CCN 2025 ensemble perception models
Description
Individual and ensemble perception are crucial for interacting with objects in our environment. Individual perception processes single objects, while ensemble perception extracts summary information from object groups. To investigate how these two modes of perception work with different set sizes (3, 6, 10) in naturalistic settings, we compare two bayesian models on our data. The first model, a variant of the summation model, is the 'Individual Encoding Model'. The second model is the 'Ensemble Encoding Model', which is related to the automatic averaging model. We conducted an experiment in which participants encoded the position of an individual object or an ensemble position that summarized multiple objects in a 3D rendered scene and indicated its remembered position by mouse click on the screen. The 'Individual Encoding Model' assumes that each object's position is encoded in memory, the ensemble position is only evaluated on demand. In the 'Ensemble Encoding Model', the ensemble position is part of the process that generates the scene and is inferred from the observable object locations. We found that the accuracy of reproducing individual object positions increased as set size increased, while the estimation of the ensemble position (arithmetic mean) only differed between the 6- and 10-object set size conditions, with smaller deviations observed for scenes with 6 objects. The Ensemble Encoding Model generally explains the human behavioral data better. The subject-specific bayes factors in its favor increase with set size. We conclude that in naturalistic scenes the choice between individual versus ensemble encoding is likely driven by the more compact scene representation of the ensemble model.
