Attention increases representational distance near task-relevant orientations

Background: Attention can alter not just visual performance, but also the appearance of different stimulus features, such as contrast, spatial position, and color. Such perceptual effects can be characterized by their representational geometry: a framework that quantifies dissimilarities in the perceptual representations of different stimuli. Attention has been hypothesized to increase the representational distance (perceived dissimilarity between nearby orientations) around the attended stimulus, while decreasing distance around unattended stimuli. Goal: To date, studies have focused on how attention changes representations of specific target and non-target stimuli. Here we investigated how attention affects the entire representational space of orientation. Methods: On each trial, participants performed two interleaved tasks: a feature-based attention task and a triad similarity judgment task. In the attention task, participants attended to one of two spatially overlapping arrays of oriented bars (45° clockwise or counterclockwise from vertical) to detect small tilts in their orientations. In the triad task, participants judged the perceived similarity among an array of three Gabor patches that varied in orientation. To assess how attention affected the representation of orientation, we performed a multidimensional scaling analysis on the pairwise orientation distances estimated from the triad task under each attention condition. Results: Multidimensional scaling revealed a 1-dimensional manifold of orientation that was best represented in a 3-dimensional space, with two primary dimensions aligned to cardinal and intercardinal orientations and a third dimension that differentiated cardinals from intercardinals. Along this third dimension, attention significantly modulated orientation representations, with attention to -45° orientations expanding the representational space around that orientation. Attention to +45° led to a similar effect, though it was not significant. Conclusions: Our findings support the hypothesis that attention can distort representational geometries in perception. Attention altered participants’ similarity judgments, exaggerating the distance between orientations near the attended angle.