Prediction of aperiodic target sequences by saccades

Abstract

Through recording of saccadic eye movements, we investigated whether humans can achieve prediction of aperiodic target sequences which cannot be predicted based solely on memorizing short-length patterns of the target sequence. We proposed a novel experimental paradigm in which Auto-Regressive (AR) processes are used to generate aperiodic target sequences. If subjects can fully utilize the knowledge on the AR dynamics that have generated the target sequence, optimal prediction can be made. As a control task, a completely unpredictable (random) target sequence was generated by shuffling the AR sequences. Behavioral analysis suggested that the prediction of the next target position in the AR sequence was significantly more successful than that by the random guess or the optimal guess for the random sequence. Although their performances were not optimal, learning of the AR dynamics was observed for first-order AR sequences, suggesting that the subjects attempted to predict the next target position based on partially identified AR dynamics.