Encoding multiple orientations in a recurrent network

Richard S. Zemel, Jonathan Pillow

Research output: Contribution to journalArticle

4 Scopus citations

Abstract

Models containing recurrent connections amongst the cells within a population can account for a range of empirical data on orientation selectivity in striate cortex. However, existing recurrent models are unable to veridically encode more than one orientation at a time. Underlying this inability is an inherent limitation in the variety of activity profiles that can be stably maintained. We propose a new recurrent model that can form a broader range of stable population activity patterns. We demonstrate that these patterns preserve information about multiple orientations present in the population inputs. This preservation has significant computational consequences when information encoded in several populations must be integrated to perform behavioral tasks, such as visual discrimination. (C) 2000 Elsevier Science B.V. All rights reserved.

Original languageEnglish (US)
Article number1088
Pages (from-to)609-616
Number of pages8
JournalNeurocomputing
Volume32-33
DOIs
StatePublished - Jan 1 2000
Externally publishedYes

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Cognitive Neuroscience
  • Artificial Intelligence

Keywords

  • Orientation selectivity
  • Population codes
  • Recurrent network models

Fingerprint Dive into the research topics of 'Encoding multiple orientations in a recurrent network'. Together they form a unique fingerprint.

  • Cite this