Implementation of a modified Nesterov's Accelerated quasi-Newton Method on Tensorflow

被引:8
|
作者
Indrapriyadarsini, S. [1 ]
Mahboubi, Shahrzad [2 ]
Ninomiya, Hiroshi [2 ]
Asai, Hideki [3 ]
机构
[1] Shizuoka Univ, Grad Sch Integrated Sci & Technol, Hamamatsu, Shizuoka, Japan
[2] Shonan Inst Technol, Grad Sch Elect & Informat Engn, Fujisawa, Kanagawa, Japan
[3] Shizuoka Univ, Res Inst Elect, Hamamatsu, Shizuoka, Japan
关键词
Neural networks; training algorithm; quasi-Newton method; Nesterov's accelerated gradient; Global convergence; Tensorflow; highly-nonlinear function modeling;
D O I
10.1109/ICMLA.2018.00185
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recent studies incorporate Nesterov's accelerated gradient method for the acceleration of gradient based training. The Nesterov's Accelerated Quasi-Newton (NAQ) method has shown to drastically improve the convergence speed compared to the conventional quasi-Newton method. This paper implements NAQ for non-convex optimization on Tensorflow. Two modifications have been proposed to the original NAQ algorithm to ensure global convergence and eliminate linesearch. The performance of the proposed algorithm - mNAQ is evaluated on standard non-convex function approximation benchmark problems and microwave circuit modelling problems. The results show that the improved algorithm converges better and faster compared to first order optimizers such as AdaGrad, RMSProp, Adam, and the second order methods such as the quasi-Newton method.
引用
收藏
页码:1147 / 1154
页数:8
相关论文
共 50 条
  • [41] A quasi-Newton trust-region method
    E. Michael Gertz
    [J]. Mathematical Programming, 2004, 100 : 447 - 470
  • [42] A Switching Algorithm Based on Modified Quasi-Newton Equation
    Yueting Yang and Chengxian Xu Faculty of Science
    [J]. Numerical Mathematics(Theory,Methods and Applications), 2006, (03) : 257 - 267
  • [43] Maximization of Efficiency of IPMSM by Quasi-Newton Method
    Baek, Sung-min
    Kim, Gyu-tak
    Shin, Heung-kyo
    [J]. 2018 21ST INTERNATIONAL CONFERENCE ON ELECTRICAL MACHINES AND SYSTEMS (ICEMS), 2018, : 472 - 475
  • [44] A quasi-Newton type method for equilibrium problems
    Sousa, Leonardo A.
    Scheimberg, Susana
    Santos, Pedro Jorge S.
    Santos, Paulo Sergio M.
    [J]. NUMERICAL ALGORITHMS, 2022, 89 (03) : 1129 - 1143
  • [45] Modified quasi-Newton methods for training neural networks
    Robitaille, B
    Marcos, B
    Veillette, M
    Payre, G
    [J]. COMPUTERS & CHEMICAL ENGINEERING, 1996, 20 (09) : 1133 - 1140
  • [46] A Modified Orthant-Wise Limited Memory Quasi-Newton Method with Convergence Analysis
    Gong, Pinghua
    Ye, Jieping
    [J]. INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 37, 2015, 37 : 276 - 284
  • [47] AN IMPLEMENTATION OF QUASI-NEWTON METHODS FOR SOLVING SETS OF NONLINEAR EQUATIONS
    PALOSCHI, JR
    PERKINS, JD
    [J]. COMPUTERS & CHEMICAL ENGINEERING, 1988, 12 (08) : 767 - 776
  • [48] Implementation of quasi-Newton algorithm on FPGA for IoT endpoint devices
    Huang, Shizhen
    Guo, Anhua
    Su, Kaikai
    Chen, Siyu
    Chen, Ruiqi
    [J]. International Journal of Security and Networks, 2022, 17 (02) : 124 - 134
  • [49] Optimal control with accelerated convergence: Combining the Krotov and quasi-Newton methods
    Eitan, Reuven
    Mundt, Michael
    Tannor, David J.
    [J]. PHYSICAL REVIEW A, 2011, 83 (05):
  • [50] A Hybrid of Quasi-Newton Method with CG Method for Unconstrained Optimization
    Aini, N.
    Mamat, M.
    Rivaie, M.
    Sulaiman, I. M.
    [J]. 2ND INTERNATIONAL CONFERENCE ON APPLIED & INDUSTRIAL MATHEMATICS AND STATISTICS, 2019, 1366