Influences of Prior Experience on Current-trial Ensemble Perception

Abstract

Introduction: Object ensembles are summarized quickly and efficiently by their summary statistics. While studies focused on extraction of current-view and current-trial statistics (i.e., short glimpses), little attention was paid to how ensemble perception is affected by prior experience (Crawford et al., 2018) and some studies were designed to avoid or eliminate influences of previous exposure to ensemble stimuli. However, statistical information about objects is learned over time, and Bayesian theories analyzed how priors influence perception. We now devised methods to directly test prior experience effects on ensemble averaging. Methods: To characterize influences of current-trial, recent-trial and long-term statistics (i.e., experimental session stimulus distribution), we measured implicit effects of these different statistics in a circle-size membership task. 100 Amazon MTurks participated. On each trial, circles of different sizes were presented in serial sequence. Then, two test circles were presented and participants chose which was present in the sequence. With this visual memory task, where it is hard to keep individual circle sizes in memory, we assume that participants unconsciously rely on ensemble statistics and choose test images closer to the ensemble mean. To differentiate influences from the current and from previous trials, in some trial blocks the 2 test circles were equally distant from the current trial sequence mean, isolating longer-term influences. Results: Participants biased membership judgements towards the ensemble mean of the current trial, when available – the largest effect found– and also towards the mean of the preceding trial sequence, when isolated. Lastly, participants also showed preferences to the longer-term (global) stimulus-distribution mean of the entire experimental session. Conclusion: Ensemble perception is influenced not only by the statistics of the set of objects presented in the current trial, but also by recently presented stimuli and by the long-term mean (prototype) which was learned and formed through the experiment.