Best Convex Lower Approximations of the l 0 Pseudonorm on Unit Balls

Thomas Bittar (1) , Jean-Philippe Chancelier , Michel de Lara

1 ENPC - École des Ponts ParisTech

Whereas the norm of a vector measures amplitude (and is a 1-homogeneous function), sparsity is measured by the 0-homogeneous l0 pseudonorm, which counts the number of nonzero components. We propose a family of conjugacies suitable for the analysis of 0-homogeneous functions. These conjugacies are derived from couplings between vectors, given by their scalar product divided by a 1-homogeneous normalizing factor. With this, we characterize the best convex lower approximation of a 0-homogeneous function on the unit "ball" of a normalization function (i.e. a norm without the requirement of subadditivity). We do the same with the best convex and 1-homogeneous lower approximation. In particular, we provide expressions for the tightest convex lower approximation of the l0 pseudonorm on any unit ball, and we show that the tightest norm which minorizes the l0 pseudonorm on the unit ball of any lp-norm is the l1-norm. We also provide the tightest convex lower convex approximation of the l0 pseudonorm on the unit ball of any norm.

Mots clés: l0 pseudonorm; convexity; Capra conjugacy; generalized top-k and k-support norms; sparsity inducing norm

Domaines: Optimisation et contrôle [math.OC]

Thomas Bittar, Jean-Philippe Chancelier, Michel de Lara. Best Convex Lower Approximations of the l 0 Pseudonorm on Unit Balls. 2021. ⟨hal-03240878⟩