Abstract
Training has been shown to improve perceptual performance on limited sets of stimuli. However, whether training can generally improve top-down biasing of visual search in a target-nonspecific manner remains unknown. We trained subjects over ten days on a visual search task, challenging them with a novel target (top-down goal) on every trial, while bottom-up uncertainty (distribution of distractors) remained constant. We analyzed the changes in saccade statistics and visual behavior over the course of training by recording eye movements as subjects performed the task. Subjects became experts at this task, with twofold increased performance, decreased fixation duration, and stronger tendency to guide gaze toward items with color and spatial frequency (but not necessarily orientation) that resembled the target, suggesting improved general top-down biasing of search.
Introduction
Perceptual learning typically results in improvements that are highly specific to the trained stimuli, such as a particular orientation or spatial location. While valuable, this specificity limits the real-world utility of such training, as natural environments require searching for ever-changing targets. This study, published in PLoS ONE by Farhan Baluch and Laurent Itti, investigates whether the general mechanism of top-down attention—the ability to use a mental goal to bias visual processing—can be improved through training.
Methodology
The researchers employed a "triple-conjunction" search task designed to prevent stimulus-specific memorization. Five subjects participated in a 10-day training regimen, completing a total of 1,000 trials.
The Search Task
In each trial, subjects were tasked with finding a specific Gabor patch (a striped, colored fragment) hidden among 31 distractors. These stimuli were embedded in 1/f noise, which replicates the statistical properties of natural scenes. Crucially, the target's color (hue), spatial frequency (stripe thickness), and orientation changed on every trial. This forced subjects to rely on a dynamic top-down goal rather than long-term memory of a specific target.
Data Collection and Analysis
Eye movements were recorded at 240Hz to track search strategies. The team developed "Feature Similarity Maps" to quantify how closely each distractor matched the target across the three feature dimensions. By correlating these maps with actual eye fixations, the researchers could measure the effectiveness of top-down guidance.
Key Results and Insights
The training led to a significant "process-based" improvement in search efficiency, where subjects didn't just get better at seeing, but better at the act of searching.
Performance and Efficiency
- Accuracy: Mean performance increased from 44% on Day 1 to 71% by Day 10.
- Reaction Time: Subjects became significantly faster. Analysis showed that reaction time was primarily driven by the number of saccades (eye jumps) rather than the speed of individual movements.
- Dwell Time: The intersaccadic interval (the duration of fixations) decreased, indicating that subjects could reject distractors more rapidly as they became more expert.
Feature-Based Guidance
The study revealed a hierarchy in how the brain uses features for top-down guidance. Subjects showed a marked increase in their ability to guide their gaze toward items sharing the target's color and spatial frequency. Interestingly, orientation did not show the same level of improvement, suggesting it may be a more difficult feature to use for top-down biasing in noisy environments.
Integration of Features
The findings support a "linear combination rule" for saliency. The brain appears to merge information from different feature channels (e.g., color and frequency) into a unified map that guides eye movements toward the most likely target candidates.
Conclusion
This research demonstrates that top-down attention is a plastic mechanism that can be trained. By practicing search with variable targets, subjects improved the efficiency of the "neural handshake" between goal-setting regions and visual processing areas. While the study's sample size is small, the results suggest that "learning to search" is a generalizable skill that could have applications in professional fields requiring high-stakes visual inspection, such as radiology or security screening.
Summary updated