Title:Revisiting Invex Functions: Explicit Kernel Constructions and Applications

Abstract:An invex function generalizes a convex function in the sense that every stationary point is a global minimizer. Recently, invex functions and related concepts have attracted attention in signal processing and machine learning. However, proving that a function is invex is not straightforward, because the definition involves an unknown function called a kernel function. This paper develops several methods for constructing explicit kernel functions, which have been missing from the literature. These methods support proving invexity of new functions, and they would also be useful in the development of optimization algorithms for invex problems. We also clarify connections to pseudoconvex functions and present examples of nonsmooth, non-pseudoconvex invex functions that arise in signal processing.