Estimating the strength of unlabeled information during semi-supervised learning

Brenden M. Lake, James L. McClelland

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Semi-supervised category learning is when participants make classification judgements while receiving feedback about the right answers on some trials (labeled stimuli) but not others (unlabeled stimuli). Sporadic feedback is common outside the laboratory, and it is important to understand how people learn in this setting. While there are numerous recent studies, the strength and robustness of semi-supervised learning effects remain unclear, particularly when labeled and unlabeled stimuli are dispersed across learning. We designed an experiment, using simple unidimensional category learning, that allows us to measure the relative contribution of labeled and unlabeled experience. Based on an analysis of this task, we find that an unlabeled stimulus is worth more than 40% of a labeled stimulus.

Original languageEnglish (US)
Title of host publicationExpanding the Space of Cognitive Science - Proceedings of the 33rd Annual Meeting of the Cognitive Science Society, CogSci 2011
EditorsLaura Carlson, Christoph Hoelscher, Thomas F. Shipley
PublisherThe Cognitive Science Society
Pages1400-1405
Number of pages6
ISBN (Electronic)9780976831877
StatePublished - 2011
Event33rd Annual Meeting of the Cognitive Science Society: Expanding the Space of Cognitive Science, CogSci 2011 - Boston, United States
Duration: Jul 20 2011Jul 23 2011

Publication series

NameExpanding the Space of Cognitive Science - Proceedings of the 33rd Annual Meeting of the Cognitive Science Society, CogSci 2011

Conference

Conference33rd Annual Meeting of the Cognitive Science Society: Expanding the Space of Cognitive Science, CogSci 2011
Country/TerritoryUnited States
CityBoston
Period7/20/117/23/11

Keywords

  • categorization
  • semi-supervised learning

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Science Applications
  • Human-Computer Interaction
  • Cognitive Neuroscience

Fingerprint

Dive into the research topics of 'Estimating the strength of unlabeled information during semi-supervised learning'. Together they form a unique fingerprint.

Cite this