A Class of Convergent Parallel Algorithms for SVMs Training

The training of Support Vector Machines may be a very difficult task when dealing with very large datasets. The memory requirement and the time consumption of the SVMs algorithms grow rapidly with the increase of the data. To overcome these drawbacks a lot of parallel algorithms have been implemented, but they lack of convergence properties. In this work we propose a generic parallel algorithmic scheme for SVMs and we state its asymptotical global convergence under suitable conditions. We outline how these assumptions can be satisfied in practice and we suggest various specific implementations exploiting the adaptable structure of the algorithmic model.