Scalable Projection-Free Optimization Methods via MultiRadial Duality Theory

Published in ArXiv Preprint, 2024

This paper presents a first order method for solving constrained convex optimization problems. The presented method, called the multiradial method, does not rely on projections to the feasible region. Instead, the method only requires normal vector computations and interior reference points of each constraint set defining the feasible region. In addition, the method works for general convex objectives, without the need to assume smoothness or Lipschitz continuity.

Recommended citation: Thabo Samakhoana, Benjamin Grimmer. (2024). "Scalable Projection-Free Optimization Methods via MultiRadial Duality Theory." ArXiv .
Download Paper