Purchase this article with an account.
Andrey Chetverikov, Gianluca Campana, Árni Kristjánsson; Probabilistic perceptual landscapes. Journal of Vision 2018;18(10):529. doi: https://doi.org/10.1167/18.10.529.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
How do humans represent the visual world? Previous work suggests that observers' representations are probabilistic and incorporate the uncertainty associated with stimuli. However, this work was usually limited to simple stimuli (e.g., a single grating) arguably rare in the real world. Recently, we found that probabilistic representations of distractors in odd-one-out visual search follow the shape of the probability distribution used to generate the distractors. Here we demonstrate how this effect allows the decoding of the representations of complex stimuli, such as an array of lines with orientations drawn from different probability distributions. We show how the distributions of features by locations in a physical world are transformed into probabilistic landscapes in observers' minds. We collected observers' responses during odd-one-out orientation search with distractors generated from two different distributions. These distributions were either mixed randomly, separated into different halves of the visual field, or presented in stripes. Trials were organized in sequences. Each "prime" sequence (during which the distractor distributions stayed the same) was followed by test trials with varying similarity between test target and prime distractors. We then estimated via bootstrapping the most expected distractor orientation corresponding to the slowest response times in test trials at different locations. Thus, we obtained 2D (horizontal position and orientation) and 3D (horizontal and vertical positions and orientation) maps showing observers' probabilistic representations. The representations generally followed the generative probability distributions, with some exceptions: even when the distributions had sharp boundaries, representations were smoothed, with intermediate features expected on borderline locations. We introduce a new method allowing decoding of representations (including associated uncertainty) of complex heterogeneous stimuli showing that they are encoded probabilistically just as simpler stimuli are. This provides strong support for the idea of hierarchical probabilistic representations in the brain and shows how detailed statistics of the environment can be acquired.
Meeting abstract presented at VSS 2018
This PDF is available to Subscribers Only