Abstract
Single-unit recording studies suggest that when making a coarse-discrimination (e.g., find a 90° target among 180° distractors), attention enhances the gain of neurons that are tuned to the target feature. However, when making a fine-discrimination (e.g. find a 90° target among 85° distractors), boosting the gain of neurons tuned to the target feature is suboptimal because these neurons respond about equally well to the target and to distractors. Gain should instead be applied to neurons that are tuned to an ‘exaggerated’ target feature (e.g. 95°) because these neurons will undergo a larger firing rate modulation in response to the small difference between the target and distractors (Navalpakkam & Itti, Neuron, 2007). Thus, the distribution of attentional gain across feature space — or the attentional gain function — should critically depend on the nature of the behavioral task.
Here, we estimated the shape of the attentional gain function during a fine-discrimination task. On every trial, a central precue indicated the distractor orientation as well as the directional offset of the target from the distractors (±5°). On 70% of the trials, four oriented Gabors were presented simultaneously, one in each quadrant; subjects indicated the spatial position of the target with a button press response. Given the high frequency of these trials, subjects should always anticipate a fine-discrimination trial. However, we occasionally (30% of trials) probed the distribution of attentional gain by measuring contrast detection thresholds for one oriented Gabor offset by 0°, ±5°, ±10°, ±20°, or ±40° from the anticipated target orientation. Contrast thresholds were lowest either at the target orientation or at the ‘exaggerated’ target orientation (+5° from target), suggesting that attentional gain was generally applied to the most informative sensory neurons. Moreover, the shape of the attentional gain function reveals the accuracy and precision of feature-based attention on a subject-by-subject basis.