Review on Learning to Optimize

Abstract

Algorithm design was a laborious process and requires much iteration of ideation and validation. The paper, explore automating algorithm design, present a method to learn and optimization algorithm, which we `believe to be the first method that can automatically discover a better algorithm. The approach for problem from a reinforcement learning perspective and represent any particular optimization algorithm as a policy. We thus learn and optimization algorithm using guided policy search, demonstrate that the resulting algorithm outperforms existing hand-engineered algorithms in terms of convergence speed or the final objective value.

Keywords: Algorithm design, hyper-parameter, optimization algorithm