn) method, which, for the first time, solves the two problems simultaneously, while successfully maintaining the same convergence rate as the vanilla SFW. 

6: We implement our algorithm in python (with MPI) to run on Amazon EC2, 

7: and demonstrate that SFW-asyn yields speed-ups almost linear to the number of machines compared to the vanilla SFW. %Our distributed framework could also be generalized to other optimization enhancements like variance reduction; we specifically extend it to asynchronous Stochastic Variance Reduced Frank-Wolfe with theoretical convergence analysis.

8: 

9: \end{abstract}

10: