1: \begin{abstract}

2: In this paper, we provide a subgradient based algorithm to solve general constrained convex optimization without taking projections onto the domain set. The well studied Frank-Wolfe type algorithms also avoid projections. However, they are only designed to handle smooth objective functions. The proposed algorithm treats both smooth and nonsmooth problems and achieves an $O(1/\sqrt{T})$ convergence rate (which matches existing lower bounds). The algorithm yields similar performance in expectation when the deterministic subgradients are replaced by stochastic subgradients. Thus, the proposed algorithm is a projection-free alternative to the Projected subgradient Descent (PGD) and Stochastic projected subgradient Descent (SGD) algorithms.

3: \end{abstract}

4: