Figure 1.

Hypotheses and methods. (A) Schematics describing predominant hypotheses about links between attention, visual cortical activity, and behavior. The left plot depicts MT population responses to two visual stimuli plotted along two dimensions in population response space (e.g. the first two principal components; see Methods) and a readout dimension which represents the visual information that is communicated to neuronal populations involved in planning behavior during the uncued condition. The insets depict projections of the population responses onto the readout dimension. Hypothesis 1 is that the MT representations of the two stimuli become more easily distinguishable (e.g. by separating the distributions of responses to the two stimuli). In this scenario, the distributions of projections along even a suboptimal readout axis may also be more separable. Hypothesis 2 suggests that attention changes the way visual information is read out from MT such that projections of MT population responses to the two stimuli onto the readout dimension are more separable. (B) Our new hypothesis: attention reshapes population responses so they are better aligned with relatively static readout dimensions. This alignment could be a direct result of widely observed attention-related changes in firing rates and response variability. (C) Direction change-detection task with cued attention. The drifting Gabor stimuli before the change were identical on every trial within an experimental session and can be thought of as stimulus A, while the changed stimulus can be thought of as stimulus B in the schematics in (A). (D) Psychometric curves from two example sessions (monkey ST, top, monkey HO, bottom) with best-fitting Weibull functions. Attention improved detection of median difficulty trials by 25% on average across all experiments (cued 76.5% detected across sessions, uncued 51.8% detected; N=15 sessions, two-tailed Wilcoxon signed-rank test, p=1.8X10⁻⁴). (E) Receptive field (RF) centers of recorded units from the same example session as in the top plot in (D). Dots represent the RF center (red, MT; blue, SC). The circle represents the size and location of the median RF from each area.