Purchase this article with an account.
Benjamin Backus; Bayesian model of cue combination for ambiguous stimuli. Journal of Vision 2007;7(9):66. https://doi.org/10.1167/7.9.66.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Perceptually ambiguous stimuli have the property that observers can accurately report how they look, which makes them useful for studying the construction of visual percepts (appearance, qualia, or “how things look”). Recent work on cue recruitment (Haijiang et al., 2006) makes clear the need for theory to quantify the effectiveness of factors that influence a dichotomous perceptual decision. Here we propose a simple Bayesian model for dichotomous perceptual decisions in the tradition of Brunswik's probabilistic functionalism and Savage's personalistic view of probability: the Mixture of Bernoulli Experts or MBE model. The MBE equation describes the system's reliance on each factor as being equivalent to having observed a certain number trials in a binomial experiment. Biasing factors include a noise term, an overall bias (the prior), and a term for each cue. The MBE equation distinguishes between the system's reliance on a given expert and the expert's estimate of the Bernoulli probability. If this distinction is not made, then the equation simplifies to a form that resembles a log odds decomposition or the linear weighting of evidence. Previous work supports such a model: several cues can exert independent effects that bias the apparent direction of rotation of a Necker Cube (Dosher, Sperling and Wurst, 1986). MBE provides a method for measuring the effectiveness of cues in the special situation where the perceptual system makes a probabilistic choice between exactly two representations. MBE is not a model to explain bistability. Instead, it presupposes bistability. MBE is model only for the effects that weak cues have on the appearance of a stimulus, causing it to appear in one or the other of its two possible forms.
This PDF is available to Subscribers Only