Ideal Observer Theory

https://doi.org/10.1016/B978-008045046-9.01435-2Get rights and content

Ideal observer models are applications of Bayesian statistical decision theory to problems of neural information transduction, transmission, and utilization. A basic motivation is that, because sensory inputs provide noisy or ambiguous information about states of the world, probabilistic methods are required to understand how reliable decisions can be made. Thus, the focus is first on modeling the information for a task, independent of the observer under study, and second on comparisons of that model with a test observer, such as a human or neuron. A key rationale for such comparisons is that the ideal observer can be used to normalize performance for task difficulty. An ideal observer can also provide a starting point for modeling perceptual performance.

References (0)

Cited by (4)

  • Desirable properties of Bayesian models in perceptual psychology. A discussion of distribution families closed under multiplication

    2019, Journal of Mathematical Psychology
    Citation Excerpt :

    The behavior of the Beta law is comparatively poor, and no improvements appear if it is treated as a one-parameter family, that is, a limitation is imposed consisting of combining Beta distributions that share the same value for one of the two parameters of the law. A warning that occasionally appears in the literature is that some expected properties of quantitative models of cue integration may lose validity when passing from Normal to non-Normal Bayesian models (e.g., Knill, 2007, p. 17; Kersten & Mamassian, 2009, p. 91). This study aimed to put this question under closer scrutiny.

View full text