Nonsmooth Projection-Free Optimization with Functional Constraints
This addresses optimization problems with nonsmooth objectives and functional constraints, offering a projection-free method that matches lower bounds, though it is incremental relative to existing smooth projection-free methods.
The paper tackles constrained nonsmooth convex optimization by proposing a subgradient-based algorithm that avoids projections onto the feasible set, achieving an ε-suboptimal solution in O(ε^{-2}) iterations with a single Linear Minimization Oracle call per iteration.
This paper presents a subgradient-based algorithm for constrained nonsmooth convex optimization that does not require projections onto the feasible set. While the well-established Frank-Wolfe algorithm and its variants already avoid projections, they are primarily designed for smooth objective functions. In contrast, our proposed algorithm can handle nonsmooth problems with general convex functional inequality constraints. It achieves an $ε$-suboptimal solution in $\mathcal{O}(ε^{-2})$ iterations, with each iteration requiring only a single (potentially inexact) Linear Minimization Oracle (LMO) call and a (possibly inexact) subgradient computation. This performance is consistent with existing lower bounds. Similar performance is observed when deterministic subgradients are replaced with stochastic subgradients. In the special case where there are no functional inequality constraints, our algorithm competes favorably with a recent nonsmooth projection-free method designed for constraint-free problems. Our approach utilizes a simple separation scheme in conjunction with a new Lagrange multiplier update rule.