Abstract
Objects have a variety of different features that can be represented as probability distributions. Previous findings show that besides mean and variance, the visual system also encodes the distribution shape, both for color and orientation ensembles (Chetverikov et al. 2016, Cognition; Chetverikov et al. 2017, Psych. Science). In an odd-one-out search task we investigated observers' ability to learn multiple feature distributions simultaneously. Our stimuli were defined by two distinct features (color and orientation) while only one was relevant to the search task. We investigated whether the irrelevant feature distribution influences learning of the task-relevant distribution and whether observers also encode the irrelevant distribution. Subjects participated in blocks consisting of learning and test streaks. During learning streaks (3-4 trials) orientation and color of each distractor were drawn from a pre-defined distribution (Gaussian or uniform) with constant mean and variance. The target was either distinguishable by color or by orientation based on a pre-defined distance from the distractor mean in feature space. The irrelevant feature of the target was drawn from within the distractor distribution. During test trials targets were probed at particular distances from the mean of the previously learned distractor distribution. The underlying shape of the distractor distribution was assessed through changes in RT as a function of the distance between the target during testing and the distractor mean during learning. Our preliminary results show that properties of the irrelevant distribution impaired the encoding of the relevant one. Data hint towards an asymmetry between the two different features: searching for the oddly-oriented target was more difficult than searching for the oddly-colored target. Moreover, subjects also encode mean and variance of the irrelevant feature distribution during learning but not the distribution shape. Our study demonstrates both an ability to encode information of multiple feature distributions simultaneously but also encoding limitations.
Meeting abstract presented at VSS 2018