Executive Summary : | Alternating Direction Method of Multiplier (ADMM) has been successfully applied in many conventional image processing and machine learning problems. Due to the various limitation of Stochastic Gradient Descent and its variant, ADMM is used in machine learning as well as in the image processing problems. In ADMM, the model problem is divided into many sub-problems, and after solving individually, all the solutions are coordinated to solve the original problem. The advantages of ADMM are numerous and versatile: as data are processed parallelly in different cores, it saves time; it does not require checking the gradient vanishing criteria; it also handles the problems having poor conditioning quite well. |