Summary
The recent visual past has a strong impact on our current perception. Recent studies of serial dependence in perception show that low-level adaptation repels our current perception away from previous stimuli1–5 whereas post-perceptual decision attracts perceptual report toward the immediate past6–12. In their studies, these repulsive and attractive biases were observed with different task demands perturbing ongoing sequential process. Therefore, it is unclear whether the opposite biases arise naturally in navigating complex real-life environments. Here we only manipulated the environmental statistics to characterize how serially dependent perceptual decisions unfold in spatiotemporally changing visual environments. During sequential mean orientation adjustment task on the array of Gabor patches, we found that the repulsion effect dominated only when ensemble variance increased across consecutive trials whereas the attraction effect prevailed when ensemble variance decreased or remained the same. The observed attractive bias by high- to-low-variance stimuli and repulsive bias by low-to-high-variance stimuli were reinforced by the repeated exposure to the low and the high ensemble variance, respectively. Further, this variance-dependent differential pattern of serial dependence in ensemble representation remained the same regardless of whether observers had a prior knowledge of environmental statistics or not. We used a Bayesian observer model constrained by visual adaptation13,14 to provide a unifying account of both attractive and repulsive bias in perception. Our results establish that the temporal integration and segregation of visual information is flexibly adjusted through variance adaptation.
Competing Interest Statement
The authors have declared no competing interest.