1: \begin{abstract}

2: A class of generalized conditional gradient algorithms for the solution of optimization

3: problem in spaces of Radon measures is presented.

4: The method iteratively inserts additional Dirac-delta functions and optimizes

5: the corresponding coefficients. Under general assumptions, a sub-linear

6: $\Ob(1/k)$ rate in the objective functional is obtained, which is sharp in

7: most cases. To improve efficiency, one can fully resolve the

8: finite-dimensional subproblems occurring in each iteration of the method. We

9: provide an analysis for the resulting procedure: under a structural

10: assumption on the optimal solution, a linear $\Ob(\zeta^k)$ convergence

11: rate is obtained locally.

12: \end{abstract}

13: