A Simple Self-Tuning Technique for Adaptive Proximal Forward-Backward Splitting

摘要

We propose a self-tuning technique for the regularization parameter of the Adaptive proximal forward-backward splitting (APFBS) algorithm. The proposed technique is motivated by TREX, a self-tuning technique for the regularization parameter of LASSO. The self-tuning is implemented at each iteration of APFBS, with an appropriate choice given in TREX. The effectiveness of the proposed technique is shown by numerical experiments.