NeurIPS 2020

Walking in the Shadow: A New Perspective on Descent Directions for Constrained Minimization

Meta Review

The paper analyzes a different descent directions for the Frank-Wolfe algorithm which relies on introducing the best local feasible direction of descent. The theoretical part is solid and is viewed as interesting and novel by the reviewers. There are some (not major) concerns regarding the organization of the papers and also additional experimental results that could potentially better highlight the empirical benefits of the newly introduced shadow step. I encourage the authors to revise the paper accordingly and recommend acceptance.