Abstract
Serial dependence promotes perceptual stability by integrating similar features of successive stimuli over time. A Recent study demonstrated serial dependence in ensemble representations. This suggests that our visual system tends to represent sets of similar items as statistical summaries across both space and time. We investigated if such temporal smoothing process would be disrupted when ensemble variability changes over time. Observers were asked to adjust the average orientation of an array of 16 randomly oriented Gabors in the peripheral visual field. Each array had either high (H) or low (L) standard deviation. There were two neutral sessions with random stimulus sequence of H and L and two repetitive sessions with stimulus sequence of either L or H. We analyzed serial dependence for the nearest sequential pair in each session. Serial dependence was observed in L stimulus regardless of previous stimulus variability. Serial dependence was also observed in H stimulus when the previous stimulus had H. However, the negative bias was observed in H stimulus when the previous stimulus had L. Internal expectancy of ensemble variability might cause this negative bias in LH pair. It might be the case that internal transition probability from L to H would be smaller than actual stimulus transition probability(≈0.5), making LH pair less predictable. To test this hypothesis, Hidden Markov Model was fitted to the behavioral errors categorized into H and L group. HMM showed that LH transition probability was significantly smaller than 0.5. More importantly, negative bias in LH pair was stronger for unpredicted trials, while predicted trials did not show negative bias. These results suggest that unpredictably increased variability makes sensory system adaptive to maximize change sensitivity by negatively biasing away from previous ensemble representation.
Meeting abstract presented at VSS 2018