Visual search is a goal-oriented activity we perform daily. It involves active scanning of the environment, with the goal of locating objects of interest in the background of irrelevant distractors. Our research focuses on studying the computational mechanism underlying pop-out visual search.

In a pop-out task, the deviant object swiftly stands out and can be located with high fidelity. A widely accepted theory asserts that pop-out is computed by a winner-take-all competition between contextually modulated cells. However, past studies have shown that the ability of the winner-take-all mechanisms to accumulate information from large populations of neurons is limited; thus, raising the question of whether winner-take-all can underlie pop-out visual search.

To address this question, we studied the pop-out task from the perspective of a readout mechanism and investigated the accuracy with which a winner-take-all mechanism can detect the deviant stimulus. We find that the performance of the winner-take-all deteriorates rapidly when the number of distractors is increased while improving slowly with the number of neurons in each population. Moreover, applying the winner-take-all readout to electrophysiological data reveals that the inherent neuronal heterogeneity prevents the winner-take-all mechanism from achieving reasonable performance even in simple tasks with a small number of distractors.

Finally, we show that a generalized population-code competitive readout can achieve the high performance expected from a pop-out mechanism. Nevertheless, our study indicates that individual bias (i.e. performance variability among different subjects) and deterioration of the performance with the number of distractors are expected to be found in the generalized population model too.