STATISTICAL PHYSICS ESTIMATES FOR THE COMPLEXITY OF FEEDFORWARD NEURAL NETWORKS

被引：18
|
作者：
OPPER, M
机构：
[1] Institut für Theoretische Physik, Universität Würzburg, D-97074 Würzburg, Am Hubland
来源：
PHYSICAL REVIEW E | 1995年 / 51卷 / 04期
关键词：
D O I：
10.1103/PhysRevE.51.3613
中图分类号：
O35 [流体力学]; O53 [等离子体物理学];
学科分类号：
070204 ; 080103 ; 080704 ;
摘要：
Using simple information theoretic inequalities, a lower bound to the Vapnik-Chervonenkis (VC) complexity of neural networks is investigated. This bound is expressed by the average entropy used in the statistical mechanics approach to the network's generalization problem. Within the annealed theory, exact bounds to the VC dimension or the storage capacity can be calculated explicitly, without using the replica method. For the parity machine, the estimates of capacities match known upper bounds asymptotically, when the number of hidden units grows large. © 1995 The American Physical Society.
引用
收藏
页码：3613 / 3618
页数：6
相关论文
共 50 条
  • [1] CLASSES OF FEEDFORWARD NEURAL NETWORKS AND THEIR CIRCUIT COMPLEXITY
    SHAWETAYLOR, JS
    ANTHONY, MHG
    KERN, W
    NEURAL NETWORKS, 1992, 5 (06) : 971 - 977
  • [2] On the circuit complexity of sigmoid feedforward neural networks
    Beiu, V
    Taylor, JG
    NEURAL NETWORKS, 1996, 9 (07) : 1155 - 1171
  • [3] Statistical physics of neural networks
    Kinzel, W
    COMPUTER PHYSICS COMMUNICATIONS, 1999, 121 : 86 - 93
  • [4] Low Complexity Algorithmic Trading by Feedforward Neural Networks
    J. Levendovszky
    I. Reguly
    A. Olah
    A. Ceffer
    Computational Economics, 2019, 54 : 267 - 279
  • [5] Low Complexity Algorithmic Trading by Feedforward Neural Networks
    Levendovszky, J.
    Reguly, I
    Olah, A.
    Ceffer, A.
    COMPUTATIONAL ECONOMICS, 2019, 54 (01) : 267 - 279
  • [6] NEURAL NETWORKS - A STATISTICAL PHYSICS APPROACH
    NADAL, JP
    JOURNAL DE PHYSIQUE, 1989, 50 (C-3): : 223 - 227
  • [7] Statistical physics of interacting neural networks
    Kinzel, W
    Metzler, R
    Kanter, I
    PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS, 2001, 302 (1-4) : 44 - 55
  • [8] Generalized feedforward control using physics informed neural networks
    Bolderman, M.
    Fan, D.
    Lazar, M.
    Butler, H.
    IFAC PAPERSONLINE, 2022, 55 (16): : 148 - 153
  • [9] Statistical estimation of the number of hidden units for feedforward neural networks
    Fujita, O
    NEURAL NETWORKS, 1998, 11 (05) : 851 - 859
  • [10] Statistical physics, neural networks, brain studies
    Toulouse, Gerard
    Turkish Journal of Physics, 23 (01): : 29 - 38
12345