Conditional gradient algorithms for norm-regularized smooth convex optimization (Q494314)

Let \(K\subset E\) be a closed convex cone in the Euclidean space \(E\), let \(f: K\to\mathbb{R}\) be a smooth convex function and let \(\| \cdot \|\) be an arbitrary norm on \(E\). The authors discuss the norm minimization problem \[ \text{Minimize }\| x\|\quad\text{s.t.}\quad x\in K,\;f(x)\leq, 0 \] which can be rewritten by \[ \text{Minimize }\rho\geq 0\quad\text{s.t.}\quad \text{Opt}(\rho):= \min\{f(x):\| x\|\leq\rho,\;x\in K\}\leq 0 \] and the penalized norm minimization problem \[ \text{Minimize }f(x)+ \kappa\| x\|, \quad\text{s.t.} \quad x\in K. \] Both optimization problems are of a large interest for the theory of signal processing and for machine learning. The authors provide special conditional gradient algorithms for these problems to find \(\varepsilon\)-solutions and include numerical results. Some applications are presented regarding large-scale optimization problems in the space of \((p\times q)\) matrices.