Representation of trajectories within visual cortical maps
The pioneering work of Hubel and Wiesel established a paradigmatic model of visual processing, in which neurons work as feature detectors at each stage with progressively more elaborate receptive field selectivity. In this framework, low-level visual information (such as retinotopy and orientation) is extracted within local receptive fields and rapidly cascades to higher cortical areas. This hierarchical model remains dominant to this day. One specific consequence of this is that visual motion processing has been mainly investigated using stimuli presented within a fixed region centered on their receptive field, i.e. a stationary aperture, hereby focusing on a steady-state and piecewise information confined to the receptive field. In contrast, real-world visual inputs are intrinsically dynamical, often ambiguous and non-stationary, such as objects moving along trajectories that can be partially occluded. How such stimuli are processed by the visual system remains an open and challenging question. A moving stimulus will generate non-stationary sequences of feedforward inputs in various positions within the retinotopic maps of visual areas. It is thus crucial to have access to the evoked neuronal activity dynamics at the proper scale (within cortical maps), using multi-unit array or voltage-sensitive dye imaging. Using these methods in the awake monkey, we demonstrated a first key result: any local stationary stimulus is, in itself, generating traveling waves that propagate over retinotopic maps. As a consequence, the feedforward activation sequence induced by a moving stimulus will be relayed, at each position of the sequence, by an intricate interplay of propagations within, but also between, cortical retinotopic maps. Our recordings indeed show the existence of non-trivial recurrent interactions between feedforward, intra- and inter-cortical inputs that shape the cortical representation of moving stimuli implementing the framework for spatio-temporal predictive computation. These results call for the emergence of new computational paradigms beyond the standard feedforward model.