TL;DR
This study investigates how feature-based attention (FBA) affects visual performance, specifically how it influences our ability to detect stimuli at different contrasts. The researchers found that FBA primarily boosts performance by increasing the "response gain" of visual neurons, making them more sensitive to stimuli, rather than by changing their sensitivity to contrast itself. This finding supports a prominent computational model of attention.
What Problem Does This Paper Address?
Visual attention is crucial for filtering the overwhelming amount of sensory information we receive, allowing us to focus on what's important. While we often think of attention as being directed to specific locations (spatial attention), it can also be directed to specific features, like a particular color or orientation, regardless of where they appear in our visual field (feature-based attention, FBA).
Previous research on spatial attention showed that it could improve performance by either increasing "contrast gain" (making neurons more sensitive to the overall contrast of a stimulus) or "response gain" (making neurons fire more strongly for a given stimulus). The "normalization model of attention" is a computational framework that attempts to explain these different effects based on the relationship between the size of the attended feature and the stimulus itself. A key prediction of this model is that FBA should only affect response gain, regardless of the size of the attended feature. However, experimental evidence for this specific prediction regarding FBA and its contrast dependence was lacking. This paper aims to fill that gap by systematically measuring how FBA affects performance across a range of stimulus contrasts and by manipulating the "featural extent" of attention.
Methods
The researchers employed a psychophysical approach using a feature-based attention (FBA) task. Participants were presented with two brief visual displays. The first display contained a random arrangement of grating patches with two distinct orientations (e.g., one tilted right, one tilted left). The second display contained patches all with the same orientation. Participants' task was to discriminate whether the orientation in the second display was slightly clockwise or counter-clockwise of the corresponding orientation in the first display.
To manipulate FBA, a pre-cue was used. This cue (a line at fixation) indicated which of the two orientations in the first display was the "target" orientation. A valid cue meant the pre-cue correctly predicted the target orientation, while an invalid cue meant it did not. This allowed the researchers to compare performance when attention was directed to the relevant feature versus when it was not.
Crucially, they manipulated the "featural extent" of attention by varying the uncertainty about the upcoming stimulus orientations. In the low-uncertainty condition, the pre-cue was very specific, and the possible orientations were narrowly defined. In the high-uncertainty condition, the pre-cue was broader, and the possible orientations spanned a wider range. This manipulation was designed to test the model's prediction that FBA effects would be consistent across different featural extents of attention.
Performance was measured across nine different contrast levels, and the data were analyzed by fitting psychometric functions. These functions describe how performance changes with stimulus contrast. The parameters of these functions, specifically d'max (asymptotic performance at high contrast) and C50 (contrast at half-maximum performance), are interpreted as measures of response gain and contrast gain, respectively.
Datasets
Not discussed in the paper. The study used custom-generated visual stimuli presented on a monitor.
Key Insights and Ideas
- FBA primarily boosts response gain, not contrast gain: This is the central finding. Unlike spatial attention, which can affect both, FBA seems to work by making neurons more responsive to stimuli, rather than changing their fundamental sensitivity to contrast.
- The "featural extent" of attention is flexible: By manipulating uncertainty about the target orientation, the researchers demonstrated that participants could adjust the "spread" of their feature-based attention. This is a novel finding for FBA, extending previous work on spatial attention.
- Support for the Normalization Model: The results align well with the predictions of the normalization model of attention, particularly its prediction that FBA should primarily modulate response gain. This model provides a unified framework for understanding different attentional effects.
- Spatial and feature-based attention might operate differently: While both are forms of attention, their underlying neural mechanisms, as revealed by their effects on contrast response, appear to differ.
Key Results
- Increased d'max with FBA: In both low and high uncertainty conditions, performance accuracy (d') was significantly higher for validly cued trials (attending to the target feature) compared to invalidly cued trials. This increase in d'max reflects an improvement in response gain.
- No significant change in C50: The contrast at which performance reached half its maximum (C50), a measure of contrast gain, did not significantly differ between valid and invalid cue conditions. This indicates that FBA did not alter the stimulus contrast required for a given level of performance.
- Uncertainty manipulation was effective: Participants performed worse in the high-uncertainty condition compared to the low-uncertainty condition when attention was validly cued, confirming that the uncertainty manipulation successfully influenced how broadly attention was deployed.
- Featural extent influences attention spread: In the low-uncertainty condition, performance was best for orientations slightly different from the cued one, suggesting a narrow attentional focus. In the high-uncertainty condition, performance was more uniform across a wider range of orientations, indicating a broader spread of attention.
Limitations and Future Work
- Model Simulations vs. Human Performance: While the study uses a computational model to guide predictions, the direct link between simulated neural responses and psychophysical performance is an inference. The authors acknowledge that performance accuracy (d') is proportional to the signal-to-noise ratio, which is influenced by response gain.
- Biological Plausibility of Feature Domain: The authors note that for FBA to induce contrast gain changes in the normalization model, the attention field would need to be unrealistically narrow in the feature domain. This suggests that the feature space might have different constraints than the spatial domain.
- Future Work: The authors suggest exploring whether FBA also sharpens neural tuning, as has been observed in some studies. They also propose testing whether FBA to directions of motion yields similar changes in response gain, given that neurons in area MT have broader tuning bandwidths. Further investigation into the interplay between featural attention field size and neural tuning width is also suggested.
Summary updated