Abstract
Although it is well established that attention modulates the visual responses of single neurons, how attention alters neuronal representations remains controversial. One set of experiments, largely based on measurements with single stimuli within receptive fields, suggests that attention simply increases the gain of visual responses. Another set of experiments, largely based on measurements with multiple stimuli within receptive fields, suggests that neuronal responses are altered by the filtering unattended stimuli. To reconcile these results, we recorded from V4 neurons in 3 animals trained to perform a reaction-time orientation change detection in which either single or multiple stimuli were placed in the receptive field. We constructed quantitative models of spatial summation for each neuron. We confirm previous results in finding that attention acts as a gain factor on responses to single stimuli. However, the paired stimulus responses showed less consistency: while responses to paired stimuli strongly depend on which stimulus was attended in 2 animals (N=159), in the third animal paired stimulus responses were multiplicatively increased regardless of which stimulus was attended (N=43). By taking into account spatial summation of each neuron, we found that the best fitting model of attention was one in which facilitation of attended inputs was paired with weak suppression of unattended inputs. A DOG model of the spatial distribution of attentional gain in which the spatial scale was free to vary was able to explain the data from all three animals. These results are consistent with a gain model of attention in which the spatial scale of facilitation and suppression reflects flexible and task-specific windows of attention.