A dynamical model of feature-based attention with strong lateral inhibition to resolve competition among candidate feature locations

摘要

When subjects are instructed to attend to an object defined by a feature in the visual field, e.g. a shape element, a colour, or a direction of motion, one finds correlates of this feature in neural activity across all retinotopic locations in low level areas in visual cortex, even in locations far removed from the neurons representing the object of interest. Next to the well established phenomenon of spatial attention, there is another form of attention which is featurebased, but not location specific. While feature-based attention has been demonstrated convincingly, its role is not well established. In earlier work we have suggested that is role is to help to prepare eye movements (or other motor actions) to the object of interest. In order to be able to do so, the location of the defining feature must first be established. In a single-object scene this is trivial, but in a multiobject scene the feature must be located among distractors. While we showed there that the retinotopic position of the object can be retrieved by an interaction between top-down attention-driven activation and stimulus-driven activation in lower visual areas, we did not model the dynamics of the selection process explicitly. Here we show that if one does this, one must resolve a competition process between the location of interest and those of the distractors. The process is decided by feature-based attention, but in order to implement this competition lateral inhibitory connections are required, in line with the'biased-competition model'. We will show that this process enables an unambiguous determination of the location of the target object and that the effect of the lateral inhibition is the effective shrinkage of receptive fields around an attended object which has been reported in several experiments.