Kuo-Wei Chen¹ (kchen172@asu.edu), Juhyun Ha, Ling-Jing Liu, Scott Janetsky, Gi-Yeul Bae; ¹Arizona State University

Achieving perceptual stability over time in the face of noisy visual inputs is a fundamental goal of visual information processing. Studies have proposed that the visual system achieves this goal by biasing perception of a new stimulus toward the past stimulus (i.e., attractive serial dependence). However, the visual system must also maintain perceptual sensitivity over time to detect differences in stimuli encountered at different time points (i.e., repulsive serial dependence). Here, we investigated how the visual system balances these competing demands in the context of motion perception. In two experiments, participants viewed a brief random-dot kinematogram with varying levels of motion coherence and reproduced the perceived motion direction using a computer mouse. We found that direction reports exhibited an attractive serial bias when the motion stimulus was noisier (i.e., lower coherence), and that this attractive bias decreased and eventually reversed into a repulsive bias as the motion stimulus became less noisy (i.e., higher coherence). To understand how these opposing biases emerged, we examined the temporal dynamics underlying the bias effects by analyzing mouse trajectories. We found that, for both attractive and repulsive biases, mouse reports began with a strong repulsive bias away from the prior stimulus, reflecting adaptation to the prior stimulus (or decision), and that this initial repulsion was substantially decreased over the course of the response. These results show that both the repulsive and attractive serial biases are driven by post-perceptual correction mechanism that reduces the strong repulsive bias driven by adaptation, with the direction of the serial bias effect determined by the quality of the stimulus information. Together, our findings suggest that the visual system optimizes perceptual behavior under stimulus uncertainty by balancing stability and sensitivity through a single quantitative decision mechanism, rather than through qualitatively distinct multiple mechanisms proposed by prevailing models of serial dependence.