Set Summary Perception, Outlier Pop Out, and Categorization: A Common Underlying Computation?

Seminar | October 29 | 11:10 a.m.-12:30 p.m. | 489 Minor Hall

 Shaul Hochstein, Professor, The Hebrew University of Jerusalem

 Neuroscience Institute, Helen Wills

Recent research has focused on perception of set statistics. Presented briefly with a group of elements, simultaneously or successively, observers report precisely the mean of a variety of set features, but are unaware of individual element values. This has been shown for both low and high level features, from circle size to facial expression. A remaining puzzle is how can the perceptual system compute the mean of element values without first knowing the individual values. We performed a series of studies to extend these findings and shed light on this conundrum. We found that set mean computation is performed automatically and implicitly, affecting performance of an unrelated task, and that it is performed on-the-fly for each psychophysical trial, independently. We find that observers also rapidly identify outliers within a set, indicating that they perceive the range of stimulus sets. We find that range perception, too, is automatic, implicit, and on-the-fly. In purposely-designed parallel studies, we find similar characteristics for set and category perception: category prototype and boundary correspond to set mean and range. Our matching findings suggest that categorization and set summary perception might share computational elements. We suggest and analyze a fundamental computational procedure, based on population encoding, that encompasses all the features of set summary perception, and might also underlie categorization. This computational procedure precludes the classic debate concerning category representation by prototype or boundary.

 nrterranova@berkeley.edu