Linear separability in classification learning

Douglas L. Medin*, Paula J. Schwanenflugel

*Corresponding author for this work

Research output: Contribution to journalArticle

144 Scopus citations

Abstract

Performed 4 experiments to determine whether linearly separable (LS) categories (which can be perfectly partitioned on the basis of a weighted, additive combination of component information) are easier to learn than non-LS categories. 224 Ss, predominantly undergraduates, participated. Independent cue models (e.g., prototype theories) predict that, with average between-category similarity held constant, LS categories will be easier to master. Relational coding models (e.g., the context model) imply that individual cases of high similarity of exemplars to exemplars in contrasting categories are the major determinant of task difficulty. These experiments varied stimulus type (geometric shapes or photographs of faces), category size (small or infinitely large), and instructions. With average similarity controlled, instances of high similarity of exemplars to exemplars influenced performance, but there was no evidence that LS categories were learned more readily than non-LS categories. (29 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved).

Original languageEnglish (US)
Pages (from-to)355-368
Number of pages14
JournalJournal of Experimental Psychology: Human Learning and Memory
Volume7
Issue number5
DOIs
StatePublished - Sep 1 1981

Keywords

  • category size &
  • instructions, classification learning, college students
  • linear separability of stimulus category type &

Fingerprint Dive into the research topics of 'Linear separability in classification learning'. Together they form a unique fingerprint.

  • Cite this