DocumentCode :
239653
Title :
Multi-task nonconvex optimization with total budget constraint: A distributed algorithm using Monte Carlo estimates
Author :
Mengdi Wang ; Yunjian Xu ; Yuntao Gu
Author_Institution :
Dept. of Oper. Res. & Financial Eng., Princeton Univ., Princeton, NJ, USA
fYear :
2014
fDate :
20-23 Aug. 2014
Firstpage :
793
Lastpage :
796
Abstract :
Multi-task optimization is common in machine learning, filtering, communication and network problems. We focus on the nonconvex separable problem where the objective is the sum of N individual utility functions subject to a total budget constraint. By leveraging the Lagrangian dual decomposition, the dual ascent method naturally applies and can be implemented distributively. For stochastic versions of multi-task problems, we propose a simulation-based dual ascent algorithm. According to a classical result from convex geometry, the average-per-task duality gap between the primal and dual problems is bounded by O (1/N). This suggests that the nonconvex multi-task problem is getting “convexified” as the number of tasks increases. As a result, the proposed distributed dual algorithm recovers the optimal solution of the nonconvex problem with very small error.
Keywords :
Monte Carlo methods; concave programming; convergence; distributed algorithms; duality (mathematics); stochastic processes; Lagrangian dual decomposition; Monte Carlo estimates; average-per-task duality gap; convex geometry; distributed dual algorithm; dual ascent method; dual problems; individual utility functions; multitask optimization; nonconvex multitask; nonconvex separable problem; primal problems; stochastic versions; total budget constraint; Approximation algorithms; Approximation methods; Convergence; Digital signal processing; Monte Carlo methods; Optimization; Signal processing algorithms; Monte Carlo; distributed algorithms; dual decomposition; duality gap; multi-task learning; nonconvex optimization; simulation;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Digital Signal Processing (DSP), 2014 19th International Conference on
Conference_Location :
Hong Kong
Type :
conf
DOI :
10.1109/ICDSP.2014.6900773
Filename :
6900773
Link To Document :
بازگشت