Unsupervised temporal learning underlies the development of complex cells in primary visual cortex
In mammals, visual perception relies on cortical representations of visual objects that remain relatively stable with respect to the tremendous variation in object appearance that is typically experienced during natural vision (e.g., because of position, size or viewpoint changes). Such stability, known as transformation tolerance, is built incrementally along the visual cortical hierarchy devoted to shape processing, but early instances of position tolerance can be found already in primary visual cortex (V1), where the neurons known as complex cells maintain their orientation tuning over the entire span of their receptive fields.
To date, it remains unclear what mechanisms are at the origin of such tolerance, in V1 as well as in higher-order visual cortical areas. One of the leading theories, known as unsupervised temporal learning, postulates that visual neurons exploit the temporal continuity of visual experience (i.e. the natural tendency of different object views to occur nearby in time) to associatively link temporally-contiguous stimuli, so as to factor out object identity from other faster-varying, lower-level visual attributes (e.g., object positions, size, etc.).
In my lab, we causally tested this hypothesis by rearing newborn rats in visually controlled environments, where the animals were exposed either to a battery of natural movies (control group) or to frame-scrambled versions of such movies (resulting in temporally unstructured visual input; experimental group) for the whole duration of the critical period (60 days). Following this controlled rearing phase, we performed multi-electrode extracellular recordings from V1 of each animal, finding a reduction of the proportion of complex cells in the experimental group and a concomitant decrease in the ability of such neurons to code stimulus orientation in a position-invariant way.
These findings causally demonstrate that temporal continuity of the visual input plays an important role in the development of complex cells in V1, thus providing a critical experimental validation of the hypothesis that visual cortex achieves transformation tolerance through unsupervised temporal learning.