October 2020
Volume 20, Issue 11
Open Access
Vision Sciences Society Annual Meeting Abstract  |   October 2020
Supervised learning enables generalization across dissimilar appearances of the same identity by conceptual rather than perceptual mechanisms
Author Affiliations
  • Galit Yovel
    Tel Aviv University
  • Maya Gotlieb
    Tel Aviv University
  • Naphtali Abudarham
    Tel Aviv University
  • Yarden Shir
    Tel Aviv University
Journal of Vision October 2020, Vol.20, 501. doi:https://doi.org/10.1167/jov.20.11.501
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Galit Yovel, Maya Gotlieb, Naphtali Abudarham, Yarden Shir; Supervised learning enables generalization across dissimilar appearances of the same identity by conceptual rather than perceptual mechanisms. Journal of Vision 2020;20(11):501. https://doi.org/10.1167/jov.20.11.501.

      Download citation file:


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

      ×
  • Supplements
Abstract

Humans effortlessly generalize across highly dissimilar appearances of the same identity, as well as discriminate similarly looking different identities for familiar but not for unfamiliar faces. Here we propose that this remarkable performance is enabled by supervised learning, which links perceptually dissimilar appearances of the same identity to the same label (e.g., name). These learned appearances are stored in memory as separated representations, that are linked to the same identity label, and can therefore be matched conceptually rather than perceptually. To test this hypothesis, we first presented subjects with pairs of perceptually dissimilar or perceptually similar faces during supervised (i.e. with labels) or unsupervised (i.e., without labels) learning task, and found that perceptually different faces were matched to the same identity following supervised but not unsupervised learning. Next, we examined whether two perceptually different appearances of the same identity are stored as a single prototype, or multiple sub-prototypes. We hypothesized that in a single prototype representation, an average face of the learned faces would be recognized as the learned identity, but not in multiple sub-prototype representation. To test this hypothesis, participants learned pairs of perceptually similar or dissimilar faces of the same identity with name labels. During test, they were presented with the learned faces, and also with their unlearned average faces. For perceptually similar pairs, the average face was recognized as the learned identity, indicating a single prototype for perceptually similar faces. However, for the perceptually dissimilar faces, their average face was not recognized, indicating that perceptually dissimilar faces were represented by separate prototypes, that are nevertheless linked to the same identity conceptually. This representation takes advantage of the conceptual information that is typically associated with familiar faces during learning and retrieval, to account for human remarkable recognition of familiar faces.

×
×

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.

×