Learning of unconscious scene-target spatial associations involves the sharpening of a distributed network of visual areas

Abstract

 

Scene context is known to facilitate attentional deployment in visual search. How are scene-target associations learned in this task? We developped a modified version of the contextual cueing paradigm. In the predictive condition, each layout of distractors (scene) was always associated with the same target position. In the non predictive condition, each scene was associated with each possible target position on successive presentations. A regular scene-target association was present only in the predictive condition. The learning of this association is expressed behaviorally after just 5 presentations of each scene: reaction times become shorter in the predictive than in the non predictive condition. At the beginning of the experiment, when behavior is not yet influenced, the brain is forming a new representation of the scene-target association. Predictive - non predictive differences in the magnetoencephalographic (MEG) signal are seen in the occipito-parietal cortex at 115–175 ms, followed by a difference in the left ventral occipital cortex at 175–225 ms. Spanning these two time windows (100–400 ms), we found an increase of induced oscillatory gamma activity (30–48Hz) over parietal sites in the predictive condition. Scene-target associations are thus first detected in a distributed network of visual areas. Gamma oscillations could coordinate this network in a synchronous neural assembly. In contrast, at the end of the experiment, when scene-target associations influence behavior, the earliest difference in the MEG signal appears as soon as 100 ms in the calcarine region. A rapid extraction of the learned scene-target associations thus occurs locally, possibly in a feedforward manner. How the associations first detected in a distributed network are finally transferred to early visual cortex remains an open question. We suggest that gamma oscillations trigger the neural plasticity that sharpens the distributed representation in a more efficient network representing scene-target associations in early visual cortices.