August 2016
Volume 16, Issue 12
Open Access
Vision Sciences Society Annual Meeting Abstract  |   September 2016
Humans implement nonlinear computations to achieve near optimality in the face of scalar variability.
Author Affiliations
  • Seth Egger
    Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, MA
  • Mehrdad Jazayeri
    Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, MA
Journal of Vision September 2016, Vol.16, 583. doi:10.1167/16.12.583
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Seth Egger, Mehrdad Jazayeri; Humans implement nonlinear computations to achieve near optimality in the face of scalar variability.. Journal of Vision 2016;16(12):583. doi: 10.1167/16.12.583.

      Download citation file:


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

      ×
  • Supplements
Abstract

Humans seek to optimize sensory estimates by integrating (1) prior experience and (2) available sensory measurements. For example, when multiple cues are present, humans can linearly combine independent measurements according to their reliability. This linear weighting, however, does not optimize estimates of sensory quantities for which the noise scales with stimulus strength (scalar variability) such as elapsed time, numerosity and heaviness. For these stimuli, the optimal integration rule is significantly more challenging as it requires a multidimensional and highly nonlinear operation. We asked if humans can implement an optimal integration strategy in timing behavior for which noise is thought to scale with duration. Subjects performed a time interval reproduction task in which we controlled the number of measurements subjects made. On each trial, we visually presented either one or two identical sample intervals drawn from a fixed prior distribution, and asked subjects to accurately reproduce that interval. Production intervals were systematically biased toward the mean of the prior distribution demonstrating that humans used knowledge of the prior to improve their performance. Additionally, errors decreased with two measurements compared to one showing that subjects combined information from multiple measurements. We then used a Bayesian model to assess the optimality of subjects' performance. Although the model captured the overall trade off between bias and variance, the improvement furnished by an additional measurement fell short of the model predictions. This modest suboptimality motivated us to ask whether an alternative model that combines measurements linearly accounts for the suboptimal behavior. However, the improvement in performance almost invariably exceeded the predictions of linear integration, suggesting humans integrate interval measurements according to a nonlinear rule. Given the prevalence of scalar variability in perception and cognition, our results have far-reaching implications for the brain's capacity to integrate sensory information in the control of behavior.

Meeting abstract presented at VSS 2016

×
×

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.

×