Deep Radial-Basis Value Functions for Continuous Control

被引:0
|
作者
Asadi, Kavosh [1 ,2 ]
Parikh, Neev [2 ]
Parr, Ronald E. [3 ]
Konidaris, George D. [2 ]
Littman, Michael L. [2 ]
机构
[1] Amazon Web Serv, Seattle, WA 98109 USA
[2] Brown Univ, Providence, RI 02912 USA
[3] Duke Univ, Durham, NC 27706 USA
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
A core operation in reinforcement learning (RL) is finding an action that is optimal with respect to a learned value function. This operation is often challenging when the learned value function takes continuous actions as input. We introduce deep radial-basis value functions (RBVFs): value functions learned using a deep network with a radial-basis function (RBF) output layer. We show that the maximum action-value with respect to a deep RBVF can be approximated easily and accurately. Moreover, deep RBVFs can represent any true value function owing to their support for universal function approximation. We extend the standard DQN algorithm to continuous control by endowing the agent with a deep RBVF. We show that the resultant agent, called RBF-DQN, significantly outperforms value-function-only baselines, and is competitive with state-of-the-art actor-critic algorithms.
引用
收藏
页码:6696 / 6704
页数:9
相关论文
共 50 条
  • [1] MULTIVARIATE CARDINAL INTERPOLATION WITH RADIAL-BASIS FUNCTIONS
    BUHMANN, MD
    CONSTRUCTIVE APPROXIMATION, 1990, 6 (03) : 225 - 255
  • [2] Estimates of approximation rates by Gaussian radial-basis functions
    Kainen, Paul C.
    Kurkova, Vera
    Sanguineti, Marcello
    ADAPTIVE AND NATURAL COMPUTING ALGORITHMS, PT 2, 2007, 4432 : 11 - +
  • [3] Nonuniform Sampling and Recovery of Multidimensional Bandlimited Functions by Gaussian Radial-Basis Functions
    Bailey, B. A.
    Schlumprecht, T.
    Sivakumar, N.
    JOURNAL OF FOURIER ANALYSIS AND APPLICATIONS, 2011, 17 (03) : 519 - 533
  • [4] Adaptive control of multidimensional nonlinear objects on the basis of radial-basis networks
    Rudenko O.G.
    Bessonov A.A.
    Cybernetics and Systems Analysis, 2005, 41 (2) : 302 - 308
  • [5] Nonuniform Sampling and Recovery of Multidimensional Bandlimited Functions by Gaussian Radial-Basis Functions
    B. A. Bailey
    T. Schlumprecht
    N. Sivakumar
    Journal of Fourier Analysis and Applications, 2011, 17 : 519 - 533
  • [6] Active tension control of the conical winding system based on the neural network control algorithm of radial-basis functions
    Zhang, Hua
    Wang, Jiangtao
    Wu, Jie
    Bian, Huoding
    Wei, Yikun
    TEXTILE RESEARCH JOURNAL, 2023, 93 (15-16) : 3443 - 3458
  • [7] Adaptive Sampling Methodology for Structural Identification Using Radial-Basis Functions
    Proverbio, Marco
    Costa, Alberto
    Smith, Ian F. C.
    JOURNAL OF COMPUTING IN CIVIL ENGINEERING, 2018, 32 (03)
  • [8] Hardware implementation of radial-basis neural networks with Gaussian activation functions on FPGA
    Shymkovych, Volodymyr
    Telenyk, Sergii
    Kravets, Petro
    NEURAL COMPUTING & APPLICATIONS, 2021, 33 (15): : 9467 - 9479
  • [9] On cardinal interpolation by Gaussian radial-basis functions: Properties of fundamental functions and estimates for Lebesgue constants
    S. D. Riemenschneider
    N. Sivakumar
    Journal d’Analyse Mathématique, 1999, 79 : 33 - 61
  • [10] Hardware implementation of radial-basis neural networks with Gaussian activation functions on FPGA
    Volodymyr Shymkovych
    Sergii Telenyk
    Petro Kravets
    Neural Computing and Applications, 2021, 33 : 9467 - 9479