Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes

In this paper, we introduce the discretized-Vapnik-Chervonenkis (VC) dimension for studying the complexity of a real function class, and then analyze properties of real function classes and neural networks. We first prove that a countable traversal set is enough to achieve the VC dimension for a rea...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhang, Chao, Bian, Wei, Tao, Dacheng, Lin, Weisi
Other Authors: School of Computer Engineering
Format: Article
Language:English
Published: 2013
Subjects:
Online Access:https://hdl.handle.net/10356/99545
http://hdl.handle.net/10220/13524
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-99545
record_format dspace
spelling sg-ntu-dr.10356-995452020-05-28T07:17:32Z Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes Zhang, Chao Bian, Wei Tao, Dacheng Lin, Weisi School of Computer Engineering DRNTU::Engineering::Computer science and engineering In this paper, we introduce the discretized-Vapnik-Chervonenkis (VC) dimension for studying the complexity of a real function class, and then analyze properties of real function classes and neural networks. We first prove that a countable traversal set is enough to achieve the VC dimension for a real function class, whereas its classical definition states that the traversal set is the output range of the function class. Based on this result, we propose the discretized-VC dimension defined by using a countable traversal set consisting of rational numbers in the range of a real function class. By using the discretized-VC dimension, we show that if a real function class has a finite VC dimension, only a finite traversal set is needed to achieve the VC dimension. We then point out that the real function classes, which have the infinite VC dimension, can be grouped into two categories: TYPE-A and TYPE-B. Subsequently, based on the obtained results, we discuss the relationship between the VC dimension of an indicator-output network and that of the real-output network, when both networks have the same structure except for the output activation functions. Finally, we present the risk bound based on the discretized-VC dimension for a real function class that has infinite VC dimension and is of TYPE-A. We prove that, with such a function class, the empirical risk minimization (ERM) principle for the function class is still consistent with overwhelming probability. This is a development of the existing knowledge that the ERM learning is consistent if and only if the function class has a finite VC dimension. 2013-09-19T01:25:57Z 2019-12-06T20:08:34Z 2013-09-19T01:25:57Z 2019-12-06T20:08:34Z 2012 2012 Journal Article Zhang, C., Bian, W., Tao, D., & Lin, W. (2012). Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes. IEEE transactions on neural networks and learning systems, 23(9), 1461-1472. 2162-237X https://hdl.handle.net/10356/99545 http://hdl.handle.net/10220/13524 10.1109/TNNLS.2012.2204773 en IEEE transactions on neural networks and learning systems © 2012 IEEE
institution Nanyang Technological University
building NTU Library
country Singapore
collection DR-NTU
language English
topic DRNTU::Engineering::Computer science and engineering
spellingShingle DRNTU::Engineering::Computer science and engineering
Zhang, Chao
Bian, Wei
Tao, Dacheng
Lin, Weisi
Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes
description In this paper, we introduce the discretized-Vapnik-Chervonenkis (VC) dimension for studying the complexity of a real function class, and then analyze properties of real function classes and neural networks. We first prove that a countable traversal set is enough to achieve the VC dimension for a real function class, whereas its classical definition states that the traversal set is the output range of the function class. Based on this result, we propose the discretized-VC dimension defined by using a countable traversal set consisting of rational numbers in the range of a real function class. By using the discretized-VC dimension, we show that if a real function class has a finite VC dimension, only a finite traversal set is needed to achieve the VC dimension. We then point out that the real function classes, which have the infinite VC dimension, can be grouped into two categories: TYPE-A and TYPE-B. Subsequently, based on the obtained results, we discuss the relationship between the VC dimension of an indicator-output network and that of the real-output network, when both networks have the same structure except for the output activation functions. Finally, we present the risk bound based on the discretized-VC dimension for a real function class that has infinite VC dimension and is of TYPE-A. We prove that, with such a function class, the empirical risk minimization (ERM) principle for the function class is still consistent with overwhelming probability. This is a development of the existing knowledge that the ERM learning is consistent if and only if the function class has a finite VC dimension.
author2 School of Computer Engineering
author_facet School of Computer Engineering
Zhang, Chao
Bian, Wei
Tao, Dacheng
Lin, Weisi
format Article
author Zhang, Chao
Bian, Wei
Tao, Dacheng
Lin, Weisi
author_sort Zhang, Chao
title Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes
title_short Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes
title_full Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes
title_fullStr Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes
title_full_unstemmed Discretized-Vapnik-Chervonenkis dimension for analyzing complexity of real function classes
title_sort discretized-vapnik-chervonenkis dimension for analyzing complexity of real function classes
publishDate 2013
url https://hdl.handle.net/10356/99545
http://hdl.handle.net/10220/13524
_version_ 1681056071336067072