August 2012
Volume 12, Issue 9
Vision Sciences Society Annual Meeting Abstract  |   August 2012
Neural Representations of Object Categories at Multiple Taxonomic Levels
Author Affiliations
  • Marius Cătălin Iordan
    Computer Science Department, Stanford University
  • Michelle R. Greene
    Computer Science Department, Stanford University
  • Diane M. Beck
    Psychology Department and Beckman Institute, University of Illinois Urbana-Champaign
  • Li Fei-Fei
    Computer Science Department, Stanford University
Journal of Vision August 2012, Vol.12, 1105. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Marius Cătălin Iordan, Michelle R. Greene, Diane M. Beck, Li Fei-Fei; Neural Representations of Object Categories at Multiple Taxonomic Levels. Journal of Vision 2012;12(9):1105. doi:

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

PROBLEM STATEMENT: Objects can be described at multiple taxonomic levels; i.e. Fluffy is simultaneously a tabby, a cat, an animal, and a living organism. Yet, some descriptions are more basic than others (in this case: 'cat') in that they are generated first and maximize within-category similarity (Rosch 1976). Despite numerous behavioral findings confirming this observation, very little is known about the neural representation of objects with respect to different taxonomic levels. METHODS: We address this question with an fMRI study in which subjects passively view stimuli from 32 subordinate-level categories grouped into 4 basic-level categories (8 breeds of dogs, 8 types of planes, 8 types of flowers, 8 types of shoes). ANALYSIS AND RESULTS: We restrict our analysis to early visual areas and object-, scene-, and face-selective areas: V1, V2, VP, V4, LOC, TOS, PPA, RSC, FFA. Dissimilarity matrices based on patterns of fMRI activity (Kriegeskorte et. al. 2008) were computed for our 32 stimulus categories. Interestingly, clear clusters corresponding to basic-level categories do not emerge until V4, LOC, and TOS. However, a correlation classifier could decode categories at the basic-level well above chance for all 9 areas considered, with LOC and TOS approaching 90% accuracy (chance is 25%). In fact, our basic-level classifier performs above chance in all areas even when it is trained and tested on disjoint subsets of subordinate categories. Decoding at the subordinate level, however, was considerably poorer even when controlling for sample size bias between the two levels of classification. SUMMARY: Although information regarding basic level category is present throughout visual cortex, activity patterns within a basic-level category are most similar in later visual areas. Moreover, our results indicate that the basic-level is privileged over the subordinate level, in accordance with Rosch’s view, and this representation is most saliently identified in LOC and TOS.

Meeting abstract presented at VSS 2012


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.