On the approximation of rough functions with deep neural networks

被引：0
|
作者：
De Ryck T. [1 ]
Mishra S. [1 ]
Ray D. [2 ]
机构：
[1] Seminar for Applied Mathematics, ETH Zürich, Rämistrasse 101, Zurich
[2] University of Southern California, Los Angeles
来源：
SeMA Journal | 2022年 / 79卷 / 3期
基金：
欧盟地平线“2020”;
关键词：
Data compression; Deep ReLU networks; ENO; Interpolation; Rough functions;
D O I：
10.1007/s40324-022-00299-w
中图分类号：
学科分类号：
摘要：
The essentially non-oscillatory (ENO) procedure and its variant, the ENO-SR procedure, are very efficient algorithms for interpolating (reconstructing) rough functions. We prove that the ENO (and ENO-SR) procedure are equivalent to deep ReLU neural networks. This demonstrates the ability of deep ReLU neural networks to approximate rough functions to high-order of accuracy. Numerical tests for the resulting trained neural networks show excellent performance for interpolating functions, approximating solutions of nonlinear conservation laws and at data compression. © 2022, The Author(s).
引用
收藏
页码：399 / 440
页数：41
相关论文
共 50 条
  • [41] Optimal deep neural networks by maximization of the approximation power
    Calvo-Pardo, Hector
    Mancini, Tullio
    Olmo, Jose
    COMPUTERS & OPERATIONS RESEARCH, 2023, 156
  • [42] Deep Neural Networks with Multistate Activation Functions
    Cai, Chenghao
    Xu, Yanyan
    Ke, Dengfeng
    Su, Kaile
    COMPUTATIONAL INTELLIGENCE AND NEUROSCIENCE, 2015, 2015
  • [43] Activation Functions and Their Characteristics in Deep Neural Networks
    Ding, Bin
    Qian, Huimin
    Zhou, Jun
    PROCEEDINGS OF THE 30TH CHINESE CONTROL AND DECISION CONFERENCE (2018 CCDC), 2018, : 1836 - 1841
  • [44] Deep Convolutional Neural Networks on Cartoon Functions
    Grohs, Philipp
    Wiatowski, Thomas
    Bolcskei, Helmut
    2016 IEEE INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY, 2016, : 1163 - 1167
  • [45] Towards a Visualization of Deep Neural Networks For Rough Line Images
    Chaudhary, Narendra
    Savari, Serap A.
    35TH EUROPEAN MASK AND LITHOGRAPHY CONFERENCE (EMLC 2019), 2019, 11177
  • [46] Stability of Deep Neural Networks via Discrete Rough Paths
    Bayer, Christian
    Friz, Peter K.
    Tapia, Nikolas
    SIAM JOURNAL ON MATHEMATICS OF DATA SCIENCE, 2023, 5 (01): : 50 - 76
  • [47] Exponential convergence of the deep neural network approximation for analytic functions
    E, Weinan
    Wang, Qingcan
    SCIENCE CHINA-MATHEMATICS, 2018, 61 (10) : 1733 - 1740
  • [48] Exponential convergence of the deep neural network approximation for analytic functions
    Weinan E
    Qingcan Wang
    Science China Mathematics, 2018, 61 : 1733 - 1740
  • [49] On the near optimality of the stochastic approximation of smooth functions by neural networks
    Maiorov, VE
    Meir, R
    ADVANCES IN COMPUTATIONAL MATHEMATICS, 2000, 13 (01) : 79 - 103
  • [50] On Approximation by Neural Networks with Optimized Activation Functions and Fixed Weights
    Yu, Dansheng
    Qian, Yunyou
    Li, Fengjun
    ANALYSIS IN THEORY AND APPLICATIONS, 2023, 39 (01): : 93 - 104
12345