Universal Approximation of Input-Output Maps by Temporal Convolutional Nets

Abstract

There has been a recent shift in sequence-to-sequence modeling from recurrent network architectures to convolutional network architectures due to computational advantages in training and operation while still achieving competitive performance. For systems having limited long-term temporal dependencies, the approximation capability of recurrent networks is essentially equivalent to that of temporal convolutional nets (TCNs). We prove that TCNs can approximate a large class of input-output maps having approximately finite memory to arbitrary error tolerance. Furthermore, we derive quantitative approximation rates for deep ReLU TCNs in terms of the width and depth of the network and modulus of continuity of the original input-output map, and apply these results to input-output maps of systems that admit finite-dimensional state-space realizations (i.e., recurrent models).

arXiv commentfinal version to appear in NeurIPS 2019

References in corpus (8)

• An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling
• WaveNet: A Generative Model for Raw Audio
• Convolutional Sequence to Sequence Learning
• Comparative Study of CNN and RNN for Natural Language Processing
• Neural Machine Translation in Linear Time
• Why Deep Neural Networks for Function Approximation?
• Approximating Continuous Functions by ReLU Nets of Minimal Width
• Stable Recurrent Models