Abstract
While it is well-known that cognition influences visual information acquisition via large eye movements, it is unclear whether cognition also exerts effects on ocular drifts, the continuous, Brownian-like eye movements between saccades. These drifts act as critical computational steps for visual processing, but little is known about their control. To identify possible cognitive influences, we examined how task knowledge affects ocular drifts in a letter discrimination task. Subject reported which of two letters was displayed, with a different letter pair used in separate blocks of trials (H vs. N and E vs. F). Letters subtended 1.5 deg, were masked with 1/f noise, and had a contrast that yielded ~75% correct performance. To separate top-down and sensory influences, trials with no letter were randomly introduced (20% of trials). Drift was characterized by the velocity covariance matrix; we examined how this covariance differed between HN and EF blocks. Drifts differed significantly in all subjects in the letter-absent trials and in 4 out 6 subjects in the letter-present trials. For both conditions, analysis across subjects revealed a shared transformation accounting for much of the difference between the HN and EF blocks. This shared transformation shifts from diagonally-oriented drifts in HN trials to vertically-oriented drifts in EF trials, appropriate for increasing sensitivity to the critical distinguishing feature in each block. Lastly, we decoded single-trial drift trajectories by comparing individual-trial covariances to the overall covariance matrices. This decoder identified the task (HN vs. EF) from the drift trajectory at above chance levels (p<0.05) in 5 of the 6 subjects. Thus, task knowledge influences the dominant directions of ocular drifts. This influence is present both under open-loop conditions, when only cognitive factors are present, and under closed-loop conditions, when cognitive factors interact with incoming visual information.