COMPSTAT 2016: Start Registration
View Submission - COMPSTAT
Title: Solving large scale penalized $l_{0}$-norm regression problems via parallel proximal algorithms Authors:  Tso-Jung Yen - Academia Sinica (Taiwan) [presenting]
Abstract: An algorithm is developed for solving a regression estimation problem involving a structured $l_{0}$-norm penalty function. This algorithm incorporates the ideas of the proximal gradient method and iterative hard-thresholding. It decomposes the computational task into several sub-tasks that can be carried out separately in parallel. It obtains updates for parameters by using a closed form representation for the proximal operator of the structured $l_{0}$-norm penalty function. It is scalable in terms of sample size or the number of parameters, and is able to be implemented under a data parallelism framework. We demonstrate performance of the algorithm by conducting several simulation studies.