Experimental Design Networks: A Paradigm for Serving Heterogeneous Learners Under Networking Constraints

被引:0
作者
Li, Yuanyuan [1 ]
Liu, Yuezhou [1 ]
Su, Lili [1 ]
Yeh, Edmund [1 ]
Ioannidis, Stratis [1 ]
机构
[1] Northeastern Univ, Dept Elect & Comp Engn, Boston, MA 02115 USA
基金
美国国家科学基金会;
关键词
Linear regression; Task analysis; Optimization; Bayes methods; Data models; Training; Network topology; Experimental design; DR-submodularity; Bayesian linear regression; PLACEMENT;
D O I
10.1109/TNET.2023.3243534
中图分类号
TP3 [计算技术、计算机技术];
学科分类号
0812 ;
摘要
Significant advances in edge computing capabilities enable learning to occur at geographically diverse locations. In general, the training data needed in those learning tasks are not only heterogeneous but also not fully generated locally. In this paper, we propose an experimental design network paradigm, wherein learner nodes train possibly different Bayesian linear regression models via consuming data streams generated by data source nodes over a network. We formulate this problem as a social welfare optimization problem in which the global objective is defined as the sum of experimental design objectives of individual learners, and the decision variables are the data transmission strategies subject to network constraints. We first show that, assuming Poisson data streams in steady state, the global objective is a continuous DR-submodular function. We then propose a Frank-Wolfe type algorithm that outputs a solution within a 1-1/e factor from the optimal. Our algorithm contains a novel gradient estimation component which is carefully designed based on Poisson tail bounds and sampling. Finally, we complement our theoretical findings through extensive experiments. Our numerical evaluation shows that the proposed algorithm outperforms several baseline algorithms both in maximizing the global objective and in the quality of the trained models.
引用
收藏
页码:2236 / 2250
页数:15
相关论文
共 52 条
  • [1] Mobile Edge Computing: A Survey
    Abbas, Nasir
    Zhang, Yan
    Taherkordi, Amir
    Skeie, Tor
    [J]. IEEE INTERNET OF THINGS JOURNAL, 2018, 5 (01): : 450 - 465
  • [2] Pipage rounding: A new method of constructing algorithms with proven performance guarantee
    Ageev, AA
    Sviridenko, MI
    [J]. JOURNAL OF COMBINATORIAL OPTIMIZATION, 2004, 8 (03) : 307 - 328
  • [3] Smart Cities: Definitions, Dimensions, Performance, and Initiatives
    Albino, Vito
    Berardi, Umberto
    Dangelico, Rosa Maria
    [J]. JOURNAL OF URBAN TECHNOLOGY, 2015, 22 (01) : 3 - 21
  • [4] Alon N., 2016, WILEY SERIES DISCRET
  • [5] [Anonymous], 2011, Tech. Rep.
  • [6] Bian AA, 2017, PR MACH LEARN RES, V54, P111
  • [7] Boyd S., 2004, CONVEX OPTIMIZATION, DOI [DOI 10.1017/CBO9780511804441, 10.1017/CBO9780511804441]
  • [8] MAXIMIZING A MONOTONE SUBMODULAR FUNCTION SUBJECT TO A MATROID CONSTRAINT
    Calinescu, Gruia
    Chekuri, Chandra
    Pal, Martin
    Vondrak, Jan
    [J]. SIAM JOURNAL ON COMPUTING, 2011, 40 (06) : 1740 - 1766
  • [9] Canonne C.L., 2016, A short note on Poisson tail bounds
  • [10] Dependent Randomized Rounding via Exchange Properties of Combinatorial Structures
    Chekuri, Chandra
    Vondrak, Jan
    Zenklusen, Rico
    [J]. 2010 IEEE 51ST ANNUAL SYMPOSIUM ON FOUNDATIONS OF COMPUTER SCIENCE, 2010, : 575 - 584