Cléo Schoeffel¹, Anna Montagnini¹, Guillaume S Masson¹; ¹Aix-marseille Université - Institut de Neurosciences de la Timone - CNRS - Marseille, France

Smooth pursuit eye movements reflect several biases identified in velocity estimation, such as the aperture problem for direction and the spatial frequency-dependent bias for speed. These biases are best explained by the properties of prior knowledge within the Bayesian inference framework. Some, but not all, previous studies have shown that global motion perception can be adapted through repeated exposure, suggesting that perceptual Priors may be updated based on target motion statistics. We investigated whether overrepresenting visual feature-speed associations could affect pursuit initiation. 11 participants tracked a Gabor (5°) moving horizontally at 10°/s, with varying orientations (0°,+/-45°) or spatial frequencies (0.3, 1cpd). Across five sessions, pre- and post-training baseline blocks (240 trials each) surrounded training blocks where one orientation or spatial frequency was paired with a higher speed (20°/s) and repeated across 720 trials. We compared pre- and post-training pursuit latency and initial eye velocity. Results showed that exposure to motion stimuli with a particular feature/speed association led to changes in pursuit initiation. There was no learning effect on latencies or initial eye velocity for the orientation/speed pairing. However, low spatial frequencies had later pursuit initiation compared to high spatial frequencies. After training, this delay disappeared as (i) latencies for low spatial frequencies were shortened over time and (ii) latency distributions now overlapped between the two spatial frequency conditions. Moreover, after latency realignment of pursuit responses, repeating high spatial frequencies at a higher speed increased initial pursuit eye velocity for high spatial frequency at slow speed. These results suggest that overrepresenting visual features during pursuit could bias pursuit initiation, consistent with an update of the sensory Priors used for 2D global motion estimation. The impact of target spatial frequency, but not its orientation, suggests that biases in speed and direction might be associated with different Priors and/or decoding of global motion.