Abstract
How does the brain make sense of variable and unreliable sensory input from an external world that is largely stable over short periods of time? A statistically optimal observer would capitalize on the world's stability by integrating past and present sensory inputs, weighting each by the uncertainty in its neural representation. Here, we use fMRI in combination with a probabilistic decoding algorithm to test this prediction. Participants viewed sequences of randomly oriented gratings, and reported their orientation. Consistent with previous behavioral work (Fischer & Whitney, 2014), we found that the stimulus orientation observers reported on the current trial was biased towards the orientation presented on the preceding trial, suggesting that their perception reflected a combination of current and previous sensory input. To test whether previously seen stimuli influenced perceptual decisions more strongly when current sensory information was less reliable, we used a probabilistic decoding algorithm to estimate sensory uncertainty from stimulus representations in early visual cortex (V1-V3 combined). Interestingly, comparing stimulus uncertainty between consecutive trials revealed that behavioral biases towards previously seen gratings were larger when the cortical stimulus representation on the present trial was more uncertain. This suggests that serial dependence effects in behavior are underpinned by a statistically optimal sensory integration process, in which uncertain sensory information is given less weight. Meeting abstract presented at VSS 2017
