Feature-based attention enhances performance by increasing response gain

Abstract

Objective: Characterize the contrast dependence of feature-based attention. Methods: Observers performed a 2IFC orientation-discrimination task. The first interval contained 32 randomly positioned Gabors (16 had identical orientation θ_R near 45° and 16 had identical orientation θ_L near −45°). In the second interval, 32 Gabors all shared identical orientation θ_test chosen randomly from θ_R ± 1 JND or θ_L ± 1 JND. Observers indicated whether θ_test was slightly clockwise or counter-clockwise of the corresponding orientation (θ_R or θ_L) in the first interval. Feature-based attention was manipulated with a 75%-valid pre-cue (a ±45° line at fixation) resulting in valid (pre-cue and θ_test roughly matched) and invalid (mismatched) cue conditions. The precise values of θ_R and θ_L were randomly and independently varied over trials so that the first-interval display was uninformative as to which orientation would be queried. Performance (d′) was measured for each cue condition and several contrasts, randomly interleaved. Results: We fit psychometric functions with five free parameters (full model) to the data: asymptotic performance at high contrast (d′_max) for each cue condition, contrast yielding half-maximum performance (c₅₀) for each cue condition, and an exponent shared by both cue conditions. Feature-based attention significantly changed d′_max, but not c₅₀. Nested F-tests showed a similarly good fit of a restricted model using one c₅₀ for both cue conditions (compared to the full model). A restricted model with one d′_max for both cue conditions fit significantly worse than to the full model. Conclusion: We manipulated feature-based attention while minimizing effects of spatial attention. Our results are consistent with an increase in response gain and support a key prediction of the normalization model of attention (Figure 4c of Reynolds & Heeger, Neuron, 2009).

NIH R01-EY019693 to DJH and MC.