Extended Data Fig. 3: An outline of the acuity trade-off model.
From: Expectation-driven sensory adaptations support enhanced acuity during categorical perception
(A) A decrease in measurement/representational noise reduces similarity and improves discriminability between stimuli. (B) When stimuli are sampled from regions of stimulus space that are sufficiently close to one another, similarity increases in the task-relevant dimension. (C) The difference between similarity matrices for the left-cued and right-cued syllables, based upon the 1D task-relevant model. The example from (A) and (B) are marked as dots with arrows pointing towards them. (D) Empirical results from our study. The observed shift in spike train vector cosine similarity for left-cued minus right-cued trials. The shift is depicted here is averaged across units and morphs. Compare to (C), where the diagonal does not match the predictions from the 1D model. (E) Predictions of the acuity trade-off model. If there are 0 task-irrelevant dimensions, points that are close to each other in stimulus space will become more similar because noise in measurement is reduced. As more task-relevant dimensions are added, the similarity of close points decreases. (F) A scatterplot of the noise in measurement for task-relevant and irrelevant dimensions under the acuity trade-off model. When a stimulus is cued, the noise in measurement is reduced in a task-relevant dimension (here the morph dimension) and noise is increased in another dimension.
