Forward stagewise and Frank Wolfe are popular gradient based projection free optimization algorithms which both require convex constraints. We propose a method to extend the applicability of these algorithms to problems of the form (Formula Presented) where f(x) is an invex (Invexity is a generalization of convexity and ensures that all local optima are also global optima.) objective function and g(x) is a non-convex constraint. We provide a theorem which defines a class of monotone component-wise transformation functions (Formula Presented). These transformations lead to a convex constraint function (Formula Presented). Assuming invexity of the original function f(x) that same transformation (Formula Presented) will lead to a transformed objective function (Formula Presented) which is also invex. For algorithms that rely on a non-zero gradient (Formula Presented) to produce new update steps invexity ensures that these algorithms will move forward as long as a descent direction exists.
CITATION STYLE
Keller, S. M., Murezzan, D., & Roth, V. (2019). Invexity Preserving Transformations for Projection Free Optimization with Sparsity Inducing Non-convex Constraints. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11269 LNCS, pp. 682–697). Springer Verlag. https://doi.org/10.1007/978-3-030-12939-2_47
Mendeley helps you to discover research relevant for your work.