August 2012
Volume 12, Issue 9
Vision Sciences Society Annual Meeting Abstract  |   August 2012
Learning to predict: unsupervised training of temporal sequences
Author Affiliations
  • Yang Zhang
    School of Psychology, University of Birmingham, UK
  • Tom Hardwicke
    School of Psychology, University of Birmingham, UK
  • Aimee Goldstone
    School of Psychology, University of Birmingham, UK
  • Josie Harding
    School of Psychology, University of Birmingham, UK
  • Matthew Dexter
    School of Psychology, University of Birmingham, UK
  • Zoe Kourtzi
    School of Psychology, University of Birmingham, UK\nLaboratory for Neuro- and Psychophysiology, K.U.Leuven, Belgium
Journal of Vision August 2012, Vol.12, 692. doi:10.1167/12.9.692
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Yang Zhang, Tom Hardwicke, Aimee Goldstone, Josie Harding, Matthew Dexter, Zoe Kourtzi; Learning to predict: unsupervised training of temporal sequences. Journal of Vision 2012;12(9):692. doi: 10.1167/12.9.692.

      Download citation file:

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

  • Supplements

Experience and training are known to facilitate our ability to extract regularities that are critical for visual recognition. However, the brain mechanisms that enable us to exploit previous knowledge to predict upcoming events remain largely unknown. Here, we combine behavioral and fMRI measurements to investigate the neural mechanisms that mediate predictive learning. Human observers were presented with a temporal sequence of leftwards vs. rightwards oriented gratings followed by a test grating. We asked observers to judge whether the orientation of the test stimulus matched their prediction based on the preceding sequence. Observers were trained on two structured sequences that gave rise to opposite predictions but were composed of common temporal segments (e.g. pairs of orientations). To ensure judgments were not based on memory, the orientation of the first stimulus in the sequence was randomly selected and the last three stimuli were the same for both sequences. Unsupervised training (three to five sessions of exposure to the sequences without feedback) improved observers’ performance for structured, but not random, sequences. This predictive learning generalized to a) sequences with the same temporal structure but different grating orientations and b) sequences with the same temporal segments but presented in a different order. After training, we measured fMRI responses, contrasting activity from the trained structured sequences against random sequences. We found significantly stronger activations for structured than random sequences in visual areas and a network of frontal, medial temporal and subcortical (i.e. thalamic) regions. These results suggest that training enhances our ability to exploit temporal regularities for predicting future events. This predictive learning ability is mediated by a network of brain areas known to be involved in associative learning that may, in turn, modulate early visual processing.

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.