Convergence of linesearch-based generalized conditional gradient methods without smoothness assumptions

TL;DR

This paper proves convergence of linesearch-based generalized conditional gradient methods without requiring the smoothness of the objective function's components, broadening their applicability to nonsmooth problems.

Contribution

It introduces a parameter-free variant that adapts to the H"older exponent, ensuring convergence without smoothness assumptions.

Findings

Convergence is guaranteed without Lipschitz continuity of the gradient.

The method automatically adapts to the H"older continuity of the gradient.

Applicable to a wider class of nonsmooth composite problems.

Abstract

The generalized conditional gradient method is a popular algorithm for solving composite problems whose objective function is the sum of a smooth function and a nonsmooth convex function. Many convergence analyses of the algorithm rely on smoothness assumptions, such as the Lipschitz continuity of the gradient of the smooth part. This paper provides convergence results of linesearch-based generalized conditional gradient methods without smoothness assumptions. In particular, we show that a parameter-free variant, which automatically adapts to the H\"older exponent, guarantees convergence even when the gradient of the smooth part of the objective is not H\"older continuous.