University of Washington
Many algorithms have been developed to solve large-scale box constrained optimization problems over the past 15 years. The primary goal of these methods is to devise techniques that reduce the per iteration computational cost while preserving overall efficiency. L-BFGS-B and TRON are two popular algorithms for this class of problems that attempt to reduce the per iteration costs in different ways. L-BFGS-B developed by Nocedal et al. uses an approximate Hessian that induces a reduction in dimension, and TRON developed by More et al. employs active constraint techniques again with the goal of dimension reduction. In this talk we provide a brief introduction to these methods.
University of Washington