@inproceedings{0dac38ab0ca44249a3923f96a13dcd04,
title = "Multi-task nonconvex optimization with total budget constraint: A distributed algorithm using Monte Carlo estimates",
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 = "Distributed algorithms, Dual decomposition, Duality gap, Monte Carlo, Multi-task learning, Nonconvex optimization, Simulation",
author = "Mengdi Wang and Yunjian Xu and Yuntao Gu",
note = "Publisher Copyright: {\textcopyright} 2014 IEEE.; 2014 19th International Conference on Digital Signal Processing, DSP 2014 ; Conference date: 20-08-2014 Through 23-08-2014",
year = "2014",
doi = "10.1109/ICDSP.2014.6900773",
language = "English (US)",
series = "International Conference on Digital Signal Processing, DSP",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "793--796",
booktitle = "2014 19th International Conference on Digital Signal Processing, DSP 2014",
address = "United States",
}