September 2017
Volume 17, Issue 10
Open Access
Vision Sciences Society Annual Meeting Abstract  |   August 2017
The Easy-to-Hard Advantage with Real-World Visual Categories
Author Affiliations
  • Brett Roads
    Department of Computer Science, University of Colorado Boulder
    Institute of Cognitive Science, University of Colorado Boulder
  • Buyun Xu
    Department of Psychology, University of Victoria
  • June Robinson
    Department of Dermatology, Feinberg School of Medicine, Northwestern University
  • James Tanaka
    Department of Psychology, University of Victoria
Journal of Vision August 2017, Vol.17, 1234. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Brett Roads, Buyun Xu, June Robinson, James Tanaka; The Easy-to-Hard Advantage with Real-World Visual Categories. Journal of Vision 2017;17(10):1234.

      Download citation file:

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

  • Supplements

Previous work on visual category learning has demonstrated that the easy-to-hard approach (i.e., train with easier items earlier and difficult items later) can benefit learning outcomes. However, it is unclear whether previous findings generalize to the complex category structure found in real-world categorization tasks. We investigate the easy-to-hard advantage by training participants to distinguish between benign and malignant skin lesions. To determine item difficulty, we employ a novel approach that leverages an exemplar-level psychological representation and model of human similarity judgments. The representation and similarity function allow us to compute an ease value for every exemplar. Ease is defined as the probability that a participant will classify a given exemplar correctly. Specifically, ease is proportional to the sum of the similarity between the given exemplar and all other exemplars in the same category. Ease takes into account multiple aspects of category structure, such as within- and between-category variability. The stimuli for each category are binned into easy, intermediate, and hard items based on the ease value. In the easy-to-hard (hard-to-easy) training policy, easy (hard) items are learned on day 1, easy and intermediate (hard and intermediate) items are learned on day 2, and all items are learned on day 3. On each day, training continues until 90% accuracy is achieved. Participants were given a pre-test before the first training day and a post-test one day after the third day of training. Pre- and post-test items consisted of images not seen during training. Both groups were able to learn the task as indicated by a significant d' improvement from 0.35 prior to training to 1.61 after training. Critically, participants in the easy-to-hard condition required almost 30% fewer training trials than participants in the hard-to-easy condition.

Meeting abstract presented at VSS 2017


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.