Abstract
Ensemble perception has been shown to be resistant to the influence of outlier stimuli. However, outlier rejection is inherently paradoxical—how can ensemble processes disregard outliers, when outliers are defined based on ensemble properties? We propose that the solution is that ensemble perception operates iteratively, continuously updating and refining perceived summary statistics. We tested this hypothesis with three experiments. In experiment 1, participants reported the average orientation of groups of lines, in displays with or without outliers, using an adjustable probe. Extending prior results using facial emotion, the weight of outliers in reported averages was reduced, although error remained higher compared to conditions with no outliers present. In experiment 2 we directly tested the timing of this process using a speeded response task. Here we found that responses were slowed significantly when outliers were present in the ensemble. In experiment 3 we tested the precise timing of outlier discounting, using masking at 50, 100, 200, 300 and 500ms post stimulus onset. Results showed that the influence of outliers on the reported average decreased linearly over time, with outliers seemingly rejected fully by 500ms. Altogether these findings support the hypothesis that ensemble perception works as an iterative process. The idea of ensemble perception as a continuously updating iterative process provides a useful explanatory framework for a number of findings: it bridges results that show ensemble perception is rapid with those that suggest it incorporates complex properties that require more extensive processing. It also sheds light on how statistics are collected from streams of stimuli, as well how summary statistics can guide attention towards or away from specific objects in visual scenes. We discuss the framework of iterative ensemble coding in the broader context of the literature and propose ways in which it can be rigorously tested using behavioral and neuroimaging experiments.