A General Solution to Least Squares Problems with Box Constraints and Its Applications

The main contribution of this paper is presenting a flexible solution to the box-constrained least squares problems. This solution is applicable to many existing problems, such as nonnegative matrix factorization, support vector machine, signal deconvolution, and computed tomography reconstruction. The key concept of the proposed algorithm is to replace the minimization of the cost function at each iteration by the minimization of a surrogate, leading to a guaranteed decrease in the cost function. In addition to the monotonicity, the proposed algorithm also owns a few good features including the self-constraint in the feasible region and the absence of a predetermined step size. This paper theoretically proves the global convergence for a special case of below-bounded constraints. Using the proposed mechanism, some valuable algorithms can be derived. The simulation results demonstrate that the proposed algorithm provides performance that is comparable to that of other commonly used methods in numerical experiment and computed tomography reconstruction.