A decomposition method for large-scale box constrained optimization

A decomposition method for solving large-scale box constrained optimization is proposed. The algorithm is motivated by the successful use of the decomposition method presented by Joachims for training support vector machines. In particular, a new technique, based on the new definition “KKT-violating index”, is introduced for working set identification. Finally, the numerical experiments and implementation details show that this method is practical for large-scale problems.