Composite patterns

D. Horn*

*Corresponding author for this work

Research output: Contribution to journalConference articlepeer-review


We address the question how to construct a composite pattern in terms of elementary patterns within the context of neural-network models. This problem suggests itself when one considers the issue of error correction of letters on a printed page. Given a neural-network which stores in it all letters as elementary patterns it can serve as a tool for this error correction purpose provided it has the additional property of shift-invariance, i.e. it has to be able to recognize the letter at any location on the page. The result can be obtained by a suitable generalization of the Hopfield model.

Original languageEnglish
Pages (from-to)27
Number of pages1
JournalNeural Networks
Issue number1 SUPPL
StatePublished - 1988
EventInternational Neural Network Society 1988 First Annual Meeting - Boston, MA, USA
Duration: 6 Sep 198810 Sep 1988


Dive into the research topics of 'Composite patterns'. Together they form a unique fingerprint.

Cite this