Abstract
We can extract general trends as well as momentary values when we look at ongoing natural and social events. We constantly make such judgments all the way through the duration of events. To understand computational principles underlying the perceptual decision on global trends (and momentary values) of temporally varying data, we analyzed human judgments for a Gabor patch with temporally fluctuating orientation or motion direction. The stimulus was presented for 2-4 seconds, and either its orientation or motion velocity temporally fluctuated according to a Gaussian distribution with particular mean and variance. Observers were asked to judge the temporal average of orientation (right / left, not the direction of rotation) or motion direction (rightward / leftward). They were asked to respond as soon as possible after the stimulus onset in one condition, or to respond after the presentation in the other condition. In a separate experiment, the observer judged not only the temporal average, but also the momentary value of orientation or motion direction at the moment that was specified by a flash of small square. By using image classification and multiple logit analysis, we calculated the impact of feature values (orientation, phase, and velocity) at each temporal frame upon the observers’ responses. The results revealed five basic characteristics of human decisions on dynamic events. (1) Primacy-recency effects: judgments on global trend strongly depend on local information at around stimulus onset, stimulus offset, and the observer's responses. (2) Low-pass property: features at rapid change are ignored for calculating global trends. (3) Robust averaging: outliers are ignored. (4) Temporal crowding: judgments on local values are inaccurate and based on information over approximately ±150ms. (5) Local-global interactions (leak effect): judgments on temporal average are influenced by the value at the cued moment.
Meeting abstract presented at VSS 2013