Visuo-motor sequential predictions guiding the representation of visual objects

Traditionally, vision has been largely treated as a bottom-up system that tries to make sense of the incoming input. However, contemporary accounts of active inference and predictive coding see the visual system as a goal-oriented system in which internal knowledge and learning fundamentally influence perception through top-down interactions. To study how internal visuo-motor predictions guide object visual processing, we combined high-density electrophysiology and a sequential learning task (N=41) in which subjects statistically learn the association between a sequence of objects (faces, scenes, etc) and keypresses with different fingers (e.g., face-index, scene-middle finger, and so on). Our analysis focuses on two issues. First, we aim to record neural correlates of the development of an internal model. Second, to examine where/how sequential predictions change visual object processing. Behaviourally, we show that predictive sequences of object-finger presses are parametrically faster and more accurate than random sequences. Interestingly, this probability is independent of whether blocks of predictive vs. random sequences occur early or late during the experiment. Late event related potentials (ERPs)(latency 0.2s-0.6s) source localized to frontal and parietal regions correlate with sequence probability and behaviour, hence representing a potential internal model for the task. Furthermore, sequential predictions modify early visual processing components specific to object categories (i.e., faces, scenes). We conclude that top-down visuo-motor predictions are important for the processing of ecological objects, while sequence coding is one possible brain mechanism that allows for navigating in complex everyday environments.