Published on December 13, 2018 | Updated on January 29, 2019

CORTEX conference by Frédéric Chavane

December 10th, 2015

Representation of motion trajectories within visual cortical maps.

Since the pioneering work of the Hubel and Wiesel, our understanding of biological visual systems has been dominated by the feed-forward hierarchical approach. Accordingly, low-level visual information (such as position and orientation) is extracted locally within stationary receptive fields and is cascaded in a hierarchical feedforward sequence to encode more complex features and shapes. This approach is dominant in most fields of vision, including visual motion. As a consequence, motion has been investigated using stimuli presented within a stationary aperture, hereby focusing on a steady state and piecewise information. However naturalistic inputs are intrinsically dynamical, often ambiguous and non-stationary, such as objects moving along trajectories that can be partially occluded. To date, we still have a very poor knowledge of how objects moving along trajectories are processed by the visual system. Such stimulus will generate sequences of feed-forward inputs, relayed by an intricate interplay of propagations within and between cortical retinotopic maps that can propagate the information faster than the feedforward sequence. In awake monkeys, we investigated how these nested propagations shape the cortical mapping of a motion trajectory. The population response dynamics of moving stimuli, recorded using voltage-sensitive dye imaging, showed the existence of important non-linear interactions between feedforward input and lateral interactions that shape the spatio-temporal representation of the stimulus velocity. In response to continuous motion along a trajectory, these interactions lead gradually to the emergence of a strong anticipatory spiking activity. Predictive and accurate representation of non-stationary motion signals along trajectories thus results from the convergent non-linear interplay of intra- and inter-cortical inputs propagating information faster than the feed-forward sequence, potentially implementing a spatiotemporal predictive computation.