Symmetries and Discriminability in Feedforward Network Architectures
Symmetries and Discriminability in Feedforward Network Architectures
This paper investigates the effects of introducing symmetries into feedforward neural networks in what are termed symmetry networks. This technique allows more efficient training for problems in which we require the output of a network to be invariant under a set of tranformations of the input. The particular problem of graph recognition is considered. In this case the network is designed to deliver the same output for isomorphic graphs. This leads to the question fo which inputs can be distinguished by such artitectures. A theorem characterizing when two inputs can be distinguished by a symmetry network is given. As a consequence, a particular network design is shown to be able to distinguish nonisomorphic graphs if and only if the graph reconstruction conjecture holds.
816-826
Shawe-Taylor, J.
c32d0ee4-b422-491f-8c28-78663851d6db
September 1993
Shawe-Taylor, J.
c32d0ee4-b422-491f-8c28-78663851d6db
(1993)
Symmetries and Discriminability in Feedforward Network Architectures.
IEEE Transactions on Neural Networks, 4 (5), .
Abstract
This paper investigates the effects of introducing symmetries into feedforward neural networks in what are termed symmetry networks. This technique allows more efficient training for problems in which we require the output of a network to be invariant under a set of tranformations of the input. The particular problem of graph recognition is considered. In this case the network is designed to deliver the same output for isomorphic graphs. This leads to the question fo which inputs can be distinguished by such artitectures. A theorem characterizing when two inputs can be distinguished by a symmetry network is given. As a consequence, a particular network design is shown to be able to distinguish nonisomorphic graphs if and only if the graph reconstruction conjecture holds.
PDF
SymmetriesAndDiscInFeedforwardnetworkArch.pdf
- Other
More information
Published date: September 1993
Organisations:
Electronics & Computer Science
Identifiers
Local EPrints ID: 259819
URI: http://eprints.soton.ac.uk/id/eprint/259819
PURE UUID: dda9aa0d-c40d-4fad-91d6-dcde52e09478
Catalogue record
Date deposited: 03 Sep 2004
Last modified: 18 Jul 2017 09:19
Export record
Contributors
Author:
J. Shawe-Taylor
University divisions
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics